diff options
Diffstat (limited to 'drivers/lguest')
| -rw-r--r-- | drivers/lguest/Kconfig | 7 | ||||
| -rw-r--r-- | drivers/lguest/Makefile | 2 | ||||
| -rw-r--r-- | drivers/lguest/core.c | 217 | ||||
| -rw-r--r-- | drivers/lguest/hypercalls.c | 155 | ||||
| -rw-r--r-- | drivers/lguest/interrupts_and_traps.c | 393 | ||||
| -rw-r--r-- | drivers/lguest/lg.h | 86 | ||||
| -rw-r--r-- | drivers/lguest/lguest_device.c | 335 | ||||
| -rw-r--r-- | drivers/lguest/lguest_user.c | 381 | ||||
| -rw-r--r-- | drivers/lguest/page_tables.c | 1086 | ||||
| -rw-r--r-- | drivers/lguest/segments.c | 147 | ||||
| -rw-r--r-- | drivers/lguest/x86/core.c | 425 | ||||
| -rw-r--r-- | drivers/lguest/x86/switcher_32.S | 22 |
12 files changed, 2236 insertions, 1020 deletions
diff --git a/drivers/lguest/Kconfig b/drivers/lguest/Kconfig index 6b8dbb9ba73..ee035ec4526 100644 --- a/drivers/lguest/Kconfig +++ b/drivers/lguest/Kconfig @@ -1,12 +1,13 @@ config LGUEST tristate "Linux hypervisor example code" - depends on X86_32 && EXPERIMENTAL && !X86_PAE && FUTEX && !(X86_VISWS || X86_VOYAGER) + depends on X86_32 && EVENTFD && TTY select HVC_DRIVER ---help--- This is a very simple module which allows you to run multiple instances of the same Linux kernel, using the - "lguest" command found in the Documentation/lguest directory. + "lguest" command found in the tools/lguest directory. + Note that "lguest" is pronounced to rhyme with "fell quest", - not "rustyvisor". See Documentation/lguest/lguest.txt. + not "rustyvisor". See tools/lguest/lguest.txt. If unsure, say N. If curious, say M. If masochistic, say Y. diff --git a/drivers/lguest/Makefile b/drivers/lguest/Makefile index 7d463c26124..c4197503900 100644 --- a/drivers/lguest/Makefile +++ b/drivers/lguest/Makefile @@ -18,7 +18,7 @@ Mastery: PREFIX=M Beer: @for f in Preparation Guest Drivers Launcher Host Switcher Mastery; do echo "{==- $$f -==}"; make -s $$f; done; echo "{==-==}" Preparation Preparation! Guest Drivers Launcher Host Switcher Mastery: - @sh ../../Documentation/lguest/extract $(PREFIX) `find ../../* -name '*.[chS]' -wholename '*lguest*'` + @sh ../../tools/lguest/extract $(PREFIX) `find ../../* -name '*.[chS]' -wholename '*lguest*'` Puppy: @clear @printf " __ \n (___()'\`;\n /, /\`\n \\\\\\\"--\\\\\\ \n" diff --git a/drivers/lguest/core.c b/drivers/lguest/core.c index 5eea4356d70..0bf1e4edf04 100644 --- a/drivers/lguest/core.c +++ b/drivers/lguest/core.c @@ -1,6 +1,8 @@ -/*P:400 This contains run_guest() which actually calls into the Host<->Guest +/*P:400 + * This contains run_guest() which actually calls into the Host<->Guest * Switcher and analyzes the return, such as determining if the Guest wants the - * Host to do something. This file also contains useful helper routines. :*/ + * Host to do something. This file also contains useful helper routines. +:*/ #include <linux/module.h> #include <linux/stringify.h> #include <linux/stddef.h> @@ -10,6 +12,7 @@ #include <linux/cpu.h> #include <linux/freezer.h> #include <linux/highmem.h> +#include <linux/slab.h> #include <asm/paravirt.h> #include <asm/pgtable.h> #include <asm/uaccess.h> @@ -17,14 +20,15 @@ #include <asm/asm-offsets.h> #include "lg.h" - +unsigned long switcher_addr; +struct page **lg_switcher_pages; static struct vm_struct *switcher_vma; -static struct page **switcher_page; /* This One Big lock protects all inter-guest data structures. */ DEFINE_MUTEX(lguest_lock); -/*H:010 We need to set up the Switcher at a high virtual address. Remember the +/*H:010 + * We need to set up the Switcher at a high virtual address. Remember the * Switcher is a few hundred bytes of assembler code which actually changes the * CPU to run the Guest, and then changes back to the Host when a trap or * interrupt happens. @@ -33,7 +37,8 @@ DEFINE_MUTEX(lguest_lock); * Host since it will be running as the switchover occurs. * * Trying to map memory at a particular address is an unusual thing to do, so - * it's not a simple one-liner. */ + * it's not a simple one-liner. + */ static __init int map_switcher(void) { int i, err; @@ -47,41 +52,53 @@ static __init int map_switcher(void) * easy. */ - /* We allocate an array of struct page pointers. map_vm_area() wants - * this, rather than just an array of pages. */ - switcher_page = kmalloc(sizeof(switcher_page[0])*TOTAL_SWITCHER_PAGES, - GFP_KERNEL); - if (!switcher_page) { + /* We assume Switcher text fits into a single page. */ + if (end_switcher_text - start_switcher_text > PAGE_SIZE) { + printk(KERN_ERR "lguest: switcher text too large (%zu)\n", + end_switcher_text - start_switcher_text); + return -EINVAL; + } + + /* + * We allocate an array of struct page pointers. map_vm_area() wants + * this, rather than just an array of pages. + */ + lg_switcher_pages = kmalloc(sizeof(lg_switcher_pages[0]) + * TOTAL_SWITCHER_PAGES, + GFP_KERNEL); + if (!lg_switcher_pages) { err = -ENOMEM; goto out; } - /* Now we actually allocate the pages. The Guest will see these pages, - * so we make sure they're zeroed. */ + /* + * Now we actually allocate the pages. The Guest will see these pages, + * so we make sure they're zeroed. + */ for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) { - unsigned long addr = get_zeroed_page(GFP_KERNEL); - if (!addr) { + lg_switcher_pages[i] = alloc_page(GFP_KERNEL|__GFP_ZERO); + if (!lg_switcher_pages[i]) { err = -ENOMEM; goto free_some_pages; } - switcher_page[i] = virt_to_page(addr); } - /* First we check that the Switcher won't overlap the fixmap area at - * the top of memory. It's currently nowhere near, but it could have - * very strange effects if it ever happened. */ - if (SWITCHER_ADDR + (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE > FIXADDR_START){ - err = -ENOMEM; - printk("lguest: mapping switcher would thwack fixmap\n"); - goto free_pages; - } + /* + * We place the Switcher underneath the fixmap area, which is the + * highest virtual address we can get. This is important, since we + * tell the Guest it can't access this memory, so we want its ceiling + * as high as possible. + */ + switcher_addr = FIXADDR_START - (TOTAL_SWITCHER_PAGES+1)*PAGE_SIZE; - /* Now we reserve the "virtual memory area" we want: 0xFFC00000 - * (SWITCHER_ADDR). We might not get it in theory, but in practice - * it's worked so far. The end address needs +1 because __get_vm_area - * allocates an extra guard page, so we need space for that. */ + /* + * Now we reserve the "virtual memory area" we want. We might + * not get it in theory, but in practice it's worked so far. + * The end address needs +1 because __get_vm_area allocates an + * extra guard page, so we need space for that. + */ switcher_vma = __get_vm_area(TOTAL_SWITCHER_PAGES * PAGE_SIZE, - VM_ALLOC, SWITCHER_ADDR, SWITCHER_ADDR + VM_ALLOC, switcher_addr, switcher_addr + (TOTAL_SWITCHER_PAGES+1) * PAGE_SIZE); if (!switcher_vma) { err = -ENOMEM; @@ -89,20 +106,24 @@ static __init int map_switcher(void) goto free_pages; } - /* This code actually sets up the pages we've allocated to appear at - * SWITCHER_ADDR. map_vm_area() takes the vma we allocated above, the + /* + * This code actually sets up the pages we've allocated to appear at + * switcher_addr. map_vm_area() takes the vma we allocated above, the * kind of pages we're mapping (kernel pages), and a pointer to our * array of struct pages. It increments that pointer, but we don't - * care. */ - pagep = switcher_page; - err = map_vm_area(switcher_vma, PAGE_KERNEL, &pagep); + * care. + */ + pagep = lg_switcher_pages; + err = map_vm_area(switcher_vma, PAGE_KERNEL_EXEC, &pagep); if (err) { printk("lguest: map_vm_area failed: %i\n", err); goto free_vma; } - /* Now the Switcher is mapped at the right address, we can't fail! - * Copy in the compiled-in Switcher code (from <arch>_switcher.S). */ + /* + * Now the Switcher is mapped at the right address, we can't fail! + * Copy in the compiled-in Switcher code (from x86/switcher_32.S). + */ memcpy(switcher_vma->addr, start_switcher_text, end_switcher_text - start_switcher_text); @@ -117,15 +138,14 @@ free_pages: i = TOTAL_SWITCHER_PAGES; free_some_pages: for (--i; i >= 0; i--) - __free_pages(switcher_page[i], 0); - kfree(switcher_page); + __free_pages(lg_switcher_pages[i], 0); + kfree(lg_switcher_pages); out: return err; } /*:*/ -/* Cleaning up the mapping when the module is unloaded is almost... - * too easy. */ +/* Cleaning up the mapping when the module is unloaded is almost... too easy. */ static void unmap_switcher(void) { unsigned int i; @@ -134,7 +154,8 @@ static void unmap_switcher(void) vunmap(switcher_vma->addr); /* Now we just need to free the pages we copied the switcher into */ for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) - __free_pages(switcher_page[i], 0); + __free_pages(lg_switcher_pages[i], 0); + kfree(lg_switcher_pages); } /*H:032 @@ -150,16 +171,19 @@ static void unmap_switcher(void) * But we can't trust the Guest: it might be trying to access the Launcher * code. We have to check that the range is below the pfn_limit the Launcher * gave us. We have to make sure that addr + len doesn't give us a false - * positive by overflowing, too. */ -int lguest_address_ok(const struct lguest *lg, - unsigned long addr, unsigned long len) + * positive by overflowing, too. + */ +bool lguest_address_ok(const struct lguest *lg, + unsigned long addr, unsigned long len) { return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr); } -/* This routine copies memory from the Guest. Here we can see how useful the +/* + * This routine copies memory from the Guest. Here we can see how useful the * kill_lguest() routine we met in the Launcher can be: we return a random - * value (all zeroes) instead of needing to return an error. */ + * value (all zeroes) instead of needing to return an error. + */ void __lgread(struct lg_cpu *cpu, void *b, unsigned long addr, unsigned bytes) { if (!lguest_address_ok(cpu->lg, addr, bytes) @@ -180,58 +204,87 @@ void __lgwrite(struct lg_cpu *cpu, unsigned long addr, const void *b, } /*:*/ -/*H:030 Let's jump straight to the the main loop which runs the Guest. +/*H:030 + * Let's jump straight to the the main loop which runs the Guest. * Remember, this is called by the Launcher reading /dev/lguest, and we keep - * going around and around until something interesting happens. */ + * going around and around until something interesting happens. + */ int run_guest(struct lg_cpu *cpu, unsigned long __user *user) { /* We stop running once the Guest is dead. */ while (!cpu->lg->dead) { + unsigned int irq; + bool more; + /* First we run any hypercalls the Guest wants done. */ if (cpu->hcall) do_hypercalls(cpu); - /* It's possible the Guest did a NOTIFY hypercall to the - * Launcher, in which case we return from the read() now. */ + /* + * It's possible the Guest did a NOTIFY hypercall to the + * Launcher. + */ if (cpu->pending_notify) { - if (put_user(cpu->pending_notify, user)) - return -EFAULT; - return sizeof(cpu->pending_notify); + /* + * Does it just needs to write to a registered + * eventfd (ie. the appropriate virtqueue thread)? + */ + if (!send_notify_to_eventfd(cpu)) { + /* OK, we tell the main Launcher. */ + if (put_user(cpu->pending_notify, user)) + return -EFAULT; + return sizeof(cpu->pending_notify); + } } + /* + * All long-lived kernel loops need to check with this horrible + * thing called the freezer. If the Host is trying to suspend, + * it stops us. + */ + try_to_freeze(); + /* Check for signals */ if (signal_pending(current)) return -ERESTARTSYS; - /* If Waker set break_out, return to Launcher. */ - if (cpu->break_out) - return -EAGAIN; - - /* Check if there are any interrupts which can be delivered now: + /* + * Check if there are any interrupts which can be delivered now: * if so, this sets up the hander to be executed when we next - * run the Guest. */ - maybe_do_interrupt(cpu); - - /* All long-lived kernel loops need to check with this horrible - * thing called the freezer. If the Host is trying to suspend, - * it stops us. */ - try_to_freeze(); - - /* Just make absolutely sure the Guest is still alive. One of - * those hypercalls could have been fatal, for example. */ + * run the Guest. + */ + irq = interrupt_pending(cpu, &more); + if (irq < LGUEST_IRQS) + try_deliver_interrupt(cpu, irq, more); + + /* + * Just make absolutely sure the Guest is still alive. One of + * those hypercalls could have been fatal, for example. + */ if (cpu->lg->dead) break; - /* If the Guest asked to be stopped, we sleep. The Guest's - * clock timer or LHCALL_BREAK from the Waker will wake us. */ + /* + * If the Guest asked to be stopped, we sleep. The Guest's + * clock timer will wake us. + */ if (cpu->halted) { set_current_state(TASK_INTERRUPTIBLE); - schedule(); + /* + * Just before we sleep, make sure no interrupt snuck in + * which we should be doing. + */ + if (interrupt_pending(cpu, &more) < LGUEST_IRQS) + set_current_state(TASK_RUNNING); + else + schedule(); continue; } - /* OK, now we're ready to jump into the Guest. First we put up - * the "Do Not Disturb" sign: */ + /* + * OK, now we're ready to jump into the Guest. First we put up + * the "Do Not Disturb" sign: + */ local_irq_disable(); /* Actually run the Guest until something happens. */ @@ -265,7 +318,7 @@ static int __init init(void) int err; /* Lguest can't run under Xen, VMI or itself. It does Tricky Stuff. */ - if (paravirt_enabled()) { + if (get_kernel_rpl() != 0) { printk("lguest is afraid of being a guest\n"); return -EPERM; } @@ -275,15 +328,10 @@ static int __init init(void) if (err) goto out; - /* Now we set up the pagetable implementation for the Guests. */ - err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES); - if (err) - goto unmap; - /* We might need to reserve an interrupt vector. */ err = init_interrupts(); if (err) - goto free_pgtables; + goto unmap; /* /dev/lguest needs to be registered. */ err = lguest_device_init(); @@ -298,8 +346,6 @@ static int __init init(void) free_interrupts: free_interrupts(); -free_pgtables: - free_pagetables(); unmap: unmap_switcher(); out: @@ -311,15 +357,16 @@ static void __exit fini(void) { lguest_device_remove(); free_interrupts(); - free_pagetables(); unmap_switcher(); lguest_arch_host_fini(); } /*:*/ -/* The Host side of lguest can be a module. This is a nice way for people to - * play with it. */ +/* + * The Host side of lguest can be a module. This is a nice way for people to + * play with it. + */ module_init(init); module_exit(fini); MODULE_LICENSE("GPL"); diff --git a/drivers/lguest/hypercalls.c b/drivers/lguest/hypercalls.c index 54d66f05fef..83511eb0923 100644 --- a/drivers/lguest/hypercalls.c +++ b/drivers/lguest/hypercalls.c @@ -1,8 +1,10 @@ -/*P:500 Just as userspace programs request kernel operations through a system +/*P:500 + * Just as userspace programs request kernel operations through a system * call, the Guest requests Host operations through a "hypercall". You might * notice this nomenclature doesn't really follow any logic, but the name has * been around for long enough that we're stuck with it. As you'd expect, this - * code is basically a one big switch statement. :*/ + * code is basically a one big switch statement. +:*/ /* Copyright (C) 2006 Rusty Russell IBM Corporation @@ -28,26 +30,41 @@ #include <asm/pgtable.h> #include "lg.h" -/*H:120 This is the core hypercall routine: where the Guest gets what it wants. - * Or gets killed. Or, in the case of LHCALL_SHUTDOWN, both. */ +/*H:120 + * This is the core hypercall routine: where the Guest gets what it wants. + * Or gets killed. Or, in the case of LHCALL_SHUTDOWN, both. + */ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) { switch (args->arg0) { case LHCALL_FLUSH_ASYNC: - /* This call does nothing, except by breaking out of the Guest - * it makes us process all the asynchronous hypercalls. */ + /* + * This call does nothing, except by breaking out of the Guest + * it makes us process all the asynchronous hypercalls. + */ + break; + case LHCALL_SEND_INTERRUPTS: + /* + * This call does nothing too, but by breaking out of the Guest + * it makes us process any pending interrupts. + */ break; case LHCALL_LGUEST_INIT: - /* You can't get here unless you're already initialized. Don't - * do that. */ + /* + * You can't get here unless you're already initialized. Don't + * do that. + */ kill_guest(cpu, "already have lguest_data"); break; case LHCALL_SHUTDOWN: { - /* Shutdown is such a trivial hypercall that we do it in four - * lines right here. */ char msg[128]; - /* If the lgread fails, it will call kill_guest() itself; the - * kill_guest() with the message will be ignored. */ + /* + * Shutdown is such a trivial hypercall that we do it in five + * lines right here. + * + * If the lgread fails, it will call kill_guest() itself; the + * kill_guest() with the message will be ignored. + */ __lgread(cpu, msg, args->arg1, sizeof(msg)); msg[sizeof(msg)-1] = '\0'; kill_guest(cpu, "CRASH: %s", msg); @@ -56,16 +73,17 @@ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) break; } case LHCALL_FLUSH_TLB: - /* FLUSH_TLB comes in two flavors, depending on the - * argument: */ + /* FLUSH_TLB comes in two flavors, depending on the argument: */ if (args->arg1) guest_pagetable_clear_all(cpu); else guest_pagetable_flush_user(cpu); break; - /* All these calls simply pass the arguments through to the right - * routines. */ + /* + * All these calls simply pass the arguments through to the right + * routines. + */ case LHCALL_NEW_PGTABLE: guest_new_pagetable(cpu, args->arg1); break; @@ -73,11 +91,21 @@ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) guest_set_stack(cpu, args->arg1, args->arg2, args->arg3); break; case LHCALL_SET_PTE: +#ifdef CONFIG_X86_PAE + guest_set_pte(cpu, args->arg1, args->arg2, + __pte(args->arg3 | (u64)args->arg4 << 32)); +#else guest_set_pte(cpu, args->arg1, args->arg2, __pte(args->arg3)); +#endif break; + case LHCALL_SET_PGD: + guest_set_pgd(cpu->lg, args->arg1, args->arg2); + break; +#ifdef CONFIG_X86_PAE case LHCALL_SET_PMD: guest_set_pmd(cpu->lg, args->arg1, args->arg2); break; +#endif case LHCALL_SET_CLOCKEVENT: guest_set_clockevent(cpu, args->arg1); break; @@ -98,15 +126,16 @@ static void do_hcall(struct lg_cpu *cpu, struct hcall_args *args) kill_guest(cpu, "Bad hypercall %li\n", args->arg0); } } -/*:*/ -/*H:124 Asynchronous hypercalls are easy: we just look in the array in the +/*H:124 + * Asynchronous hypercalls are easy: we just look in the array in the * Guest's "struct lguest_data" to see if any new ones are marked "ready". * * We are careful to do these in order: obviously we respect the order the * Guest put them in the ring, but we also promise the Guest that they will * happen before any normal hypercall (which is why we check this before - * checking for a normal hcall). */ + * checking for a normal hcall). + */ static void do_async_hcalls(struct lg_cpu *cpu) { unsigned int i; @@ -119,22 +148,28 @@ static void do_async_hcalls(struct lg_cpu *cpu) /* We process "struct lguest_data"s hcalls[] ring once. */ for (i = 0; i < ARRAY_SIZE(st); i++) { struct hcall_args args; - /* We remember where we were up to from last time. This makes + /* + * We remember where we were up to from last time. This makes * sure that the hypercalls are done in the order the Guest - * places them in the ring. */ + * places them in the ring. + */ unsigned int n = cpu->next_hcall; /* 0xFF means there's no call here (yet). */ if (st[n] == 0xFF) break; - /* OK, we have hypercall. Increment the "next_hcall" cursor, - * and wrap back to 0 if we reach the end. */ + /* + * OK, we have hypercall. Increment the "next_hcall" cursor, + * and wrap back to 0 if we reach the end. + */ if (++cpu->next_hcall == LHCALL_RING_SIZE) cpu->next_hcall = 0; - /* Copy the hypercall arguments into a local copy of - * the hcall_args struct. */ + /* + * Copy the hypercall arguments into a local copy of the + * hcall_args struct. + */ if (copy_from_user(&args, &cpu->lg->lguest_data->hcalls[n], sizeof(struct hcall_args))) { kill_guest(cpu, "Fetching async hypercalls"); @@ -150,19 +185,25 @@ static void do_async_hcalls(struct lg_cpu *cpu) break; } - /* Stop doing hypercalls if they want to notify the Launcher: - * it needs to service this first. */ + /* + * Stop doing hypercalls if they want to notify the Launcher: + * it needs to service this first. + */ if (cpu->pending_notify) break; } } -/* Last of all, we look at what happens first of all. The very first time the - * Guest makes a hypercall, we end up here to set things up: */ +/* + * Last of all, we look at what happens first of all. The very first time the + * Guest makes a hypercall, we end up here to set things up: + */ static void initialize(struct lg_cpu *cpu) { - /* You can't do anything until you're initialized. The Guest knows the - * rules, so we're unforgiving here. */ + /* + * You can't do anything until you're initialized. The Guest knows the + * rules, so we're unforgiving here. + */ if (cpu->hcall->arg0 != LHCALL_LGUEST_INIT) { kill_guest(cpu, "hypercall %li before INIT", cpu->hcall->arg0); return; @@ -171,32 +212,44 @@ static void initialize(struct lg_cpu *cpu) if (lguest_arch_init_hypercalls(cpu)) kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); - /* The Guest tells us where we're not to deliver interrupts by putting - * the range of addresses into "struct lguest_data". */ + /* + * The Guest tells us where we're not to deliver interrupts by putting + * the range of addresses into "struct lguest_data". + */ if (get_user(cpu->lg->noirq_start, &cpu->lg->lguest_data->noirq_start) || get_user(cpu->lg->noirq_end, &cpu->lg->lguest_data->noirq_end)) kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); - /* We write the current time into the Guest's data page once so it can - * set its clock. */ + /* + * We write the current time into the Guest's data page once so it can + * set its clock. + */ write_timestamp(cpu); /* page_tables.c will also do some setup. */ page_table_guest_data_init(cpu); - /* This is the one case where the above accesses might have been the + /* + * This is the one case where the above accesses might have been the * first write to a Guest page. This may have caused a copy-on-write * fault, but the old page might be (read-only) in the Guest - * pagetable. */ + * pagetable. + */ guest_pagetable_clear_all(cpu); } /*:*/ -/*M:013 If a Guest reads from a page (so creates a mapping) that it has never +/*M:013 + * If a Guest reads from a page (so creates a mapping) that it has never * written to, and then the Launcher writes to it (ie. the output of a virtual * device), the Guest will still see the old page. In practice, this never * happens: why would the Guest read a page which it has never written to? But - * a similar scenario might one day bite us, so it's worth mentioning. :*/ + * a similar scenario might one day bite us, so it's worth mentioning. + * + * Note that if we used a shared anonymous mapping in the Launcher instead of + * mapping /dev/zero private, we wouldn't worry about cop-on-write. And we + * need that to switch the Launcher to processes (away from threads) anyway. +:*/ /*H:100 * Hypercalls @@ -215,17 +268,22 @@ void do_hypercalls(struct lg_cpu *cpu) return; } - /* The Guest has initialized. + /* + * The Guest has initialized. * - * Look in the hypercall ring for the async hypercalls: */ + * Look in the hypercall ring for the async hypercalls: + */ do_async_hcalls(cpu); - /* If we stopped reading the hypercall ring because the Guest did a + /* + * If we stopped reading the hypercall ring because the Guest did a * NOTIFY to the Launcher, we want to return now. Otherwise we do - * the hypercall. */ + * the hypercall. + */ if (!cpu->pending_notify) { do_hcall(cpu, cpu->hcall); - /* Tricky point: we reset the hcall pointer to mark the + /* + * Tricky point: we reset the hcall pointer to mark the * hypercall as "done". We use the hcall pointer rather than * the trap number to indicate a hypercall is pending. * Normally it doesn't matter: the Guest will run again and @@ -234,13 +292,16 @@ void do_hypercalls(struct lg_cpu *cpu) * However, if we are signalled or the Guest sends I/O to the * Launcher, the run_guest() loop will exit without running the * Guest. When it comes back it would try to re-run the - * hypercall. Finding that bug sucked. */ + * hypercall. Finding that bug sucked. + */ cpu->hcall = NULL; } } -/* This routine supplies the Guest with time: it's used for wallclock time at - * initial boot and as a rough time source if the TSC isn't available. */ +/* + * This routine supplies the Guest with time: it's used for wallclock time at + * initial boot and as a rough time source if the TSC isn't available. + */ void write_timestamp(struct lg_cpu *cpu) { struct timespec now; diff --git a/drivers/lguest/interrupts_and_traps.c b/drivers/lguest/interrupts_and_traps.c index 0414ddf8758..70dfcdc29f1 100644 --- a/drivers/lguest/interrupts_and_traps.c +++ b/drivers/lguest/interrupts_and_traps.c @@ -1,4 +1,5 @@ -/*P:800 Interrupts (traps) are complicated enough to earn their own file. +/*P:800 + * Interrupts (traps) are complicated enough to earn their own file. * There are three classes of interrupts: * * 1) Real hardware interrupts which occur while we're running the Guest, @@ -10,10 +11,12 @@ * just like real hardware would deliver them. Traps from the Guest can be set * up to go directly back into the Guest, but sometimes the Host wants to see * them first, so we also have a way of "reflecting" them into the Guest as if - * they had been delivered to it directly. :*/ + * they had been delivered to it directly. +:*/ #include <linux/uaccess.h> #include <linux/interrupt.h> #include <linux/module.h> +#include <linux/sched.h> #include "lg.h" /* Allow Guests to use a non-128 (ie. non-Linux) syscall trap. */ @@ -26,21 +29,25 @@ static unsigned long idt_address(u32 lo, u32 hi) return (lo & 0x0000FFFF) | (hi & 0xFFFF0000); } -/* The "type" of the interrupt handler is a 4 bit field: we only support a - * couple of types. */ +/* + * The "type" of the interrupt handler is a 4 bit field: we only support a + * couple of types. + */ static int idt_type(u32 lo, u32 hi) { return (hi >> 8) & 0xF; } /* An IDT entry can't be used unless the "present" bit is set. */ -static int idt_present(u32 lo, u32 hi) +static bool idt_present(u32 lo, u32 hi) { return (hi & 0x8000); } -/* We need a helper to "push" a value onto the Guest's stack, since that's a - * big part of what delivering an interrupt does. */ +/* + * We need a helper to "push" a value onto the Guest's stack, since that's a + * big part of what delivering an interrupt does. + */ static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val) { /* Stack grows upwards: move stack then write value. */ @@ -48,7 +55,8 @@ static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val) lgwrite(cpu, *gstack, u32, val); } -/*H:210 The set_guest_interrupt() routine actually delivers the interrupt or +/*H:210 + * The set_guest_interrupt() routine actually delivers the interrupt or * trap. The mechanics of delivering traps and interrupts to the Guest are the * same, except some traps have an "error code" which gets pushed onto the * stack as well: the caller tells us if this is one. @@ -59,27 +67,35 @@ static void push_guest_stack(struct lg_cpu *cpu, unsigned long *gstack, u32 val) * * We set up the stack just like the CPU does for a real interrupt, so it's * identical for the Guest (and the standard "iret" instruction will undo - * it). */ -static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi, int has_err) + * it). + */ +static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi, + bool has_err) { unsigned long gstack, origstack; u32 eflags, ss, irq_enable; unsigned long virtstack; - /* There are two cases for interrupts: one where the Guest is already + /* + * There are two cases for interrupts: one where the Guest is already * in the kernel, and a more complex one where the Guest is in - * userspace. We check the privilege level to find out. */ + * userspace. We check the privilege level to find out. + */ if ((cpu->regs->ss&0x3) != GUEST_PL) { - /* The Guest told us their kernel stack with the SET_STACK - * hypercall: both the virtual address and the segment */ + /* + * The Guest told us their kernel stack with the SET_STACK + * hypercall: both the virtual address and the segment. + */ virtstack = cpu->esp1; ss = cpu->ss1; origstack = gstack = guest_pa(cpu, virtstack); - /* We push the old stack segment and pointer onto the new + /* + * We push the old stack segment and pointer onto the new * stack: when the Guest does an "iret" back from the interrupt * handler the CPU will notice they're dropping privilege - * levels and expect these here. */ + * levels and expect these here. + */ push_guest_stack(cpu, &gstack, cpu->regs->ss); push_guest_stack(cpu, &gstack, cpu->regs->esp); } else { @@ -90,18 +106,22 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi, int has_err) origstack = gstack = guest_pa(cpu, virtstack); } - /* Remember that we never let the Guest actually disable interrupts, so + /* + * Remember that we never let the Guest actually disable interrupts, so * the "Interrupt Flag" bit is always set. We copy that bit from the * Guest's "irq_enabled" field into the eflags word: we saw the Guest - * copy it back in "lguest_iret". */ + * copy it back in "lguest_iret". + */ eflags = cpu->regs->eflags; if (get_user(irq_enable, &cpu->lg->lguest_data->irq_enabled) == 0 && !(irq_enable & X86_EFLAGS_IF)) eflags &= ~X86_EFLAGS_IF; - /* An interrupt is expected to push three things on the stack: the old + /* + * An interrupt is expected to push three things on the stack: the old * "eflags" word, the old code segment, and the old instruction - * pointer. */ + * pointer. + */ push_guest_stack(cpu, &gstack, eflags); push_guest_stack(cpu, &gstack, cpu->regs->cs); push_guest_stack(cpu, &gstack, cpu->regs->eip); @@ -110,15 +130,29 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi, int has_err) if (has_err) push_guest_stack(cpu, &gstack, cpu->regs->errcode); - /* Now we've pushed all the old state, we change the stack, the code - * segment and the address to execute. */ + /* + * Now we've pushed all the old state, we change the stack, the code + * segment and the address to execute. + */ cpu->regs->ss = ss; cpu->regs->esp = virtstack + (gstack - origstack); cpu->regs->cs = (__KERNEL_CS|GUEST_PL); cpu->regs->eip = idt_address(lo, hi); - /* There are two kinds of interrupt handlers: 0xE is an "interrupt - * gate" which expects interrupts to be disabled on entry. */ + /* + * Trapping always clears these flags: + * TF: Trap flag + * VM: Virtual 8086 mode + * RF: Resume + * NT: Nested task. + */ + cpu->regs->eflags &= + ~(X86_EFLAGS_TF|X86_EFLAGS_VM|X86_EFLAGS_RF|X86_EFLAGS_NT); + + /* + * There are two kinds of interrupt handlers: 0xE is an "interrupt + * gate" which expects interrupts to be disabled on entry. + */ if (idt_type(lo, hi) == 0xE) if (put_user(0, &cpu->lg->lguest_data->irq_enabled)) kill_guest(cpu, "Disabling interrupts"); @@ -127,33 +161,49 @@ static void set_guest_interrupt(struct lg_cpu *cpu, u32 lo, u32 hi, int has_err) /*H:205 * Virtual Interrupts. * - * maybe_do_interrupt() gets called before every entry to the Guest, to see if - * we should divert the Guest to running an interrupt handler. */ -void maybe_do_interrupt(struct lg_cpu *cpu) + * interrupt_pending() returns the first pending interrupt which isn't blocked + * by the Guest. It is called before every entry to the Guest, and just before + * we go to sleep when the Guest has halted itself. + */ +unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more) { unsigned int irq; DECLARE_BITMAP(blk, LGUEST_IRQS); - struct desc_struct *idt; /* If the Guest hasn't even initialized yet, we can do nothing. */ if (!cpu->lg->lguest_data) - return; + return LGUEST_IRQS; - /* Take our "irqs_pending" array and remove any interrupts the Guest - * wants blocked: the result ends up in "blk". */ + /* + * Take our "irqs_pending" array and remove any interrupts the Guest + * wants blocked: the result ends up in "blk". + */ if (copy_from_user(&blk, cpu->lg->lguest_data->blocked_interrupts, sizeof(blk))) - return; + return LGUEST_IRQS; bitmap_andnot(blk, cpu->irqs_pending, blk, LGUEST_IRQS); /* Find the first interrupt. */ irq = find_first_bit(blk, LGUEST_IRQS); - /* None? Nothing to do */ - if (irq >= LGUEST_IRQS) - return; + *more = find_next_bit(blk, LGUEST_IRQS, irq+1); + + return irq; +} + +/* + * This actually diverts the Guest to running an interrupt handler, once an + * interrupt has been identified by interrupt_pending(). + */ +void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more) +{ + struct desc_struct *idt; + + BUG_ON(irq >= LGUEST_IRQS); - /* They may be in the middle of an iret, where they asked us never to - * deliver interrupts. */ + /* + * They may be in the middle of an iret, where they asked us never to + * deliver interrupts. + */ if (cpu->regs->eip >= cpu->lg->noirq_start && (cpu->regs->eip < cpu->lg->noirq_end)) return; @@ -169,39 +219,75 @@ void maybe_do_interrupt(struct lg_cpu *cpu) u32 irq_enabled; if (get_user(irq_enabled, &cpu->lg->lguest_data->irq_enabled)) irq_enabled = 0; - if (!irq_enabled) + if (!irq_enabled) { + /* Make sure they know an IRQ is pending. */ + put_user(X86_EFLAGS_IF, + &cpu->lg->lguest_data->irq_pending); return; + } } - /* Look at the IDT entry the Guest gave us for this interrupt. The + /* + * Look at the IDT entry the Guest gave us for this interrupt. The * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip - * over them. */ + * over them. + */ idt = &cpu->arch.idt[FIRST_EXTERNAL_VECTOR+irq]; /* If they don't have a handler (yet?), we just ignore it */ if (idt_present(idt->a, idt->b)) { /* OK, mark it no longer pending and deliver it. */ clear_bit(irq, cpu->irqs_pending); - /* set_guest_interrupt() takes the interrupt descriptor and a + /* + * set_guest_interrupt() takes the interrupt descriptor and a * flag to say whether this interrupt pushes an error code onto - * the stack as well: virtual interrupts never do. */ - set_guest_interrupt(cpu, idt->a, idt->b, 0); + * the stack as well: virtual interrupts never do. + */ + set_guest_interrupt(cpu, idt->a, idt->b, false); } - /* Every time we deliver an interrupt, we update the timestamp in the + /* + * Every time we deliver an interrupt, we update the timestamp in the * Guest's lguest_data struct. It would be better for the Guest if we * did this more often, but it can actually be quite slow: doing it * here is a compromise which means at least it gets updated every - * timer interrupt. */ + * timer interrupt. + */ write_timestamp(cpu); + + /* + * If there are no other interrupts we want to deliver, clear + * the pending flag. + */ + if (!more) + put_user(0, &cpu->lg->lguest_data->irq_pending); +} + +/* And this is the routine when we want to set an interrupt for the Guest. */ +void set_interrupt(struct lg_cpu *cpu, unsigned int irq) +{ + /* + * Next time the Guest runs, the core code will see if it can deliver + * this interrupt. + */ + set_bit(irq, cpu->irqs_pending); + + /* + * Make sure it sees it; it might be asleep (eg. halted), or running + * the Guest right now, in which case kick_process() will knock it out. + */ + if (!wake_up_process(cpu->tsk)) + kick_process(cpu->tsk); } /*:*/ -/* Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent +/* + * Linux uses trap 128 for system calls. Plan9 uses 64, and Ron Minnich sent * me a patch, so we support that too. It'd be a big step for lguest if half * the Plan 9 user base were to start using it. * * Actually now I think of it, it's possible that Ron *is* half the Plan 9 - * userbase. Oh well. */ + * userbase. Oh well. + */ static bool could_be_syscall(unsigned int num) { /* Normal Linux SYSCALL_VECTOR or reserved vector? */ @@ -222,11 +308,16 @@ bool check_syscall_vector(struct lguest *lg) int init_interrupts(void) { /* If they want some strange system call vector, reserve it now */ - if (syscall_vector != SYSCALL_VECTOR - && test_and_set_bit(syscall_vector, used_vectors)) { - printk("lg: couldn't reserve syscall %u\n", syscall_vector); - return -EBUSY; + if (syscall_vector != SYSCALL_VECTOR) { + if (test_bit(syscall_vector, used_vectors) || + vector_used_by_percpu_irq(syscall_vector)) { + printk(KERN_ERR "lg: couldn't reserve syscall %u\n", + syscall_vector); + return -EBUSY; + } + set_bit(syscall_vector, used_vectors); } + return 0; } @@ -236,32 +327,39 @@ void free_interrupts(void) clear_bit(syscall_vector, used_vectors); } -/*H:220 Now we've got the routines to deliver interrupts, delivering traps like +/*H:220 + * Now we've got the routines to deliver interrupts, delivering traps like * page fault is easy. The only trick is that Intel decided that some traps - * should have error codes: */ -static int has_err(unsigned int trap) + * should have error codes: + */ +static bool has_err(unsigned int trap) { return (trap == 8 || (trap >= 10 && trap <= 14) || trap == 17); } /* deliver_trap() returns true if it could deliver the trap. */ -int deliver_trap(struct lg_cpu *cpu, unsigned int num) +bool deliver_trap(struct lg_cpu *cpu, unsigned int num) { - /* Trap numbers are always 8 bit, but we set an impossible trap number - * for traps inside the Switcher, so check that here. */ + /* + * Trap numbers are always 8 bit, but we set an impossible trap number + * for traps inside the Switcher, so check that here. + */ if (num >= ARRAY_SIZE(cpu->arch.idt)) - return 0; + return false; - /* Early on the Guest hasn't set the IDT entries (or maybe it put a - * bogus one in): if we fail here, the Guest will be killed. */ + /* + * Early on the Guest hasn't set the IDT entries (or maybe it put a + * bogus one in): if we fail here, the Guest will be killed. + */ if (!idt_present(cpu->arch.idt[num].a, cpu->arch.idt[num].b)) - return 0; + return false; set_guest_interrupt(cpu, cpu->arch.idt[num].a, cpu->arch.idt[num].b, has_err(num)); - return 1; + return true; } -/*H:250 Here's the hard part: returning to the Host every time a trap happens +/*H:250 + * Here's the hard part: returning to the Host every time a trap happens * and then calling deliver_trap() and re-entering the Guest is slow. * Particularly because Guest userspace system calls are traps (usually trap * 128). @@ -273,68 +371,85 @@ int deliver_trap(struct lg_cpu *cpu, unsigned int num) * the other hypervisors would beat it up at lunchtime. * * This routine indicates if a particular trap number could be delivered - * directly. */ -static int direct_trap(unsigned int num) + * directly. + */ +static bool direct_trap(unsigned int num) { - /* Hardware interrupts don't go to the Guest at all (except system - * call). */ + /* + * Hardware interrupts don't go to the Guest at all (except system + * call). + */ if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num)) - return 0; + return false; - /* The Host needs to see page faults (for shadow paging and to save the + /* + * The Host needs to see page faults (for shadow paging and to save the * fault address), general protection faults (in/out emulation) and - * device not available (TS handling), and of course, the hypercall - * trap. */ + * device not available (TS handling) and of course, the hypercall trap. + */ return num != 14 && num != 13 && num != 7 && num != LGUEST_TRAP_ENTRY; } /*:*/ -/*M:005 The Guest has the ability to turn its interrupt gates into trap gates, +/*M:005 + * The Guest has the ability to turn its interrupt gates into trap gates, * if it is careful. The Host will let trap gates can go directly to the * Guest, but the Guest needs the interrupts atomically disabled for an * interrupt gate. It can do this by pointing the trap gate at instructions - * within noirq_start and noirq_end, where it can safely disable interrupts. */ + * within noirq_start and noirq_end, where it can safely disable interrupts. + */ -/*M:006 The Guests do not use the sysenter (fast system call) instruction, +/*M:006 + * The Guests do not use the sysenter (fast system call) instruction, * because it's hardcoded to enter privilege level 0 and so can't go direct. * It's about twice as fast as the older "int 0x80" system call, so it might * still be worthwhile to handle it in the Switcher and lcall down to the * Guest. The sysenter semantics are hairy tho: search for that keyword in - * entry.S :*/ + * entry.S +:*/ -/*H:260 When we make traps go directly into the Guest, we need to make sure +/*H:260 + * When we make traps go directly into the Guest, we need to make sure * the kernel stack is valid (ie. mapped in the page tables). Otherwise, the * CPU trying to deliver the trap will fault while trying to push the interrupt * words on the stack: this is called a double fault, and it forces us to kill * the Guest. * - * Which is deeply unfair, because (literally!) it wasn't the Guests' fault. */ + * Which is deeply unfair, because (literally!) it wasn't the Guests' fault. + */ void pin_stack_pages(struct lg_cpu *cpu) { unsigned int i; - /* Depending on the CONFIG_4KSTACKS option, the Guest can have one or - * two pages of stack space. */ + /* + * Depending on the CONFIG_4KSTACKS option, the Guest can have one or + * two pages of stack space. + */ for (i = 0; i < cpu->lg->stack_pages; i++) - /* The stack grows *upwards*, so the address we're given is the + /* + * The stack grows *upwards*, so the address we're given is the * start of the page after the kernel stack. Subtract one to * get back onto the first stack page, and keep subtracting to - * get to the rest of the stack pages. */ + * get to the rest of the stack pages. + */ pin_page(cpu, cpu->esp1 - 1 - i * PAGE_SIZE); } -/* Direct traps also mean that we need to know whenever the Guest wants to use - * a different kernel stack, so we can change the IDT entries to use that - * stack. The IDT entries expect a virtual address, so unlike most addresses +/* + * Direct traps also mean that we need to know whenever the Guest wants to use + * a different kernel stack, so we can change the guest TSS to use that + * stack. The TSS entries expect a virtual address, so unlike most addresses * the Guest gives us, the "esp" (stack pointer) value here is virtual, not * physical. * * In Linux each process has its own kernel stack, so this happens a lot: we - * change stacks on each context switch. */ + * change stacks on each context switch. + */ void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages) { - /* You are not allowed have a stack segment with privilege level 0: bad - * Guest! */ + /* + * You're not allowed a stack segment with privilege level 0: bad Guest! + */ if ((seg & 0x3) != GUEST_PL) kill_guest(cpu, "bad stack segment %i", seg); /* We only expect one or two stack pages. */ @@ -348,11 +463,15 @@ void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages) pin_stack_pages(cpu); } -/* All this reference to mapping stacks leads us neatly into the other complex - * part of the Host: page table handling. */ +/* + * All this reference to mapping stacks leads us neatly into the other complex + * part of the Host: page table handling. + */ -/*H:235 This is the routine which actually checks the Guest's IDT entry and - * transfers it into the entry in "struct lguest": */ +/*H:235 + * This is the routine which actually checks the Guest's IDT entry and + * transfers it into the entry in "struct lguest": + */ static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap, unsigned int num, u32 lo, u32 hi) { @@ -368,30 +487,38 @@ static void set_trap(struct lg_cpu *cpu, struct desc_struct *trap, if (type != 0xE && type != 0xF) kill_guest(cpu, "bad IDT type %i", type); - /* We only copy the handler address, present bit, privilege level and + /* + * We only copy the handler address, present bit, privilege level and * type. The privilege level controls where the trap can be triggered * manually with an "int" instruction. This is usually GUEST_PL, - * except for system calls which userspace can use. */ + * except for system calls which userspace can use. + */ trap->a = ((__KERNEL_CS|GUEST_PL)<<16) | (lo&0x0000FFFF); trap->b = (hi&0xFFFFEF00); } -/*H:230 While we're here, dealing with delivering traps and interrupts to the +/*H:230 + * While we're here, dealing with delivering traps and interrupts to the * Guest, we might as well complete the picture: how the Guest tells us where * it wants them to go. This would be simple, except making traps fast * requires some tricks. * * We saw the Guest setting Interrupt Descriptor Table (IDT) entries with the - * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here. */ + * LHCALL_LOAD_IDT_ENTRY hypercall before: that comes here. + */ void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi) { - /* Guest never handles: NMI, doublefault, spurious interrupt or - * hypercall. We ignore when it tries to set them. */ + /* + * Guest never handles: NMI, doublefault, spurious interrupt or + * hypercall. We ignore when it tries to set them. + */ if (num == 2 || num == 8 || num == 15 || num == LGUEST_TRAP_ENTRY) return; - /* Mark the IDT as changed: next time the Guest runs we'll know we have - * to copy this again. */ + /* + * Mark the IDT as changed: next time the Guest runs we'll know we have + * to copy this again. + */ cpu->changed |= CHANGED_IDT; /* Check that the Guest doesn't try to step outside the bounds. */ @@ -401,20 +528,31 @@ void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int num, u32 lo, u32 hi) set_trap(cpu, &cpu->arch.idt[num], num, lo, hi); } -/* The default entry for each interrupt points into the Switcher routines which +/* + * The default entry for each interrupt points into the Switcher routines which * simply return to the Host. The run_guest() loop will then call - * deliver_trap() to bounce it back into the Guest. */ + * deliver_trap() to bounce it back into the Guest. + */ static void default_idt_entry(struct desc_struct *idt, int trap, - const unsigned long handler) + const unsigned long handler, + const struct desc_struct *base) { /* A present interrupt gate. */ u32 flags = 0x8e00; - /* Set the privilege level on the entry for the hypercall: this allows - * the Guest to use the "int" instruction to trigger it. */ + /* + * Set the privilege level on the entry for the hypercall: this allows + * the Guest to use the "int" instruction to trigger it. + */ if (trap == LGUEST_TRAP_ENTRY) flags |= (GUEST_PL << 13); + else if (base) + /* + * Copy privilege level from what Guest asked for. This allows + * debug (int 3) traps from Guest userspace, for example. + */ + flags |= (base->b & 0x6000); /* Now pack it into the IDT entry in its weird format. */ idt->a = (LGUEST_CS<<16) | (handler&0x0000FFFF); @@ -428,33 +566,44 @@ void setup_default_idt_entries(struct lguest_ro_state *state, unsigned int i; for (i = 0; i < ARRAY_SIZE(state->guest_idt); i++) - default_idt_entry(&state->guest_idt[i], i, def[i]); + default_idt_entry(&state->guest_idt[i], i, def[i], NULL); } -/*H:240 We don't use the IDT entries in the "struct lguest" directly, instead +/*H:240 + * We don't use the IDT entries in the "struct lguest" directly, instead * we copy them into the IDT which we've set up for Guests on this CPU, just - * before we run the Guest. This routine does that copy. */ + * before we run the Guest. This routine does that copy. + */ void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, const unsigned long *def) { unsigned int i; - /* We can simply copy the direct traps, otherwise we use the default - * ones in the Switcher: they will return to the Host. */ + /* + * We can simply copy the direct traps, otherwise we use the default + * ones in the Switcher: they will return to the Host. + */ for (i = 0; i < ARRAY_SIZE(cpu->arch.idt); i++) { + const struct desc_struct *gidt = &cpu->arch.idt[i]; + /* If no Guest can ever override this trap, leave it alone. */ if (!direct_trap(i)) continue; - /* Only trap gates (type 15) can go direct to the Guest. + /* + * Only trap gates (type 15) can go direct to the Guest. * Interrupt gates (type 14) disable interrupts as they are * entered, which we never let the Guest do. Not present - * entries (type 0x0) also can't go direct, of course. */ - if (idt_type(cpu->arch.idt[i].a, cpu->arch.idt[i].b) == 0xF) - idt[i] = cpu->arch.idt[i]; + * entries (type 0x0) also can't go direct, of course. + * + * If it can't go direct, we still need to copy the priv. level: + * they might want to give userspace access to a software + * interrupt. + */ + if (idt_type(gidt->a, gidt->b) == 0xF) + idt[i] = *gidt; else - /* Reset it to the default. */ - default_idt_entry(&idt[i], i, def[i]); + default_idt_entry(&idt[i], i, def[i], gidt); } } @@ -469,7 +618,8 @@ void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, * the next timer interrupt (in nanoseconds). We use the high-resolution timer * infrastructure to set a callback at that time. * - * 0 means "turn off the clock". */ + * 0 means "turn off the clock". + */ void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta) { ktime_t expires; @@ -480,9 +630,11 @@ void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta) return; } - /* We use wallclock time here, so the Guest might not be running for + /* + * We use wallclock time here, so the Guest might not be running for * all the time between now and the timer interrupt it asked for. This - * is almost always the right thing to do. */ + * is almost always the right thing to do. + */ expires = ktime_add_ns(ktime_get_real(), delta); hrtimer_start(&cpu->hrt, expires, HRTIMER_MODE_ABS); } @@ -493,10 +645,7 @@ static enum hrtimer_restart clockdev_fn(struct hrtimer *timer) struct lg_cpu *cpu = container_of(timer, struct lg_cpu, hrt); /* Remember the first interrupt is the timer interrupt. */ - set_bit(0, cpu->irqs_pending); - /* If the Guest is actually stopped, we need to wake it up. */ - if (cpu->halted) - wake_up_process(cpu->tsk); + set_interrupt(cpu, 0); return HRTIMER_NORESTART; } diff --git a/drivers/lguest/lg.h b/drivers/lguest/lg.h index 2337e1a06f0..2eef40be4c0 100644 --- a/drivers/lguest/lg.h +++ b/drivers/lguest/lg.h @@ -10,22 +10,19 @@ #include <linux/wait.h> #include <linux/hrtimer.h> #include <linux/err.h> -#include <asm/semaphore.h> +#include <linux/slab.h> #include <asm/lguest.h> -void free_pagetables(void); -int init_pagetables(struct page **switcher_page, unsigned int pages); - -struct pgdir -{ +struct pgdir { unsigned long gpgdir; + bool switcher_mapped; + int last_host_cpu; pgd_t *pgdir; }; /* We have two pages shared with guests, per cpu. */ -struct lguest_pages -{ +struct lguest_pages { /* This is the stack page mapped rw in guest */ char spare[PAGE_SIZE - sizeof(struct lguest_regs)]; struct lguest_regs regs; @@ -39,8 +36,6 @@ struct lguest_pages #define CHANGED_GDT_TLS 4 /* Actually a subset of CHANGED_GDT */ #define CHANGED_ALL 3 -struct lguest; - struct lg_cpu { unsigned int id; struct lguest *lg; @@ -50,20 +45,22 @@ struct lg_cpu { u32 cr2; int ts; u32 esp1; - u8 ss1; + u16 ss1; /* Bitmap of what has changed: see CHANGED_* above. */ int changed; unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */ - /* At end of a page shared mapped over lguest_pages in guest. */ + /* At end of a page shared mapped over lguest_pages in guest. */ unsigned long regs_page; struct lguest_regs *regs; struct lguest_pages *last_pages; - int cpu_pgd; /* which pgd this cpu is currently using */ + /* Initialization mode: linear map everything. */ + bool linear_pages; + int cpu_pgd; /* Which pgd this cpu is currently using */ /* If a hypercall was asked for, this points to the arguments. */ struct hcall_args *hcall; @@ -72,9 +69,7 @@ struct lg_cpu { /* Virtual clock device */ struct hrtimer hrt; - /* Do we need to stop what we're doing and return to userspace? */ - int break_out; - wait_queue_head_t break_wq; + /* Did the Guest tell us to halt? */ int halted; /* Pending virtual interrupts */ @@ -83,16 +78,28 @@ struct lg_cpu { struct lg_cpu_arch arch; }; +struct lg_eventfd { + unsigned long addr; + struct eventfd_ctx *event; +}; + +struct lg_eventfd_map { + unsigned int num; + struct lg_eventfd map[]; +}; + /* The private info the thread maintains about the guest. */ -struct lguest -{ +struct lguest { struct lguest_data __user *lguest_data; struct lg_cpu cpus[NR_CPUS]; unsigned int nr_cpus; u32 pfn_limit; - /* This provides the offset to the base of guest-physical - * memory in the Launcher. */ + + /* + * This provides the offset to the base of guest-physical memory in the + * Launcher. + */ void __user *mem_base; unsigned long kernel_address; @@ -103,6 +110,8 @@ struct lguest unsigned int stack_pages; u32 tsc_khz; + struct lg_eventfd_map *eventfds; + /* Dead? */ const char *dead; }; @@ -110,16 +119,19 @@ struct lguest extern struct mutex lguest_lock; /* core.c: */ -int lguest_address_ok(const struct lguest *lg, - unsigned long addr, unsigned long len); +bool lguest_address_ok(const struct lguest *lg, + unsigned long addr, unsigned long len); void __lgread(struct lg_cpu *, void *, unsigned long, unsigned); void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned); +extern struct page **lg_switcher_pages; -/*H:035 Using memory-copy operations like that is usually inconvient, so we +/*H:035 + * Using memory-copy operations like that is usually inconvient, so we * have the following helper macros which read and write a specific type (often * an unsigned long). * - * This reads into a variable of the given type then returns that. */ + * This reads into a variable of the given type then returns that. + */ #define lgread(cpu, addr, type) \ ({ type _v; __lgread((cpu), &_v, (addr), sizeof(_v)); _v; }) @@ -133,16 +145,21 @@ void __lgwrite(struct lg_cpu *, unsigned long, const void *, unsigned); int run_guest(struct lg_cpu *cpu, unsigned long __user *user); -/* Helper macros to obtain the first 12 or the last 20 bits, this is only the +/* + * Helper macros to obtain the first 12 or the last 20 bits, this is only the * first step in the migration to the kernel types. pte_pfn is already defined - * in the kernel. */ + * in the kernel. + */ #define pgd_flags(x) (pgd_val(x) & ~PAGE_MASK) -#define pte_flags(x) (pte_val(x) & ~PAGE_MASK) #define pgd_pfn(x) (pgd_val(x) >> PAGE_SHIFT) +#define pmd_flags(x) (pmd_val(x) & ~PAGE_MASK) +#define pmd_pfn(x) (pmd_val(x) >> PAGE_SHIFT) /* interrupts_and_traps.c: */ -void maybe_do_interrupt(struct lg_cpu *cpu); -int deliver_trap(struct lg_cpu *cpu, unsigned int num); +unsigned int interrupt_pending(struct lg_cpu *cpu, bool *more); +void try_deliver_interrupt(struct lg_cpu *cpu, unsigned int irq, bool more); +void set_interrupt(struct lg_cpu *cpu, unsigned int irq); +bool deliver_trap(struct lg_cpu *cpu, unsigned int num); void load_guest_idt_entry(struct lg_cpu *cpu, unsigned int i, u32 low, u32 hi); void guest_set_stack(struct lg_cpu *cpu, u32 seg, u32 esp, unsigned int pages); @@ -152,6 +169,7 @@ void setup_default_idt_entries(struct lguest_ro_state *state, void copy_traps(const struct lg_cpu *cpu, struct desc_struct *idt, const unsigned long *def); void guest_set_clockevent(struct lg_cpu *cpu, unsigned long delta); +bool send_notify_to_eventfd(struct lg_cpu *cpu); void init_clockdev(struct lg_cpu *cpu); bool check_syscall_vector(struct lguest *lg); int init_interrupts(void); @@ -160,22 +178,26 @@ void free_interrupts(void); /* segments.c: */ void setup_default_gdt_entries(struct lguest_ro_state *state); void setup_guest_gdt(struct lg_cpu *cpu); -void load_guest_gdt(struct lg_cpu *cpu, unsigned long table, u32 num); +void load_guest_gdt_entry(struct lg_cpu *cpu, unsigned int i, + u32 low, u32 hi); void guest_load_tls(struct lg_cpu *cpu, unsigned long tls_array); void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt); void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt); /* page_tables.c: */ -int init_guest_pagetable(struct lguest *lg, unsigned long pgtable); +int init_guest_pagetable(struct lguest *lg); void free_guest_pagetable(struct lguest *lg); void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable); +void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 i); +#ifdef CONFIG_X86_PAE void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i); +#endif void guest_pagetable_clear_all(struct lg_cpu *cpu); void guest_pagetable_flush_user(struct lg_cpu *cpu); void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir, unsigned long vaddr, pte_t val); void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages); -int demand_page(struct lg_cpu *cpu, unsigned long cr2, int errcode); +bool demand_page(struct lg_cpu *cpu, unsigned long cr2, int errcode); void pin_page(struct lg_cpu *cpu, unsigned long vaddr); unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr); void page_table_guest_data_init(struct lg_cpu *cpu); diff --git a/drivers/lguest/lguest_device.c b/drivers/lguest/lguest_device.c index 2bc9bf7e88e..d0a1d8a45c8 100644 --- a/drivers/lguest/lguest_device.c +++ b/drivers/lguest/lguest_device.c @@ -1,10 +1,12 @@ -/*P:050 Lguest guests use a very simple method to describe devices. It's a +/*P:050 + * Lguest guests use a very simple method to describe devices. It's a * series of device descriptors contained just above the top of normal Guest * memory. * * We use the standard "virtio" device infrastructure, which provides us with a * console, a network and a block driver. Each one expects some configuration - * information and a "virtqueue" or two to send and receive data. :*/ + * information and a "virtqueue" or two to send and receive data. +:*/ #include <linux/init.h> #include <linux/bootmem.h> #include <linux/lguest_launcher.h> @@ -13,6 +15,8 @@ #include <linux/interrupt.h> #include <linux/virtio_ring.h> #include <linux/err.h> +#include <linux/export.h> +#include <linux/slab.h> #include <asm/io.h> #include <asm/paravirt.h> #include <asm/lguest_hcall.h> @@ -20,14 +24,13 @@ /* The pointer to our (page) of device descriptions. */ static void *lguest_devices; -/* Unique numbering for lguest devices. */ -static unsigned int dev_index; - -/* For Guests, device memory can be used as normal memory, so we cast away the - * __iomem to quieten sparse. */ +/* + * For Guests, device memory can be used as normal memory, so we cast away the + * __iomem to quieten sparse. + */ static inline void *lguest_map(unsigned long phys_addr, unsigned long pages) { - return (__force void *)ioremap(phys_addr, PAGE_SIZE*pages); + return (__force void *)ioremap_cache(phys_addr, PAGE_SIZE*pages); } static inline void lguest_unmap(void *addr) @@ -35,8 +38,10 @@ static inline void lguest_unmap(void *addr) iounmap((__force void __iomem *)addr); } -/*D:100 Each lguest device is just a virtio device plus a pointer to its entry - * in the lguest_devices page. */ +/*D:100 + * Each lguest device is just a virtio device plus a pointer to its entry + * in the lguest_devices page. + */ struct lguest_device { struct virtio_device vdev; @@ -44,9 +49,11 @@ struct lguest_device { struct lguest_device_desc *desc; }; -/* Since the virtio infrastructure hands us a pointer to the virtio_device all +/* + * Since the virtio infrastructure hands us a pointer to the virtio_device all * the time, it helps to have a curt macro to get a pointer to the struct - * lguest_device it's enclosed in. */ + * lguest_device it's enclosed in. + */ #define to_lgdev(vd) container_of(vd, struct lguest_device, vdev) /*D:130 @@ -58,7 +65,8 @@ struct lguest_device { * the driver will look at them during setup. * * A convenient routine to return the device's virtqueue config array: - * immediately after the descriptor. */ + * immediately after the descriptor. + */ static struct lguest_vqconfig *lg_vq(const struct lguest_device_desc *desc) { return (void *)(desc + 1); @@ -85,27 +93,63 @@ static unsigned desc_size(const struct lguest_device_desc *desc) + desc->config_len; } -/* This tests (and acknowleges) a feature bit. */ -static bool lg_feature(struct virtio_device *vdev, unsigned fbit) +/* This gets the device's feature bits. */ +static u32 lg_get_features(struct virtio_device *vdev) { + unsigned int i; + u32 features = 0; struct lguest_device_desc *desc = to_lgdev(vdev)->desc; - u8 *features; - - /* Obviously if they ask for a feature off the end of our feature - * bitmap, it's not set. */ - if (fbit / 8 > desc->feature_len) - return false; - - /* The feature bitmap comes after the virtqueues. */ - features = lg_features(desc); - if (!(features[fbit / 8] & (1 << (fbit % 8)))) - return false; - - /* We set the matching bit in the other half of the bitmap to tell the - * Host we want to use this feature. We don't use this yet, but we - * could in future. */ - features[desc->feature_len + fbit / 8] |= (1 << (fbit % 8)); - return true; + u8 *in_features = lg_features(desc); + + /* We do this the slow but generic way. */ + for (i = 0; i < min(desc->feature_len * 8, 32); i++) + if (in_features[i / 8] & (1 << (i % 8))) + features |= (1 << i); + + return features; +} + +/* + * To notify on reset or feature finalization, we (ab)use the NOTIFY + * hypercall, with the descriptor address of the device. + */ +static void status_notify(struct virtio_device *vdev) +{ + unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices; + + hcall(LHCALL_NOTIFY, (max_pfn << PAGE_SHIFT) + offset, 0, 0, 0); +} + +/* + * The virtio core takes the features the Host offers, and copies the ones + * supported by the driver into the vdev->features array. Once that's all + * sorted out, this routine is called so we can tell the Host which features we + * understand and accept. + */ +static void lg_finalize_features(struct virtio_device *vdev) +{ + unsigned int i, bits; + struct lguest_device_desc *desc = to_lgdev(vdev)->desc; + /* Second half of bitmap is features we accept. */ + u8 *out_features = lg_features(desc) + desc->feature_len; + + /* Give virtio_ring a chance to accept features. */ + vring_transport_features(vdev); + + /* + * The vdev->feature array is a Linux bitmask: this isn't the same as a + * the simple array of bits used by lguest devices for features. So we + * do this slow, manual conversion which is completely general. + */ + memset(out_features, 0, desc->feature_len); + bits = min_t(unsigned, desc->feature_len, sizeof(vdev->features)) * 8; + for (i = 0; i < bits; i++) { + if (test_bit(i, vdev->features)) + out_features[i / 8] |= (1 << (i % 8)); + } + + /* Tell Host we've finished with this device's feature negotiation */ + status_notify(vdev); } /* Once they've found a field, getting a copy of it is easy. */ @@ -130,8 +174,10 @@ static void lg_set(struct virtio_device *vdev, unsigned int offset, memcpy(lg_config(desc) + offset, buf, len); } -/* The operations to get and set the status word just access the status field - * of the device descriptor. */ +/* + * The operations to get and set the status word just access the status field + * of the device descriptor. + */ static u8 lg_get_status(struct virtio_device *vdev) { return to_lgdev(vdev)->desc->status; @@ -141,16 +187,17 @@ static void lg_set_status(struct virtio_device *vdev, u8 status) { BUG_ON(!status); to_lgdev(vdev)->desc->status = status; + + /* Tell Host immediately if we failed. */ + if (status & VIRTIO_CONFIG_S_FAILED) + status_notify(vdev); } -/* To reset the device, we (ab)use the NOTIFY hypercall, with the descriptor - * address of the device. The Host will zero the status and all the - * features. */ static void lg_reset(struct virtio_device *vdev) { - unsigned long offset = (void *)to_lgdev(vdev)->desc - lguest_devices; - - hcall(LHCALL_NOTIFY, (max_pfn<<PAGE_SHIFT) + offset, 0, 0); + /* 0 status means "reset" */ + to_lgdev(vdev)->desc->status = 0; + status_notify(vdev); } /* @@ -169,8 +216,7 @@ static void lg_reset(struct virtio_device *vdev) */ /*D:140 This is the information we remember about each virtqueue. */ -struct lguest_vq_info -{ +struct lguest_vq_info { /* A copy of the information contained in the device config. */ struct lguest_vqconfig config; @@ -178,19 +224,28 @@ struct lguest_vq_info void *pages; }; -/* When the virtio_ring code wants to prod the Host, it calls us here and we +/* + * When the virtio_ring code wants to prod the Host, it calls us here and we * make a hypercall. We hand the physical address of the virtqueue so the Host - * knows which virtqueue we're talking about. */ -static void lg_notify(struct virtqueue *vq) + * knows which virtqueue we're talking about. + */ +static bool lg_notify(struct virtqueue *vq) { - /* We store our virtqueue information in the "priv" pointer of the - * virtqueue structure. */ + /* + * We store our virtqueue information in the "priv" pointer of the + * virtqueue structure. + */ struct lguest_vq_info *lvq = vq->priv; - hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0); + hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0, 0); + return true; } -/* This routine finds the first virtqueue described in the configuration of +/* An extern declaration inside a C file is bad form. Don't do it. */ +extern int lguest_setup_irq(unsigned int irq); + +/* + * This routine finds the Nth virtqueue described in the configuration of * this device and sets it up. * * This is kind of an ugly duckling. It'd be nicer to have a standard @@ -198,18 +253,20 @@ static void lg_notify(struct virtqueue *vq) * everyone wants to do it differently. The KVM coders want the Guest to * allocate its own pages and tell the Host where they are, but for lguest it's * simpler for the Host to simply tell us where the pages are. - * - * So we provide drivers with a "find the Nth virtqueue and set it up" - * function. */ + */ static struct virtqueue *lg_find_vq(struct virtio_device *vdev, unsigned index, - void (*callback)(struct virtqueue *vq)) + void (*callback)(struct virtqueue *vq), + const char *name) { struct lguest_device *ldev = to_lgdev(vdev); struct lguest_vq_info *lvq; struct virtqueue *vq; int err; + if (!name) + return NULL; + /* We must have this many virtqueues. */ if (index >= ldev->desc->num_vq) return ERR_PTR(-ENOENT); @@ -218,9 +275,11 @@ static struct virtqueue *lg_find_vq(struct virtio_device *vdev, if (!lvq) return ERR_PTR(-ENOMEM); - /* Make a copy of the "struct lguest_vqconfig" entry, which sits after + /* + * Make a copy of the "struct lguest_vqconfig" entry, which sits after * the descriptor. We need a copy because the config space might not - * be aligned correctly. */ + * be aligned correctly. + */ memcpy(&lvq->config, lg_vq(ldev->desc)+index, sizeof(lvq->config)); printk("Mapping virtqueue %i addr %lx\n", index, @@ -228,37 +287,53 @@ static struct virtqueue *lg_find_vq(struct virtio_device *vdev, /* Figure out how many pages the ring will take, and map that memory */ lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT, DIV_ROUND_UP(vring_size(lvq->config.num, - PAGE_SIZE), + LGUEST_VRING_ALIGN), PAGE_SIZE)); if (!lvq->pages) { err = -ENOMEM; goto free_lvq; } - /* OK, tell virtio_ring.c to set up a virtqueue now we know its size - * and we've got a pointer to its pages. */ - vq = vring_new_virtqueue(lvq->config.num, vdev, lvq->pages, - lg_notify, callback); + /* + * OK, tell virtio_ring.c to set up a virtqueue now we know its size + * and we've got a pointer to its pages. Note that we set weak_barriers + * to 'true': the host just a(nother) SMP CPU, so we only need inter-cpu + * barriers. + */ + vq = vring_new_virtqueue(index, lvq->config.num, LGUEST_VRING_ALIGN, vdev, + true, lvq->pages, lg_notify, callback, name); if (!vq) { err = -ENOMEM; goto unmap; } - /* Tell the interrupt for this virtqueue to go to the virtio_ring - * interrupt handler. */ - /* FIXME: We used to have a flag for the Host to tell us we could use + /* Make sure the interrupt is allocated. */ + err = lguest_setup_irq(lvq->config.irq); + if (err) + goto destroy_vring; + + /* + * Tell the interrupt for this virtqueue to go to the virtio_ring + * interrupt handler. + * + * FIXME: We used to have a flag for the Host to tell us we could use * the interrupt as a source of randomness: it'd be nice to have that - * back.. */ + * back. + */ err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED, - vdev->dev.bus_id, vq); + dev_name(&vdev->dev), vq); if (err) - goto destroy_vring; + goto free_desc; - /* Last of all we hook up our 'struct lguest_vq_info" to the - * virtqueue's priv pointer. */ + /* + * Last of all we hook up our 'struct lguest_vq_info" to the + * virtqueue's priv pointer. + */ vq->priv = lvq; return vq; +free_desc: + irq_free_desc(lvq->config.irq); destroy_vring: vring_del_virtqueue(vq); unmap: @@ -284,26 +359,65 @@ static void lg_del_vq(struct virtqueue *vq) kfree(lvq); } +static void lg_del_vqs(struct virtio_device *vdev) +{ + struct virtqueue *vq, *n; + + list_for_each_entry_safe(vq, n, &vdev->vqs, list) + lg_del_vq(vq); +} + +static int lg_find_vqs(struct virtio_device *vdev, unsigned nvqs, + struct virtqueue *vqs[], + vq_callback_t *callbacks[], + const char *names[]) +{ + struct lguest_device *ldev = to_lgdev(vdev); + int i; + + /* We must have this many virtqueues. */ + if (nvqs > ldev->desc->num_vq) + return -ENOENT; + + for (i = 0; i < nvqs; ++i) { + vqs[i] = lg_find_vq(vdev, i, callbacks[i], names[i]); + if (IS_ERR(vqs[i])) + goto error; + } + return 0; + +error: + lg_del_vqs(vdev); + return PTR_ERR(vqs[i]); +} + +static const char *lg_bus_name(struct virtio_device *vdev) +{ + return ""; +} + /* The ops structure which hooks everything together. */ -static struct virtio_config_ops lguest_config_ops = { - .feature = lg_feature, +static const struct virtio_config_ops lguest_config_ops = { + .get_features = lg_get_features, + .finalize_features = lg_finalize_features, .get = lg_get, .set = lg_set, .get_status = lg_get_status, .set_status = lg_set_status, .reset = lg_reset, - .find_vq = lg_find_vq, - .del_vq = lg_del_vq, + .find_vqs = lg_find_vqs, + .del_vqs = lg_del_vqs, + .bus_name = lg_bus_name, }; -/* The root device for the lguest virtio devices. This makes them appear as - * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */ -static struct device lguest_root = { - .parent = NULL, - .bus_id = "lguest", -}; +/* + * The root device for the lguest virtio devices. This makes them appear as + * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. + */ +static struct device *lguest_root; -/*D:120 This is the core of the lguest bus: actually adding a new device. +/*D:120 + * This is the core of the lguest bus: actually adding a new device. * It's a separate function because it's neater that way, and because an * earlier version of the code supported hotplug and unplug. They were removed * early on because they were never used. @@ -311,46 +425,54 @@ static struct device lguest_root = { * As Andrew Tridgell says, "Untested code is buggy code". * * It's worth reading this carefully: we start with a pointer to the new device - * descriptor in the "lguest_devices" page. */ -static void add_lguest_device(struct lguest_device_desc *d) + * descriptor in the "lguest_devices" page, and the offset into the device + * descriptor page so we can uniquely identify it if things go badly wrong. + */ +static void add_lguest_device(struct lguest_device_desc *d, + unsigned int offset) { struct lguest_device *ldev; - /* Start with zeroed memory; Linux's device layer seems to count on - * it. */ + /* Start with zeroed memory; Linux's device layer counts on it. */ ldev = kzalloc(sizeof(*ldev), GFP_KERNEL); if (!ldev) { - printk(KERN_EMERG "Cannot allocate lguest dev %u\n", - dev_index++); + printk(KERN_EMERG "Cannot allocate lguest dev %u type %u\n", + offset, d->type); return; } /* This devices' parent is the lguest/ dir. */ - ldev->vdev.dev.parent = &lguest_root; - /* We have a unique device index thanks to the dev_index counter. */ - ldev->vdev.index = dev_index++; - /* The device type comes straight from the descriptor. There's also a + ldev->vdev.dev.parent = lguest_root; + /* + * The device type comes straight from the descriptor. There's also a * device vendor field in the virtio_device struct, which we leave as - * 0. */ + * 0. + */ ldev->vdev.id.device = d->type; - /* We have a simple set of routines for querying the device's - * configuration information and setting its status. */ + /* + * We have a simple set of routines for querying the device's + * configuration information and setting its status. + */ ldev->vdev.config = &lguest_config_ops; /* And we remember the device's descriptor for lguest_config_ops. */ ldev->desc = d; - /* register_virtio_device() sets up the generic fields for the struct + /* + * register_virtio_device() sets up the generic fields for the struct * virtio_device and calls device_register(). This makes the bus - * infrastructure look for a matching driver. */ + * infrastructure look for a matching driver. + */ if (register_virtio_device(&ldev->vdev) != 0) { - printk(KERN_ERR "Failed to register lguest device %u\n", - ldev->vdev.index); + printk(KERN_ERR "Failed to register lguest dev %u type %u\n", + offset, d->type); kfree(ldev); } } -/*D:110 scan_devices() simply iterates through the device page. The type 0 is - * reserved to mean "end of devices". */ +/*D:110 + * scan_devices() simply iterates through the device page. The type 0 is + * reserved to mean "end of devices". + */ static void scan_devices(void) { unsigned int i; @@ -365,11 +487,12 @@ static void scan_devices(void) break; printk("Device at %i has size %u\n", i, desc_size(d)); - add_lguest_device(d); + add_lguest_device(d, i); } } -/*D:105 Fairly early in boot, lguest_devices_init() is called to set up the +/*D:105 + * Fairly early in boot, lguest_devices_init() is called to set up the * lguest device infrastructure. We check that we are a Guest by checking * pv_info.name: there are other ways of checking, but this seems most * obvious to me. @@ -380,13 +503,15 @@ static void scan_devices(void) * correct sysfs incantation). * * Finally we call scan_devices() which adds all the devices found in the - * lguest_devices page. */ + * lguest_devices page. + */ static int __init lguest_devices_init(void) { if (strcmp(pv_info.name, "lguest") != 0) return 0; - if (device_register(&lguest_root) != 0) + lguest_root = root_device_register("lguest"); + if (IS_ERR(lguest_root)) panic("Could not register lguest root"); /* Devices are in a single page above top of "normal" mem */ @@ -398,11 +523,13 @@ static int __init lguest_devices_init(void) /* We do this after core stuff, but before the drivers. */ postcore_initcall(lguest_devices_init); -/*D:150 At this point in the journey we used to now wade through the lguest +/*D:150 + * At this point in the journey we used to now wade through the lguest * devices themselves: net, block and console. Since they're all now virtio * devices rather than lguest-specific, I've decided to ignore them. Mostly, * they're kind of boring. But this does mean you'll never experience the * thrill of reading the forbidden love scene buried deep in the block driver. * * "make Launcher" beckons, where we answer questions like "Where do Guests - * come from?", and "What do you do when someone asks for optimization?". */ + * come from?", and "What do you do when someone asks for optimization?". + */ diff --git a/drivers/lguest/lguest_user.c b/drivers/lguest/lguest_user.c index 645e6e040bf..4263f4cc8c5 100644 --- a/drivers/lguest/lguest_user.c +++ b/drivers/lguest/lguest_user.c @@ -1,42 +1,182 @@ -/*P:200 This contains all the /dev/lguest code, whereby the userspace launcher - * controls and communicates with the Guest. For example, the first write will - * tell us the Guest's memory layout, pagetable, entry point and kernel address - * offset. A read will run the Guest until something happens, such as a signal - * or the Guest doing a NOTIFY out to the Launcher. :*/ +/*P:200 This contains all the /dev/lguest code, whereby the userspace + * launcher controls and communicates with the Guest. For example, + * the first write will tell us the Guest's memory layout and entry + * point. A read will run the Guest until something happens, such as + * a signal or the Guest doing a NOTIFY out to the Launcher. There is + * also a way for the Launcher to attach eventfds to particular NOTIFY + * values instead of returning from the read() call. +:*/ #include <linux/uaccess.h> #include <linux/miscdevice.h> #include <linux/fs.h> #include <linux/sched.h> +#include <linux/eventfd.h> +#include <linux/file.h> +#include <linux/slab.h> +#include <linux/export.h> #include "lg.h" -/*L:055 When something happens, the Waker process needs a way to stop the - * kernel running the Guest and return to the Launcher. So the Waker writes - * LHREQ_BREAK and the value "1" to /dev/lguest to do this. Once the Launcher - * has done whatever needs attention, it writes LHREQ_BREAK and "0" to release - * the Waker. */ -static int break_guest_out(struct lg_cpu *cpu, const unsigned long __user*input) +/*L:056 + * Before we move on, let's jump ahead and look at what the kernel does when + * it needs to look up the eventfds. That will complete our picture of how we + * use RCU. + * + * The notification value is in cpu->pending_notify: we return true if it went + * to an eventfd. + */ +bool send_notify_to_eventfd(struct lg_cpu *cpu) { - unsigned long on; + unsigned int i; + struct lg_eventfd_map *map; + + /* + * This "rcu_read_lock()" helps track when someone is still looking at + * the (RCU-using) eventfds array. It's not actually a lock at all; + * indeed it's a noop in many configurations. (You didn't expect me to + * explain all the RCU secrets here, did you?) + */ + rcu_read_lock(); + /* + * rcu_dereference is the counter-side of rcu_assign_pointer(); it + * makes sure we don't access the memory pointed to by + * cpu->lg->eventfds before cpu->lg->eventfds is set. Sounds crazy, + * but Alpha allows this! Paul McKenney points out that a really + * aggressive compiler could have the same effect: + * http://lists.ozlabs.org/pipermail/lguest/2009-July/001560.html + * + * So play safe, use rcu_dereference to get the rcu-protected pointer: + */ + map = rcu_dereference(cpu->lg->eventfds); + /* + * Simple array search: even if they add an eventfd while we do this, + * we'll continue to use the old array and just won't see the new one. + */ + for (i = 0; i < map->num; i++) { + if (map->map[i].addr == cpu->pending_notify) { + eventfd_signal(map->map[i].event, 1); + cpu->pending_notify = 0; + break; + } + } + /* We're done with the rcu-protected variable cpu->lg->eventfds. */ + rcu_read_unlock(); - /* Fetch whether they're turning break on or off. */ - if (get_user(on, input) != 0) - return -EFAULT; + /* If we cleared the notification, it's because we found a match. */ + return cpu->pending_notify == 0; +} - if (on) { - cpu->break_out = 1; - /* Pop it out of the Guest (may be running on different CPU) */ - wake_up_process(cpu->tsk); - /* Wait for them to reset it */ - return wait_event_interruptible(cpu->break_wq, !cpu->break_out); - } else { - cpu->break_out = 0; - wake_up(&cpu->break_wq); - return 0; +/*L:055 + * One of the more tricksy tricks in the Linux Kernel is a technique called + * Read Copy Update. Since one point of lguest is to teach lguest journeyers + * about kernel coding, I use it here. (In case you're curious, other purposes + * include learning about virtualization and instilling a deep appreciation for + * simplicity and puppies). + * + * We keep a simple array which maps LHCALL_NOTIFY values to eventfds, but we + * add new eventfds without ever blocking readers from accessing the array. + * The current Launcher only does this during boot, so that never happens. But + * Read Copy Update is cool, and adding a lock risks damaging even more puppies + * than this code does. + * + * We allocate a brand new one-larger array, copy the old one and add our new + * element. Then we make the lg eventfd pointer point to the new array. + * That's the easy part: now we need to free the old one, but we need to make + * sure no slow CPU somewhere is still looking at it. That's what + * synchronize_rcu does for us: waits until every CPU has indicated that it has + * moved on to know it's no longer using the old one. + * + * If that's unclear, see http://en.wikipedia.org/wiki/Read-copy-update. + */ +static int add_eventfd(struct lguest *lg, unsigned long addr, int fd) +{ + struct lg_eventfd_map *new, *old = lg->eventfds; + + /* + * We don't allow notifications on value 0 anyway (pending_notify of + * 0 means "nothing pending"). + */ + if (!addr) + return -EINVAL; + + /* + * Replace the old array with the new one, carefully: others can + * be accessing it at the same time. + */ + new = kmalloc(sizeof(*new) + sizeof(new->map[0]) * (old->num + 1), + GFP_KERNEL); + if (!new) + return -ENOMEM; + + /* First make identical copy. */ + memcpy(new->map, old->map, sizeof(old->map[0]) * old->num); + new->num = old->num; + + /* Now append new entry. */ + new->map[new->num].addr = addr; + new->map[new->num].event = eventfd_ctx_fdget(fd); + if (IS_ERR(new->map[new->num].event)) { + int err = PTR_ERR(new->map[new->num].event); + kfree(new); + return err; } + new->num++; + + /* + * Now put new one in place: rcu_assign_pointer() is a fancy way of + * doing "lg->eventfds = new", but it uses memory barriers to make + * absolutely sure that the contents of "new" written above is nailed + * down before we actually do the assignment. + * + * We have to think about these kinds of things when we're operating on + * live data without locks. + */ + rcu_assign_pointer(lg->eventfds, new); + + /* + * We're not in a big hurry. Wait until no one's looking at old + * version, then free it. + */ + synchronize_rcu(); + kfree(old); + + return 0; } -/*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt - * number to /dev/lguest. */ +/*L:052 + * Receiving notifications from the Guest is usually done by attaching a + * particular LHCALL_NOTIFY value to an event filedescriptor. The eventfd will + * become readable when the Guest does an LHCALL_NOTIFY with that value. + * + * This is really convenient for processing each virtqueue in a separate + * thread. + */ +static int attach_eventfd(struct lguest *lg, const unsigned long __user *input) +{ + unsigned long addr, fd; + int err; + + if (get_user(addr, input) != 0) + return -EFAULT; + input++; + if (get_user(fd, input) != 0) + return -EFAULT; + + /* + * Just make sure two callers don't add eventfds at once. We really + * only need to lock against callers adding to the same Guest, so using + * the Big Lguest Lock is overkill. But this is setup, not a fast path. + */ + mutex_lock(&lguest_lock); + err = add_eventfd(lg, addr, fd); + mutex_unlock(&lguest_lock); + + return err; +} + +/*L:050 + * Sending an interrupt is done by writing LHREQ_IRQ and an interrupt + * number to /dev/lguest. + */ static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input) { unsigned long irq; @@ -45,14 +185,19 @@ static int user_send_irq(struct lg_cpu *cpu, const unsigned long __user *input) return -EFAULT; if (irq >= LGUEST_IRQS) return -EINVAL; - /* Next time the Guest runs, the core code will see if it can deliver - * this interrupt. */ - set_bit(irq, cpu->irqs_pending); + + /* + * Next time the Guest runs, the core code will see if it can deliver + * this interrupt. + */ + set_interrupt(cpu, irq); return 0; } -/*L:040 Once our Guest is initialized, the Launcher makes it run by reading - * from /dev/lguest. */ +/*L:040 + * Once our Guest is initialized, the Launcher makes it run by reading + * from /dev/lguest. + */ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) { struct lguest *lg = file->private_data; @@ -88,8 +233,10 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) return len; } - /* If we returned from read() last time because the Guest sent I/O, - * clear the flag. */ + /* + * If we returned from read() last time because the Guest sent I/O, + * clear the flag. + */ if (cpu->pending_notify) cpu->pending_notify = 0; @@ -97,57 +244,67 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o) return run_guest(cpu, (unsigned long __user *)user); } -/*L:025 This actually initializes a CPU. For the moment, a Guest is only - * uniprocessor, so "id" is always 0. */ +/*L:025 + * This actually initializes a CPU. For the moment, a Guest is only + * uniprocessor, so "id" is always 0. + */ static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) { - /* We have a limited number the number of CPUs in the lguest struct. */ - if (id >= NR_CPUS) + /* We have a limited number of CPUs in the lguest struct. */ + if (id >= ARRAY_SIZE(cpu->lg->cpus)) return -EINVAL; /* Set up this CPU's id, and pointer back to the lguest struct. */ cpu->id = id; - cpu->lg = container_of((cpu - id), struct lguest, cpus[0]); + cpu->lg = container_of(cpu, struct lguest, cpus[id]); cpu->lg->nr_cpus++; /* Each CPU has a timer it can set. */ init_clockdev(cpu); - /* We need a complete page for the Guest registers: they are accessible - * to the Guest and we can only grant it access to whole pages. */ + /* + * We need a complete page for the Guest registers: they are accessible + * to the Guest and we can only grant it access to whole pages. + */ cpu->regs_page = get_zeroed_page(GFP_KERNEL); if (!cpu->regs_page) return -ENOMEM; - /* We actually put the registers at the bottom of the page. */ + /* We actually put the registers at the end of the page. */ cpu->regs = (void *)cpu->regs_page + PAGE_SIZE - sizeof(*cpu->regs); - /* Now we initialize the Guest's registers, handing it the start - * address. */ + /* + * Now we initialize the Guest's registers, handing it the start + * address. + */ lguest_arch_setup_regs(cpu, start_ip); - /* Initialize the queue for the Waker to wait on */ - init_waitqueue_head(&cpu->break_wq); - - /* We keep a pointer to the Launcher task (ie. current task) for when - * other Guests want to wake this one (eg. console input). */ + /* + * We keep a pointer to the Launcher task (ie. current task) for when + * other Guests want to wake this one (eg. console input). + */ cpu->tsk = current; - /* We need to keep a pointer to the Launcher's memory map, because if + /* + * We need to keep a pointer to the Launcher's memory map, because if * the Launcher dies we need to clean it up. If we don't keep a - * reference, it is destroyed before close() is called. */ + * reference, it is destroyed before close() is called. + */ cpu->mm = get_task_mm(cpu->tsk); - /* We remember which CPU's pages this Guest used last, for optimization - * when the same Guest runs on the same CPU twice. */ + /* + * We remember which CPU's pages this Guest used last, for optimization + * when the same Guest runs on the same CPU twice. + */ cpu->last_pages = NULL; /* No error == success. */ return 0; } -/*L:020 The initialization write supplies 4 pointer sized (32 or 64 bit) - * values (in addition to the LHREQ_INITIALIZE value). These are: +/*L:020 + * The initialization write supplies 3 pointer sized (32 or 64 bit) values (in + * addition to the LHREQ_INITIALIZE value). These are: * * base: The start of the Guest-physical memory inside the Launcher memory. * @@ -155,21 +312,19 @@ static int lg_cpu_start(struct lg_cpu *cpu, unsigned id, unsigned long start_ip) * allowed to access. The Guest memory lives inside the Launcher, so it sets * this to ensure the Guest can only reach its own memory. * - * pgdir: The (Guest-physical) address of the top of the initial Guest - * pagetables (which are set up by the Launcher). - * * start: The first instruction to execute ("eip" in x86-speak). */ static int initialize(struct file *file, const unsigned long __user *input) { - /* "struct lguest" contains everything we (the Host) know about a - * Guest. */ + /* "struct lguest" contains all we (the Host) know about a Guest. */ struct lguest *lg; int err; - unsigned long args[4]; + unsigned long args[3]; - /* We grab the Big Lguest lock, which protects against multiple - * simultaneous initializations. */ + /* + * We grab the Big Lguest lock, which protects against multiple + * simultaneous initializations. + */ mutex_lock(&lguest_lock); /* You can't initialize twice! Close the device and start again... */ if (file->private_data) { @@ -188,18 +343,27 @@ static int initialize(struct file *file, const unsigned long __user *input) goto unlock; } + lg->eventfds = kmalloc(sizeof(*lg->eventfds), GFP_KERNEL); + if (!lg->eventfds) { + err = -ENOMEM; + goto free_lg; + } + lg->eventfds->num = 0; + /* Populate the easy fields of our "struct lguest" */ lg->mem_base = (void __user *)args[0]; lg->pfn_limit = args[1]; - /* This is the first cpu (cpu 0) and it will start booting at args[3] */ - err = lg_cpu_start(&lg->cpus[0], 0, args[3]); + /* This is the first cpu (cpu 0) and it will start booting at args[2] */ + err = lg_cpu_start(&lg->cpus[0], 0, args[2]); if (err) - goto release_guest; + goto free_eventfds; - /* Initialize the Guest's shadow page tables, using the toplevel - * address the Launcher gave us. This allocates memory, so can fail. */ - err = init_guest_pagetable(lg, args[2]); + /* + * Initialize the Guest's shadow page tables. This allocates + * memory, so can fail. + */ + err = init_guest_pagetable(lg); if (err) goto free_regs; @@ -214,27 +378,33 @@ static int initialize(struct file *file, const unsigned long __user *input) free_regs: /* FIXME: This should be in free_vcpu */ free_page(lg->cpus[0].regs_page); -release_guest: +free_eventfds: + kfree(lg->eventfds); +free_lg: kfree(lg); unlock: mutex_unlock(&lguest_lock); return err; } -/*L:010 The first operation the Launcher does must be a write. All writes +/*L:010 + * The first operation the Launcher does must be a write. All writes * start with an unsigned long number: for the first write this must be * LHREQ_INITIALIZE to set up the Guest. After that the Launcher can use - * writes of other values to send interrupts. + * writes of other values to send interrupts or set up receipt of notifications. * * Note that we overload the "offset" in the /dev/lguest file to indicate what - * CPU number we're dealing with. Currently this is always 0, since we only + * CPU number we're dealing with. Currently this is always 0 since we only * support uniprocessor Guests, but you can see the beginnings of SMP support - * here. */ + * here. + */ static ssize_t write(struct file *file, const char __user *in, size_t size, loff_t *off) { - /* Once the Guest is initialized, we hold the "struct lguest" in the - * file private data. */ + /* + * Once the Guest is initialized, we hold the "struct lguest" in the + * file private data. + */ struct lguest *lg = file->private_data; const unsigned long __user *input = (const unsigned long __user *)in; unsigned long req; @@ -251,17 +421,10 @@ static ssize_t write(struct file *file, const char __user *in, if (!lg || (cpu_id >= lg->nr_cpus)) return -EINVAL; cpu = &lg->cpus[cpu_id]; - if (!cpu) - return -EINVAL; /* Once the Guest is dead, you can only read() why it died. */ if (lg->dead) return -ENOENT; - - /* If you're not the task which owns the Guest, all you can do - * is break the Launcher out of running the Guest. */ - if (current != cpu->tsk && req != LHREQ_BREAK) - return -EPERM; } switch (req) { @@ -269,20 +432,22 @@ static ssize_t write(struct file *file, const char __user *in, return initialize(file, input); case LHREQ_IRQ: return user_send_irq(cpu, input); - case LHREQ_BREAK: - return break_guest_out(cpu, input); + case LHREQ_EVENTFD: + return attach_eventfd(lg, input); default: return -EINVAL; } } -/*L:060 The final piece of interface code is the close() routine. It reverses +/*L:060 + * The final piece of interface code is the close() routine. It reverses * everything done in initialize(). This is usually called because the * Launcher exited. * * Note that the close routine returns 0 or a negative error number: it can't * really fail, but it can whine. I blame Sun for this wart, and K&R C for - * letting them do it. :*/ + * letting them do it. +:*/ static int close(struct inode *inode, struct file *file) { struct lguest *lg = file->private_data; @@ -292,8 +457,10 @@ static int close(struct inode *inode, struct file *file) if (!lg) return 0; - /* We need the big lock, to protect from inter-guest I/O and other - * Launchers initializing guests. */ + /* + * We need the big lock, to protect from inter-guest I/O and other + * Launchers initializing guests. + */ mutex_lock(&lguest_lock); /* Free up the shadow page tables for the Guest. */ @@ -304,17 +471,26 @@ static int close(struct inode *inode, struct file *file) hrtimer_cancel(&lg->cpus[i].hrt); /* We can free up the register page we allocated. */ free_page(lg->cpus[i].regs_page); - /* Now all the memory cleanups are done, it's safe to release - * the Launcher's memory management structure. */ + /* + * Now all the memory cleanups are done, it's safe to release + * the Launcher's memory management structure. + */ mmput(lg->cpus[i].mm); } - /* If lg->dead doesn't contain an error code it will be NULL or a - * kmalloc()ed string, either of which is ok to hand to kfree(). */ + + /* Release any eventfds they registered. */ + for (i = 0; i < lg->eventfds->num; i++) + eventfd_ctx_put(lg->eventfds->map[i].event); + kfree(lg->eventfds); + + /* + * If lg->dead doesn't contain an error code it will be NULL or a + * kmalloc()ed string, either of which is ok to hand to kfree(). + */ if (!IS_ERR(lg->dead)) kfree(lg->dead); - /* We clear the entire structure, which also marks it as free for the - * next user. */ - memset(lg, 0, sizeof(*lg)); + /* Free the memory allocated to the lguest_struct */ + kfree(lg); /* Release lock and exit. */ mutex_unlock(&lguest_lock); @@ -334,16 +510,21 @@ static int close(struct inode *inode, struct file *file) * * We begin our understanding with the Host kernel interface which the Launcher * uses: reading and writing a character device called /dev/lguest. All the - * work happens in the read(), write() and close() routines: */ -static struct file_operations lguest_fops = { + * work happens in the read(), write() and close() routines: + */ +static const struct file_operations lguest_fops = { .owner = THIS_MODULE, .release = close, .write = write, .read = read, + .llseek = default_llseek, }; +/*:*/ -/* This is a textbook example of a "misc" character device. Populate a "struct - * miscdevice" and register it with misc_register(). */ +/* + * This is a textbook example of a "misc" character device. Populate a "struct + * miscdevice" and register it with misc_register(). + */ static struct miscdevice lguest_dev = { .minor = MISC_DYNAMIC_MINOR, .name = "lguest", diff --git a/drivers/lguest/page_tables.c b/drivers/lguest/page_tables.c index d93500f24fb..e8b55c3a617 100644 --- a/drivers/lguest/page_tables.c +++ b/drivers/lguest/page_tables.c @@ -1,13 +1,16 @@ -/*P:700 The pagetable code, on the other hand, still shows the scars of +/*P:700 + * The pagetable code, on the other hand, still shows the scars of * previous encounters. It's functional, and as neat as it can be in the * circumstances, but be wary, for these things are subtle and break easily. * The Guest provides a virtual to physical mapping, but we can neither trust * it nor use it: we verify and convert it here then point the CPU to the - * converted Guest pages when running the Guest. :*/ + * converted Guest pages when running the Guest. +:*/ -/* Copyright (C) Rusty Russell IBM Corporation 2006. +/* Copyright (C) Rusty Russell IBM Corporation 2013. * GPL v2 and any later version */ #include <linux/mm.h> +#include <linux/gfp.h> #include <linux/types.h> #include <linux/spinlock.h> #include <linux/random.h> @@ -16,18 +19,20 @@ #include <asm/uaccess.h> #include "lg.h" -/*M:008 We hold reference to pages, which prevents them from being swapped. +/*M:008 + * We hold reference to pages, which prevents them from being swapped. * It'd be nice to have a callback in the "struct mm_struct" when Linux wants * to swap out. If we had this, and a shrinker callback to trim PTE pages, we - * could probably consider launching Guests as non-root. :*/ + * could probably consider launching Guests as non-root. +:*/ /*H:300 * The Page Table Code * - * We use two-level page tables for the Guest. If you're not entirely - * comfortable with virtual addresses, physical addresses and page tables then - * I recommend you review arch/x86/lguest/boot.c's "Page Table Handling" (with - * diagrams!). + * We use two-level page tables for the Guest, or three-level with PAE. If + * you're not entirely comfortable with virtual addresses, physical addresses + * and page tables then I recommend you review arch/x86/lguest/boot.c's "Page + * Table Handling" (with diagrams!). * * The Guest keeps page tables, but we maintain the actual ones here: these are * called "shadow" page tables. Which is a very Guest-centric name: these are @@ -44,22 +49,28 @@ * (v) Flushing (throwing away) page tables, * (vi) Mapping the Switcher when the Guest is about to run, * (vii) Setting up the page tables initially. - :*/ +:*/ - -/* 1024 entries in a page table page maps 1024 pages: 4MB. The Switcher is - * conveniently placed at the top 4MB, so it uses a separate, complete PTE - * page. */ +/* + * The Switcher uses the complete top PTE page. That's 1024 PTE entries (4MB) + * or 512 PTE entries with PAE (2MB). + */ #define SWITCHER_PGD_INDEX (PTRS_PER_PGD - 1) -/* We actually need a separate PTE page for each CPU. Remember that after the - * Switcher code itself comes two pages for each CPU, and we don't want this - * CPU's guest to see the pages of any other CPU. */ -static DEFINE_PER_CPU(pte_t *, switcher_pte_pages); -#define switcher_pte_page(cpu) per_cpu(switcher_pte_pages, cpu) - -/*H:320 The page table code is curly enough to need helper functions to keep it - * clear and clean. +/* + * For PAE we need the PMD index as well. We use the last 2MB, so we + * will need the last pmd entry of the last pmd page. + */ +#ifdef CONFIG_X86_PAE +#define CHECK_GPGD_MASK _PAGE_PRESENT +#else +#define CHECK_GPGD_MASK _PAGE_TABLE +#endif + +/*H:320 + * The page table code is curly enough to need helper functions to keep it + * clear and clean. The kernel itself provides many of them; one advantage + * of insisting that the Guest and Host use the same CONFIG_X86_PAE setting. * * There are two functions which return pointers to the shadow (aka "real") * page tables. @@ -67,104 +78,160 @@ static DEFINE_PER_CPU(pte_t *, switcher_pte_pages); * spgd_addr() takes the virtual address and returns a pointer to the top-level * page directory entry (PGD) for that address. Since we keep track of several * page tables, the "i" argument tells us which one we're interested in (it's - * usually the current one). */ + * usually the current one). + */ static pgd_t *spgd_addr(struct lg_cpu *cpu, u32 i, unsigned long vaddr) { unsigned int index = pgd_index(vaddr); - /* We kill any Guest trying to touch the Switcher addresses. */ - if (index >= SWITCHER_PGD_INDEX) { - kill_guest(cpu, "attempt to access switcher pages"); - index = 0; - } /* Return a pointer index'th pgd entry for the i'th page table. */ return &cpu->lg->pgdirs[i].pgdir[index]; } -/* This routine then takes the page directory entry returned above, which +#ifdef CONFIG_X86_PAE +/* + * This routine then takes the PGD entry given above, which contains the + * address of the PMD page. It then returns a pointer to the PMD entry for the + * given address. + */ +static pmd_t *spmd_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr) +{ + unsigned int index = pmd_index(vaddr); + pmd_t *page; + + /* You should never call this if the PGD entry wasn't valid */ + BUG_ON(!(pgd_flags(spgd) & _PAGE_PRESENT)); + page = __va(pgd_pfn(spgd) << PAGE_SHIFT); + + return &page[index]; +} +#endif + +/* + * This routine then takes the page directory entry returned above, which * contains the address of the page table entry (PTE) page. It then returns a - * pointer to the PTE entry for the given address. */ -static pte_t *spte_addr(pgd_t spgd, unsigned long vaddr) + * pointer to the PTE entry for the given address. + */ +static pte_t *spte_addr(struct lg_cpu *cpu, pgd_t spgd, unsigned long vaddr) { +#ifdef CONFIG_X86_PAE + pmd_t *pmd = spmd_addr(cpu, spgd, vaddr); + pte_t *page = __va(pmd_pfn(*pmd) << PAGE_SHIFT); + + /* You should never call this if the PMD entry wasn't valid */ + BUG_ON(!(pmd_flags(*pmd) & _PAGE_PRESENT)); +#else pte_t *page = __va(pgd_pfn(spgd) << PAGE_SHIFT); /* You should never call this if the PGD entry wasn't valid */ BUG_ON(!(pgd_flags(spgd) & _PAGE_PRESENT)); - return &page[(vaddr >> PAGE_SHIFT) % PTRS_PER_PTE]; +#endif + + return &page[pte_index(vaddr)]; } -/* These two functions just like the above two, except they access the Guest - * page tables. Hence they return a Guest address. */ +/* + * These functions are just like the above, except they access the Guest + * page tables. Hence they return a Guest address. + */ static unsigned long gpgd_addr(struct lg_cpu *cpu, unsigned long vaddr) { unsigned int index = vaddr >> (PGDIR_SHIFT); return cpu->lg->pgdirs[cpu->cpu_pgd].gpgdir + index * sizeof(pgd_t); } -static unsigned long gpte_addr(pgd_t gpgd, unsigned long vaddr) +#ifdef CONFIG_X86_PAE +/* Follow the PGD to the PMD. */ +static unsigned long gpmd_addr(pgd_t gpgd, unsigned long vaddr) { unsigned long gpage = pgd_pfn(gpgd) << PAGE_SHIFT; BUG_ON(!(pgd_flags(gpgd) & _PAGE_PRESENT)); - return gpage + ((vaddr>>PAGE_SHIFT) % PTRS_PER_PTE) * sizeof(pte_t); + return gpage + pmd_index(vaddr) * sizeof(pmd_t); +} + +/* Follow the PMD to the PTE. */ +static unsigned long gpte_addr(struct lg_cpu *cpu, + pmd_t gpmd, unsigned long vaddr) +{ + unsigned long gpage = pmd_pfn(gpmd) << PAGE_SHIFT; + + BUG_ON(!(pmd_flags(gpmd) & _PAGE_PRESENT)); + return gpage + pte_index(vaddr) * sizeof(pte_t); +} +#else +/* Follow the PGD to the PTE (no mid-level for !PAE). */ +static unsigned long gpte_addr(struct lg_cpu *cpu, + pgd_t gpgd, unsigned long vaddr) +{ + unsigned long gpage = pgd_pfn(gpgd) << PAGE_SHIFT; + + BUG_ON(!(pgd_flags(gpgd) & _PAGE_PRESENT)); + return gpage + pte_index(vaddr) * sizeof(pte_t); } +#endif /*:*/ -/*M:014 get_pfn is slow; it takes the mmap sem and calls get_user_pages. We - * could probably try to grab batches of pages here as an optimization - * (ie. pre-faulting). :*/ +/*M:007 + * get_pfn is slow: we could probably try to grab batches of pages here as + * an optimization (ie. pre-faulting). +:*/ -/*H:350 This routine takes a page number given by the Guest and converts it to +/*H:350 + * This routine takes a page number given by the Guest and converts it to * an actual, physical page number. It can fail for several reasons: the * virtual address might not be mapped by the Launcher, the write flag is set * and the page is read-only, or the write flag was set and the page was * shared so had to be copied, but we ran out of memory. * * This holds a reference to the page, so release_pte() is careful to put that - * back. */ + * back. + */ static unsigned long get_pfn(unsigned long virtpfn, int write) { struct page *page; - /* This value indicates failure. */ - unsigned long ret = -1UL; - /* get_user_pages() is a complex interface: it gets the "struct - * vm_area_struct" and "struct page" assocated with a range of pages. - * It also needs the task's mmap_sem held, and is not very quick. - * It returns the number of pages it got. */ - down_read(¤t->mm->mmap_sem); - if (get_user_pages(current, current->mm, virtpfn << PAGE_SHIFT, - 1, write, 1, &page, NULL) == 1) - ret = page_to_pfn(page); - up_read(¤t->mm->mmap_sem); - return ret; + /* gup me one page at this address please! */ + if (get_user_pages_fast(virtpfn << PAGE_SHIFT, 1, write, &page) == 1) + return page_to_pfn(page); + + /* This value indicates failure. */ + return -1UL; } -/*H:340 Converting a Guest page table entry to a shadow (ie. real) page table +/*H:340 + * Converting a Guest page table entry to a shadow (ie. real) page table * entry can be a little tricky. The flags are (almost) the same, but the * Guest PTE contains a virtual page number: the CPU needs the real page - * number. */ + * number. + */ static pte_t gpte_to_spte(struct lg_cpu *cpu, pte_t gpte, int write) { unsigned long pfn, base, flags; - /* The Guest sets the global flag, because it thinks that it is using + /* + * The Guest sets the global flag, because it thinks that it is using * PGE. We only told it to use PGE so it would tell us whether it was * flushing a kernel mapping or a userspace mapping. We don't actually - * use the global bit, so throw it away. */ + * use the global bit, so throw it away. + */ flags = (pte_flags(gpte) & ~_PAGE_GLOBAL); /* The Guest's pages are offset inside the Launcher. */ base = (unsigned long)cpu->lg->mem_base / PAGE_SIZE; - /* We need a temporary "unsigned long" variable to hold the answer from + /* + * We need a temporary "unsigned long" variable to hold the answer from * get_pfn(), because it returns 0xFFFFFFFF on failure, which wouldn't * fit in spte.pfn. get_pfn() finds the real physical number of the - * page, given the virtual number. */ + * page, given the virtual number. + */ pfn = get_pfn(base + pte_pfn(gpte), write); if (pfn == -1UL) { kill_guest(cpu, "failed to get page %lu", pte_pfn(gpte)); - /* When we destroy the Guest, we'll go through the shadow page + /* + * When we destroy the Guest, we'll go through the shadow page * tables and release_pte() them. Make sure we don't think - * this one is valid! */ + * this one is valid! + */ flags = 0; } /* Now we assemble our shadow PTE from the page number and flags. */ @@ -174,25 +241,127 @@ static pte_t gpte_to_spte(struct lg_cpu *cpu, pte_t gpte, int write) /*H:460 And to complete the chain, release_pte() looks like this: */ static void release_pte(pte_t pte) { - /* Remember that get_user_pages() took a reference to the page, in - * get_pfn()? We have to put it back now. */ + /* + * Remember that get_user_pages_fast() took a reference to the page, in + * get_pfn()? We have to put it back now. + */ if (pte_flags(pte) & _PAGE_PRESENT) - put_page(pfn_to_page(pte_pfn(pte))); + put_page(pte_page(pte)); } /*:*/ -static void check_gpte(struct lg_cpu *cpu, pte_t gpte) +static bool check_gpte(struct lg_cpu *cpu, pte_t gpte) { if ((pte_flags(gpte) & _PAGE_PSE) || - pte_pfn(gpte) >= cpu->lg->pfn_limit) + pte_pfn(gpte) >= cpu->lg->pfn_limit) { kill_guest(cpu, "bad page table entry"); + return false; + } + return true; } -static void check_gpgd(struct lg_cpu *cpu, pgd_t gpgd) +static bool check_gpgd(struct lg_cpu *cpu, pgd_t gpgd) { - if ((pgd_flags(gpgd) & ~_PAGE_TABLE) || - (pgd_pfn(gpgd) >= cpu->lg->pfn_limit)) + if ((pgd_flags(gpgd) & ~CHECK_GPGD_MASK) || + (pgd_pfn(gpgd) >= cpu->lg->pfn_limit)) { kill_guest(cpu, "bad page directory entry"); + return false; + } + return true; +} + +#ifdef CONFIG_X86_PAE +static bool check_gpmd(struct lg_cpu *cpu, pmd_t gpmd) +{ + if ((pmd_flags(gpmd) & ~_PAGE_TABLE) || + (pmd_pfn(gpmd) >= cpu->lg->pfn_limit)) { + kill_guest(cpu, "bad page middle directory entry"); + return false; + } + return true; +} +#endif + +/*H:331 + * This is the core routine to walk the shadow page tables and find the page + * table entry for a specific address. + * + * If allocate is set, then we allocate any missing levels, setting the flags + * on the new page directory and mid-level directories using the arguments + * (which are copied from the Guest's page table entries). + */ +static pte_t *find_spte(struct lg_cpu *cpu, unsigned long vaddr, bool allocate, + int pgd_flags, int pmd_flags) +{ + pgd_t *spgd; + /* Mid level for PAE. */ +#ifdef CONFIG_X86_PAE + pmd_t *spmd; +#endif + + /* Get top level entry. */ + spgd = spgd_addr(cpu, cpu->cpu_pgd, vaddr); + if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) { + /* No shadow entry: allocate a new shadow PTE page. */ + unsigned long ptepage; + + /* If they didn't want us to allocate anything, stop. */ + if (!allocate) + return NULL; + + ptepage = get_zeroed_page(GFP_KERNEL); + /* + * This is not really the Guest's fault, but killing it is + * simple for this corner case. + */ + if (!ptepage) { + kill_guest(cpu, "out of memory allocating pte page"); + return NULL; + } + /* + * And we copy the flags to the shadow PGD entry. The page + * number in the shadow PGD is the page we just allocated. + */ + set_pgd(spgd, __pgd(__pa(ptepage) | pgd_flags)); + } + + /* + * Intel's Physical Address Extension actually uses three levels of + * page tables, so we need to look in the mid-level. + */ +#ifdef CONFIG_X86_PAE + /* Now look at the mid-level shadow entry. */ + spmd = spmd_addr(cpu, *spgd, vaddr); + + if (!(pmd_flags(*spmd) & _PAGE_PRESENT)) { + /* No shadow entry: allocate a new shadow PTE page. */ + unsigned long ptepage; + + /* If they didn't want us to allocate anything, stop. */ + if (!allocate) + return NULL; + + ptepage = get_zeroed_page(GFP_KERNEL); + + /* + * This is not really the Guest's fault, but killing it is + * simple for this corner case. + */ + if (!ptepage) { + kill_guest(cpu, "out of memory allocating pmd page"); + return NULL; + } + + /* + * And we copy the flags to the shadow PMD entry. The page + * number in the shadow PMD is the page we just allocated. + */ + set_pmd(spmd, __pmd(__pa(ptepage) | pmd_flags)); + } +#endif + + /* Get the pointer to the shadow PTE entry we're going to set. */ + return spte_addr(cpu, *spgd, vaddr); } /*H:330 @@ -204,60 +373,98 @@ static void check_gpgd(struct lg_cpu *cpu, pgd_t gpgd) * and return to the Guest without it knowing. * * If we fixed up the fault (ie. we mapped the address), this routine returns - * true. Otherwise, it was a real fault and we need to tell the Guest. */ -int demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode) + * true. Otherwise, it was a real fault and we need to tell the Guest. + */ +bool demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode) { - pgd_t gpgd; - pgd_t *spgd; unsigned long gpte_ptr; pte_t gpte; pte_t *spte; + pmd_t gpmd; + pgd_t gpgd; + + /* We never demand page the Switcher, so trying is a mistake. */ + if (vaddr >= switcher_addr) + return false; /* First step: get the top-level Guest page table entry. */ - gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t); - /* Toplevel not present? We can't map it in. */ - if (!(pgd_flags(gpgd) & _PAGE_PRESENT)) - return 0; + if (unlikely(cpu->linear_pages)) { + /* Faking up a linear mapping. */ + gpgd = __pgd(CHECK_GPGD_MASK); + } else { + gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t); + /* Toplevel not present? We can't map it in. */ + if (!(pgd_flags(gpgd) & _PAGE_PRESENT)) + return false; + + /* + * This kills the Guest if it has weird flags or tries to + * refer to a "physical" address outside the bounds. + */ + if (!check_gpgd(cpu, gpgd)) + return false; + } - /* Now look at the matching shadow entry. */ - spgd = spgd_addr(cpu, cpu->cpu_pgd, vaddr); - if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) { - /* No shadow entry: allocate a new shadow PTE page. */ - unsigned long ptepage = get_zeroed_page(GFP_KERNEL); - /* This is not really the Guest's fault, but killing it is - * simple for this corner case. */ - if (!ptepage) { - kill_guest(cpu, "out of memory allocating pte page"); - return 0; - } - /* We check that the Guest pgd is OK. */ - check_gpgd(cpu, gpgd); - /* And we copy the flags to the shadow PGD entry. The page - * number in the shadow PGD is the page we just allocated. */ - *spgd = __pgd(__pa(ptepage) | pgd_flags(gpgd)); + /* This "mid-level" entry is only used for non-linear, PAE mode. */ + gpmd = __pmd(_PAGE_TABLE); + +#ifdef CONFIG_X86_PAE + if (likely(!cpu->linear_pages)) { + gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t); + /* Middle level not present? We can't map it in. */ + if (!(pmd_flags(gpmd) & _PAGE_PRESENT)) + return false; + + /* + * This kills the Guest if it has weird flags or tries to + * refer to a "physical" address outside the bounds. + */ + if (!check_gpmd(cpu, gpmd)) + return false; } - /* OK, now we look at the lower level in the Guest page table: keep its - * address, because we might update it later. */ - gpte_ptr = gpte_addr(gpgd, vaddr); - gpte = lgread(cpu, gpte_ptr, pte_t); + /* + * OK, now we look at the lower level in the Guest page table: keep its + * address, because we might update it later. + */ + gpte_ptr = gpte_addr(cpu, gpmd, vaddr); +#else + /* + * OK, now we look at the lower level in the Guest page table: keep its + * address, because we might update it later. + */ + gpte_ptr = gpte_addr(cpu, gpgd, vaddr); +#endif + + if (unlikely(cpu->linear_pages)) { + /* Linear? Make up a PTE which points to same page. */ + gpte = __pte((vaddr & PAGE_MASK) | _PAGE_RW | _PAGE_PRESENT); + } else { + /* Read the actual PTE value. */ + gpte = lgread(cpu, gpte_ptr, pte_t); + } /* If this page isn't in the Guest page tables, we can't page it in. */ if (!(pte_flags(gpte) & _PAGE_PRESENT)) - return 0; + return false; - /* Check they're not trying to write to a page the Guest wants - * read-only (bit 2 of errcode == write). */ + /* + * Check they're not trying to write to a page the Guest wants + * read-only (bit 2 of errcode == write). + */ if ((errcode & 2) && !(pte_flags(gpte) & _PAGE_RW)) - return 0; + return false; /* User access to a kernel-only page? (bit 3 == user access) */ if ((errcode & 4) && !(pte_flags(gpte) & _PAGE_USER)) - return 0; + return false; - /* Check that the Guest PTE flags are OK, and the page number is below - * the pfn_limit (ie. not mapping the Launcher binary). */ - check_gpte(cpu, gpte); + /* + * Check that the Guest PTE flags are OK, and the page number is below + * the pfn_limit (ie. not mapping the Launcher binary). + */ + if (!check_gpte(cpu, gpte)) + return false; /* Add the _PAGE_ACCESSED and (for a write) _PAGE_DIRTY flag */ gpte = pte_mkyoung(gpte); @@ -265,31 +472,45 @@ int demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode) gpte = pte_mkdirty(gpte); /* Get the pointer to the shadow PTE entry we're going to set. */ - spte = spte_addr(*spgd, vaddr); - /* If there was a valid shadow PTE entry here before, we release it. - * This can happen with a write to a previously read-only entry. */ + spte = find_spte(cpu, vaddr, true, pgd_flags(gpgd), pmd_flags(gpmd)); + if (!spte) + return false; + + /* + * If there was a valid shadow PTE entry here before, we release it. + * This can happen with a write to a previously read-only entry. + */ release_pte(*spte); - /* If this is a write, we insist that the Guest page is writable (the - * final arg to gpte_to_spte()). */ + /* + * If this is a write, we insist that the Guest page is writable (the + * final arg to gpte_to_spte()). + */ if (pte_dirty(gpte)) *spte = gpte_to_spte(cpu, gpte, 1); else - /* If this is a read, don't set the "writable" bit in the page + /* + * If this is a read, don't set the "writable" bit in the page * table entry, even if the Guest says it's writable. That way * we will come back here when a write does actually occur, so - * we can update the Guest's _PAGE_DIRTY flag. */ - *spte = gpte_to_spte(cpu, pte_wrprotect(gpte), 0); - - /* Finally, we write the Guest PTE entry back: we've set the - * _PAGE_ACCESSED and maybe the _PAGE_DIRTY flags. */ - lgwrite(cpu, gpte_ptr, pte_t, gpte); - - /* The fault is fixed, the page table is populated, the mapping + * we can update the Guest's _PAGE_DIRTY flag. + */ + set_pte(spte, gpte_to_spte(cpu, pte_wrprotect(gpte), 0)); + + /* + * Finally, we write the Guest PTE entry back: we've set the + * _PAGE_ACCESSED and maybe the _PAGE_DIRTY flags. + */ + if (likely(!cpu->linear_pages)) + lgwrite(cpu, gpte_ptr, pte_t, gpte); + + /* + * The fault is fixed, the page table is populated, the mapping * manipulated, the result returned and the code complete. A small * delay and a trace of alliteration are the only indications the Guest - * has that a page fault occurred at all. */ - return 1; + * has that a page fault occurred at all. + */ + return true; } /*H:360 @@ -301,42 +522,92 @@ int demand_page(struct lg_cpu *cpu, unsigned long vaddr, int errcode) * mapped, so it's overkill. * * This is a quick version which answers the question: is this virtual address - * mapped by the shadow page tables, and is it writable? */ -static int page_writable(struct lg_cpu *cpu, unsigned long vaddr) + * mapped by the shadow page tables, and is it writable? + */ +static bool page_writable(struct lg_cpu *cpu, unsigned long vaddr) { - pgd_t *spgd; + pte_t *spte; unsigned long flags; - /* Look at the current top level entry: is it present? */ - spgd = spgd_addr(cpu, cpu->cpu_pgd, vaddr); - if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) - return 0; + /* You can't put your stack in the Switcher! */ + if (vaddr >= switcher_addr) + return false; - /* Check the flags on the pte entry itself: it must be present and - * writable. */ - flags = pte_flags(*(spte_addr(*spgd, vaddr))); + /* If there's no shadow PTE, it's not writable. */ + spte = find_spte(cpu, vaddr, false, 0, 0); + if (!spte) + return false; + /* + * Check the flags on the pte entry itself: it must be present and + * writable. + */ + flags = pte_flags(*spte); return (flags & (_PAGE_PRESENT|_PAGE_RW)) == (_PAGE_PRESENT|_PAGE_RW); } -/* So, when pin_stack_pages() asks us to pin a page, we check if it's already +/* + * So, when pin_stack_pages() asks us to pin a page, we check if it's already * in the page tables, and if not, we call demand_page() with error code 2 - * (meaning "write"). */ + * (meaning "write"). + */ void pin_page(struct lg_cpu *cpu, unsigned long vaddr) { if (!page_writable(cpu, vaddr) && !demand_page(cpu, vaddr, 2)) kill_guest(cpu, "bad stack page %#lx", vaddr); } +/*:*/ + +#ifdef CONFIG_X86_PAE +static void release_pmd(pmd_t *spmd) +{ + /* If the entry's not present, there's nothing to release. */ + if (pmd_flags(*spmd) & _PAGE_PRESENT) { + unsigned int i; + pte_t *ptepage = __va(pmd_pfn(*spmd) << PAGE_SHIFT); + /* For each entry in the page, we might need to release it. */ + for (i = 0; i < PTRS_PER_PTE; i++) + release_pte(ptepage[i]); + /* Now we can free the page of PTEs */ + free_page((long)ptepage); + /* And zero out the PMD entry so we never release it twice. */ + set_pmd(spmd, __pmd(0)); + } +} + +static void release_pgd(pgd_t *spgd) +{ + /* If the entry's not present, there's nothing to release. */ + if (pgd_flags(*spgd) & _PAGE_PRESENT) { + unsigned int i; + pmd_t *pmdpage = __va(pgd_pfn(*spgd) << PAGE_SHIFT); + + for (i = 0; i < PTRS_PER_PMD; i++) + release_pmd(&pmdpage[i]); + + /* Now we can free the page of PMDs */ + free_page((long)pmdpage); + /* And zero out the PGD entry so we never release it twice. */ + set_pgd(spgd, __pgd(0)); + } +} -/*H:450 If we chase down the release_pgd() code, it looks like this: */ -static void release_pgd(struct lguest *lg, pgd_t *spgd) +#else /* !CONFIG_X86_PAE */ +/*H:450 + * If we chase down the release_pgd() code, the non-PAE version looks like + * this. The PAE version is almost identical, but instead of calling + * release_pte it calls release_pmd(), which looks much like this. + */ +static void release_pgd(pgd_t *spgd) { /* If the entry's not present, there's nothing to release. */ if (pgd_flags(*spgd) & _PAGE_PRESENT) { unsigned int i; - /* Converting the pfn to find the actual PTE page is easy: turn + /* + * Converting the pfn to find the actual PTE page is easy: turn * the page number into a physical address, then convert to a - * virtual address (easy for kernel pages like this one). */ + * virtual address (easy for kernel pages like this one). + */ pte_t *ptepage = __va(pgd_pfn(*spgd) << PAGE_SHIFT); /* For each entry in the page, we might need to release it. */ for (i = 0; i < PTRS_PER_PTE; i++) @@ -347,22 +618,27 @@ static void release_pgd(struct lguest *lg, pgd_t *spgd) *spgd = __pgd(0); } } +#endif -/*H:445 We saw flush_user_mappings() twice: once from the flush_user_mappings() +/*H:445 + * We saw flush_user_mappings() twice: once from the flush_user_mappings() * hypercall and once in new_pgdir() when we re-used a top-level pgdir page. - * It simply releases every PTE page from 0 up to the Guest's kernel address. */ + * It simply releases every PTE page from 0 up to the Guest's kernel address. + */ static void flush_user_mappings(struct lguest *lg, int idx) { unsigned int i; /* Release every pgd entry up to the kernel's address. */ for (i = 0; i < pgd_index(lg->kernel_address); i++) - release_pgd(lg, lg->pgdirs[idx].pgdir + i); + release_pgd(lg->pgdirs[idx].pgdir + i); } -/*H:440 (v) Flushing (throwing away) page tables, +/*H:440 + * (v) Flushing (throwing away) page tables, * * The Guest has a hypercall to throw away the page tables: it's used when a - * large number of mappings have been changed. */ + * large number of mappings have been changed. + */ void guest_pagetable_flush_user(struct lg_cpu *cpu) { /* Drop the userspace part of the current page table. */ @@ -375,23 +651,43 @@ unsigned long guest_pa(struct lg_cpu *cpu, unsigned long vaddr) { pgd_t gpgd; pte_t gpte; +#ifdef CONFIG_X86_PAE + pmd_t gpmd; +#endif + + /* Still not set up? Just map 1:1. */ + if (unlikely(cpu->linear_pages)) + return vaddr; /* First step: get the top-level Guest page table entry. */ gpgd = lgread(cpu, gpgd_addr(cpu, vaddr), pgd_t); /* Toplevel not present? We can't map it in. */ - if (!(pgd_flags(gpgd) & _PAGE_PRESENT)) + if (!(pgd_flags(gpgd) & _PAGE_PRESENT)) { kill_guest(cpu, "Bad address %#lx", vaddr); + return -1UL; + } - gpte = lgread(cpu, gpte_addr(gpgd, vaddr), pte_t); +#ifdef CONFIG_X86_PAE + gpmd = lgread(cpu, gpmd_addr(gpgd, vaddr), pmd_t); + if (!(pmd_flags(gpmd) & _PAGE_PRESENT)) { + kill_guest(cpu, "Bad address %#lx", vaddr); + return -1UL; + } + gpte = lgread(cpu, gpte_addr(cpu, gpmd, vaddr), pte_t); +#else + gpte = lgread(cpu, gpte_addr(cpu, gpgd, vaddr), pte_t); +#endif if (!(pte_flags(gpte) & _PAGE_PRESENT)) kill_guest(cpu, "Bad address %#lx", vaddr); return pte_pfn(gpte) * PAGE_SIZE | (vaddr & ~PAGE_MASK); } -/* We keep several page tables. This is a simple routine to find the page +/* + * We keep several page tables. This is a simple routine to find the page * table (if any) corresponding to this top-level address the Guest has given - * us. */ + * us. + */ static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable) { unsigned int i; @@ -401,18 +697,22 @@ static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable) return i; } -/*H:435 And this is us, creating the new page directory. If we really do +/*H:435 + * And this is us, creating the new page directory. If we really do * allocate a new one (and so the kernel parts are not there), we set - * blank_pgdir. */ + * blank_pgdir. + */ static unsigned int new_pgdir(struct lg_cpu *cpu, unsigned long gpgdir, int *blank_pgdir) { unsigned int next; - /* We pick one entry at random to throw out. Choosing the Least - * Recently Used might be better, but this is easy. */ - next = random32() % ARRAY_SIZE(cpu->lg->pgdirs); + /* + * We pick one entry at random to throw out. Choosing the Least + * Recently Used might be better, but this is easy. + */ + next = prandom_u32() % ARRAY_SIZE(cpu->lg->pgdirs); /* If it's never been allocated at all before, try now. */ if (!cpu->lg->pgdirs[next].pgdir) { cpu->lg->pgdirs[next].pgdir = @@ -420,76 +720,160 @@ static unsigned int new_pgdir(struct lg_cpu *cpu, /* If the allocation fails, just keep using the one we have */ if (!cpu->lg->pgdirs[next].pgdir) next = cpu->cpu_pgd; - else - /* This is a blank page, so there are no kernel - * mappings: caller must map the stack! */ + else { + /* + * This is a blank page, so there are no kernel + * mappings: caller must map the stack! + */ *blank_pgdir = 1; + } } /* Record which Guest toplevel this shadows. */ cpu->lg->pgdirs[next].gpgdir = gpgdir; /* Release all the non-kernel mappings. */ flush_user_mappings(cpu->lg, next); + /* This hasn't run on any CPU at all. */ + cpu->lg->pgdirs[next].last_host_cpu = -1; + return next; } -/*H:430 (iv) Switching page tables +/*H:501 + * We do need the Switcher code mapped at all times, so we allocate that + * part of the Guest page table here. We map the Switcher code immediately, + * but defer mapping of the guest register page and IDT/LDT etc page until + * just before we run the guest in map_switcher_in_guest(). * - * Now we've seen all the page table setting and manipulation, let's see what - * what happens when the Guest changes page tables (ie. changes the top-level - * pgdir). This occurs on almost every context switch. */ -void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable) + * We *could* do this setup in map_switcher_in_guest(), but at that point + * we've interrupts disabled, and allocating pages like that is fraught: we + * can't sleep if we need to free up some memory. + */ +static bool allocate_switcher_mapping(struct lg_cpu *cpu) { - int newpgdir, repin = 0; - - /* Look to see if we have this one already. */ - newpgdir = find_pgdir(cpu->lg, pgtable); - /* If not, we allocate or mug an existing one: if it's a fresh one, - * repin gets set to 1. */ - if (newpgdir == ARRAY_SIZE(cpu->lg->pgdirs)) - newpgdir = new_pgdir(cpu, pgtable, &repin); - /* Change the current pgd index to the new one. */ - cpu->cpu_pgd = newpgdir; - /* If it was completely blank, we map in the Guest kernel stack */ - if (repin) - pin_stack_pages(cpu); + int i; + + for (i = 0; i < TOTAL_SWITCHER_PAGES; i++) { + pte_t *pte = find_spte(cpu, switcher_addr + i * PAGE_SIZE, true, + CHECK_GPGD_MASK, _PAGE_TABLE); + if (!pte) + return false; + + /* + * Map the switcher page if not already there. It might + * already be there because we call allocate_switcher_mapping() + * in guest_set_pgd() just in case it did discard our Switcher + * mapping, but it probably didn't. + */ + if (i == 0 && !(pte_flags(*pte) & _PAGE_PRESENT)) { + /* Get a reference to the Switcher page. */ + get_page(lg_switcher_pages[0]); + /* Create a read-only, exectuable, kernel-style PTE */ + set_pte(pte, + mk_pte(lg_switcher_pages[0], PAGE_KERNEL_RX)); + } + } + cpu->lg->pgdirs[cpu->cpu_pgd].switcher_mapped = true; + return true; } -/*H:470 Finally, a routine which throws away everything: all PGD entries in all +/*H:470 + * Finally, a routine which throws away everything: all PGD entries in all * the shadow page tables, including the Guest's kernel mappings. This is used - * when we destroy the Guest. */ + * when we destroy the Guest. + */ static void release_all_pagetables(struct lguest *lg) { unsigned int i, j; /* Every shadow pagetable this Guest has */ - for (i = 0; i < ARRAY_SIZE(lg->pgdirs); i++) - if (lg->pgdirs[i].pgdir) - /* Every PGD entry except the Switcher at the top */ - for (j = 0; j < SWITCHER_PGD_INDEX; j++) - release_pgd(lg, lg->pgdirs[i].pgdir + j); + for (i = 0; i < ARRAY_SIZE(lg->pgdirs); i++) { + if (!lg->pgdirs[i].pgdir) + continue; + + /* Every PGD entry. */ + for (j = 0; j < PTRS_PER_PGD; j++) + release_pgd(lg->pgdirs[i].pgdir + j); + lg->pgdirs[i].switcher_mapped = false; + lg->pgdirs[i].last_host_cpu = -1; + } } -/* We also throw away everything when a Guest tells us it's changed a kernel +/* + * We also throw away everything when a Guest tells us it's changed a kernel * mapping. Since kernel mappings are in every page table, it's easiest to * throw them all away. This traps the Guest in amber for a while as - * everything faults back in, but it's rare. */ + * everything faults back in, but it's rare. + */ void guest_pagetable_clear_all(struct lg_cpu *cpu) { release_all_pagetables(cpu->lg); /* We need the Guest kernel stack mapped again. */ pin_stack_pages(cpu); + /* And we need Switcher allocated. */ + if (!allocate_switcher_mapping(cpu)) + kill_guest(cpu, "Cannot populate switcher mapping"); +} + +/*H:430 + * (iv) Switching page tables + * + * Now we've seen all the page table setting and manipulation, let's see + * what happens when the Guest changes page tables (ie. changes the top-level + * pgdir). This occurs on almost every context switch. + */ +void guest_new_pagetable(struct lg_cpu *cpu, unsigned long pgtable) +{ + int newpgdir, repin = 0; + + /* + * The very first time they call this, we're actually running without + * any page tables; we've been making it up. Throw them away now. + */ + if (unlikely(cpu->linear_pages)) { + release_all_pagetables(cpu->lg); + cpu->linear_pages = false; + /* Force allocation of a new pgdir. */ + newpgdir = ARRAY_SIZE(cpu->lg->pgdirs); + } else { + /* Look to see if we have this one already. */ + newpgdir = find_pgdir(cpu->lg, pgtable); + } + + /* + * If not, we allocate or mug an existing one: if it's a fresh one, + * repin gets set to 1. + */ + if (newpgdir == ARRAY_SIZE(cpu->lg->pgdirs)) + newpgdir = new_pgdir(cpu, pgtable, &repin); + /* Change the current pgd index to the new one. */ + cpu->cpu_pgd = newpgdir; + /* + * If it was completely blank, we map in the Guest kernel stack and + * the Switcher. + */ + if (repin) + pin_stack_pages(cpu); + + if (!cpu->lg->pgdirs[cpu->cpu_pgd].switcher_mapped) { + if (!allocate_switcher_mapping(cpu)) + kill_guest(cpu, "Cannot populate switcher mapping"); + } } /*:*/ -/*M:009 Since we throw away all mappings when a kernel mapping changes, our + +/*M:009 + * Since we throw away all mappings when a kernel mapping changes, our * performance sucks for guests using highmem. In fact, a guest with * PAGE_OFFSET 0xc0000000 (the default) and more than about 700MB of RAM is * usually slower than a Guest with less memory. * * This, of course, cannot be fixed. It would take some kind of... well, I - * don't know, but the term "puissant code-fu" comes to mind. :*/ + * don't know, but the term "puissant code-fu" comes to mind. +:*/ -/*H:420 This is the routine which actually sets the page table entry for then +/*H:420 + * This is the routine which actually sets the page table entry for then * "idx"'th shadow page table. * * Normally, we can just throw out the old entry and replace it with 0: if they @@ -503,33 +887,52 @@ void guest_pagetable_clear_all(struct lg_cpu *cpu) * _PAGE_ACCESSED then we can put a read-only PTE entry in immediately, and if * they set _PAGE_DIRTY then we can put a writable PTE entry in immediately. */ -static void do_set_pte(struct lg_cpu *cpu, int idx, +static void __guest_set_pte(struct lg_cpu *cpu, int idx, unsigned long vaddr, pte_t gpte) { /* Look up the matching shadow page directory entry. */ pgd_t *spgd = spgd_addr(cpu, idx, vaddr); +#ifdef CONFIG_X86_PAE + pmd_t *spmd; +#endif /* If the top level isn't present, there's no entry to update. */ if (pgd_flags(*spgd) & _PAGE_PRESENT) { - /* Otherwise, we start by releasing the existing entry. */ - pte_t *spte = spte_addr(*spgd, vaddr); - release_pte(*spte); - - /* If they're setting this entry as dirty or accessed, we might - * as well put that entry they've given us in now. This shaves - * 10% off a copy-on-write micro-benchmark. */ - if (pte_flags(gpte) & (_PAGE_DIRTY | _PAGE_ACCESSED)) { - check_gpte(cpu, gpte); - *spte = gpte_to_spte(cpu, gpte, - pte_flags(gpte) & _PAGE_DIRTY); - } else - /* Otherwise kill it and we can demand_page() it in - * later. */ - *spte = __pte(0); +#ifdef CONFIG_X86_PAE + spmd = spmd_addr(cpu, *spgd, vaddr); + if (pmd_flags(*spmd) & _PAGE_PRESENT) { +#endif + /* Otherwise, start by releasing the existing entry. */ + pte_t *spte = spte_addr(cpu, *spgd, vaddr); + release_pte(*spte); + + /* + * If they're setting this entry as dirty or accessed, + * we might as well put that entry they've given us in + * now. This shaves 10% off a copy-on-write + * micro-benchmark. + */ + if (pte_flags(gpte) & (_PAGE_DIRTY | _PAGE_ACCESSED)) { + if (!check_gpte(cpu, gpte)) + return; + set_pte(spte, + gpte_to_spte(cpu, gpte, + pte_flags(gpte) & _PAGE_DIRTY)); + } else { + /* + * Otherwise kill it and we can demand_page() + * it in later. + */ + set_pte(spte, __pte(0)); + } +#ifdef CONFIG_X86_PAE + } +#endif } } -/*H:410 Updating a PTE entry is a little trickier. +/*H:410 + * Updating a PTE entry is a little trickier. * * We keep track of several different page tables (the Guest uses one for each * process, so it makes sense to cache at least a few). Each of these have @@ -538,23 +941,32 @@ static void do_set_pte(struct lg_cpu *cpu, int idx, * all the page tables, not just the current one. This is rare. * * The benefit is that when we have to track a new page table, we can keep all - * the kernel mappings. This speeds up context switch immensely. */ + * the kernel mappings. This speeds up context switch immensely. + */ void guest_set_pte(struct lg_cpu *cpu, unsigned long gpgdir, unsigned long vaddr, pte_t gpte) { - /* Kernel mappings must be changed on all top levels. Slow, but doesn't - * happen often. */ + /* We don't let you remap the Switcher; we need it to get back! */ + if (vaddr >= switcher_addr) { + kill_guest(cpu, "attempt to set pte into Switcher pages"); + return; + } + + /* + * Kernel mappings must be changed on all top levels. Slow, but doesn't + * happen often. + */ if (vaddr >= cpu->lg->kernel_address) { unsigned int i; for (i = 0; i < ARRAY_SIZE(cpu->lg->pgdirs); i++) if (cpu->lg->pgdirs[i].pgdir) - do_set_pte(cpu, i, vaddr, gpte); + __guest_set_pte(cpu, i, vaddr, gpte); } else { /* Is this page table one we have a shadow for? */ int pgdir = find_pgdir(cpu->lg, gpgdir); if (pgdir != ARRAY_SIZE(cpu->lg->pgdirs)) /* If so, do the update. */ - do_set_pte(cpu, pgdir, vaddr, gpte); + __guest_set_pte(cpu, pgdir, vaddr, gpte); } } @@ -570,56 +982,101 @@ void guest_set_pte(struct lg_cpu *cpu, * tells us they've changed. When the Guest tries to use the new entry it will * fault and demand_page() will fix it up. * - * So with that in mind here's our code to to update a (top-level) PGD entry: + * So with that in mind here's our code to update a (top-level) PGD entry: */ -void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 idx) +void guest_set_pgd(struct lguest *lg, unsigned long gpgdir, u32 idx) { int pgdir; - /* The kernel seems to try to initialize this early on: we ignore its - * attempts to map over the Switcher. */ - if (idx >= SWITCHER_PGD_INDEX) + if (idx > PTRS_PER_PGD) { + kill_guest(&lg->cpus[0], "Attempt to set pgd %u/%u", + idx, PTRS_PER_PGD); return; + } /* If they're talking about a page table we have a shadow for... */ pgdir = find_pgdir(lg, gpgdir); - if (pgdir < ARRAY_SIZE(lg->pgdirs)) + if (pgdir < ARRAY_SIZE(lg->pgdirs)) { /* ... throw it away. */ - release_pgd(lg, lg->pgdirs[pgdir].pgdir + idx); + release_pgd(lg->pgdirs[pgdir].pgdir + idx); + /* That might have been the Switcher mapping, remap it. */ + if (!allocate_switcher_mapping(&lg->cpus[0])) { + kill_guest(&lg->cpus[0], + "Cannot populate switcher mapping"); + } + lg->pgdirs[pgdir].last_host_cpu = -1; + } } -/*H:500 (vii) Setting up the page tables initially. +#ifdef CONFIG_X86_PAE +/* For setting a mid-level, we just throw everything away. It's easy. */ +void guest_set_pmd(struct lguest *lg, unsigned long pmdp, u32 idx) +{ + guest_pagetable_clear_all(&lg->cpus[0]); +} +#endif + +/*H:500 + * (vii) Setting up the page tables initially. + * + * When a Guest is first created, set initialize a shadow page table which + * we will populate on future faults. The Guest doesn't have any actual + * pagetables yet, so we set linear_pages to tell demand_page() to fake it + * for the moment. * - * When a Guest is first created, the Launcher tells us where the toplevel of - * its first page table is. We set some things up here: */ -int init_guest_pagetable(struct lguest *lg, unsigned long pgtable) -{ - /* We start on the first shadow page table, and give it a blank PGD - * page. */ - lg->pgdirs[0].gpgdir = pgtable; - lg->pgdirs[0].pgdir = (pgd_t *)get_zeroed_page(GFP_KERNEL); - if (!lg->pgdirs[0].pgdir) + * We do need the Switcher to be mapped at all times, so we allocate that + * part of the Guest page table here. + */ +int init_guest_pagetable(struct lguest *lg) +{ + struct lg_cpu *cpu = &lg->cpus[0]; + int allocated = 0; + + /* lg (and lg->cpus[]) starts zeroed: this allocates a new pgdir */ + cpu->cpu_pgd = new_pgdir(cpu, 0, &allocated); + if (!allocated) return -ENOMEM; - lg->cpus[0].cpu_pgd = 0; + + /* We start with a linear mapping until the initialize. */ + cpu->linear_pages = true; + + /* Allocate the page tables for the Switcher. */ + if (!allocate_switcher_mapping(cpu)) { + release_all_pagetables(lg); + return -ENOMEM; + } + return 0; } -/* When the Guest calls LHCALL_LGUEST_INIT we do more setup. */ +/*H:508 When the Guest calls LHCALL_LGUEST_INIT we do more setup. */ void page_table_guest_data_init(struct lg_cpu *cpu) { + /* + * We tell the Guest that it can't use the virtual addresses + * used by the Switcher. This trick is equivalent to 4GB - + * switcher_addr. + */ + u32 top = ~switcher_addr + 1; + /* We get the kernel address: above this is all kernel memory. */ if (get_user(cpu->lg->kernel_address, &cpu->lg->lguest_data->kernel_address) - /* We tell the Guest that it can't use the top 4MB of virtual - * addresses used by the Switcher. */ - || put_user(4U*1024*1024, &cpu->lg->lguest_data->reserve_mem) - || put_user(cpu->lg->pgdirs[0].gpgdir, &cpu->lg->lguest_data->pgdir)) + /* + * We tell the Guest that it can't use the top virtual + * addresses (used by the Switcher). + */ + || put_user(top, &cpu->lg->lguest_data->reserve_mem)) { kill_guest(cpu, "bad guest page %p", cpu->lg->lguest_data); + return; + } - /* In flush_user_mappings() we loop from 0 to + /* + * In flush_user_mappings() we loop from 0 to * "pgd_index(lg->kernel_address)". This assumes it won't hit the - * Switcher mappings, so check that now. */ - if (pgd_index(cpu->lg->kernel_address) >= SWITCHER_PGD_INDEX) + * Switcher mappings, so check that now. + */ + if (cpu->lg->kernel_address >= switcher_addr) kill_guest(cpu, "bad kernel address %#lx", cpu->lg->kernel_address); } @@ -636,77 +1093,97 @@ void free_guest_pagetable(struct lguest *lg) free_page((long)lg->pgdirs[i].pgdir); } -/*H:480 (vi) Mapping the Switcher when the Guest is about to run. - * - * The Switcher and the two pages for this CPU need to be visible in the - * Guest (and not the pages for other CPUs). We have the appropriate PTE pages - * for each CPU already set up, we just need to hook them in now we know which - * Guest is about to run on this CPU. */ -void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages) +/*H:481 + * This clears the Switcher mappings for cpu #i. + */ +static void remove_switcher_percpu_map(struct lg_cpu *cpu, unsigned int i) { - pte_t *switcher_pte_page = __get_cpu_var(switcher_pte_pages); - pgd_t switcher_pgd; - pte_t regs_pte; - unsigned long pfn; - - /* Make the last PGD entry for this Guest point to the Switcher's PTE - * page for this CPU (with appropriate flags). */ - switcher_pgd = __pgd(__pa(switcher_pte_page) | __PAGE_KERNEL); - - cpu->lg->pgdirs[cpu->cpu_pgd].pgdir[SWITCHER_PGD_INDEX] = switcher_pgd; + unsigned long base = switcher_addr + PAGE_SIZE + i * PAGE_SIZE*2; + pte_t *pte; - /* We also change the Switcher PTE page. When we're running the Guest, - * we want the Guest's "regs" page to appear where the first Switcher - * page for this CPU is. This is an optimization: when the Switcher - * saves the Guest registers, it saves them into the first page of this - * CPU's "struct lguest_pages": if we make sure the Guest's register - * page is already mapped there, we don't have to copy them out - * again. */ - pfn = __pa(cpu->regs_page) >> PAGE_SHIFT; - regs_pte = pfn_pte(pfn, __pgprot(__PAGE_KERNEL)); - switcher_pte_page[(unsigned long)pages/PAGE_SIZE%PTRS_PER_PTE] = regs_pte; -} -/*:*/ + /* Clear the mappings for both pages. */ + pte = find_spte(cpu, base, false, 0, 0); + release_pte(*pte); + set_pte(pte, __pte(0)); -static void free_switcher_pte_pages(void) -{ - unsigned int i; - - for_each_possible_cpu(i) - free_page((long)switcher_pte_page(i)); + pte = find_spte(cpu, base + PAGE_SIZE, false, 0, 0); + release_pte(*pte); + set_pte(pte, __pte(0)); } -/*H:520 Setting up the Switcher PTE page for given CPU is fairly easy, given - * the CPU number and the "struct page"s for the Switcher code itself. +/*H:480 + * (vi) Mapping the Switcher when the Guest is about to run. + * + * The Switcher and the two pages for this CPU need to be visible in the Guest + * (and not the pages for other CPUs). * - * Currently the Switcher is less than a page long, so "pages" is always 1. */ -static __init void populate_switcher_pte_page(unsigned int cpu, - struct page *switcher_page[], - unsigned int pages) + * The pages for the pagetables have all been allocated before: we just need + * to make sure the actual PTEs are up-to-date for the CPU we're about to run + * on. + */ +void map_switcher_in_guest(struct lg_cpu *cpu, struct lguest_pages *pages) { - unsigned int i; - pte_t *pte = switcher_pte_page(cpu); + unsigned long base; + struct page *percpu_switcher_page, *regs_page; + pte_t *pte; + struct pgdir *pgdir = &cpu->lg->pgdirs[cpu->cpu_pgd]; + + /* Switcher page should always be mapped by now! */ + BUG_ON(!pgdir->switcher_mapped); + + /* + * Remember that we have two pages for each Host CPU, so we can run a + * Guest on each CPU without them interfering. We need to make sure + * those pages are mapped correctly in the Guest, but since we usually + * run on the same CPU, we cache that, and only update the mappings + * when we move. + */ + if (pgdir->last_host_cpu == raw_smp_processor_id()) + return; - /* The first entries are easy: they map the Switcher code. */ - for (i = 0; i < pages; i++) { - pte[i] = mk_pte(switcher_page[i], - __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED)); + /* -1 means unknown so we remove everything. */ + if (pgdir->last_host_cpu == -1) { + unsigned int i; + for_each_possible_cpu(i) + remove_switcher_percpu_map(cpu, i); + } else { + /* We know exactly what CPU mapping to remove. */ + remove_switcher_percpu_map(cpu, pgdir->last_host_cpu); } - /* The only other thing we map is this CPU's pair of pages. */ - i = pages + cpu*2; - - /* First page (Guest registers) is writable from the Guest */ - pte[i] = pfn_pte(page_to_pfn(switcher_page[i]), - __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED|_PAGE_RW)); - - /* The second page contains the "struct lguest_ro_state", and is - * read-only. */ - pte[i+1] = pfn_pte(page_to_pfn(switcher_page[i+1]), - __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED)); + /* + * When we're running the Guest, we want the Guest's "regs" page to + * appear where the first Switcher page for this CPU is. This is an + * optimization: when the Switcher saves the Guest registers, it saves + * them into the first page of this CPU's "struct lguest_pages": if we + * make sure the Guest's register page is already mapped there, we + * don't have to copy them out again. + */ + /* Find the shadow PTE for this regs page. */ + base = switcher_addr + PAGE_SIZE + + raw_smp_processor_id() * sizeof(struct lguest_pages); + pte = find_spte(cpu, base, false, 0, 0); + regs_page = pfn_to_page(__pa(cpu->regs_page) >> PAGE_SHIFT); + get_page(regs_page); + set_pte(pte, mk_pte(regs_page, __pgprot(__PAGE_KERNEL & ~_PAGE_GLOBAL))); + + /* + * We map the second page of the struct lguest_pages read-only in + * the Guest: the IDT, GDT and other things it's not supposed to + * change. + */ + pte = find_spte(cpu, base + PAGE_SIZE, false, 0, 0); + percpu_switcher_page + = lg_switcher_pages[1 + raw_smp_processor_id()*2 + 1]; + get_page(percpu_switcher_page); + set_pte(pte, mk_pte(percpu_switcher_page, + __pgprot(__PAGE_KERNEL_RO & ~_PAGE_GLOBAL))); + + pgdir->last_host_cpu = raw_smp_processor_id(); } -/* We've made it through the page table code. Perhaps our tired brains are +/*H:490 + * We've made it through the page table code. Perhaps our tired brains are * still processing the details, or perhaps we're simply glad it's over. * * If nothing else, note that all this complexity in juggling shadow page tables @@ -715,28 +1192,5 @@ static __init void populate_switcher_pte_page(unsigned int cpu, * uses exotic direct Guest pagetable manipulation, and why both Intel and AMD * have implemented shadow page table support directly into hardware. * - * There is just one file remaining in the Host. */ - -/*H:510 At boot or module load time, init_pagetables() allocates and populates - * the Switcher PTE page for each CPU. */ -__init int init_pagetables(struct page **switcher_page, unsigned int pages) -{ - unsigned int i; - - for_each_possible_cpu(i) { - switcher_pte_page(i) = (pte_t *)get_zeroed_page(GFP_KERNEL); - if (!switcher_pte_page(i)) { - free_switcher_pte_pages(); - return -ENOMEM; - } - populate_switcher_pte_page(i, switcher_page, pages); - } - return 0; -} -/*:*/ - -/* Cleaning up simply involves freeing the PTE page for each CPU. */ -void free_pagetables(void) -{ - free_switcher_pte_pages(); -} + * There is just one file remaining in the Host. + */ diff --git a/drivers/lguest/segments.c b/drivers/lguest/segments.c index ec6aa3f1c36..c4fb424dfdd 100644 --- a/drivers/lguest/segments.c +++ b/drivers/lguest/segments.c @@ -1,4 +1,5 @@ -/*P:600 The x86 architecture has segments, which involve a table of descriptors +/*P:600 + * The x86 architecture has segments, which involve a table of descriptors * which can be used to do funky things with virtual address interpretation. * We originally used to use segments so the Guest couldn't alter the * Guest<->Host Switcher, and then we had to trim Guest segments, and restore @@ -8,7 +9,8 @@ * * In these modern times, the segment handling code consists of simple sanity * checks, and the worst you'll experience reading this code is butterfly-rash - * from frolicking through its parklike serenity. :*/ + * from frolicking through its parklike serenity. +:*/ #include "lg.h" /*H:600 @@ -41,11 +43,13 @@ * begin. */ -/* There are several entries we don't let the Guest set. The TSS entry is the +/* + * There are several entries we don't let the Guest set. The TSS entry is the * "Task State Segment" which controls all kinds of delicate things. The * LGUEST_CS and LGUEST_DS entries are reserved for the Switcher, and the - * the Guest can't be trusted to deal with double faults. */ -static int ignored_gdt(unsigned int num) + * the Guest can't be trusted to deal with double faults. + */ +static bool ignored_gdt(unsigned int num) { return (num == GDT_ENTRY_TSS || num == GDT_ENTRY_LGUEST_CS @@ -53,42 +57,52 @@ static int ignored_gdt(unsigned int num) || num == GDT_ENTRY_DOUBLEFAULT_TSS); } -/*H:630 Once the Guest gave us new GDT entries, we fix them up a little. We +/*H:630 + * Once the Guest gave us new GDT entries, we fix them up a little. We * don't care if they're invalid: the worst that can happen is a General * Protection Fault in the Switcher when it restores a Guest segment register * which tries to use that entry. Then we kill the Guest for causing such a - * mess: the message will be "unhandled trap 256". */ + * mess: the message will be "unhandled trap 256". + */ static void fixup_gdt_table(struct lg_cpu *cpu, unsigned start, unsigned end) { unsigned int i; for (i = start; i < end; i++) { - /* We never copy these ones to real GDT, so we don't care what - * they say */ + /* + * We never copy these ones to real GDT, so we don't care what + * they say + */ if (ignored_gdt(i)) continue; - /* Segment descriptors contain a privilege level: the Guest is + /* + * Segment descriptors contain a privilege level: the Guest is * sometimes careless and leaves this as 0, even though it's - * running at privilege level 1. If so, we fix it here. */ - if ((cpu->arch.gdt[i].b & 0x00006000) == 0) - cpu->arch.gdt[i].b |= (GUEST_PL << 13); + * running at privilege level 1. If so, we fix it here. + */ + if (cpu->arch.gdt[i].dpl == 0) + cpu->arch.gdt[i].dpl |= GUEST_PL; - /* Each descriptor has an "accessed" bit. If we don't set it + /* + * Each descriptor has an "accessed" bit. If we don't set it * now, the CPU will try to set it when the Guest first loads * that entry into a segment register. But the GDT isn't - * writable by the Guest, so bad things can happen. */ - cpu->arch.gdt[i].b |= 0x00000100; + * writable by the Guest, so bad things can happen. + */ + cpu->arch.gdt[i].type |= 0x1; } } -/*H:610 Like the IDT, we never simply use the GDT the Guest gives us. We keep +/*H:610 + * Like the IDT, we never simply use the GDT the Guest gives us. We keep * a GDT for each CPU, and copy across the Guest's entries each time we want to * run the Guest on that CPU. * * This routine is called at boot or modprobe time for each CPU to set up the * constant GDT entries: the ones which are the same no matter what Guest we're - * running. */ + * running. + */ void setup_default_gdt_entries(struct lguest_ro_state *state) { struct desc_struct *gdt = state->guest_gdt; @@ -98,30 +112,43 @@ void setup_default_gdt_entries(struct lguest_ro_state *state) gdt[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT; gdt[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT; - /* The TSS segment refers to the TSS entry for this particular CPU. - * Forgive the magic flags: the 0x8900 means the entry is Present, it's - * privilege level 0 Available 386 TSS system segment, and the 0x67 - * means Saturn is eclipsed by Mercury in the twelfth house. */ - gdt[GDT_ENTRY_TSS].a = 0x00000067 | (tss << 16); - gdt[GDT_ENTRY_TSS].b = 0x00008900 | (tss & 0xFF000000) - | ((tss >> 16) & 0x000000FF); + /* + * The TSS segment refers to the TSS entry for this particular CPU. + */ + gdt[GDT_ENTRY_TSS].a = 0; + gdt[GDT_ENTRY_TSS].b = 0; + + gdt[GDT_ENTRY_TSS].limit0 = 0x67; + gdt[GDT_ENTRY_TSS].base0 = tss & 0xFFFF; + gdt[GDT_ENTRY_TSS].base1 = (tss >> 16) & 0xFF; + gdt[GDT_ENTRY_TSS].base2 = tss >> 24; + gdt[GDT_ENTRY_TSS].type = 0x9; /* 32-bit TSS (available) */ + gdt[GDT_ENTRY_TSS].p = 0x1; /* Entry is present */ + gdt[GDT_ENTRY_TSS].dpl = 0x0; /* Privilege level 0 */ + gdt[GDT_ENTRY_TSS].s = 0x0; /* system segment */ + } -/* This routine sets up the initial Guest GDT for booting. All entries start - * as 0 (unusable). */ +/* + * This routine sets up the initial Guest GDT for booting. All entries start + * as 0 (unusable). + */ void setup_guest_gdt(struct lg_cpu *cpu) { - /* Start with full 0-4G segments... */ + /* + * Start with full 0-4G segments...except the Guest is allowed to use + * them, so set the privilege level appropriately in the flags. + */ cpu->arch.gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT; cpu->arch.gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT; - /* ...except the Guest is allowed to use them, so set the privilege - * level appropriately in the flags. */ - cpu->arch.gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13); - cpu->arch.gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13); + cpu->arch.gdt[GDT_ENTRY_KERNEL_CS].dpl |= GUEST_PL; + cpu->arch.gdt[GDT_ENTRY_KERNEL_DS].dpl |= GUEST_PL; } -/*H:650 An optimization of copy_gdt(), for just the three "thead-local storage" - * entries. */ +/*H:650 + * An optimization of copy_gdt(), for just the three "thead-local storage" + * entries. + */ void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt) { unsigned int i; @@ -130,41 +157,56 @@ void copy_gdt_tls(const struct lg_cpu *cpu, struct desc_struct *gdt) gdt[i] = cpu->arch.gdt[i]; } -/*H:640 When the Guest is run on a different CPU, or the GDT entries have - * changed, copy_gdt() is called to copy the Guest's GDT entries across to this - * CPU's GDT. */ +/*H:640 + * When the Guest is run on a different CPU, or the GDT entries have changed, + * copy_gdt() is called to copy the Guest's GDT entries across to this CPU's + * GDT. + */ void copy_gdt(const struct lg_cpu *cpu, struct desc_struct *gdt) { unsigned int i; - /* The default entries from setup_default_gdt_entries() are not - * replaced. See ignored_gdt() above. */ + /* + * The default entries from setup_default_gdt_entries() are not + * replaced. See ignored_gdt() above. + */ for (i = 0; i < GDT_ENTRIES; i++) if (!ignored_gdt(i)) gdt[i] = cpu->arch.gdt[i]; } -/*H:620 This is where the Guest asks us to load a new GDT (LHCALL_LOAD_GDT). - * We copy it from the Guest and tweak the entries. */ -void load_guest_gdt(struct lg_cpu *cpu, unsigned long table, u32 num) +/*H:620 + * This is where the Guest asks us to load a new GDT entry + * (LHCALL_LOAD_GDT_ENTRY). We tweak the entry and copy it in. + */ +void load_guest_gdt_entry(struct lg_cpu *cpu, u32 num, u32 lo, u32 hi) { - /* We assume the Guest has the same number of GDT entries as the - * Host, otherwise we'd have to dynamically allocate the Guest GDT. */ - if (num > ARRAY_SIZE(cpu->arch.gdt)) + /* + * We assume the Guest has the same number of GDT entries as the + * Host, otherwise we'd have to dynamically allocate the Guest GDT. + */ + if (num >= ARRAY_SIZE(cpu->arch.gdt)) { kill_guest(cpu, "too many gdt entries %i", num); + return; + } - /* We read the whole thing in, then fix it up. */ - __lgread(cpu, cpu->arch.gdt, table, num * sizeof(cpu->arch.gdt[0])); - fixup_gdt_table(cpu, 0, ARRAY_SIZE(cpu->arch.gdt)); - /* Mark that the GDT changed so the core knows it has to copy it again, - * even if the Guest is run on the same CPU. */ + /* Set it up, then fix it. */ + cpu->arch.gdt[num].a = lo; + cpu->arch.gdt[num].b = hi; + fixup_gdt_table(cpu, num, num+1); + /* + * Mark that the GDT changed so the core knows it has to copy it again, + * even if the Guest is run on the same CPU. + */ cpu->changed |= CHANGED_GDT; } -/* This is the fast-track version for just changing the three TLS entries. +/* + * This is the fast-track version for just changing the three TLS entries. * Remember that this happens on every context switch, so it's worth * optimizing. But wouldn't it be neater to have a single hypercall to cover - * both cases? */ + * both cases? + */ void guest_load_tls(struct lg_cpu *cpu, unsigned long gtls) { struct desc_struct *tls = &cpu->arch.gdt[GDT_ENTRY_TLS_MIN]; @@ -174,7 +216,6 @@ void guest_load_tls(struct lg_cpu *cpu, unsigned long gtls) /* Note that just the TLS entries have changed. */ cpu->changed |= CHANGED_GDT_TLS; } -/*:*/ /*H:660 * With this, we have finished the Host. diff --git a/drivers/lguest/x86/core.c b/drivers/lguest/x86/core.c index 5126d5d9ea0..922a1acbf65 100644 --- a/drivers/lguest/x86/core.c +++ b/drivers/lguest/x86/core.c @@ -17,13 +17,15 @@ * along with this program; if not, write to the Free Software * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. */ -/*P:450 This file contains the x86-specific lguest code. It used to be all +/*P:450 + * This file contains the x86-specific lguest code. It used to be all * mixed in with drivers/lguest/core.c but several foolhardy code slashers * wrestled most of the dependencies out to here in preparation for porting * lguest to other architectures (see what I mean by foolhardy?). * * This also contains a couple of non-obvious setup and teardown pieces which - * were implemented after days of debugging pain. :*/ + * were implemented after days of debugging pain. +:*/ #include <linux/kernel.h> #include <linux/start_kernel.h> #include <linux/string.h> @@ -57,17 +59,16 @@ static struct { /* Offset from where switcher.S was compiled to where we've copied it */ static unsigned long switcher_offset(void) { - return SWITCHER_ADDR - (unsigned long)start_switcher_text; + return switcher_addr - (unsigned long)start_switcher_text; } -/* This cpu's struct lguest_pages. */ +/* This cpu's struct lguest_pages (after the Switcher text page) */ static struct lguest_pages *lguest_pages(unsigned int cpu) { - return &(((struct lguest_pages *) - (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]); + return &(((struct lguest_pages *)(switcher_addr + PAGE_SIZE))[cpu]); } -static DEFINE_PER_CPU(struct lg_cpu *, last_cpu); +static DEFINE_PER_CPU(struct lg_cpu *, lg_last_cpu); /*S:010 * We approach the Switcher. @@ -82,25 +83,33 @@ static DEFINE_PER_CPU(struct lg_cpu *, last_cpu); */ static void copy_in_guest_info(struct lg_cpu *cpu, struct lguest_pages *pages) { - /* Copying all this data can be quite expensive. We usually run the + /* + * Copying all this data can be quite expensive. We usually run the * same Guest we ran last time (and that Guest hasn't run anywhere else * meanwhile). If that's not the case, we pretend everything in the - * Guest has changed. */ - if (__get_cpu_var(last_cpu) != cpu || cpu->last_pages != pages) { - __get_cpu_var(last_cpu) = cpu; + * Guest has changed. + */ + if (__this_cpu_read(lg_last_cpu) != cpu || cpu->last_pages != pages) { + __this_cpu_write(lg_last_cpu, cpu); cpu->last_pages = pages; cpu->changed = CHANGED_ALL; } - /* These copies are pretty cheap, so we do them unconditionally: */ - /* Save the current Host top-level page directory. */ + /* + * These copies are pretty cheap, so we do them unconditionally: */ + /* Save the current Host top-level page directory. + */ pages->state.host_cr3 = __pa(current->mm->pgd); - /* Set up the Guest's page tables to see this CPU's pages (and no - * other CPU's pages). */ + /* + * Set up the Guest's page tables to see this CPU's pages (and no + * other CPU's pages). + */ map_switcher_in_guest(cpu, pages); - /* Set up the two "TSS" members which tell the CPU what stack to use + /* + * Set up the two "TSS" members which tell the CPU what stack to use * for traps which do directly into the Guest (ie. traps at privilege - * level 1). */ + * level 1). + */ pages->state.guest_tss.sp1 = cpu->esp1; pages->state.guest_tss.ss1 = cpu->ss1; @@ -125,94 +134,132 @@ static void run_guest_once(struct lg_cpu *cpu, struct lguest_pages *pages) /* This is a dummy value we need for GCC's sake. */ unsigned int clobber; - /* Copy the guest-specific information into this CPU's "struct - * lguest_pages". */ + /* + * Copy the guest-specific information into this CPU's "struct + * lguest_pages". + */ copy_in_guest_info(cpu, pages); - /* Set the trap number to 256 (impossible value). If we fault while + /* + * Set the trap number to 256 (impossible value). If we fault while * switching to the Guest (bad segment registers or bug), this will - * cause us to abort the Guest. */ + * cause us to abort the Guest. + */ cpu->regs->trapnum = 256; - /* Now: we push the "eflags" register on the stack, then do an "lcall". + /* + * Now: we push the "eflags" register on the stack, then do an "lcall". * This is how we change from using the kernel code segment to using * the dedicated lguest code segment, as well as jumping into the * Switcher. * * The lcall also pushes the old code segment (KERNEL_CS) onto the * stack, then the address of this call. This stack layout happens to - * exactly match the stack layout created by an interrupt... */ - asm volatile("pushf; lcall *lguest_entry" - /* This is how we tell GCC that %eax ("a") and %ebx ("b") - * are changed by this routine. The "=" means output. */ + * exactly match the stack layout created by an interrupt... + */ + asm volatile("pushf; lcall *%4" + /* + * This is how we tell GCC that %eax ("a") and %ebx ("b") + * are changed by this routine. The "=" means output. + */ : "=a"(clobber), "=b"(clobber) - /* %eax contains the pages pointer. ("0" refers to the + /* + * %eax contains the pages pointer. ("0" refers to the * 0-th argument above, ie "a"). %ebx contains the * physical address of the Guest's top-level page - * directory. */ - : "0"(pages), "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir)) - /* We tell gcc that all these registers could change, + * directory. + */ + : "0"(pages), + "1"(__pa(cpu->lg->pgdirs[cpu->cpu_pgd].pgdir)), + "m"(lguest_entry) + /* + * We tell gcc that all these registers could change, * which means we don't have to save and restore them in - * the Switcher. */ + * the Switcher. + */ : "memory", "%edx", "%ecx", "%edi", "%esi"); } /*:*/ -/*M:002 There are hooks in the scheduler which we can register to tell when we +/*M:002 + * There are hooks in the scheduler which we can register to tell when we * get kicked off the CPU (preempt_notifier_register()). This would allow us * to lazily disable SYSENTER which would regain some performance, and should * also simplify copy_in_guest_info(). Note that we'd still need to restore * things when we exit to Launcher userspace, but that's fairly easy. * - * We could also try using this hooks for PGE, but that might be too expensive. + * We could also try using these hooks for PGE, but that might be too expensive. * - * The hooks were designed for KVM, but we can also put them to good use. :*/ + * The hooks were designed for KVM, but we can also put them to good use. +:*/ -/*H:040 This is the i386-specific code to setup and run the Guest. Interrupts - * are disabled: we own the CPU. */ +/*H:040 + * This is the i386-specific code to setup and run the Guest. Interrupts + * are disabled: we own the CPU. + */ void lguest_arch_run_guest(struct lg_cpu *cpu) { - /* Remember the awfully-named TS bit? If the Guest has asked to set it + /* + * Remember the awfully-named TS bit? If the Guest has asked to set it * we set it now, so we can trap and pass that trap to the Guest if it - * uses the FPU. */ - if (cpu->ts) - lguest_set_ts(); + * uses the FPU. + */ + if (cpu->ts && user_has_fpu()) + stts(); - /* SYSENTER is an optimized way of doing system calls. We can't allow + /* + * SYSENTER is an optimized way of doing system calls. We can't allow * it because it always jumps to privilege level 0. A normal Guest * won't try it because we don't advertise it in CPUID, but a malicious * Guest (or malicious Guest userspace program) could, so we tell the - * CPU to disable it before running the Guest. */ + * CPU to disable it before running the Guest. + */ if (boot_cpu_has(X86_FEATURE_SEP)) wrmsr(MSR_IA32_SYSENTER_CS, 0, 0); - /* Now we actually run the Guest. It will return when something + /* + * Now we actually run the Guest. It will return when something * interesting happens, and we can examine its registers to see what it - * was doing. */ + * was doing. + */ run_guest_once(cpu, lguest_pages(raw_smp_processor_id())); - /* Note that the "regs" structure contains two extra entries which are + /* + * Note that the "regs" structure contains two extra entries which are * not really registers: a trap number which says what interrupt or * trap made the switcher code come back, and an error code which some - * traps set. */ + * traps set. + */ + + /* Restore SYSENTER if it's supposed to be on. */ + if (boot_cpu_has(X86_FEATURE_SEP)) + wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0); + + /* Clear the host TS bit if it was set above. */ + if (cpu->ts && user_has_fpu()) + clts(); - /* If the Guest page faulted, then the cr2 register will tell us the + /* + * If the Guest page faulted, then the cr2 register will tell us the * bad virtual address. We have to grab this now, because once we * re-enable interrupts an interrupt could fault and thus overwrite - * cr2, or we could even move off to a different CPU. */ + * cr2, or we could even move off to a different CPU. + */ if (cpu->regs->trapnum == 14) cpu->arch.last_pagefault = read_cr2(); - /* Similarly, if we took a trap because the Guest used the FPU, - * we have to restore the FPU it expects to see. */ - else if (cpu->regs->trapnum == 7) + /* + * Similarly, if we took a trap because the Guest used the FPU, + * we have to restore the FPU it expects to see. + * math_state_restore() may sleep and we may even move off to + * a different CPU. So all the critical stuff should be done + * before this. + */ + else if (cpu->regs->trapnum == 7 && !user_has_fpu()) math_state_restore(); - - /* Restore SYSENTER if it's supposed to be on. */ - if (boot_cpu_has(X86_FEATURE_SEP)) - wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0); } -/*H:130 Now we've examined the hypercall code; our Guest can make requests. +/*H:130 + * Now we've examined the hypercall code; our Guest can make requests. * Our Guest is usually so well behaved; it never tries to do things it isn't * allowed to, and uses hypercalls instead. Unfortunately, Linux's paravirtual * infrastructure isn't quite complete, because it doesn't contain replacements @@ -222,35 +269,55 @@ void lguest_arch_run_guest(struct lg_cpu *cpu) * * When the Guest uses one of these instructions, we get a trap (General * Protection Fault) and come here. We see if it's one of those troublesome - * instructions and skip over it. We return true if we did. */ + * instructions and skip over it. We return true if we did. + */ static int emulate_insn(struct lg_cpu *cpu) { u8 insn; - unsigned int insnlen = 0, in = 0, shift = 0; - /* The eip contains the *virtual* address of the Guest's instruction: - * guest_pa just subtracts the Guest's page_offset. */ + unsigned int insnlen = 0, in = 0, small_operand = 0; + /* + * The eip contains the *virtual* address of the Guest's instruction: + * walk the Guest's page tables to find the "physical" address. + */ unsigned long physaddr = guest_pa(cpu, cpu->regs->eip); - /* This must be the Guest kernel trying to do something, not userspace! + /* + * This must be the Guest kernel trying to do something, not userspace! * The bottom two bits of the CS segment register are the privilege - * level. */ + * level. + */ if ((cpu->regs->cs & 3) != GUEST_PL) return 0; /* Decoding x86 instructions is icky. */ insn = lgread(cpu, physaddr, u8); - /* 0x66 is an "operand prefix". It means it's using the upper 16 bits - of the eax register. */ + /* + * Around 2.6.33, the kernel started using an emulation for the + * cmpxchg8b instruction in early boot on many configurations. This + * code isn't paravirtualized, and it tries to disable interrupts. + * Ignore it, which will Mostly Work. + */ + if (insn == 0xfa) { + /* "cli", or Clear Interrupt Enable instruction. Skip it. */ + cpu->regs->eip++; + return 1; + } + + /* + * 0x66 is an "operand prefix". It means a 16, not 32 bit in/out. + */ if (insn == 0x66) { - shift = 16; + small_operand = 1; /* The instruction is 1 byte so far, read the next byte. */ insnlen = 1; insn = lgread(cpu, physaddr + insnlen, u8); } - /* We can ignore the lower bit for the moment and decode the 4 opcodes - * we need to emulate. */ + /* + * We can ignore the lower bit for the moment and decode the 4 opcodes + * we need to emulate. + */ switch (insn & 0xFE) { case 0xE4: /* in <next byte>,%al */ insnlen += 2; @@ -271,15 +338,20 @@ static int emulate_insn(struct lg_cpu *cpu) return 0; } - /* If it was an "IN" instruction, they expect the result to be read + /* + * If it was an "IN" instruction, they expect the result to be read * into %eax, so we change %eax. We always return all-ones, which - * traditionally means "there's nothing there". */ + * traditionally means "there's nothing there". + */ if (in) { - /* Lower bit tells is whether it's a 16 or 32 bit access */ - if (insn & 0x1) - cpu->regs->eax = 0xFFFFFFFF; - else - cpu->regs->eax |= (0xFFFF << shift); + /* Lower bit tells means it's a 32/16 bit access */ + if (insn & 0x1) { + if (small_operand) + cpu->regs->eax |= 0xFFFF; + else + cpu->regs->eax = 0xFFFFFFFF; + } else + cpu->regs->eax |= 0xFF; } /* Finally, we've "done" the instruction, so move past it. */ cpu->regs->eip += insnlen; @@ -292,76 +364,91 @@ void lguest_arch_handle_trap(struct lg_cpu *cpu) { switch (cpu->regs->trapnum) { case 13: /* We've intercepted a General Protection Fault. */ - /* Check if this was one of those annoying IN or OUT + /* + * Check if this was one of those annoying IN or OUT * instructions which we need to emulate. If so, we just go - * back into the Guest after we've done it. */ + * back into the Guest after we've done it. + */ if (cpu->regs->errcode == 0) { if (emulate_insn(cpu)) return; } break; case 14: /* We've intercepted a Page Fault. */ - /* The Guest accessed a virtual address that wasn't mapped. + /* + * The Guest accessed a virtual address that wasn't mapped. * This happens a lot: we don't actually set up most of the page * tables for the Guest at all when we start: as it runs it asks * for more and more, and we set them up as required. In this * case, we don't even tell the Guest that the fault happened. * * The errcode tells whether this was a read or a write, and - * whether kernel or userspace code. */ + * whether kernel or userspace code. + */ if (demand_page(cpu, cpu->arch.last_pagefault, cpu->regs->errcode)) return; - /* OK, it's really not there (or not OK): the Guest needs to + /* + * OK, it's really not there (or not OK): the Guest needs to * know. We write out the cr2 value so it knows where the * fault occurred. * * Note that if the Guest were really messed up, this could * happen before it's done the LHCALL_LGUEST_INIT hypercall, so - * lg->lguest_data could be NULL */ + * lg->lguest_data could be NULL + */ if (cpu->lg->lguest_data && put_user(cpu->arch.last_pagefault, &cpu->lg->lguest_data->cr2)) kill_guest(cpu, "Writing cr2"); break; case 7: /* We've intercepted a Device Not Available fault. */ - /* If the Guest doesn't want to know, we already restored the - * Floating Point Unit, so we just continue without telling - * it. */ + /* + * If the Guest doesn't want to know, we already restored the + * Floating Point Unit, so we just continue without telling it. + */ if (!cpu->ts) return; break; case 32 ... 255: - /* These values mean a real interrupt occurred, in which case - * the Host handler has already been run. We just do a + /* + * These values mean a real interrupt occurred, in which case + * the Host handler has already been run. We just do a * friendly check if another process should now be run, then - * return to run the Guest again */ + * return to run the Guest again. + */ cond_resched(); return; case LGUEST_TRAP_ENTRY: - /* Our 'struct hcall_args' maps directly over our regs: we set - * up the pointer now to indicate a hypercall is pending. */ + /* + * Our 'struct hcall_args' maps directly over our regs: we set + * up the pointer now to indicate a hypercall is pending. + */ cpu->hcall = (struct hcall_args *)cpu->regs; return; } /* We didn't handle the trap, so it needs to go to the Guest. */ if (!deliver_trap(cpu, cpu->regs->trapnum)) - /* If the Guest doesn't have a handler (either it hasn't + /* + * If the Guest doesn't have a handler (either it hasn't * registered any yet, or it's one of the faults we don't let - * it handle), it dies with this cryptic error message. */ + * it handle), it dies with this cryptic error message. + */ kill_guest(cpu, "unhandled trap %li at %#lx (%#lx)", cpu->regs->trapnum, cpu->regs->eip, cpu->regs->trapnum == 14 ? cpu->arch.last_pagefault : cpu->regs->errcode); } -/* Now we can look at each of the routines this calls, in increasing order of +/* + * Now we can look at each of the routines this calls, in increasing order of * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(), * deliver_trap() and demand_page(). After all those, we'll be ready to * examine the Switcher, and our philosophical understanding of the Host/Guest - * duality will be complete. :*/ + * duality will be complete. +:*/ static void adjust_pge(void *on) { if (on) @@ -370,13 +457,16 @@ static void adjust_pge(void *on) write_cr4(read_cr4() & ~X86_CR4_PGE); } -/*H:020 Now the Switcher is mapped and every thing else is ready, we need to do - * some more i386-specific initialization. */ +/*H:020 + * Now the Switcher is mapped and every thing else is ready, we need to do + * some more i386-specific initialization. + */ void __init lguest_arch_host_init(void) { int i; - /* Most of the i386/switcher.S doesn't care that it's been moved; on + /* + * Most of the x86/switcher_32.S doesn't care that it's been moved; on * Intel, jumps are relative, and it doesn't access any references to * external code or data. * @@ -384,7 +474,8 @@ void __init lguest_arch_host_init(void) * addresses are placed in a table (default_idt_entries), so we need to * update the table with the new addresses. switcher_offset() is a * convenience function which returns the distance between the - * compiled-in switcher code and the high-mapped copy we just made. */ + * compiled-in switcher code and the high-mapped copy we just made. + */ for (i = 0; i < IDT_ENTRIES; i++) default_idt_entries[i] += switcher_offset(); @@ -399,63 +490,81 @@ void __init lguest_arch_host_init(void) for_each_possible_cpu(i) { /* lguest_pages() returns this CPU's two pages. */ struct lguest_pages *pages = lguest_pages(i); - /* This is a convenience pointer to make the code fit one - * statement to a line. */ + /* This is a convenience pointer to make the code neater. */ struct lguest_ro_state *state = &pages->state; - /* The Global Descriptor Table: the Host has a different one + /* + * The Global Descriptor Table: the Host has a different one * for each CPU. We keep a descriptor for the GDT which says * where it is and how big it is (the size is actually the last - * byte, not the size, hence the "-1"). */ + * byte, not the size, hence the "-1"). + */ state->host_gdt_desc.size = GDT_SIZE-1; state->host_gdt_desc.address = (long)get_cpu_gdt_table(i); - /* All CPUs on the Host use the same Interrupt Descriptor + /* + * All CPUs on the Host use the same Interrupt Descriptor * Table, so we just use store_idt(), which gets this CPU's IDT - * descriptor. */ + * descriptor. + */ store_idt(&state->host_idt_desc); - /* The descriptors for the Guest's GDT and IDT can be filled + /* + * The descriptors for the Guest's GDT and IDT can be filled * out now, too. We copy the GDT & IDT into ->guest_gdt and - * ->guest_idt before actually running the Guest. */ + * ->guest_idt before actually running the Guest. + */ state->guest_idt_desc.size = sizeof(state->guest_idt)-1; state->guest_idt_desc.address = (long)&state->guest_idt; state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1; state->guest_gdt_desc.address = (long)&state->guest_gdt; - /* We know where we want the stack to be when the Guest enters + /* + * We know where we want the stack to be when the Guest enters * the Switcher: in pages->regs. The stack grows upwards, so - * we start it at the end of that structure. */ + * we start it at the end of that structure. + */ state->guest_tss.sp0 = (long)(&pages->regs + 1); - /* And this is the GDT entry to use for the stack: we keep a - * couple of special LGUEST entries. */ + /* + * And this is the GDT entry to use for the stack: we keep a + * couple of special LGUEST entries. + */ state->guest_tss.ss0 = LGUEST_DS; - /* x86 can have a finegrained bitmap which indicates what I/O + /* + * x86 can have a finegrained bitmap which indicates what I/O * ports the process can use. We set it to the end of our - * structure, meaning "none". */ + * structure, meaning "none". + */ state->guest_tss.io_bitmap_base = sizeof(state->guest_tss); - /* Some GDT entries are the same across all Guests, so we can - * set them up now. */ + /* + * Some GDT entries are the same across all Guests, so we can + * set them up now. + */ setup_default_gdt_entries(state); /* Most IDT entries are the same for all Guests, too.*/ setup_default_idt_entries(state, default_idt_entries); - /* The Host needs to be able to use the LGUEST segments on this - * CPU, too, so put them in the Host GDT. */ + /* + * The Host needs to be able to use the LGUEST segments on this + * CPU, too, so put them in the Host GDT. + */ get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT; get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT; } - /* In the Switcher, we want the %cs segment register to use the + /* + * In the Switcher, we want the %cs segment register to use the * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so * it will be undisturbed when we switch. To change %cs and jump we - * need this structure to feed to Intel's "lcall" instruction. */ + * need this structure to feed to Intel's "lcall" instruction. + */ lguest_entry.offset = (long)switch_to_guest + switcher_offset(); lguest_entry.segment = LGUEST_CS; - /* Finally, we need to turn off "Page Global Enable". PGE is an + /* + * Finally, we need to turn off "Page Global Enable". PGE is an * optimization where page table entries are specially marked to show * they never change. The Host kernel marks all the kernel pages this * way because it's always present, even when userspace is running. @@ -465,22 +574,27 @@ void __init lguest_arch_host_init(void) * you'll get really weird bugs that you'll chase for two days. * * I used to turn PGE off every time we switched to the Guest and back - * on when we return, but that slowed the Switcher down noticibly. */ + * on when we return, but that slowed the Switcher down noticibly. + */ - /* We don't need the complexity of CPUs coming and going while we're - * doing this. */ + /* + * We don't need the complexity of CPUs coming and going while we're + * doing this. + */ get_online_cpus(); if (cpu_has_pge) { /* We have a broader idea of "global". */ /* Remember that this was originally set (for cleanup). */ cpu_had_pge = 1; - /* adjust_pge is a helper function which sets or unsets the PGE - * bit on its CPU, depending on the argument (0 == unset). */ - on_each_cpu(adjust_pge, (void *)0, 0, 1); + /* + * adjust_pge is a helper function which sets or unsets the PGE + * bit on its CPU, depending on the argument (0 == unset). + */ + on_each_cpu(adjust_pge, (void *)0, 1); /* Turn off the feature in the global feature set. */ - clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability); + clear_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE); } put_online_cpus(); -}; +} /*:*/ void __exit lguest_arch_host_fini(void) @@ -488,9 +602,9 @@ void __exit lguest_arch_host_fini(void) /* If we had PGE before we started, turn it back on now. */ get_online_cpus(); if (cpu_had_pge) { - set_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability); + set_cpu_cap(&boot_cpu_data, X86_FEATURE_PGE); /* adjust_pge's argument "1" means set PGE. */ - on_each_cpu(adjust_pge, (void *)1, 0, 1); + on_each_cpu(adjust_pge, (void *)1, 1); } put_online_cpus(); } @@ -500,8 +614,8 @@ void __exit lguest_arch_host_fini(void) int lguest_arch_do_hcall(struct lg_cpu *cpu, struct hcall_args *args) { switch (args->arg0) { - case LHCALL_LOAD_GDT: - load_guest_gdt(cpu, args->arg1, args->arg2); + case LHCALL_LOAD_GDT_ENTRY: + load_guest_gdt_entry(cpu, args->arg1, args->arg2, args->arg3); break; case LHCALL_LOAD_IDT_ENTRY: load_guest_idt_entry(cpu, args->arg1, args->arg2, args->arg3); @@ -521,26 +635,32 @@ int lguest_arch_init_hypercalls(struct lg_cpu *cpu) { u32 tsc_speed; - /* The pointer to the Guest's "struct lguest_data" is the only argument. - * We check that address now. */ + /* + * The pointer to the Guest's "struct lguest_data" is the only argument. + * We check that address now. + */ if (!lguest_address_ok(cpu->lg, cpu->hcall->arg1, sizeof(*cpu->lg->lguest_data))) return -EFAULT; - /* Having checked it, we simply set lg->lguest_data to point straight + /* + * Having checked it, we simply set lg->lguest_data to point straight * into the Launcher's memory at the right place and then use * copy_to_user/from_user from now on, instead of lgread/write. I put * this in to show that I'm not immune to writing stupid - * optimizations. */ + * optimizations. + */ cpu->lg->lguest_data = cpu->lg->mem_base + cpu->hcall->arg1; - /* We insist that the Time Stamp Counter exist and doesn't change with + /* + * We insist that the Time Stamp Counter exist and doesn't change with * cpu frequency. Some devious chip manufacturers decided that TSC * changes could be handled in software. I decided that time going * backwards might be good for benchmarks, but it's bad for users. * * We also insist that the TSC be stable: the kernel detects unreliable - * TSCs for its own purposes, and we use that here. */ + * TSCs for its own purposes, and we use that here. + */ if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable()) tsc_speed = tsc_khz; else @@ -556,38 +676,45 @@ int lguest_arch_init_hypercalls(struct lg_cpu *cpu) } /*:*/ -/*L:030 lguest_arch_setup_regs() - * +/*L:030 * Most of the Guest's registers are left alone: we used get_zeroed_page() to - * allocate the structure, so they will be 0. */ + * allocate the structure, so they will be 0. + */ void lguest_arch_setup_regs(struct lg_cpu *cpu, unsigned long start) { struct lguest_regs *regs = cpu->regs; - /* There are four "segment" registers which the Guest needs to boot: + /* + * There are four "segment" registers which the Guest needs to boot: * The "code segment" register (cs) refers to the kernel code segment * __KERNEL_CS, and the "data", "extra" and "stack" segment registers * refer to the kernel data segment __KERNEL_DS. * * The privilege level is packed into the lower bits. The Guest runs - * at privilege level 1 (GUEST_PL).*/ + * at privilege level 1 (GUEST_PL). + */ regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL; regs->cs = __KERNEL_CS|GUEST_PL; - /* The "eflags" register contains miscellaneous flags. Bit 1 (0x002) + /* + * The "eflags" register contains miscellaneous flags. Bit 1 (0x002) * is supposed to always be "1". Bit 9 (0x200) controls whether * interrupts are enabled. We always leave interrupts enabled while - * running the Guest. */ - regs->eflags = X86_EFLAGS_IF | 0x2; + * running the Guest. + */ + regs->eflags = X86_EFLAGS_IF | X86_EFLAGS_FIXED; - /* The "Extended Instruction Pointer" register says where the Guest is - * running. */ + /* + * The "Extended Instruction Pointer" register says where the Guest is + * running. + */ regs->eip = start; - /* %esi points to our boot information, at physical address 0, so don't - * touch it. */ + /* + * %esi points to our boot information, at physical address 0, so don't + * touch it. + */ - /* There are a couple of GDT entries the Guest expects when first - * booting. */ + /* There are a couple of GDT entries the Guest expects at boot. */ setup_guest_gdt(cpu); } diff --git a/drivers/lguest/x86/switcher_32.S b/drivers/lguest/x86/switcher_32.S index 3fc15318a80..40634b0db9f 100644 --- a/drivers/lguest/x86/switcher_32.S +++ b/drivers/lguest/x86/switcher_32.S @@ -1,12 +1,15 @@ -/*P:900 This is the Switcher: code which sits at 0xFFC00000 astride both the - * Host and Guest to do the low-level Guest<->Host switch. It is as simple as - * it can be made, but it's naturally very specific to x86. +/*P:900 + * This is the Switcher: code which sits at 0xFFC00000 (or 0xFFE00000) astride + * both the Host and Guest to do the low-level Guest<->Host switch. It is as + * simple as it can be made, but it's naturally very specific to x86. * * You have now completed Preparation. If this has whet your appetite; if you * are feeling invigorated and refreshed then the next, more challenging stage - * can be found in "make Guest". :*/ + * can be found in "make Guest". + :*/ -/*M:012 Lguest is meant to be simple: my rule of thumb is that 1% more LOC must +/*M:012 + * Lguest is meant to be simple: my rule of thumb is that 1% more LOC must * gain at least 1% more performance. Since neither LOC nor performance can be * measured beforehand, it generally means implementing a feature then deciding * if it's worth it. And once it's implemented, who can say no? @@ -31,11 +34,14 @@ * Host (which is actually really easy). * * Two questions remain. Would the performance gain outweigh the complexity? - * And who would write the verse documenting it? :*/ + * And who would write the verse documenting it? +:*/ -/*M:011 Lguest64 handles NMI. This gave me NMI envy (until I looked at their +/*M:011 + * Lguest64 handles NMI. This gave me NMI envy (until I looked at their * code). It's worth doing though, since it would let us use oprofile in the - * Host when a Guest is running. :*/ + * Host when a Guest is running. +:*/ /*S:100 * Welcome to the Switcher itself! |