diff options
Diffstat (limited to 'arch/arm/vfp')
| -rw-r--r-- | arch/arm/vfp/Makefile | 6 | ||||
| -rw-r--r-- | arch/arm/vfp/entry.S | 44 | ||||
| -rw-r--r-- | arch/arm/vfp/vfpdouble.c | 2 | ||||
| -rw-r--r-- | arch/arm/vfp/vfphw.S | 143 | ||||
| -rw-r--r-- | arch/arm/vfp/vfpmodule.c | 416 | ||||
| -rw-r--r-- | arch/arm/vfp/vfpsingle.c | 2 |
6 files changed, 426 insertions, 187 deletions
diff --git a/arch/arm/vfp/Makefile b/arch/arm/vfp/Makefile index 39f6d8e1af7..a81404c09d5 100644 --- a/arch/arm/vfp/Makefile +++ b/arch/arm/vfp/Makefile @@ -4,10 +4,10 @@ # Copyright (C) 2001 ARM Limited # -# EXTRA_CFLAGS := -DDEBUG -# EXTRA_AFLAGS := -DDEBUG +# ccflags-y := -DDEBUG +# asflags-y := -DDEBUG -KBUILD_AFLAGS :=$(KBUILD_AFLAGS:-msoft-float=-Wa,-mfpu=softvfp+vfp) +KBUILD_AFLAGS :=$(KBUILD_AFLAGS:-msoft-float=-Wa,-mfpu=softvfp+vfp -mfloat-abi=soft) LDFLAGS +=--no-warn-mismatch obj-y += vfp.o diff --git a/arch/arm/vfp/entry.S b/arch/arm/vfp/entry.S index 4fa9903b83c..fe6ca574d09 100644 --- a/arch/arm/vfp/entry.S +++ b/arch/arm/vfp/entry.S @@ -7,25 +7,25 @@ * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 as * published by the Free Software Foundation. - * - * Basic entry code, called from the kernel's undefined instruction trap. - * r0 = faulted instruction - * r5 = faulted PC+4 - * r9 = successful return - * r10 = thread_info structure - * lr = failure return */ +#include <linux/init.h> +#include <linux/linkage.h> #include <asm/thread_info.h> #include <asm/vfpmacros.h> -#include "../kernel/entry-header.S" +#include <asm/assembler.h> +#include <asm/asm-offsets.h> +@ VFP entry point. +@ +@ r0 = instruction opcode (32-bit ARM or two 16-bit Thumb) +@ r2 = PC value to resume execution after successful emulation +@ r9 = normal "successful" return address +@ r10 = this threads thread_info structure +@ lr = unrecognised instruction return address +@ IRQs enabled. +@ ENTRY(do_vfp) -#ifdef CONFIG_PREEMPT - ldr r4, [r10, #TI_PREEMPT] @ get preempt count - add r11, r4, #1 @ increment it - str r11, [r10, #TI_PREEMPT] -#endif - enable_irq + inc_preempt_count r10, r4 ldr r4, .LCvfp ldr r11, [r10, #TI_CPU] @ CPU number add r10, r10, #TI_VFPSTATE @ r10 = workspace @@ -33,12 +33,7 @@ ENTRY(do_vfp) ENDPROC(do_vfp) ENTRY(vfp_null_entry) -#ifdef CONFIG_PREEMPT - get_thread_info r10 - ldr r4, [r10, #TI_PREEMPT] @ get preempt count - sub r11, r4, #1 @ decrement it - str r11, [r10, #TI_PREEMPT] -#endif + dec_preempt_count_ti r10, r4 mov pc, lr ENDPROC(vfp_null_entry) @@ -51,14 +46,9 @@ ENDPROC(vfp_null_entry) __INIT ENTRY(vfp_testing_entry) -#ifdef CONFIG_PREEMPT - get_thread_info r10 - ldr r4, [r10, #TI_PREEMPT] @ get preempt count - sub r11, r4, #1 @ decrement it - str r11, [r10, #TI_PREEMPT] -#endif + dec_preempt_count_ti r10, r4 ldr r0, VFP_arch_address - str r5, [r0] @ known non-zero value + str r0, [r0] @ set to non-zero value mov pc, r9 @ we have handled the fault ENDPROC(vfp_testing_entry) diff --git a/arch/arm/vfp/vfpdouble.c b/arch/arm/vfp/vfpdouble.c index 6cac43bd1d8..423f56dd402 100644 --- a/arch/arm/vfp/vfpdouble.c +++ b/arch/arm/vfp/vfpdouble.c @@ -866,6 +866,8 @@ vfp_double_multiply_accumulate(int dd, int dn, int dm, u32 fpscr, u32 negate, ch vdp.sign = vfp_sign_negate(vdp.sign); vfp_double_unpack(&vdn, vfp_get_double(dd)); + if (vdn.exponent == 0 && vdn.significand) + vfp_double_normalise_denormal(&vdn); if (negate & NEG_SUBTRACT) vdn.sign = vfp_sign_negate(vdn.sign); diff --git a/arch/arm/vfp/vfphw.S b/arch/arm/vfp/vfphw.S index 9897dcfc16d..be807625ed8 100644 --- a/arch/arm/vfp/vfphw.S +++ b/arch/arm/vfp/vfphw.S @@ -14,19 +14,25 @@ * r10 points at the start of the private FP workspace in the thread structure * sp points to a struct pt_regs (as defined in include/asm/proc/ptrace.h) */ +#include <linux/init.h> +#include <linux/linkage.h> #include <asm/thread_info.h> #include <asm/vfpmacros.h> -#include "../kernel/entry-header.S" +#include <linux/kern_levels.h> +#include <asm/assembler.h> +#include <asm/asm-offsets.h> .macro DBGSTR, str #ifdef DEBUG stmfd sp!, {r0-r3, ip, lr} - add r0, pc, #4 + ldr r0, =1f bl printk - b 1f - .asciz "<7>VFP: \str\n" - .balign 4 -1: ldmfd sp!, {r0-r3, ip, lr} + ldmfd sp!, {r0-r3, ip, lr} + + .pushsection .rodata, "a" +1: .ascii KERN_DEBUG "VFP: \str\n" + .byte 0 + .previous #endif .endm @@ -34,12 +40,14 @@ #ifdef DEBUG stmfd sp!, {r0-r3, ip, lr} mov r1, \arg - add r0, pc, #4 + ldr r0, =1f bl printk - b 1f - .asciz "<7>VFP: \str\n" - .balign 4 -1: ldmfd sp!, {r0-r3, ip, lr} + ldmfd sp!, {r0-r3, ip, lr} + + .pushsection .rodata, "a" +1: .ascii KERN_DEBUG "VFP: \str\n" + .byte 0 + .previous #endif .endm @@ -49,55 +57,62 @@ mov r3, \arg3 mov r2, \arg2 mov r1, \arg1 - add r0, pc, #4 + ldr r0, =1f bl printk - b 1f - .asciz "<7>VFP: \str\n" - .balign 4 -1: ldmfd sp!, {r0-r3, ip, lr} + ldmfd sp!, {r0-r3, ip, lr} + + .pushsection .rodata, "a" +1: .ascii KERN_DEBUG "VFP: \str\n" + .byte 0 + .previous #endif .endm @ VFP hardware support entry point. @ -@ r0 = faulted instruction -@ r2 = faulted PC+4 -@ r9 = successful return +@ r0 = instruction opcode (32-bit ARM or two 16-bit Thumb) +@ r2 = PC value to resume execution after successful emulation +@ r9 = normal "successful" return address @ r10 = vfp_state union @ r11 = CPU number -@ lr = failure return - +@ lr = unrecognised instruction return address +@ IRQs enabled. ENTRY(vfp_support_entry) DBGSTR3 "instr %08x pc %08x state %p", r0, r2, r10 + ldr r3, [sp, #S_PSR] @ Neither lazy restore nor FP exceptions + and r3, r3, #MODE_MASK @ are supported in kernel mode + teq r3, #USR_MODE + bne vfp_kmode_exception @ Returns through lr + VFPFMRX r1, FPEXC @ Is the VFP enabled? DBGSTR1 "fpexc %08x", r1 tst r1, #FPEXC_EN bne look_for_VFP_exceptions @ VFP is already enabled DBGSTR1 "enable %x", r10 - ldr r3, last_VFP_context_address + ldr r3, vfp_current_hw_state_address orr r1, r1, #FPEXC_EN @ user FPEXC has the enable bit set - ldr r4, [r3, r11, lsl #2] @ last_VFP_context pointer + ldr r4, [r3, r11, lsl #2] @ vfp_current_hw_state pointer bic r5, r1, #FPEXC_EX @ make sure exceptions are disabled - cmp r4, r10 - beq check_for_exception @ we are returning to the same - @ process, so the registers are - @ still there. In this case, we do - @ not want to drop a pending exception. + cmp r4, r10 @ this thread owns the hw context? +#ifndef CONFIG_SMP + @ For UP, checking that this thread owns the hw context is + @ sufficient to determine that the hardware state is valid. + beq vfp_hw_state_valid + + @ On UP, we lazily save the VFP context. As a different + @ thread wants ownership of the VFP hardware, save the old + @ state if there was a previous (valid) owner. VFPFMXR FPEXC, r5 @ enable VFP, disable any pending @ exceptions, so we can get at the @ rest of it -#ifndef CONFIG_SMP - @ Save out the current registers to the old thread state - @ No need for SMP since this is not done lazily - DBGSTR1 "save old state %p", r4 - cmp r4, #0 - beq no_old_VFP_process + cmp r4, #0 @ if the vfp_current_hw_state is NULL + beq vfp_reload_hw @ then the hw state needs reloading VFPFSTMIA r4, r5 @ save the working registers VFPFMRX r5, FPSCR @ current status #ifndef CONFIG_CPU_FEROCEON @@ -110,13 +125,35 @@ ENTRY(vfp_support_entry) 1: #endif stmia r4, {r1, r5, r6, r8} @ save FPEXC, FPSCR, FPINST, FPINST2 - @ and point r4 at the word at the - @ start of the register dump +vfp_reload_hw: + +#else + @ For SMP, if this thread does not own the hw context, then we + @ need to reload it. No need to save the old state as on SMP, + @ we always save the state when we switch away from a thread. + bne vfp_reload_hw + + @ This thread has ownership of the current hardware context. + @ However, it may have been migrated to another CPU, in which + @ case the saved state is newer than the hardware context. + @ Check this by looking at the CPU number which the state was + @ last loaded onto. + ldr ip, [r10, #VFP_CPU] + teq ip, r11 + beq vfp_hw_state_valid + +vfp_reload_hw: + @ We're loading this threads state into the VFP hardware. Update + @ the CPU number which contains the most up to date VFP context. + str r11, [r10, #VFP_CPU] + + VFPFMXR FPEXC, r5 @ enable VFP, disable any pending + @ exceptions, so we can get at the + @ rest of it #endif -no_old_VFP_process: DBGSTR1 "load state %p", r10 - str r10, [r3, r11, lsl #2] @ update the last_VFP_context pointer + str r10, [r3, r11, lsl #2] @ update the vfp_current_hw_state pointer @ Load the saved state back into the VFP VFPFLDMIA r10, r5 @ reload the working registers while @ FPEXC is in a safe state @@ -132,21 +169,20 @@ no_old_VFP_process: #endif VFPFMXR FPSCR, r5 @ restore status -check_for_exception: +@ The context stored in the VFP hardware is up to date with this thread +vfp_hw_state_valid: tst r1, #FPEXC_EX bne process_exception @ might as well handle the pending @ exception before retrying branch @ out before setting an FPEXC that @ stops us reading stuff - VFPFMXR FPEXC, r1 @ restore FPEXC last - sub r2, r2, #4 - str r2, [sp, #S_PC] @ retry the instruction -#ifdef CONFIG_PREEMPT - get_thread_info r10 - ldr r4, [r10, #TI_PREEMPT] @ get preempt count - sub r11, r4, #1 @ decrement it - str r11, [r10, #TI_PREEMPT] -#endif + VFPFMXR FPEXC, r1 @ Restore FPEXC last + sub r2, r2, #4 @ Retry current instruction - if Thumb + str r2, [sp, #S_PC] @ mode it's two 16-bit instructions, + @ else it's one 32-bit instruction, so + @ always subtract 4 from the following + @ instruction address. + dec_preempt_count_ti r10, r4 mov pc, r9 @ we think we have handled things @@ -165,12 +201,7 @@ look_for_VFP_exceptions: @ not recognised by VFP DBGSTR "not VFP" -#ifdef CONFIG_PREEMPT - get_thread_info r10 - ldr r4, [r10, #TI_PREEMPT] @ get preempt count - sub r11, r4, #1 @ decrement it - str r11, [r10, #TI_PREEMPT] -#endif + dec_preempt_count_ti r10, r4 mov pc, lr process_exception: @@ -207,8 +238,8 @@ ENTRY(vfp_save_state) ENDPROC(vfp_save_state) .align -last_VFP_context_address: - .word last_VFP_context +vfp_current_hw_state_address: + .word vfp_current_hw_state .macro tbl_branch, base, tmp, shift #ifdef CONFIG_THUMB2_KERNEL diff --git a/arch/arm/vfp/vfpmodule.c b/arch/arm/vfp/vfpmodule.c index 0797cb528b4..2f37e1d6cb4 100644 --- a/arch/arm/vfp/vfpmodule.c +++ b/arch/arm/vfp/vfpmodule.c @@ -8,17 +8,23 @@ * it under the terms of the GNU General Public License version 2 as * published by the Free Software Foundation. */ -#include <linux/module.h> #include <linux/types.h> #include <linux/cpu.h> +#include <linux/cpu_pm.h> +#include <linux/hardirq.h> #include <linux/kernel.h> #include <linux/notifier.h> #include <linux/signal.h> #include <linux/sched.h> #include <linux/smp.h> #include <linux/init.h> +#include <linux/uaccess.h> +#include <linux/user.h> +#include <linux/export.h> +#include <asm/cp15.h> #include <asm/cputype.h> +#include <asm/system_info.h> #include <asm/thread_notify.h> #include <asm/vfp.h> @@ -33,7 +39,6 @@ void vfp_support_entry(void); void vfp_null_entry(void); void (*vfp_vector)(void) = vfp_null_entry; -union vfp_state *last_VFP_context[NR_CPUS]; /* * Dual-use variable. @@ -43,6 +48,46 @@ union vfp_state *last_VFP_context[NR_CPUS]; unsigned int VFP_arch; /* + * The pointer to the vfpstate structure of the thread which currently + * owns the context held in the VFP hardware, or NULL if the hardware + * context is invalid. + * + * For UP, this is sufficient to tell which thread owns the VFP context. + * However, for SMP, we also need to check the CPU number stored in the + * saved state too to catch migrations. + */ +union vfp_state *vfp_current_hw_state[NR_CPUS]; + +/* + * Is 'thread's most up to date state stored in this CPUs hardware? + * Must be called from non-preemptible context. + */ +static bool vfp_state_in_hw(unsigned int cpu, struct thread_info *thread) +{ +#ifdef CONFIG_SMP + if (thread->vfpstate.hard.cpu != cpu) + return false; +#endif + return vfp_current_hw_state[cpu] == &thread->vfpstate; +} + +/* + * Force a reload of the VFP context from the thread structure. We do + * this by ensuring that access to the VFP hardware is disabled, and + * clear vfp_current_hw_state. Must be called from non-preemptible context. + */ +static void vfp_force_reload(unsigned int cpu, struct thread_info *thread) +{ + if (vfp_state_in_hw(cpu, thread)) { + fmxr(FPEXC, fmrx(FPEXC) & ~FPEXC_EN); + vfp_current_hw_state[cpu] = NULL; + } +#ifdef CONFIG_SMP + thread->vfpstate.hard.cpu = NR_CPUS; +#endif +} + +/* * Per-thread VFP initialization. */ static void vfp_thread_flush(struct thread_info *thread) @@ -50,21 +95,27 @@ static void vfp_thread_flush(struct thread_info *thread) union vfp_state *vfp = &thread->vfpstate; unsigned int cpu; - memset(vfp, 0, sizeof(union vfp_state)); - - vfp->hard.fpexc = FPEXC_EN; - vfp->hard.fpscr = FPSCR_ROUND_NEAREST; - /* * Disable VFP to ensure we initialize it first. We must ensure - * that the modification of last_VFP_context[] and hardware disable - * are done for the same CPU and without preemption. + * that the modification of vfp_current_hw_state[] and hardware + * disable are done for the same CPU and without preemption. + * + * Do this first to ensure that preemption won't overwrite our + * state saving should access to the VFP be enabled at this point. */ cpu = get_cpu(); - if (last_VFP_context[cpu] == vfp) - last_VFP_context[cpu] = NULL; + if (vfp_current_hw_state[cpu] == vfp) + vfp_current_hw_state[cpu] = NULL; fmxr(FPEXC, fmrx(FPEXC) & ~FPEXC_EN); put_cpu(); + + memset(vfp, 0, sizeof(union vfp_state)); + + vfp->hard.fpexc = FPEXC_EN; + vfp->hard.fpscr = FPSCR_ROUND_NEAREST; +#ifdef CONFIG_SMP + vfp->hard.cpu = NR_CPUS; +#endif } static void vfp_thread_exit(struct thread_info *thread) @@ -73,11 +124,22 @@ static void vfp_thread_exit(struct thread_info *thread) union vfp_state *vfp = &thread->vfpstate; unsigned int cpu = get_cpu(); - if (last_VFP_context[cpu] == vfp) - last_VFP_context[cpu] = NULL; + if (vfp_current_hw_state[cpu] == vfp) + vfp_current_hw_state[cpu] = NULL; put_cpu(); } +static void vfp_thread_copy(struct thread_info *thread) +{ + struct thread_info *parent = current_thread_info(); + + vfp_sync_hwstate(parent); + thread->vfpstate = parent->vfpstate; +#ifdef CONFIG_SMP + thread->vfpstate.hard.cpu = NR_CPUS; +#endif +} + /* * When this function is called with the following 'cmd's, the following * is true while this function is being run: @@ -104,29 +166,25 @@ static void vfp_thread_exit(struct thread_info *thread) static int vfp_notifier(struct notifier_block *self, unsigned long cmd, void *v) { struct thread_info *thread = v; + u32 fpexc; +#ifdef CONFIG_SMP + unsigned int cpu; +#endif - if (likely(cmd == THREAD_NOTIFY_SWITCH)) { - u32 fpexc = fmrx(FPEXC); + switch (cmd) { + case THREAD_NOTIFY_SWITCH: + fpexc = fmrx(FPEXC); #ifdef CONFIG_SMP - unsigned int cpu = thread->cpu; + cpu = thread->cpu; /* * On SMP, if VFP is enabled, save the old state in * case the thread migrates to a different CPU. The * restoring is done lazily. */ - if ((fpexc & FPEXC_EN) && last_VFP_context[cpu]) { - vfp_save_state(last_VFP_context[cpu], fpexc); - last_VFP_context[cpu]->hard.cpu = cpu; - } - /* - * Thread migration, just force the reloading of the - * state on the new CPU in case the VFP registers - * contain stale data. - */ - if (thread->vfpstate.hard.cpu != cpu) - last_VFP_context[cpu] = NULL; + if ((fpexc & FPEXC_EN) && vfp_current_hw_state[cpu]) + vfp_save_state(vfp_current_hw_state[cpu], fpexc); #endif /* @@ -134,13 +192,20 @@ static int vfp_notifier(struct notifier_block *self, unsigned long cmd, void *v) * old state. */ fmxr(FPEXC, fpexc & ~FPEXC_EN); - return NOTIFY_DONE; - } + break; - if (cmd == THREAD_NOTIFY_FLUSH) + case THREAD_NOTIFY_FLUSH: vfp_thread_flush(thread); - else + break; + + case THREAD_NOTIFY_EXIT: vfp_thread_exit(thread); + break; + + case THREAD_NOTIFY_COPY: + vfp_thread_copy(thread); + break; + } return NOTIFY_DONE; } @@ -153,7 +218,7 @@ static struct notifier_block vfp_notifier_block = { * Raise a SIGFPE for the current process. * sicode describes the signal being raised. */ -void vfp_raise_sigfpe(unsigned int sicode, struct pt_regs *regs) +static void vfp_raise_sigfpe(unsigned int sicode, struct pt_regs *regs) { siginfo_t info; @@ -177,11 +242,11 @@ static void vfp_panic(char *reason, u32 inst) { int i; - printk(KERN_ERR "VFP: Error: %s\n", reason); - printk(KERN_ERR "VFP: EXC 0x%08x SCR 0x%08x INST 0x%08x\n", + pr_err("VFP: Error: %s\n", reason); + pr_err("VFP: EXC 0x%08x SCR 0x%08x INST 0x%08x\n", fmrx(FPEXC), fmrx(FPSCR), inst); for (i = 0; i < 32; i += 2) - printk(KERN_ERR "VFP: s%2u: 0x%08x s%2u: 0x%08x\n", + pr_err("VFP: s%2u: 0x%08x s%2u: 0x%08x\n", i, vfp_get_float(i), i+1, vfp_get_float(i+1)); } @@ -349,7 +414,7 @@ void VFP_bounce(u32 trigger, u32 fpexc, struct pt_regs *regs) * If there isn't a second FP instruction, exit now. Note that * the FPEXC.FP2V bit is valid only if FPEXC.EX is 1. */ - if (fpexc ^ (FPEXC_EX | FPEXC_FP2V)) + if ((fpexc & (FPEXC_EX | FPEXC_FP2V)) != (FPEXC_EX | FPEXC_FP2V)) goto exit; /* @@ -369,7 +434,10 @@ void VFP_bounce(u32 trigger, u32 fpexc, struct pt_regs *regs) static void vfp_enable(void *unused) { - u32 access = get_copro_access(); + u32 access; + + BUG_ON(preemptible()); + access = get_copro_access(); /* * Enable full access to VFP (cp10 and cp11) @@ -377,70 +445,79 @@ static void vfp_enable(void *unused) set_copro_access(access | CPACC_FULL(10) | CPACC_FULL(11)); } -#ifdef CONFIG_PM -#include <linux/sysdev.h> - -static int vfp_pm_suspend(struct sys_device *dev, pm_message_t state) +#ifdef CONFIG_CPU_PM +static int vfp_pm_suspend(void) { struct thread_info *ti = current_thread_info(); u32 fpexc = fmrx(FPEXC); /* if vfp is on, then save state for resumption */ if (fpexc & FPEXC_EN) { - printk(KERN_DEBUG "%s: saving vfp state\n", __func__); + pr_debug("%s: saving vfp state\n", __func__); vfp_save_state(&ti->vfpstate, fpexc); /* disable, just in case */ fmxr(FPEXC, fmrx(FPEXC) & ~FPEXC_EN); + } else if (vfp_current_hw_state[ti->cpu]) { +#ifndef CONFIG_SMP + fmxr(FPEXC, fpexc | FPEXC_EN); + vfp_save_state(vfp_current_hw_state[ti->cpu], fpexc); + fmxr(FPEXC, fpexc); +#endif } /* clear any information we had about last context state */ - memset(last_VFP_context, 0, sizeof(last_VFP_context)); + vfp_current_hw_state[ti->cpu] = NULL; return 0; } -static int vfp_pm_resume(struct sys_device *dev) +static void vfp_pm_resume(void) { /* ensure we have access to the vfp */ vfp_enable(NULL); /* and disable it to ensure the next usage restores the state */ fmxr(FPEXC, fmrx(FPEXC) & ~FPEXC_EN); - - return 0; } -static struct sysdev_class vfp_pm_sysclass = { - .name = "vfp", - .suspend = vfp_pm_suspend, - .resume = vfp_pm_resume, -}; +static int vfp_cpu_pm_notifier(struct notifier_block *self, unsigned long cmd, + void *v) +{ + switch (cmd) { + case CPU_PM_ENTER: + vfp_pm_suspend(); + break; + case CPU_PM_ENTER_FAILED: + case CPU_PM_EXIT: + vfp_pm_resume(); + break; + } + return NOTIFY_OK; +} -static struct sys_device vfp_pm_sysdev = { - .cls = &vfp_pm_sysclass, +static struct notifier_block vfp_cpu_pm_notifier_block = { + .notifier_call = vfp_cpu_pm_notifier, }; static void vfp_pm_init(void) { - sysdev_class_register(&vfp_pm_sysclass); - sysdev_register(&vfp_pm_sysdev); + cpu_pm_register_notifier(&vfp_cpu_pm_notifier_block); } - #else static inline void vfp_pm_init(void) { } -#endif /* CONFIG_PM */ +#endif /* CONFIG_CPU_PM */ +/* + * Ensure that the VFP state stored in 'thread->vfpstate' is up to date + * with the hardware state. + */ void vfp_sync_hwstate(struct thread_info *thread) { unsigned int cpu = get_cpu(); - /* - * If the thread we're interested in is the current owner of the - * hardware VFP state, then we need to save its state. - */ - if (last_VFP_context[cpu] == &thread->vfpstate) { + if (vfp_state_in_hw(cpu, thread)) { u32 fpexc = fmrx(FPEXC); /* @@ -454,43 +531,110 @@ void vfp_sync_hwstate(struct thread_info *thread) put_cpu(); } +/* Ensure that the thread reloads the hardware VFP state on the next use. */ void vfp_flush_hwstate(struct thread_info *thread) { unsigned int cpu = get_cpu(); + vfp_force_reload(cpu, thread); + + put_cpu(); +} + +/* + * Save the current VFP state into the provided structures and prepare + * for entry into a new function (signal handler). + */ +int vfp_preserve_user_clear_hwstate(struct user_vfp __user *ufp, + struct user_vfp_exc __user *ufp_exc) +{ + struct thread_info *thread = current_thread_info(); + struct vfp_hard_struct *hwstate = &thread->vfpstate.hard; + int err = 0; + + /* Ensure that the saved hwstate is up-to-date. */ + vfp_sync_hwstate(thread); + /* - * If the thread we're interested in is the current owner of the - * hardware VFP state, then we need to save its state. + * Copy the floating point registers. There can be unused + * registers see asm/hwcap.h for details. */ - if (last_VFP_context[cpu] == &thread->vfpstate) { - u32 fpexc = fmrx(FPEXC); + err |= __copy_to_user(&ufp->fpregs, &hwstate->fpregs, + sizeof(hwstate->fpregs)); + /* + * Copy the status and control register. + */ + __put_user_error(hwstate->fpscr, &ufp->fpscr, err); - fmxr(FPEXC, fpexc & ~FPEXC_EN); + /* + * Copy the exception registers. + */ + __put_user_error(hwstate->fpexc, &ufp_exc->fpexc, err); + __put_user_error(hwstate->fpinst, &ufp_exc->fpinst, err); + __put_user_error(hwstate->fpinst2, &ufp_exc->fpinst2, err); - /* - * Set the context to NULL to force a reload the next time - * the thread uses the VFP. - */ - last_VFP_context[cpu] = NULL; - } + if (err) + return -EFAULT; + + /* Ensure that VFP is disabled. */ + vfp_flush_hwstate(thread); -#ifdef CONFIG_SMP /* - * For SMP we still have to take care of the case where the thread - * migrates to another CPU and then back to the original CPU on which - * the last VFP user is still the same thread. Mark the thread VFP - * state as belonging to a non-existent CPU so that the saved one will - * be reloaded in the above case. + * As per the PCS, clear the length and stride bits for function + * entry. */ - thread->vfpstate.hard.cpu = NR_CPUS; -#endif - put_cpu(); + hwstate->fpscr &= ~(FPSCR_LENGTH_MASK | FPSCR_STRIDE_MASK); + return 0; +} + +/* Sanitise and restore the current VFP state from the provided structures. */ +int vfp_restore_user_hwstate(struct user_vfp __user *ufp, + struct user_vfp_exc __user *ufp_exc) +{ + struct thread_info *thread = current_thread_info(); + struct vfp_hard_struct *hwstate = &thread->vfpstate.hard; + unsigned long fpexc; + int err = 0; + + /* Disable VFP to avoid corrupting the new thread state. */ + vfp_flush_hwstate(thread); + + /* + * Copy the floating point registers. There can be unused + * registers see asm/hwcap.h for details. + */ + err |= __copy_from_user(&hwstate->fpregs, &ufp->fpregs, + sizeof(hwstate->fpregs)); + /* + * Copy the status and control register. + */ + __get_user_error(hwstate->fpscr, &ufp->fpscr, err); + + /* + * Sanitise and restore the exception registers. + */ + __get_user_error(fpexc, &ufp_exc->fpexc, err); + + /* Ensure the VFP is enabled. */ + fpexc |= FPEXC_EN; + + /* Ensure FPINST2 is invalid and the exception flag is cleared. */ + fpexc &= ~(FPEXC_EX | FPEXC_FP2V); + hwstate->fpexc = fpexc; + + __get_user_error(hwstate->fpinst, &ufp_exc->fpinst, err); + __get_user_error(hwstate->fpinst2, &ufp_exc->fpinst2, err); + + return err ? -EFAULT : 0; } /* * VFP hardware can lose all context when a CPU goes offline. - * Safely clear our held state when a CPU has been killed, and - * re-enable access to VFP when the CPU comes back online. + * As we will be running in SMP mode with CPU hotplug, we will save the + * hardware state at every thread switch. We clear our held state when + * a CPU has been killed, indicating that the VFP hardware doesn't contain + * a threads VFP state. When a CPU starts up, we re-enable access to the + * VFP hardware. * * Both CPU_DYING and CPU_STARTING are called on the CPU which * is being offlined/onlined. @@ -498,14 +642,79 @@ void vfp_flush_hwstate(struct thread_info *thread) static int vfp_hotplug(struct notifier_block *b, unsigned long action, void *hcpu) { - if (action == CPU_DYING || action == CPU_DYING_FROZEN) { - unsigned int cpu = (long)hcpu; - last_VFP_context[cpu] = NULL; - } else if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) + if (action == CPU_DYING || action == CPU_DYING_FROZEN) + vfp_current_hw_state[(long)hcpu] = NULL; + else if (action == CPU_STARTING || action == CPU_STARTING_FROZEN) vfp_enable(NULL); return NOTIFY_OK; } +void vfp_kmode_exception(void) +{ + /* + * If we reach this point, a floating point exception has been raised + * while running in kernel mode. If the NEON/VFP unit was enabled at the + * time, it means a VFP instruction has been issued that requires + * software assistance to complete, something which is not currently + * supported in kernel mode. + * If the NEON/VFP unit was disabled, and the location pointed to below + * is properly preceded by a call to kernel_neon_begin(), something has + * caused the task to be scheduled out and back in again. In this case, + * rebuilding and running with CONFIG_DEBUG_ATOMIC_SLEEP enabled should + * be helpful in localizing the problem. + */ + if (fmrx(FPEXC) & FPEXC_EN) + pr_crit("BUG: unsupported FP instruction in kernel mode\n"); + else + pr_crit("BUG: FP instruction issued in kernel mode with FP unit disabled\n"); +} + +#ifdef CONFIG_KERNEL_MODE_NEON + +/* + * Kernel-side NEON support functions + */ +void kernel_neon_begin(void) +{ + struct thread_info *thread = current_thread_info(); + unsigned int cpu; + u32 fpexc; + + /* + * Kernel mode NEON is only allowed outside of interrupt context + * with preemption disabled. This will make sure that the kernel + * mode NEON register contents never need to be preserved. + */ + BUG_ON(in_interrupt()); + cpu = get_cpu(); + + fpexc = fmrx(FPEXC) | FPEXC_EN; + fmxr(FPEXC, fpexc); + + /* + * Save the userland NEON/VFP state. Under UP, + * the owner could be a task other than 'current' + */ + if (vfp_state_in_hw(cpu, thread)) + vfp_save_state(&thread->vfpstate, fpexc); +#ifndef CONFIG_SMP + else if (vfp_current_hw_state[cpu] != NULL) + vfp_save_state(vfp_current_hw_state[cpu], fpexc); +#endif + vfp_current_hw_state[cpu] = NULL; +} +EXPORT_SYMBOL(kernel_neon_begin); + +void kernel_neon_end(void) +{ + /* Disable the NEON/VFP unit. */ + fmxr(FPEXC, fmrx(FPEXC) & ~FPEXC_EN); + put_cpu(); +} +EXPORT_SYMBOL(kernel_neon_end); + +#endif /* CONFIG_KERNEL_MODE_NEON */ + /* * VFP support code initialisation. */ @@ -515,7 +724,7 @@ static int __init vfp_init(void) unsigned int cpu_arch = cpu_architecture(); if (cpu_arch >= CPU_ARCH_ARMv6) - vfp_enable(NULL); + on_each_cpu(vfp_enable, NULL, 1); /* * First check that there is a VFP that we can use. @@ -528,18 +737,16 @@ static int __init vfp_init(void) barrier(); vfp_vector = vfp_null_entry; - printk(KERN_INFO "VFP support v0.3: "); + pr_info("VFP support v0.3: "); if (VFP_arch) - printk("not present\n"); + pr_cont("not present\n"); else if (vfpsid & FPSID_NODOUBLE) { - printk("no double precision support\n"); + pr_cont("no double precision support\n"); } else { hotcpu_notifier(vfp_hotplug, 0); - smp_call_function(vfp_enable, NULL, 1); - VFP_arch = (vfpsid & FPSID_ARCH_MASK) >> FPSID_ARCH_BIT; /* Extract the architecture version */ - printk("implementor %02x architecture %d part %02x variant %x rev %x\n", + pr_cont("implementor %02x architecture %d part %02x variant %x rev %x\n", (vfpsid & FPSID_IMPLEMENTER_MASK) >> FPSID_IMPLEMENTER_BIT, (vfpsid & FPSID_ARCH_MASK) >> FPSID_ARCH_BIT, (vfpsid & FPSID_PART_MASK) >> FPSID_PART_BIT, @@ -561,14 +768,16 @@ static int __init vfp_init(void) elf_hwcap |= HWCAP_VFPv3; /* - * Check for VFPv3 D16. CPUs in this configuration - * only have 16 x 64bit registers. + * Check for VFPv3 D16 and VFPv4 D16. CPUs in + * this configuration only have 16 x 64bit + * registers. */ if (((fmrx(MVFR0) & MVFR0_A_SIMD_MASK)) == 1) - elf_hwcap |= HWCAP_VFPv3D16; + elf_hwcap |= HWCAP_VFPv3D16; /* also v4-D16 */ + else + elf_hwcap |= HWCAP_VFPD32; } #endif -#ifdef CONFIG_NEON /* * Check for the presence of the Advanced SIMD * load/store instructions, integer and single @@ -576,12 +785,17 @@ static int __init vfp_init(void) * for NEON if the hardware has the MVFR registers. */ if ((read_cpuid_id() & 0x000f0000) == 0x000f0000) { +#ifdef CONFIG_NEON if ((fmrx(MVFR1) & 0x000fff00) == 0x00011100) elf_hwcap |= HWCAP_NEON; - } #endif +#ifdef CONFIG_VFPv3 + if ((fmrx(MVFR1) & 0xf0000000) == 0x10000000) + elf_hwcap |= HWCAP_VFPv4; +#endif + } } return 0; } -late_initcall(vfp_init); +core_initcall(vfp_init); diff --git a/arch/arm/vfp/vfpsingle.c b/arch/arm/vfp/vfpsingle.c index b252631b406..4f96c1617aa 100644 --- a/arch/arm/vfp/vfpsingle.c +++ b/arch/arm/vfp/vfpsingle.c @@ -915,6 +915,8 @@ vfp_single_multiply_accumulate(int sd, int sn, s32 m, u32 fpscr, u32 negate, cha v = vfp_get_float(sd); pr_debug("VFP: s%u = %08x\n", sd, v); vfp_single_unpack(&vsn, v); + if (vsn.exponent == 0 && vsn.significand) + vfp_single_normalise_denormal(&vsn); if (negate & NEG_SUBTRACT) vsn.sign = vfp_sign_negate(vsn.sign); |