From b03b08ba9c7235861adf4dde712dade0bb756fe0 Mon Sep 17 00:00:00 2001 From: Robin Getz Date: Sun, 23 Dec 2007 22:57:01 +0800 Subject: [Blackfin] arch: Clean up dump_bfin_mem Clean up dump_bfin_mem so that it will display content from the kernel, as well as l1 instruction, when deferred HW errors happen, print out the last frame info if it makes sense. Signed-off-by: Robin Getz Signed-off-by: Bryan Wu --- arch/blackfin/kernel/early_printk.c | 2 +- arch/blackfin/kernel/process.c | 1 + arch/blackfin/kernel/traps.c | 132 +++++++++++++++++++++++------------- 3 files changed, 86 insertions(+), 49 deletions(-) (limited to 'arch/blackfin') diff --git a/arch/blackfin/kernel/early_printk.c b/arch/blackfin/kernel/early_printk.c index 724f4a5a1d4..c892777267d 100644 --- a/arch/blackfin/kernel/early_printk.c +++ b/arch/blackfin/kernel/early_printk.c @@ -205,7 +205,7 @@ asmlinkage void __init early_trap_c(struct pt_regs *fp, void *retaddr) if (likely(early_console == NULL)) setup_early_printk(DEFAULT_EARLY_PORT); - dump_bfin_mem((void *)fp->retx); + dump_bfin_mem(fp); show_regs(fp); dump_bfin_trace_buffer(); diff --git a/arch/blackfin/kernel/process.c b/arch/blackfin/kernel/process.c index 5bf15125f0d..fff3dd9eadd 100644 --- a/arch/blackfin/kernel/process.c +++ b/arch/blackfin/kernel/process.c @@ -327,6 +327,7 @@ void finish_atomic_sections (struct pt_regs *regs) } #if defined(CONFIG_ACCESS_CHECK) +/* Return 1 if access to memory range is OK, 0 otherwise */ int _access_ok(unsigned long addr, unsigned long size) { if (size == 0) diff --git a/arch/blackfin/kernel/traps.c b/arch/blackfin/kernel/traps.c index 8bbfef31666..c90f16825f9 100644 --- a/arch/blackfin/kernel/traps.c +++ b/arch/blackfin/kernel/traps.c @@ -39,6 +39,7 @@ #include #include #include +#include #ifdef CONFIG_KGDB # include @@ -171,7 +172,7 @@ asmlinkage void double_fault_c(struct pt_regs *fp) oops_in_progress = 1; printk(KERN_EMERG "\n" KERN_EMERG "Double Fault\n"); dump_bfin_process(fp); - dump_bfin_mem((void *)fp->retx); + dump_bfin_mem(fp); show_regs(fp); panic("Double Fault - unrecoverable event\n"); @@ -196,6 +197,10 @@ asmlinkage void trap_c(struct pt_regs *fp) * we will kernel panic, so the system reboots. * If KGDB is enabled, don't set this for kernel breakpoints */ + + /* TODO: check to see if we are in some sort of deferred HWERR + * that we should be able to recover from, not kernel panic + */ if ((bfin_read_IPEND() & 0xFFC0) #ifdef CONFIG_KGDB && trapnr != VEC_EXCPT02 @@ -478,11 +483,7 @@ asmlinkage void trap_c(struct pt_regs *fp) if (sig != SIGTRAP) { unsigned long stack; dump_bfin_process(fp); - /* Is it an interrupt, or an exception? */ - if (trapnr == VEC_HWERR) - dump_bfin_mem((void *)fp->pc); - else - dump_bfin_mem((void *)fp->retx); + dump_bfin_mem(fp); show_regs(fp); /* Print out the trace buffer if it makes sense */ @@ -644,8 +645,10 @@ void dump_bfin_process(struct pt_regs *fp) if (oops_in_progress) printk(KERN_EMERG "Kernel OOPS in progress\n"); - if (context & 0x0020) - printk(KERN_NOTICE "Deferred excecption or HW Error context\n"); + if (context & 0x0020 && (fp->seqstat & SEQSTAT_EXCAUSE) == VEC_HWERR) + printk(KERN_NOTICE "HW Error context\n"); + else if (context & 0x0020) + printk(KERN_NOTICE "Defered Exception context\n"); else if (context & 0x3FC0) printk(KERN_NOTICE "Interrupt context\n"); else if (context & 0x4000) @@ -673,49 +676,82 @@ void dump_bfin_process(struct pt_regs *fp) "No Valid process in current context\n"); } -void dump_bfin_mem(void *retaddr) +void dump_bfin_mem(struct pt_regs *fp) { + unsigned short *addr, *erraddr, val = 0, err = 0; + char sti = 0, buf[6]; - if (retaddr >= (void *)FIXED_CODE_START && retaddr < (void *)physical_mem_end -#if L1_CODE_LENGTH != 0 - /* FIXME: Copy the code out of L1 Instruction SRAM through dma - memcpy. */ - && !(retaddr >= (void *)L1_CODE_START - && retaddr < (void *)(L1_CODE_START + L1_CODE_LENGTH)) -#endif - ) { - int i = ((unsigned int)retaddr & 0xFFFFFFF0) - 32; - unsigned short x = 0; - printk(KERN_NOTICE "return address: [0x%p]; contents of:", retaddr); - for (; i < ((unsigned int)retaddr & 0xFFFFFFF0) + 32; i += 2) { - if (!(i & 0xF)) - printk("\n" KERN_NOTICE "0x%08x: ", i); - - if (get_user(x, (unsigned short *)i)) - break; + if (unlikely((fp->seqstat & SEQSTAT_EXCAUSE) == VEC_HWERR)) + erraddr = (void *)fp->pc; + else + erraddr = (void *)fp->retx; + + printk(KERN_NOTICE "return address: [0x%p]; contents of:", erraddr); + + for (addr = (unsigned short *)((unsigned long)erraddr & ~0xF) - 0x10; + addr < (unsigned short *)((unsigned long)erraddr & ~0xF) + 0x10; + addr++) { + if (!((unsigned long)addr & 0xF)) + printk("\n" KERN_NOTICE "0x%p: ", addr); + + if (get_user(val, addr)) { + if (addr >= (unsigned short *)L1_CODE_START && + addr < (unsigned short *)(L1_CODE_START + L1_CODE_LENGTH)) { + dma_memcpy(&val, addr, sizeof(val)); + sprintf(buf, "%04x", val); + } else if (addr >= (unsigned short *)FIXED_CODE_START && + addr <= (unsigned short *)memory_start) { + val = bfin_read16(addr); + sprintf(buf, "%04x", val); + } else { + val = 0; + sprintf(buf, "????"); + } + } else + sprintf(buf, "%04x", val); + + if (addr == erraddr) { + printk("[%s]", buf); + err = val; + } else + printk(" %s ", buf); + + /* Do any previous instructions turn on interrupts? */ + if (addr <= erraddr && /* in the past */ + ((val >= 0x0040 && val <= 0x0047) || /* STI instruction */ + val == 0x017b)) /* [SP++] = RETI */ + sti = 1; + } + + printk("\n"); + + /* Hardware error interrupts can be deferred */ + if (unlikely(sti && (fp->seqstat & SEQSTAT_EXCAUSE) == VEC_HWERR && + oops_in_progress)){ + printk(KERN_NOTICE "Looks like this was a deferred error - sorry\n"); #ifndef CONFIG_DEBUG_HWERR - /* If one of the last few instructions was a STI - * it is likely that the error occured awhile ago - * and we just noticed. This only happens in kernel - * context, which should mean an oops is happening - */ - if (oops_in_progress && x >= 0x0040 && x <= 0x0047 && i <= 0) - printk(KERN_EMERG "\n" - KERN_EMERG "WARNING : You should reconfigure" - " the kernel to turn on\n" - KERN_EMERG " 'Hardware error interrupt debugging'\n" - KERN_EMERG " The rest of this error is meanless\n"); -#endif - if (i == (unsigned int)retaddr) - printk("[%04x]", x); - else - printk(" %04x ", x); + printk(KERN_NOTICE "The remaining message may be meaningless\n" + KERN_NOTICE "You should enable CONFIG_DEBUG_HWERR to get a" + " better idea where it came from\n"); +#else + /* If we are handling only one peripheral interrupt + * and current mm and pid are valid, and the last error + * was in that user space process's text area + * print it out - because that is where the problem exists + */ + if ((!(((fp)->ipend & ~0x30) & (((fp)->ipend & ~0x30) - 1))) && + (current->pid && current->mm)) { + /* And the last RETI points to the current userspace context */ + if ((fp + 1)->pc >= current->mm->start_code && + (fp + 1)->pc <= current->mm->end_code) { + printk(KERN_NOTICE "It might be better to look around here : \n"); + printk(KERN_NOTICE "-------------------------------------------\n"); + show_regs(fp + 1); + printk(KERN_NOTICE "-------------------------------------------\n"); + } } - printk("\n"); - } else - printk("\n" KERN_NOTICE - "Cannot look at the [PC] <%p> for it is" - " in unreadable memory - sorry\n", retaddr); +#endif + } } void show_regs(struct pt_regs *fp) @@ -885,7 +921,7 @@ void panic_cplb_error(int cplb_panic, struct pt_regs *fp) printk(KERN_EMERG "DCPLB_FAULT_ADDR=%p\n", (void *)bfin_read_DCPLB_FAULT_ADDR()); printk(KERN_EMERG "ICPLB_FAULT_ADDR=%p\n", (void *)bfin_read_ICPLB_FAULT_ADDR()); dump_bfin_process(fp); - dump_bfin_mem((void *)fp->retx); + dump_bfin_mem(fp); show_regs(fp); dump_stack(); panic("Unrecoverable event\n"); -- cgit v1.2.3-18-g5258