reliable stack trace support (x86-64)
PATCH 2/3 reliable stack trace support (x86-64)
These are the x86_64-specific pieces to enable reliable stack traces. The only restriction with this is that it currently cannot unwind across the interrupt->normal stack boundary, as that transition is lacking proper annotation.
arch/x86_64/kernel/entry.S +33 -0
@@ -1051,3 +1051,36 @@ ENTRY(call_softirq) decl %gs:pda_irqcount ret CFI_ENDPROC + +#ifdef CONFIG_STACK_UNWIND +ENTRY(arch_unwind_init_running) + CFI_STARTPROC + movq %r15, R15(%rdi) + movq %r14, R14(%rdi) + xchgq %rsi, %rdx + movq %r13, R13(%rdi) + movq %r12, R12(%rdi) + xorl %eax, %eax + movq %rbp, RBP(%rdi) + movq %rbx, RBX(%rdi) + movq (%rsp), %rcx + movq %rax, R11(%rdi) + movq %rax, R10(%rdi) + movq %rax, R9(%rdi) + movq %rax, R8(%rdi) + movq %rax, RAX(%rdi) + movq %rax, RCX(%rdi) + movq %rax, RDX(%rdi) + movq %rax, RSI(%rdi) + movq %rax, RDI(%rdi) + movq %rax, ORIG_RAX(%rdi) + movq %rcx, RIP(%rdi) + leaq 8(%rsp), %rcx + movq $__KERNEL_CS, CS(%rdi) + movq %rax, EFLAGS(%rdi) + movq %rcx, RSP(%rdi) + movq $__KERNEL_DS, SS(%rdi) + jmpq *%rdx + CFI_ENDPROC +ENDPROC(arch_unwind_init_running) +#endif
arch/x86_64/kernel/process.c +1 -1
@@ -335,7 +335,7 @@ void show_regs(struct pt_regs *regs) {printk ("CPU %d:" ,smp_processor_id ());__show_regs (regs); -show_trace (®s->rsp); +show_trace (NULL , regs, (void *)(regs +1 )); }/*
arch/x86_64/kernel/traps.c +47 -6
103 lines not on this page · raw message
arch/x86_64/kernel/vmlinux.lds.S +9 -0
@@ -45,6 +45,15 @@ SECTIONS RODATA +#ifdef CONFIG_STACK_UNWIND + . = ALIGN(8); + .eh_frame : AT(ADDR(.eh_frame) - LOAD_OFFSET) { + __start_unwind = .; + *(.eh_frame) + __end_unwind = .; + } +#endif + /* Data */ .data : AT(ADDR(.data) - LOAD_OFFSET) { *(.data)
include/asm-x86_64/proto.h +1 -1
@@ -75,7 +75,7 @@ extern void main_timer_handler(struct ptextern unsigned long end_pfn_map; -extern void show_trace (unsigned long * rsp); +extern void show_trace (struct task_struct *,struct pt_regs *,unsigned long * rsp);extern void show_registers (struct pt_regs *regs);extern void exception_table_check (void );
include/asm-x86_64/unwind.h +99 -0
99 lines not on this page · raw message
lib/Kconfig.debug +1 -1
@@ -199,7 +199,7 @@ config UNWIND_INFO config STACK_UNWIND bool "Stack unwind support" depends on UNWIND_INFO - depends on n + depends on X86_64 help This enables more precise stack traces, omitting all unrelated occurrences of pointers into kernel code from the dump.
Quoted 5 lines · Jan Beulich
These are the x86_64-specific pieces to enable reliable stack traces. The only restriction with this is that it currently cannot unwind across the
interrupt->normalstack boundary, as that transition is lacking proper annotation.
include/asm-x86_64/unwind.h line 29
Quoted 9 lines
include/asm-x86_64/unwind.h +9
-0+#define UNW_PC (frame) (frame)->regs.rip +#define UNW_SP (frame) (frame)->regs.rsp +#ifdef CONFIG_FRAME_POINTER +#define UNW_FP (frame) (frame)->regs.rbp +#define FRAME_RETADDR_OFFSET 8 +#define FRAME_LINK_OFFSET 0 +#define STACK_BOTTOM (tsk) (((tsk)->thread.rsp0 - 1) & ~(THREAD_SIZE - 1)) +#define STACK_TOP (tsk) ((tsk)->thread.rsp0) +#endif
style: align the defines.
include/asm-x86_64/unwind.h line 83
Quoted 10 lines
include/asm-x86_64/unwind.h +10
-0+static inline int +arch_unw_user_mode (const struct unwind_frame_info *info) +{ +#if 0 /* This can only work when selector register saves/restores +are properly annotated (and tracked in UNW_REGISTER_INFO). */ +return user_mode (&info->regs); +#else +return (long )info->regs.rip >=0 ; +#endif +}
is this safe? Cannot userspace provide (long)rip < 0 by jumping to it?
Ingo
Quoted 4 lines · Jan Beulich
These are the x86_64-specific pieces to enable reliable stack traces. The only restriction with this is that it currently cannot unwind across the
interrupt->normalstack boundary, as that transition is lacking proper annotation.
It would be nice if you could submit a patch to fix that.
arch/x86_64/kernel/vmlinux.lds.S SECTIONS · line 55
Quoted 9 lines
arch/x86_64/kernel/vmlinux.lds.S +8
-0+#ifdef CONFIG_STACK_UNWIND + . = ALIGN(8); + .eh_frame : AT(ADDR(.eh_frame) - LOAD_OFFSET) { + __start_unwind = .; + *(.eh_frame) + __end_unwind = .; + } +#endif
Ah ok - it's there. Ignore my earlier question then.
include/asm-x86_64/unwind.h line 22
Quoted 2 lines
include/asm-x86_64/unwind.h +2
-0+#define UNW_PC (frame) (frame)->regs.rip +#define UNW_SP (frame) (frame)->regs.rsp
I think we alreay have instruction_pointer(). Better add a stack_pointer() in ptrace.h too.
-Andi
Quoted 13 lines
Quoted 1 line
Quoted 1 line
Ingo Molnar <[email protected]> 16.05.06 17:09 >>>
include/asm-x86_64/unwind.h line 29
Quoted 9 lines · Jan Beulich
include/asm-x86_64/unwind.h +9
-0+#define UNW_PC (frame) (frame)->regs.rip +#define UNW_SP (frame) (frame)->regs.rsp +#ifdef CONFIG_FRAME_POINTER +#define UNW_FP (frame) (frame)->regs.rbp +#define FRAME_RETADDR_OFFSET 8 +#define FRAME_LINK_OFFSET 0 +#define STACK_BOTTOM (tsk) (((tsk)->thread.rsp0 - 1) & ~(THREAD_SIZE - 1)) +#define STACK_TOP (tsk) ((tsk)->thread.rsp0) +#endif style: align the defines.
If that's important...
Quoted 4 lines
Quoted 4 lines
include/asm-x86_64/unwind.h +4
-0+static inline int +arch_unw_user_mode (const struct unwind_frame_info *info) +{ +#if 0 /* This can only work when selector register saves/restores
Quoted 8 lines
include/asm-x86_64/unwind.h line 83
Quoted 6 lines
include/asm-x86_64/unwind.h +6
-0+are properly annotated (and tracked in UNW_REGISTER_INFO). */ +return user_mode (&info->regs); +#else +return (long )info->regs.rip >=0 ; +#endif +}is this safe? Cannot userspace provide (long)rip < 0 by jumping to it?
You noted the comment, didn't you. This really should use user_mode(), but it can't as the unwind annotations don't cover selector registers. Hence the initial CS/SS will always remain in place no matter how many unwinds you do. But it's harmless for this function to return 'kernel mode' when in fact it might be user mode - nothing would crash because of that, it might only lead to ill trailing entries on a stack dump.
Jan
Quoted 1 line
Quoted 1 line
Quoted 1 line
Andi Kleen <[email protected]> 16.05.06 17:13 >>>
Quoted 6 lines · Jan Beulich
Quoted 4 lines
These are the x86_64-specific pieces to enable reliable stack traces. The only restriction with this is that it currently cannot unwind across the
interrupt->normalstack boundary, as that transition is lacking proper annotation.It would be nice if you could submit a patch to fix that.
But I don't know how to fix it. See my other mail - I have no experience with expressions, nor have I ever seen them in use.
Quoted 5 lines
include/asm-x86_64/unwind.h line 22
Quoted 2 lines
include/asm-x86_64/unwind.h +2
-0+#define UNW_PC (frame) (frame)->regs.rip +#define UNW_SP (frame) (frame)->regs.rspI think we alreay have
instruction_pointer(). Better add astack_pointer()inptrace.htoo.
I could do that, but the macros will have to remain, as they don't access pt_regs directly, so I guess it'd be pointless to change it.
Jan
Quoted 10 lines · Jan Beulich
Quoted 1 line
Quoted 1 line
Quoted 1 line
Andi Kleen <[email protected]> 16.05.06 17:13 >>>
Quoted 6 lines · Jan Beulich
Quoted 4 lines
These are the x86_64-specific pieces to enable reliable stack traces. The only restriction with this is that it currently cannot unwind across the
interrupt->normalstack boundary, as that transition is lacking proper annotation.It would be nice if you could submit a patch to fix that.
But I don't know how to fix it. See my other mail
which mail?
Quoted 2 lines
- I have no experience with expressions, nor have I ever seen them in use.
I remember Jim Houston used a hack of just loading the old stack into a register and defining that as a base register in CFI. I guess i would be willing to trade a few moves for that (should be pretty much free on a OOO CPU anyways) You think that trick would work?
Quoted 8 lines
Quoted 5 lines
include/asm-x86_64/unwind.h line 22
Quoted 2 lines
include/asm-x86_64/unwind.h +2
-0+#define UNW_PC (frame) (frame)->regs.rip +#define UNW_SP (frame) (frame)->regs.rspI think we alreay have
instruction_pointer(). Better add astack_pointer()inptrace.htoo.I could do that, but the macros will have to remain, as they don't access
pt_regsdirectly, so I guess it'd be pointless to change it.
UNW_PC() is instruction_pointer(&frame->regs), isn't it?
-Andi
Quoted 9 lines
Quoted 1 line
Quoted 1 line
Andi Kleen <[email protected]> 16.05.06 19:05 >>>
Quoted 7 lines · Jan Beulich
Quoted 1 line
Quoted 1 line
Quoted 1 line
Andi Kleen <[email protected]> 16.05.06 17:13 >>>
Quoted 5 lines · Jan Beulich
Quoted 4 lines
These are the x86_64-specific pieces to enable reliable stack traces. The only restriction with this is that it currently cannot unwind across the
interrupt->normalstack boundary, as that transition is lacking proper annotation.
Quoted 4 lines
Quoted 3 lines
Quoted 1 line
It would be nice if you could submit a patch to fix that.
But I don't know how to fix it. See my other mail
which mail?
Reply to Ingo (with you on cc) regarding patch 1/3. Just saying that I don't know much about expressions here.
Quoted 7 lines
Quoted 2 lines
- I have no experience with expressions, nor have I ever seen them in use.
I remember Jim Houston used a hack of just loading the old stack into a register and defining that as a base register in CFI. I guess i would be willing to trade a few moves for that (should be pretty much free on a OOO CPU anyways) You think that trick would work?
I don't think that would, because without CONFIG_DEBUG_INFO none of the preserved registers get saved, hence there's no register to use for this. Thus the price would not only be a move, but also a save/push and a reload/pop.
Quoted 10 lines
Quoted 8 lines
Quoted 5 lines
include/asm-x86_64/unwind.h line 22
Quoted 2 lines
include/asm-x86_64/unwind.h +2
-0+#define UNW_PC (frame) (frame)->regs.rip +#define UNW_SP (frame) (frame)->regs.rspI think we alreay have
instruction_pointer(). Better add astack_pointer()inptrace.htoo.I could do that, but the macros will have to remain, as they don't access
pt_regsdirectly, so I guess it'd be pointless to change it.
UNW_PC()isinstruction_pointer(&frame->regs), isn't it?
Yes. But the intention is that the user of UNW_PC doesn't need to know any details of what fields frame has (i.e. the parameter of UNW_PC must only be frame), so you can't replace it with instruction_pointer().
Jan
Quoted 11 lines · Jan Beulich
Quoted 11 lines
Quoted 1 line
Quoted 1 line
Andi Kleen <[email protected]> 16.05.06 19:05 >>>
Quoted 8 lines · Jan Beulich
Quoted 1 line
Quoted 1 line
Quoted 1 line
Andi Kleen <[email protected]> 16.05.06 17:13 >>>
Quoted 5 lines · Jan Beulich
Quoted 4 lines
These are the x86_64-specific pieces to enable reliable stack traces. The only restriction with this is that it currently cannot unwind across the
interrupt->normalstack boundary, as that transition is lacking proper annotation.
Quoted 9 lines
Quoted 5 lines
Quoted 3 lines
Quoted 1 line
It would be nice if you could submit a patch to fix that.
But I don't know how to fix it. See my other mail
which mail?
Reply to Ingo (with you on cc) regarding patch 1/3. Just saying that I don't know much about expressions here.
Hmm, maybe we can find somebody who does. But then they would first need to be implemented in the unwinder anyways, I guess. Could be nasty agreed.
Quoted 8 lines
Quoted 7 lines
Quoted 2 lines
- I have no experience with expressions, nor have I ever seen them in use.
I remember Jim Houston used a hack of just loading the old stack into a register and defining that as a base register in CFI. I guess i would be willing to trade a few moves for that (should be pretty much free on a OOO CPU anyways) You think that trick would work?
Quoted 4 lines
I don't think that would, because without
CONFIG_DEBUG_INFOnone of the preserved registers get saved, hence there's no register to use for this. Thus the price would not only be a move, but also a save/push and a reload/pop.
Three instructions? We might be still able to afford that.
push/pop is probably not needed because the stack frame will be already set up (ok possibly after a few instructions, but a small window might be tolerable)
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
include/asm-x86_64/unwind.h line 22
Quoted 2 lines
include/asm-x86_64/unwind.h +2
-0+#define UNW_PC (frame) (frame)->regs.rip +#define UNW_SP (frame) (frame)->regs.rspI think we alreay have
instruction_pointer(). Better add astack_pointer()inptrace.htoo.
Quoted 8 lines
Quoted 4 lines
Quoted 2 lines
I could do that, but the macros will have to remain, as they don't access
pt_regsdierectly, so I guess it'd be pointless to change it.
UNW_PC()isinstruction_pointer(&frame->regs), isn't it?Yes. But the intention is that the user of
UNW_PCdoesn't need to know any details of what fields frame has (i.e. the parameter ofUNW_PCmust only be frame), so you can't replace it withinstruction_pointer().
Maybe I'm dense but I still don't get - frame has a pt_regs so why isn't the caller allowed to know about that fact?
-Andi
Quoted 5 lines
Three instructions? We might be still able to afford that.
push/pop is probably not needed because the stack frame will be already set up (ok possibly after a few instructions, but a small window might be tolerable)
Okay, then I'll prepare a patch to that effect during the next couple of days.
Quoted 2 lines
Maybe I'm dense but I still don't get - frame has a
pt_regsso why isn't the caller allowed to know about that fact?
Because the fact that there is a regs fields and the PC is accessible through it is architecture specific, yet the caller (kernel/unwind.c) ought to be architecture independent.
Jan
Quoted 6 lines
Quoted 3 lines
Maybe I'm dense but I still don't get - frame has a
pt_regsso why isn't the caller allowed to know about that fact?Because the fact that there is a regs fields and the PC is accessible through it is architecture specific, yet the caller (
kernel/unwind.c) ought to be architecture independent.
I doubt we have any architecture where the instruction pointer is not in pt_regs, but ok.
-Andi