Skip to contentSkip to conversation

powerpc: perf hardware breakpoint fixes

PATCH New Michael Neuling 1 participant 4 messages Aug 22, 2012 – Aug 23, 2012 PowerPC

lore
Share & tools
lore
#0 lore raw

cover powerpc: perf hardware breakpoint fixes

Here are a couple of patches for perf for hardware breakpoints. Along with https://lkml.org/lkml/2012/8/16/5 hardware breakpoints now work for me powerpc. These patches don't depend on that patch though.

Michael Neuling (2):
powerpc: Remove unessasary ifdefs
powerpc: Fix null pointer deref in perf hardware breakpoints

 arch/powerpc/include/asm/cputable.h |    2 --
 arch/powerpc/kernel/hw_breakpoint.c |    3 ++-
 2 files changed, 2 insertions(+), 3 deletions(-)
Signature · 1 line
1.7.9.5
#1 lore raw

PATCH 1/1 powerpc: Remove unnecessary ifdefs

---
 arch/powerpc/include/asm/cputable.h |    2 --
 1 file changed, 2 deletions(-)
arch/powerpc/include/asm/cputable.h +0 -2
index 50d82c8..b3c083d 100644
@@ -553,9 +553,7 @@ static inline int cpu_has_feature(unsigned long feature)
 		& feature);
 }
 
-#ifdef CONFIG_HAVE_HW_BREAKPOINT
 #define HBP_NUM 1
-#endif /* CONFIG_HAVE_HW_BREAKPOINT */
 
 #endif /* !__ASSEMBLY__ */
 
Signature · 1 line
1.7.9.5
#2 lore raw

PATCH 1/1 powerpc: Fix null pointer deref in perf hardware breakpoints

In reply to Michael Neuling

Currently if you are doing a global perf recording with hardware breakpoints (ie perf record -e mem:0xdeadbeef -a), you can oops with:

Faulting instruction address: 0xc000000000738890 cpu 0xc: Vector: 300 (Data Access) at [c0000003f76af8d0]

      pc: c000000000738890: .hw_breakpoint_handler+0xa0/0x1e0
      lr: c000000000738830: .hw_breakpoint_handler+0x40/0x1e0
      sp: c0000003f76afb50
     msr: 8000000000001032
     dar: 6f0
   dsisr: 42000000
    current = 0xc0000003f765ac00
    paca    = 0xc00000000f262a00   softe: 0        irq_happened: 0x01
    pid   = 6810, comm = loop-read

enter ? for help
[c0000003f76afbe0] c00000000073cd04 .notifier_call_chain.isra.0+0x84/0xe0 [c0000003f76afc80] c00000000073cdbc .notify_die+0x3c/0x60 [c0000003f76afd20] c0000000000139f0 .do_dabr+0x40/0xf0 [c0000003f76afe30] c000000000005a9c handle_dabr_fault+0x14/0x48 --- Exception: 300 (Data Access) at 0000000010000480
SP (ff8679e0) is in userspace

This is because we don't check to see if the break point is associated with task before we deference the task_struct pointer. This checks pointer before dereferencing it.

---
 arch/powerpc/kernel/hw_breakpoint.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
arch/powerpc/kernel/hw_breakpoint.c +2 -1
index f3a82dd..768182c 100644
@@ -253,7 +253,8 @@ int __kprobes hw_breakpoint_handler(struct die_args *args)
 
 	/* Do not emulate user-space instructions, instead single-step them */
 	if (user_mode(regs)) {
-		bp->ctx->task->thread.last_hit_ubp = bp;
+		if (bp->ctx->task)
+			bp->ctx->task->thread.last_hit_ubp = bp;
 		regs->msr |= MSR_SE;
 		goto out;
 	}
Signature · 1 line
1.7.9.5
#3 lore raw

PATCH 1/1 powerpc: Fix null pointer deref in perf hardware breakpoints

Currently if you are doing a global perf recording with hardware breakpoints (ie perf record -e mem:0xdeadbeef -a), you can oops with:

Faulting instruction address: 0xc000000000738890 cpu 0xc: Vector: 300 (Data Access) at [c0000003f76af8d0]

      pc: c000000000738890: .hw_breakpoint_handler+0xa0/0x1e0
      lr: c000000000738830: .hw_breakpoint_handler+0x40/0x1e0
      sp: c0000003f76afb50
     msr: 8000000000001032
     dar: 6f0
   dsisr: 42000000
    current = 0xc0000003f765ac00
    paca    = 0xc00000000f262a00   softe: 0        irq_happened: 0x01
    pid   = 6810, comm = loop-read

enter ? for help
[c0000003f76afbe0] c00000000073cd04 .notifier_call_chain.isra.0+0x84/0xe0 [c0000003f76afc80] c00000000073cdbc .notify_die+0x3c/0x60 [c0000003f76afd20] c0000000000139f0 .do_dabr+0x40/0xf0 [c0000003f76afe30] c000000000005a9c handle_dabr_fault+0x14/0x48 --- Exception: 300 (Data Access) at 0000000010000480
SP (ff8679e0) is in userspace

This is because we don't check to see if the break point is associated with task before we deference the task_struct pointer.

This changes the update to use current.

---
v2: mpe pointed out that we should be using current here so that once we single step the next instruction, we know where to store the perf event. Confirming this, the original patch didn't actually record any data when we used -a even when we hit.

This new patch fixes this so that -a doesn't oops and records data too!

arch/powerpc/kernel/hw_breakpoint.c +1 -1
index f3a82dd..956a4c4 100644
@@ -253,7 +253,7 @@ int __kprobes hw_breakpoint_handler(struct die_args *args)
 
 	/* Do not emulate user-space instructions, instead single-step them */
 	if (user_mode(regs)) {
-		bp->ctx->task->thread.last_hit_ubp = bp;
+		current->thread.last_hit_ubp = bp;
 		regs->msr |= MSR_SE;
 		goto out;
 	}