On Mon, Feb 01, 2021 at 01:23:52PM +0100, Thomas Bogendoerfer wrote:

On Fri, Jan 29, 2021 at 12:35:07PM +0800, Huang Pei wrote:

quoted

MIPS always save kernel stack pointer in regs[29]

Signed-off-by: Huang Pei <redacted>
---
 arch/mips/include/asm/ptrace.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/mips/include/asm/ptrace.h b/arch/mips/include/asm/ptrace.h
index 1e76774b36dd..daf3cf244ea9 100644
--- a/arch/mips/include/asm/ptrace.h
+++ b/arch/mips/include/asm/ptrace.h

@@ -53,7 +53,7 @@ struct pt_regs {
 
 static inline unsigned long kernel_stack_pointer(struct pt_regs *regs)
 {
-	return regs->regs[31];
+	return regs->regs[29];

hmm, I'm still wondering where the trick is... looks like this is used
for uprobes, so nobody has ever used uprobes or I'm missing something.

How did you find that ?

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessarily a
good idea.                                                [ RFC1925, 2.3 ]

Long story for short, 

+. I think I had fix this bug in 2018, when I backported Uprobe from my
4.4 branch to CentOS 7 3.10. I just knwo it is *not* following MIPS
ABI, but I do not know how it destroy the cool function of
Kprobe/Uprobe, since the failure in porting eBPF from upstream to 3.10
just leave the fix in 3.10, totally forgotten.

+. In 2020, I was told to validate the effect of GNU XHash, and it came
to me that using Uprobe to count the number of "strcmp" called in ld.so,
so I found this fix again.

+. With more work on Kprobe/Kprobe_event/Uprobe, I found it hit only when
accessing arguments of Kprobe/Uprobe, so simple counting numbers of probe
fired would not trigger it