unwabi mismatch in 2.4.X kernels breaks oops call trace (unwind)
Oops stack unwinding was broken in 2.4.22 by this patch that was picked up from 2.5/2.6:
arch/ia64/kernel/entry.h +1 -1
@@ -14,7 +14,7 @@#define SW (f) (IA64_SWITCH_STACK_##f##_OFFSET)#define PT_REGS_SAVES (off) \ - .unwabi @svr4, 'i'; \ + .unwabi 3, 'i'; \ .fframe IA64_PT_REGS_SIZE+16+(off); \ .spillsp rp, PT(CR_IIP)+16+(off); \ .spillsp ar.pfs, PT(CR_IFS)+16+(off); \
If we want to keep this change in 2.4.X, we need to also change the test in desc_abi (unwind.c) to check for abi = 3 instead of 0.
static inline voiddesc_abi (unsigned char abi, unsigned char context, struct unw_state_record *sr){
if (abi = 0 && context = 'i') {
sr->flags |= UNW_FLAG_INTERRUPT_FRAME;
UNW_DPRINT(3, "unwind.%s: interrupt frame\n", __FUNCTION__);
}
else
UNW_DPRINT(0, "unwind%s: ignoring unwabi(abi=0x%x,context=0x%x)\n",
__FUNCTION__, abi, context);
}Otherwise we can just back the entry.h patch out. So my question
is: Is there a compelling reason to change the abi number from 0 to 3 in 2.4.X kernels? Is there an outside tool that cares, or is it sufficient for these two snippets of kernel code to merely agree on some number?
Thanks,
Bob Montgomery
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Bob Montgomery
Bob> So my question is: Is there a compelling reason to change the Bob> abi number from 0 to 3 in 2.4.X kernels?
It must change iff the syscall-streamlining patch is applied, since that patch changes the format of the pt-regs structure.
Bob> Is there an outside tool that cares
Libunwind and, for example, Ski definitely care.
--david
Quoted 7 lines · David Mosberger
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Bob Montgomery
Bob> So my question is: Is there a compelling reason to change the Bob> abi number from 0 to 3 in 2.4.X kernels?
It must change iff the syscall-streamlining patch is applied, since that patch changes the format of the pt-regs structure.
In 2.4, we have this patch:
ia64: Restructure pt_regs and optimize syscall path.
Patch by Rohit Seth, Fengua Yu, and Arun Sharma:
The main items covered by this patch are:
1) Support for 16 bytes instructions as per SDM2.1 (CSD/SSD in pt_regs)
2) f10-f11 are added as additional scratch registers for kernel's use.
3) Re-arrange pt_regs to access less cache lines in system call. Reduce scratch register saving/restoring in system call path.
4) A few instruction reorg in low-level code.
(See this link for details: http://lia64.bkbits.net:8080/linux-ia64-2.4/[email protected])
So I assume the right thing to do is to apply the following patch to 2.4. I'll do this later today unless anybody objects.
=== arch/ia64/kernel/unwind.c 1.27 vs edited ==--- 1.27/arch/ia64/kernel/unwind.c Thu Aug 14 11:26:15 2003
+++ edited/arch/ia64/kernel/unwind.c Tue Feb 10 13:41:51 2004
@@ -826,7 +826,7 @@
static inline void
desc_abi (unsigned char abi, unsigned char context, struct unw_state_record *sr) {
- if (abi = 0 && context = 'i') {
+ if (abi = 3 && context = 'i') {
sr->flags |= UNW_FLAG_INTERRUPT_FRAME;
UNW_DPRINT(3, "unwind.%s: interrupt frame\n", __FUNCTION__);
}
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Bjorn Helgaas
Bjorn> So I assume the right thing to do is to apply the following Bjorn> patch to 2.4.
Yes. Looks like that's a bit that got dropped during the backport from 2.5.
--david