Dave
Excuse me for overlooking this part of the code, I am attaching a fix
to this, hope this fixes the issue.
Dave I have few observations regarding the points you have raised
mkdumpfile -c stripping the ELF Note for pt_regs:
mkdumpfile won't be saving much by stripping ELF Notes of pt_regs
information. It will be ~256 bytes * number of cpus which is not much.
We will discuss with mkdumpfile developers to check out the possibility
of retaining this ELF Note information.
Regarding CONFIG_FRAMEPOINTER
We understand this is disabled so as to release one more
register,bp, for general purpose operations and this is default. Ideally
this information should have got saved in dwarf section, so
theoretically speaking we should be able to unwind the x86/x86_64 dump
even with out CONFIG_FRAMEPOINTER. But some how the stack unwinding is
not as direct as it is in ppc64 we are re-looking into this implementation.
Regarding Exception Frame on the top of the stack frame
As we understand if we have the pt_regs of the topmost stack we
should be able to unwind to the next stack frame, even if top most stack
frame is an exception frame, atleast in ppc64. We are not sure of x86
and x86_64 we can relook into that too.
Please let me know your thoughts on this
Thanks
Sharyathi
Dave Anderson wrote:
Dave Anderson wrote:
> That all being said, here's what I can do for you. I will take your new
> functions that you've added to netdump.c, and their protos in defs.h,
> and apply them to the next crash utility release. Since they are not being
> used by anybody but your module at the moment, it's harmless to add them,
> and then you will not have to patch the crash sources at all in your
> subsequent module patch/posts.
I take that back...
You're first going to have to fix this proposed netdump.c function:
+static void* get_x86_regs_from_elf_notes(struct task_context *tc)
+{
+ Elf32_Nhdr *note;
An x86 kdump can be either a 32-bit or 64-bit ELF vmcore, but the default
is 64-bit. You have defined only one "note" pointer as an Elf32_hdr pointer.
+ size_t len;
+ char *pt_regs;
+ ulong regs_size;
+
+ if((tc->task == tt->panic_task) ||
+ (is_task_active(tc->task))){
+ if (nd->num_prstatus_notes > 1)
+ note = (Elf32_Nhdr *)
+ nd->nt_prstatus_percpu[tc->processor];
+ else
+ note = (Elf64_Nhdr *)nd->nt_prstatus;
And then above you determine first whether there is more than one prstatus
note (i.e., more than one cpu), and then based upon that you're assigning
the "note" pointer to the appropriate location. But for the single-cpu
you make it an Elf64_Nhdr pointer? I realize it's just used as a address
value, but it's just not right. (BTW, compile crash with "make warn"
instead of just "make")
Anyway, you can conceivably have these i386 combinations:
1. single-cpu, a 32-bit ELF header, and 1 prstatus note (netdump or kdump)
2. single-cpu, a 64-bit ELF header, and 1 prstatus note (kdump only)
3. multiple-cpu, a 32-bit ELF header, and 1 prstatus note (netdump only)
4. multiple-cpu, a 32-bit ELF header, and more-than-one prstatus note (kdump)
5. multiple-cpu, a 64-bit ELF header, and more-than-one prstatus note (kdump)
But below you just default to an Elf32_Nhdr structure:
+
+ len = sizeof(Elf32_Nhdr);
+ len = roundup(len + note->n_namesz, 4);
+ pt_regs = (void *)((char *)note + len +
+ MEMBER_OFFSET("elf_prstatus", "pr_reg"));
+ /* NEED TO BE FIXED: Hack to get the proper alignment */
+ pt_regs +=4;
+ }
+ else
+ error(FATAL, "\n cannot determine register set for the task
%s"
+ "bailing out\n ", tc->comm);
+ return pt_regs;
You'll first need to determine whether it's a 32-bit or 64-bit header
being used, and then whether to to point to the single prstatus note or
to the correct one in the NR_CPUS-bounded array of prstatus notes.
Dave
--
Crash-utility mailing list
Crash-utility(a)redhat.com
https://www.redhat.com/mailman/listinfo/crash-utility