Re: [PATCH] m68knommu: add pretty back strace

From: Greg Ungerer
Date: Tue May 06 2008 - 00:52:51 EST


Hi Paulo,

Paulo Marques wrote:
Sebastian Siewior wrote:
* Paulo Marques | 2008-05-02 20:27:57 [+0100]:

Greg Ungerer wrote:
From: Sebastian Siewior <bigeasy@xxxxxxxxxxxxx>
With this patch and
CONFIG_FRAME_POINTER=y
CONFIG_KALLSYMS=y
The backtrace shows resolved function names and their numeric
address.
This is really not my area, but this patch reminds me of all the dwarf2 unwinder on x86 that caused so many problems in the beginning...

[...]
+#ifdef CONFIG_FRAME_POINTER
+ printk(KERN_EMERG "Call Trace:\n");
+
+ last_stack = stack - 1;
+ while (stack <= endstack && stack > last_stack) {
+
+ addr = *(stack + 1);
+ printk(KERN_EMERG " [%08lx] ", addr);
+ print_symbol(KERN_CONT "%s\n", addr);
+
+ last_stack = stack;
+ stack = (unsigned long *)*stack;
}
printk("\n");
+#else
+ printk(KERN_EMERG "CONFIG_FRAME_POINTER disabled, no symbolic call trace\n");
You could probably fall back to the old method in this case, no?
The old method printed every value from stack which was in the text
range (you didn't get modules AFAIK). This might be the caller as well a
function pointer as argument as well something else. I tried to find to
find a pattern without frame pointers but I had no luck. I thing the
caller fixed the stack frame or something.
Greg did not complain about removing it. If you or others want the old
method in case of no frame pointers I can send a patch.

Having output garbled with "false positive" addresses, but that have the "true positives" in between is better than no output at all.

Yes that is a reasonable argument.

But you still get the full stack hex dump, so all the raw information
is there still. Just not in a decoded form. (I tended to use the raw
dump more often myself anyway).


Also, if the stack is slightly corrupted on the top, the new method might just bail out without giving any indication about the path that lead there, when instead it could also fall back to the old method.
You mean by slightly that the first caller ORed the address with
something?

Or more commonly, a buffer overflow... (or underflow, depending on how the stack grows)

In that case we don't return safely. I don't know how I could
find out the right time for a fallback (in case of slightly corrupted
stack).

I was imagining you would get a page fault where you'd show a stack trace, but that's on a MMU processor. With no MMU I guess you don't get a chance to do that :(

As I said before, this is really not my area. I just remember that a very similar change in x86 was a real pain, and wanted to make sure that everyone here was aware of that.

I would have no problem with the original full symbolic decode
being in as a fall back for the non FRAME_POINTER case.

Regards
Greg



------------------------------------------------------------------------
Greg Ungerer -- Chief Software Dude EMAIL: gerg@xxxxxxxxxxxx
Secure Computing Corporation PHONE: +61 7 3435 2888
825 Stanley St, FAX: +61 7 3891 3630
Woolloongabba, QLD, 4102, Australia WEB: http://www.SnapGear.com
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/