RE: crash-4.0-3.14-sym.3.patch (was: modules and data/bss initialization)

["Castor Fu" <castor.fu@xxxxxxxxxxxx>]

Oops, I sent the wrong patch... 

 

I had found the second problem when reading

through the patch, but somehow grabbed the wrong file.  The 'mod -l' thing

seems like it'll take more to fix, but there's another problem which I found and

had fixed.

 

Sections which are not '.text' '.data' or '.bss' will get loaded at the wrong

addresses, which is unfortunate.  The following patch addresses this

by providing addreses for each section.

 

Happy New Year!

 

 -castor

---

From: crash-utility-bounces@xxxxxxxxxx [mailto:crash-utility-bounces@xxxxxxxxxx] On Behalf Of Dave Anderson
Sent: Tuesday, January 02, 2007 1:04 PM
To: Discussion list for crash utility usage, maintenance and development
Subject: Re: crash-4.0-3.14-sym.3.patch (was: modules and data/bss initialization)

Castor Fu wrote:

 

Hi Dave!

Sorry about the regression.

I looked further through the code, and have modified it to only add the symbols

for which it has actually identified a matching section, and to do it in a way

which supports whatever sections are defined.

This doesn't explain why a match isn't being found in the ia64 case, but shouldat least avoid the regression in the s390x case.I've also attached a test which one could run which dumps output which mighthelp me fix things by tracing through loading the 'md5' module. By default the new code is off, as requested, and can be forced witha '-l' option to 'mod', e.g.crash> mod -l -s md5Thanks again for the testing, and hopefully this will work for most people.It would be great if someone could send me output from the rarerplatforms like ppc64 or 390 if it doesn't work.Happy New Year!    -castor

Somebody forgot to QA this patch...    ;-)

Unfortunately, "mod -l -s module-name" won't work, nor will
"mod -l -S", because of the unique way that cmd_mod() handles
-s and -S -- which can optionally take additional "objfile" or "directory"
arguments respectively:

crash> mod -l -S
mod: -S is not a directory
Usage: mod [ -s module [objfile] | -d module | -S [directory] | -D | -r ]
Enter "help mod" for details.
crash> mod -l -s nfs
mod: /usr/lib/debug/lib/modules/2.4.21-37.ELsmp/kernel/fs/nfs: not an ELF format object file
Usage: mod [ -s module [objfile] | -d module | -S [directory] | -D | -r ]
Enter "help mod" for details.
crash>

So to get around that, you can just enter "mod -l" alone
prior to any -s or -S commands, which will just set the
flag.

But having done that, it actually does just the opposite,
because of backwards logic here:

@@ -7288,7 +7410,10 @@
                        strcpy(lm->mod_namelist, namelist);
                else
                        strncpy(lm->mod_namelist, namelist, MAX_MOD_NAMELIST-1);
-               goto add_symbols;
+                if (USE_V2_MOD_SYM()) {
+                        goto add_symbols;
+                }
+
        }

        if ((mbfd = bfd_openr(namelist, NULL)) == NULL)

It should be: "if (!USE_V2_MOD_SYM())".

Anyway, don't bother updating the patch for the above,
because I can keep testing with the two work-arounds.

I'd also appreciate any ppc64, s390 and s390x testers
out there...

Thanks,
  Dave
 

Attachment: crash-4.0-3.16.patch
Description: crash-4.0-3.16.patch

--
Crash-utility mailing list
Crash-utility@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/crash-utility