6:44 a.m.
Hi Dave et al.
thanks to the feedback, I have reworked my patch set. It is now much
larger than the first version, but I'm also much more happy with it.
I'm pretty sure that there are still some issues here and there, so
as always, your comments are most welcome.
Petr Tesarik
6:45 a.m.
New subject: [PATCH 1/11] Move cmd_p() main processing into its own function
Move gdb output filtering to its own function so that it can be reused
from other places.
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
6:47 a.m.
New subject: [PATCH 2/11] New convenience function: get_cpumask_buf
Make a convenience function for allocating a cpumask buffer and
put it into tools.c. Since any caller of make_cpumask() needs
such a buffer, it is better to isolate the code in a function
rather than copy it all over the place.
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
6:48 a.m.
New subject: [PATCH 3/11] Implement percpu variables for "p"
Add per-cpu specifications to the "p" command. This allows to specify
CPUs for a per-cpu symbol after a colon.
Keep in mind that a colon can be part of a valid C expression, e.g. ':'
or the ternary operator, so if the part before the colon is not a
symbol, the colon must be restored. Fortunately, a symbol name followed
by a colon cannot start a valid C expression, so there is never any
ambiguity.
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
6:49 a.m.
New subject: [PATCH 4/11] Fix percpu handling for symbols without debuginfo
When trying to show a percpu symbol without debuginfo, crash fails like
this:
crash> p fcoe_percpu:0
p: gdb request failed: p *(<data variable, no debug info>*)
0xffff88011e2d4900
This is because gdb does not really fail if the type is unknown:
crash> gdb whatis fcoe_percpu
type = <data variable, no debug info>
crash> gdb info variables fcoe_percpu
All variables matching regular expression "fcoe_percpu":
Non-debugging symbols:
0x0000000000014900 fcoe_percpu
Of course, the actual contents cannot be shown in this case, but crash
should at least display the address:
crash> p fcoe_percpu:0
&per_cpu(fcoe_percpu, 0) = $1 = (void *) 0xffff88011e214900
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
6:49 a.m.
New subject: [PATCH 5/11] Recognize "a" and "all" as CPU specifiers
Extend the syntax parsed by make_cpuset() so that "a" or "all" is
an alias for all CPUs in the system.
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
6:50 a.m.
New subject: [PATCH 6/11] Convert cmd_datatype_common variables to flags
Convert flag variables in cmd_datatype_common() to flag values, so that
fewer values have to be passed when the main printing loop is converted
to a standalone function.
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
6:50 a.m.
New subject: [PATCH 7/11] Cleanup handling of flags in cmd_datatype_common()
This patch improves handling of struct/union flags:
1. You can now specify a typedef with "*". Previously it failed like
this:
crash> *acct_t ffff8800982ad520
*: invalid argument
2. Report an error if a non-structured type is used, instead of
giving no output, e.g.:
crash> *pid_t
*: invalid argument
3. Produce an error message if the type does not match the command:
crash> union page
union: data type mismatch: page is not a union
crash> struct thread_union
struct: data type mismatch: thread_union is not a struct
Additionaly, these checks are now performed early, that is before
entering the main printing loop.
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
6:51 a.m.
New subject: [PATCH 8/11] Helper function to add member info to struct datatype_member
Instead of parsing the complete type on each iteration, move the code
to a helper function. This might be slightly faster, but most
importantly the main loop doesn't need the "structname" and
"separator"
variables after this change.
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
6:52 a.m.
New subject: [PATCH 9/11] Fix struct -o with an address and array count
Previously, the array count was ignored with the "-o" option.
While I admit that repeating code is not "good style", the goto from
the middle of a loop was not nice either, and this variant removes
a dependency of the main loop on the "aflag" variable.
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
6:53 a.m.
New subject: [PATCH 10/11] Move the main loop of cmd_datatype_common() to its own function
This change allows to reuse the main loop in a different context.
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
6:53 a.m.
New subject: [PATCH 11/11] Add percpu handling to struct/union/* commands
Extend the argument syntax of struct/union/* commands to recognize
percpu specifiers. Note that percpu addresses are not valid kernel
virtual addresses, so they must be parsed separately.
Signed-off-by: Petr Tesarik <ptesarik(a)suse.cz>
10:35 a.m.
----- Original Message -----
Hi Dave et al.
thanks to the feedback, I have reworked my patch set. It is now much
larger than the first version, but I'm also much more happy with it.
I'm pretty sure that there are still some issues here and there, so
as always, your comments are most welcome.
Petr Tesarik
Hi Petr,
I plan to get to this patch-set this afternoon, but I when I apply
them to an unmodified crash-7.0.2 tree, I get a few errors:
[1/11] p-filter-fn.patch
patching file symbols.c
[2/11] get_cpumask_buf.patch
patching file defs.h
patching file kernel.c
patching file tools.c
[3/11] percpu-p.patch
patching file symbols.c
patching file help.c
[4/11] typename-errors.patch
patching file symbols.c
[5/11] cpumask-all.patch
patching file tools.c
patching file help.c
Hunk #1 FAILED at 1089.
Hunk #2 succeeded at 2855 (offset 3 lines).
1 out of 2 hunks FAILED -- saving rejects to file help.c.rej
[6/11] datatype-flags.patch
patching file symbols.c
[7/11] datatype-cleanup.patch
patching file symbols.c
[8/11] datatype-member-offset.patch
patching file symbols.c
[9/11] datatype-offsets.patch
patching file symbols.c
Hunk #1 succeeded at 5872 (offset 1 line).
Hunk #2 succeeded at 5906 (offset 1 line).
Hunk #3 succeeded at 5942 with fuzz 1 (offset -7 lines).
[10/11] datatype-make-fn.patch
patching file symbols.c
Hunk #3 FAILED at 5902.
1 out of 5 hunks FAILED -- saving rejects to file symbols.c.rej
[11/11] percpu-struct.patch
patching file help.c
Hunk #1 succeeded at 4106 (offset 2 lines).
Hunk #2 succeeded at 4141 (offset 2 lines).
Hunk #3 succeeded at 4401 (offset 2 lines).
Hunk #4 succeeded at 4437 (offset 2 lines).
patching file symbols.c
Hunk #6 FAILED at 5944.
Hunk #7 succeeded at 6876 with fuzz 2 (offset 921 lines).
1 out of 7 hunks FAILED -- saving rejects to file symbols.c.rej
I haven't investigated the failures above (which may be obvious), but can you
maybe whip up a single patch against an unmodified crash-7.0.2 tree?
Thanks,
Dave
4054
days inactive
4057
days old
devel@lists.crash-utility.osci.io
12 comments
2 participants
participants (2)
-
Dave Anderson -
Petr Tesarik