LTTng: global_dirty_limit symbol lookup failed
A message that is not in the archive (sent off-list or not yet received): 20130125172448.GA694@Krystal
Quoted 31 lines
Hi Andrew,
Commit:
commit
b87700e318c27267890cbd6fb5e50b687279131bAuthor: Andrew Gabbasov <[email protected]> Date: Mon Dec 10 11:14:52 2012 -0500Add new kernel probes instrumentation
Add kernel probes for btrfs, compaction, ext4, printk, random, rcu, regmap, rpm, sunrpc, workqueue, writeback.
Signed-off-by Andrew Gabbasov <[email protected]>Signed-off-by Mathieu Desnoyers <[email protected]>has 2 issues:
1) it implements a wrapper around
global_dirty_limit()in probes/. All wrappers should be in /wrapper/, without exception, otherwise the code will quickly become unmaintainable (we need to know what wrappers we can eventually remove if we remove support for older kernels in the future, and we need to be aware of those wrappers for complete testing coverage). I have no problem with having a probes/Makefile using a relative path (e.g.../wrapper/somefile.o) as input if needed.2) It fills my dmesg log output with:
"LTTng:global_dirty_limitsymbol lookup failed." messages. .config follows.
Quoted 5 lines
Can you look into those issues ?
Thanks,
Mathieu
Hi Mathieu,
Sorry for a late response.
I'm sending a separate e-mail with the patch to fix these issues. Here are some details:
1) I moved the wrapper to a separate file wrapper/writeback.h. Since it is supposed to be used in writeback probe only, I think, using static variable will be harmless.
2) Since global_dirty_limit is not a function, but a global variable, its address is available to kallsyms lookup only if CONFIG_KALLSYMS_ALL is set. In order not to introduce wrapper related checks of this config into tracepoint definition include file, I decided not to exclude separate tracepoints in case of this config absence, but to omit the whole probe from building. So, writeback probe will be available only if CONFIG_KALLSYMS_ALL is set.
3) Additional issue that I noticed: in case of CONFIG_THUMB2_KERNEL setting for ARM, kallsyms_lookup_funcptr modifies the address found for the symbol. In order to avoid it for data addresses, I introduced a new kallsyms_lookup_dataptr function, that does not do any address modifications. And this new function is used for global_dirty_limit wrapper.
Thanks.
Best regards,
Andrew
A message that is not in the archive (sent off-list or not yet received): 6C5EA58090A5ED459815C4D04C2B466F8C56DAF3@EU-MBX-01.mgc.mentorg.com
Quoted 63 lines
From: Gabbasov, Andrew
Sent: Monday, February 18, 2013 14:14
To: Mathieu Desnoyers
Cc: [email protected]
Subject: RE: LTTng:global_dirty_limitsymbol lookup failedQuoted 31 lines
Hi Andrew,
Commit:
commit
b87700e318c27267890cbd6fb5e50b687279131bAuthor: Andrew Gabbasov <[email protected]> Date: Mon Dec 10 11:14:52 2012 -0500Add new kernel probes instrumentation
Add kernel probes for btrfs, compaction, ext4, printk, random, rcu, regmap, rpm, sunrpc, workqueue, writeback.
Signed-off-by Andrew Gabbasov <[email protected]>Signed-off-by Mathieu Desnoyers <[email protected]>has 2 issues:
1) it implements a wrapper around
global_dirty_limit()in probes/. All wrappers should be in /wrapper/, without exception, otherwise the code will quickly become unmaintainable (we need to know what wrappers we can eventually remove if we remove support for older kernels in the future, and we need to be aware of those wrappers for complete testing coverage). I have no problem with having a probes/Makefile using a relative path (e.g.../wrapper/somefile.o) as input if needed.2) It fills my dmesg log output with:
"LTTng:global_dirty_limitsymbol lookup failed." messages. .config follows.Quoted 5 lines
Can you look into those issues ?
Thanks,
Mathieu
Hi Mathieu,
Sorry for a late response.
I'm sending a separate e-mail with the patch to fix these issues. Here are some details:
1) I moved the wrapper to a separate file
wrapper/writeback.h. Since it is supposed to be used in writeback probe only, I think, using static variable will be harmless.2) Since
global_dirty_limitis not a function, but a global variable, its address is available to kallsyms lookup only ifCONFIG_KALLSYMS_ALLis set. In order not to introduce wrapper related checks of this config into tracepoint definition include file, I decided not to exclude separate tracepoints in case of this config absence, but to omit the whole probe from building. So, writeback probe will be available only ifCONFIG_KALLSYMS_ALLis set.3) Additional issue that I noticed: in case of
CONFIG_THUMB2_KERNELsetting for ARM,kallsyms_lookup_funcptrmodifies the address found for the symbol. In order to avoid it for data addresses, I introduced a newkallsyms_lookup_dataptrfunction, that does not do any address modifications. And this new function is used forglobal_dirty_limitwrapper.
Quoted 4 lines
Thanks.
Best regards,
Andrew
Hi Mathieu,
Any thoughts on these changes?
Do you need anything else regarding it (the patch itself was sent separately on Feb 18, 2013, with a subject "[PATCH lttng-modules] Clean up using global_dirty_limit wrapper for writeback probe").
Thanks.
Best regards,
Andrew
A message that is not in the archive (sent off-list or not yet received): 6C5EA58090A5ED459815C4D04C2B466FD931DEDB@EU-MBX-03.mgc.mentorg.com
Quoted 68 lines · * Gabbasov, Andrew ([email protected]) wrote:
Quoted 63 lines
From: Gabbasov, Andrew
Sent: Monday, February 18, 2013 14:14
To: Mathieu Desnoyers
Cc: [email protected]
Subject: RE: LTTng:global_dirty_limitsymbol lookup failedQuoted 31 lines
Hi Andrew,
Commit:
commit
b87700e318c27267890cbd6fb5e50b687279131bAuthor: Andrew Gabbasov <[email protected]> Date: Mon Dec 10 11:14:52 2012 -0500Add new kernel probes instrumentation
Add kernel probes for btrfs, compaction, ext4, printk, random, rcu, regmap, rpm, sunrpc, workqueue, writeback.
Signed-off-by Andrew Gabbasov <[email protected]>Signed-off-by Mathieu Desnoyers <[email protected]>has 2 issues:
1) it implements a wrapper around
global_dirty_limit()in probes/. All wrappers should be in /wrapper/, without exception, otherwise the code will quickly become unmaintainable (we need to know what wrappers we can eventually remove if we remove support for older kernels in the future, and we need to be aware of those wrappers for complete testing coverage). I have no problem with having a probes/Makefile using a relative path (e.g.../wrapper/somefile.o) as input if needed.2) It fills my dmesg log output with:
"LTTng:global_dirty_limitsymbol lookup failed." messages. .config follows.Quoted 5 lines
Can you look into those issues ?
Thanks,
Mathieu
Hi Mathieu,
Sorry for a late response.
I'm sending a separate e-mail with the patch to fix these issues. Here are some details:
1) I moved the wrapper to a separate file
wrapper/writeback.h. Since it is supposed to be used in writeback probe only, I think, using static variable will be harmless.2) Since
global_dirty_limitis not a function, but a global variable, its address is available to kallsyms lookup only ifCONFIG_KALLSYMS_ALLis set. In order not to introduce wrapper related checks of this config into tracepoint definition include file, I decided not to exclude separate tracepoints in case of this config absence, but to omit the whole probe from building. So, writeback probe will be available only ifCONFIG_KALLSYMS_ALLis set.3) Additional issue that I noticed: in case of
CONFIG_THUMB2_KERNELsetting for ARM,kallsyms_lookup_funcptrmodifies the address found for the symbol. In order to avoid it for data addresses, I introduced a newkallsyms_lookup_dataptrfunction, that does not do any address modifications. And this new function is used forglobal_dirty_limitwrapper.Quoted 4 lines
Thanks.
Best regards,
Andrew
Quoted 6 lines
Hi Mathieu,
Any thoughts on these changes?
Do you need anything else regarding it (the patch itself was sent separately on Feb 18, 2013, with a subject "[PATCH lttng-modules] Clean up usingglobal_dirty_limitwrapper for writeback probe").
All good! Sorry for the delay, the last months have been hectic :)
Thanks,
Mathieu
Quoted 5 lines
Thanks.
Best regards,
Andrew
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com