Re: [clear_page] 0ad07c8104 BUG: unable to handle kernel NULL pointer dereference at 0000000000000040
On Mon, Feb 20, 2017 at 11:03:52AM +0800, Fengguang Wu wrote: > We actually already test LKML patch in that way (Xiaolong maintains > this feature). Nevertheless if developers specify "base-commit:" it > could help eliminate the guessing works by the dumb robot. We'll > appreciate if the "base-commit:" or "base-patchid:" tags are listed > in the patches, especially in some non-obvious situations. Can I specify multiple base commits for testing stable backports? For example base-commit: v4.9.11, v4.4.50, v3.10.105,... > Such tags could be regarded as "explicit" test requests, where we could > send "BUILD COMPLETE" emails as a response (comparing to our normal > LKML patch tests, which only build regressions will trigger an email > notification). Yap, similar to those you guys sent when a new branch on k.org has been tested. Btw, can we make the format layout this way: patch --- <0day bot tags> --- so that when we send it to lkml, it doesn't interfere with review by slapping the tags at the beginning of the patch? Also, should we CC some special mailing list which the 0day bot parses or lkml is enough? Cool stuff. -- Regards/Gruss, Boris. SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg) --
Re: [clear_page] 0ad07c8104 BUG: unable to handle kernel NULL pointer dereference at 0000000000000040
Hi Borislav, On Sun, Feb 19, 2017 at 02:50:19PM +0100, Borislav Petkov wrote: On Sun, Feb 19, 2017 at 09:06:51AM +0800, Fengguang Wu wrote: Yes if we add it as a line below the branch URL, it could be a time saver. Right. Since it's hard to teach ALL people about the rule, it'd be best if we can work w/o any rules -- unless you want to be accurate&helpful or to customize test behaviors. Since we already tested the original patch/commit (hence the report), we should know where the fixup should be applied to. And it'd be reasonably easy to tell whether the fix is incremental or a replacement -- just try git-am onto the original commit first, if failed, continue to try the parent commit. For old bugs the fix could be against linus/master or linux-next/master, which could be tried too. Makes sense. Yes, that'd be most convenient. In general the email interface could be something like this: # "key: value" fields; if you Re: to an earlier bug report, they can be auto retrieved compiler: gcc-6 # optional base-commit: v4.10-rc8 # the robot knows kernel commits from hundreds of public git trees --- the patch --- attach kconfig files Yap, just stick those rules somewhere on a website. OK, will do when the feature is ready. According to Xiaolong, the automated test-of-fixup-patches feature is already in our plan. For introductions of the now-working build/boot test services and instructions on customization, we could probably add some markdown document under https://github.com/fengguang/lkp-tests/tree/master/doc Philip/Ying, what do you think? I can draft it. Btw, this is not only useful for a follow-on, fix patch but also for initial test request. For example, I want to backport patch to stable and would like to run it on a bunch of kernels: base-commit: v4.4-stable, v4.9-stable, ... i.e., a list of trees to apply it to. I believe people might like this a lot. Or, for example, a patch touching a bunch of arches and author doesn't necessarily have access to all those different toolchains. Shoot a mail to the 0day bot: base-commit: linus/master arch: x86_64, powerpc, sparc, ... Would be very useful too. We actually already test LKML patch in that way (Xiaolong maintains this feature). Nevertheless if developers specify "base-commit:" it could help eliminate the guessing works by the dumb robot. We'll appreciate if the "base-commit:" or "base-patchid:" tags are listed in the patches, especially in some non-obvious situations. Such tags could be regarded as "explicit" test requests, where we could send "BUILD COMPLETE" emails as a response (comparing to our normal LKML patch tests, which only build regressions will trigger an email notification). Anyway, just a couple of ideas. What would also be cool if you guys had a 0day bot howto with all those things we should pay attention to and we can go and look up. OK. Your ideas are very welcome, thanks! Regards, Fengguang
Re: [clear_page] 0ad07c8104 BUG: unable to handle kernel NULL pointer dereference at 0000000000000040
On Sun, Feb 19, 2017 at 09:06:51AM +0800, Fengguang Wu wrote: > Yes if we add it as a line below the branch URL, it could be a time saver. Right. > Since it's hard to teach ALL people about the rule, it'd be best if we > can work w/o any rules -- unless you want to be accurate&helpful or to > customize test behaviors. > > Since we already tested the original patch/commit (hence the report), > we should know where the fixup should be applied to. And it'd be > reasonably easy to tell whether the fix is incremental or a > replacement -- just try git-am onto the original commit first, if > failed, continue to try the parent commit. For old bugs the fix could > be against linus/master or linux-next/master, which could be tried too. Makes sense. > Yes, that'd be most convenient. In general the email interface could > be something like this: > ># "key: value" fields; if you Re: to an earlier bug report, they can > be auto retrieved >compiler: gcc-6 # optional >base-commit: v4.10-rc8 # the robot knows kernel commits from hundreds > of public git trees >--- >the patch >--- >attach kconfig files Yap, just stick those rules somewhere on a website. Btw, this is not only useful for a follow-on, fix patch but also for initial test request. For example, I want to backport patch to stable and would like to run it on a bunch of kernels: base-commit: v4.4-stable, v4.9-stable, ... i.e., a list of trees to apply it to. I believe people might like this a lot. Or, for example, a patch touching a bunch of arches and author doesn't necessarily have access to all those different toolchains. Shoot a mail to the 0day bot: base-commit: linus/master arch: x86_64, powerpc, sparc, ... Would be very useful too. Anyway, just a couple of ideas. What would also be cool if you guys had a 0day bot howto with all those things we should pay attention to and we can go and look up. Thanks! -- Regards/Gruss, Boris. SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg) --
Re: [clear_page] 0ad07c8104 BUG: unable to handle kernel NULL pointer dereference at 0000000000000040
On Sun, Feb 19, 2017 at 01:10:47AM +0100, Borislav Petkov wrote: Hey Fengguang, On Sun, Feb 19, 2017 at 07:29:50AM +0800, Fengguang Wu wrote: Good point! I noticed it too while sending out the report. It'll be showed as this in future: https://github.com/0day-ci/linux/commits/Borislav-Petkov/x86-Optimize-clear_page/20170210-053052 How about pointing to the patch directly? https://github.com/0day-ci/linux/commit/0ad07c8104eb5c12dfcb86581c1cc657183496cc Yes if we add it as a line below the branch URL, it could be a time saver. Sorry the 2nd report was send out manually and I only checked the emails in my _current_ mbox. Since the previous report email has been archived, it slipped through the duplication check. No worries - this was all a prelude to me hinting at the email-based talking to the bot :-) CC Xiaolong. It's possible to automate the test-of-fixup-patches. Firstly find out the original email report by the Message-ID being replied to. Then fetch all the information required for deciding where the patch should be applied to, parameters to auto-testing the patch. Sounds like a plan. It would probably even be easier for the bot if the reply-mail contained specially-formatted hints like: TEST-WITH-BELOW-PATCH: ... or so. Since it's hard to teach ALL people about the rule, it'd be best if we can work w/o any rules -- unless you want to be accurate&helpful or to customize test behaviors. Since we already tested the original patch/commit (hence the report), we should know where the fixup should be applied to. And it'd be reasonably easy to tell whether the fix is incremental or a replacement -- just try git-am onto the original commit first, if failed, continue to try the parent commit. For old bugs the fix could be against linus/master or linux-next/master, which could be tried too. Btw, another nice aspect of this talking back to the bot is that before I, as a recipient of the bug report, go and try to prepare a guest or find a machine to reproduce properly, I can send a quick diff to the bot in the meantime and say, "try this on the guest. I have a hunch it might fix it." Yes, that'd be most convenient. In general the email interface could be something like this: # "key: value" fields; if you Re: to an earlier bug report, they can be auto retrieved compiler: gcc-6 # optional base-commit: v4.10-rc8 # the robot knows kernel commits from hundreds of public git trees --- the patch --- attach kconfig files Yeah we have a TODO to do email based on-demand service, which looks close to your proposal. Cool. Ping me if you need testers. Thanks! Thanks, Fengguang
Re: [clear_page] 0ad07c8104 BUG: unable to handle kernel NULL pointer dereference at 0000000000000040
Hey Fengguang, On Sun, Feb 19, 2017 at 07:29:50AM +0800, Fengguang Wu wrote: > Good point! I noticed it too while sending out the report. It'll be > showed as this in future: > > > https://github.com/0day-ci/linux/commits/Borislav-Petkov/x86-Optimize-clear_page/20170210-053052 How about pointing to the patch directly? https://github.com/0day-ci/linux/commit/0ad07c8104eb5c12dfcb86581c1cc657183496cc > Sorry the 2nd report was send out manually and I only checked the > emails in my _current_ mbox. Since the previous report email has been > archived, it slipped through the duplication check. No worries - this was all a prelude to me hinting at the email-based talking to the bot :-) > CC Xiaolong. It's possible to automate the test-of-fixup-patches. > Firstly find out the original email report by the Message-ID being > replied to. Then fetch all the information required for deciding where > the patch should be applied to, parameters to auto-testing the patch. Sounds like a plan. It would probably even be easier for the bot if the reply-mail contained specially-formatted hints like: TEST-WITH-BELOW-PATCH: ... or so. Btw, another nice aspect of this talking back to the bot is that before I, as a recipient of the bug report, go and try to prepare a guest or find a machine to reproduce properly, I can send a quick diff to the bot in the meantime and say, "try this on the guest. I have a hunch it might fix it." > Yeah we have a TODO to do email based on-demand service, which looks > close to your proposal. Cool. Ping me if you need testers. Thanks! -- Regards/Gruss, Boris. SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg) --
Re: [clear_page] 0ad07c8104 BUG: unable to handle kernel NULL pointer dereference at 0000000000000040
Hi Borislav, On Sat, Feb 18, 2017 at 03:48:00PM +0100, Borislav Petkov wrote: Guys, please fix the 0day bot reporting. See below for more info. On Sat, Feb 18, 2017 at 01:01:53PM +0800, Fengguang Wu wrote: Greetings, 0day kernel testing robot got the below dmesg and the first bad commit is https://github.com/0day-ci/linux Borislav-Petkov/x86-Optimize-clear_page/20170210-053052 Can you make this point to the actual commit on github? Your tree there has currently 47K branches and navigating through the web interface takes forever. Good point! I noticed it too while sending out the report. It'll be showed as this in future: https://github.com/0day-ci/linux/commits/Borislav-Petkov/x86-Optimize-clear_page/20170210-053052 commit 0ad07c8104eb5c12dfcb86581c1cc657183496cc Author: Borislav Petkov AuthorDate: Thu Feb 9 20:51:25 2017 +0100 Commit: 0day robot CommitDate: Fri Feb 10 05:30:58 2017 +0800 x86: Optimize clear_page() Currently, we CALL clear_page() which then JMPs to the proper function chosen by the alternatives. What we should do instead is CALL the proper function directly. (This was something Ingo suggested a while ago). So let's do that. Measuring our favourite kernel build workload shows that there are no significant changes in performance. ... Signed-off-by: Borislav Petkov Then, this is an old patch. You already sent me a bug report, I replied with a fix but you didn't test the fix. Instead you're sending the same old report. Sorry the 2nd report was send out manually and I only checked the emails in my _current_ mbox. Since the previous report email has been archived, it slipped through the duplication check. Here's the fix: https://lists.01.org/pipermail/lkp/2017-February/005573.html And the upstream submission of the new version: https://lkml.kernel.org/r/20170215111927.emdgxf2pide3k...@pd.tnic Please fix the bot to pay attention to replies. If there is a special way I should reply with a fix so that the bot retests with the same config, please let me know. CC Xiaolong. It's possible to automate the test-of-fixup-patches. Firstly find out the original email report by the Message-ID being replied to. Then fetch all the information required for deciding where the patch should be applied to, parameters to auto-testing the patch. In general, I think it would be a very cool idea to be able to reply to the bot and say, "Dear bot, can you test this fix ontop with the exact same guest, vm, kernel .config etc. That would be lovely. Yeah we have a TODO to do email based on-demand service, which looks close to your proposal. Thanks and keep up the good work! Thanks, Fengguang
Re: [clear_page] 0ad07c8104 BUG: unable to handle kernel NULL pointer dereference at 0000000000000040
Guys, please fix the 0day bot reporting. See below for more info. On Sat, Feb 18, 2017 at 01:01:53PM +0800, Fengguang Wu wrote: > Greetings, > > 0day kernel testing robot got the below dmesg and the first bad commit is > > https://github.com/0day-ci/linux > Borislav-Petkov/x86-Optimize-clear_page/20170210-053052 Can you make this point to the actual commit on github? Your tree there has currently 47K branches and navigating through the web interface takes forever. > commit 0ad07c8104eb5c12dfcb86581c1cc657183496cc > Author: Borislav Petkov > AuthorDate: Thu Feb 9 20:51:25 2017 +0100 > Commit: 0day robot > CommitDate: Fri Feb 10 05:30:58 2017 +0800 > > x86: Optimize clear_page() > > Currently, we CALL clear_page() which then JMPs to the proper function > chosen by the alternatives. > > What we should do instead is CALL the proper function directly. (This > was something Ingo suggested a while ago). So let's do that. > > Measuring our favourite kernel build workload shows that there are no > significant changes in performance. > > ... > > Signed-off-by: Borislav Petkov Then, this is an old patch. You already sent me a bug report, I replied with a fix but you didn't test the fix. Instead you're sending the same old report. Here's the fix: https://lists.01.org/pipermail/lkp/2017-February/005573.html And the upstream submission of the new version: https://lkml.kernel.org/r/20170215111927.emdgxf2pide3k...@pd.tnic Please fix the bot to pay attention to replies. If there is a special way I should reply with a fix so that the bot retests with the same config, please let me know. In general, I think it would be a very cool idea to be able to reply to the bot and say, "Dear bot, can you test this fix ontop with the exact same guest, vm, kernel .config etc. That would be lovely. Thanks and keep up the good work! -- Regards/Gruss, Boris. SUSE Linux GmbH, GF: Felix Imendörffer, Jane Smithard, Graham Norton, HRB 21284 (AG Nürnberg) --