From: Peter Zijlstra <peterz@infradead.org> To: Will Deacon <will.deacon@arm.com> Cc: Young Xiao <92siuyang@gmail.com>, linux@armlinux.org.uk, mark.rutland@arm.com, mingo@redhat.com, bp@alien8.de, hpa@zytor.com, x86@kernel.org, kan.liang@linux.intel.com, linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org, ravi.bangoria@linux.vnet.ibm.com, mpe@ellerman.id.au, acme@redhat.com, eranian@google.com, fweisbec@gmail.com, jolsa@redhat.com Subject: Re: [PATCH] perf: Fix oops when kthread execs user process Date: Wed, 29 May 2019 18:44:07 +0200 [thread overview] Message-ID: <20190529164407.GA2623@hirez.programming.kicks-ass.net> (raw) In-Reply-To: <20190529162528.GB12420@fuggles.cambridge.arm.com> On Wed, May 29, 2019 at 05:25:28PM +0100, Will Deacon wrote: > > > > On Wed, May 29, 2019 at 02:05:21PM +0100, Will Deacon wrote: > > > > > On Wed, May 29, 2019 at 02:55:57PM +0200, Peter Zijlstra wrote: > > > > > > > > > > if (user_mode(regs)) { > > > > > > > > > > Hmm, so it just occurred to me that Mark's observation is that the regs > > > > > can be junk in some cases. In which case, should we be checking for > > > > > kthreads first? > Sorry, I'm not trying to catch you out! Just trying to understand what the > semantics are supposed to be. > > I do find the concept of user_mode(regs) bizarre for the idle task. By the > above, we definitely have a bug on arm64 (user_mode(regs) tends to be > true for the idle task), and I couldn't figure out how you avoided it on > x86. I guess it happens to work because the stack is zero-initialised or > something? So lets take the whole thing: static void perf_sample_regs_user(struct perf_regs *regs_user, struct pt_regs *regs, struct pt_regs *regs_user_copy) { if (user_mode(regs)) { regs_user->abi = perf_reg_abi(current); regs_user->regs = regs; } else if (!(current->flags & PF_KTHREAD)) { perf_get_regs_user(regs_user, regs, regs_user_copy); } else { regs_user->abi = PERF_SAMPLE_REGS_ABI_NONE; regs_user->regs = NULL; } } This is called from the perf-generate-a-sample path, which is typically an exception (IRQ/NMI/whatever) or a software/tracepoint thing. In the exception case, the @regs argument are the exception register, as provided by your entry.S to your exception handlers. In the software/tracepoint thing, it is the result of perf_arch_fetch_caller_regs(). So @regs is always 'sane' and user_mode(regs) tells us if the exception came from userspace (and software/tracepoints always fail this, they 'obviously' don't come from userspace). If we're idle, we're not from userspace, so this branch doesn't matter. Next, we test if there is a userspace part _at_all_, this is the newly minted: '!(current->flags & PF_KTHREAD)', if that passes, we use architecture magic -- task_pt_regs() -- to get the user-regs. This can be crap. But since the idle task will always fail our test (as would the old one, idle->mm is always NULL), we'll never get here for idle. Then failing the above two, as we must for idle, we'll default to ABI_NONE/NULL. Does that help?
WARNING: multiple messages have this Message-ID (diff)
From: Peter Zijlstra <peterz@infradead.org> To: Will Deacon <will.deacon@arm.com> Cc: mark.rutland@arm.com, Young Xiao <92siuyang@gmail.com>, mpe@ellerman.id.au, jolsa@redhat.com, x86@kernel.org, linux@armlinux.org.uk, eranian@google.com, linux-kernel@vger.kernel.org, acme@redhat.com, mingo@redhat.com, bp@alien8.de, hpa@zytor.com, ravi.bangoria@linux.vnet.ibm.com, fweisbec@gmail.com, linux-arm-kernel@lists.infradead.org, kan.liang@linux.intel.com Subject: Re: [PATCH] perf: Fix oops when kthread execs user process Date: Wed, 29 May 2019 18:44:07 +0200 [thread overview] Message-ID: <20190529164407.GA2623@hirez.programming.kicks-ass.net> (raw) In-Reply-To: <20190529162528.GB12420@fuggles.cambridge.arm.com> On Wed, May 29, 2019 at 05:25:28PM +0100, Will Deacon wrote: > > > > On Wed, May 29, 2019 at 02:05:21PM +0100, Will Deacon wrote: > > > > > On Wed, May 29, 2019 at 02:55:57PM +0200, Peter Zijlstra wrote: > > > > > > > > > > if (user_mode(regs)) { > > > > > > > > > > Hmm, so it just occurred to me that Mark's observation is that the regs > > > > > can be junk in some cases. In which case, should we be checking for > > > > > kthreads first? > Sorry, I'm not trying to catch you out! Just trying to understand what the > semantics are supposed to be. > > I do find the concept of user_mode(regs) bizarre for the idle task. By the > above, we definitely have a bug on arm64 (user_mode(regs) tends to be > true for the idle task), and I couldn't figure out how you avoided it on > x86. I guess it happens to work because the stack is zero-initialised or > something? So lets take the whole thing: static void perf_sample_regs_user(struct perf_regs *regs_user, struct pt_regs *regs, struct pt_regs *regs_user_copy) { if (user_mode(regs)) { regs_user->abi = perf_reg_abi(current); regs_user->regs = regs; } else if (!(current->flags & PF_KTHREAD)) { perf_get_regs_user(regs_user, regs, regs_user_copy); } else { regs_user->abi = PERF_SAMPLE_REGS_ABI_NONE; regs_user->regs = NULL; } } This is called from the perf-generate-a-sample path, which is typically an exception (IRQ/NMI/whatever) or a software/tracepoint thing. In the exception case, the @regs argument are the exception register, as provided by your entry.S to your exception handlers. In the software/tracepoint thing, it is the result of perf_arch_fetch_caller_regs(). So @regs is always 'sane' and user_mode(regs) tells us if the exception came from userspace (and software/tracepoints always fail this, they 'obviously' don't come from userspace). If we're idle, we're not from userspace, so this branch doesn't matter. Next, we test if there is a userspace part _at_all_, this is the newly minted: '!(current->flags & PF_KTHREAD)', if that passes, we use architecture magic -- task_pt_regs() -- to get the user-regs. This can be crap. But since the idle task will always fail our test (as would the old one, idle->mm is always NULL), we'll never get here for idle. Then failing the above two, as we must for idle, we'll default to ABI_NONE/NULL. Does that help? _______________________________________________ linux-arm-kernel mailing list linux-arm-kernel@lists.infradead.org http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
next prev parent reply other threads:[~2019-05-29 16:44 UTC|newest] Thread overview: 60+ messages / expand[flat|nested] mbox.gz Atom feed top 2019-05-28 12:31 [PATCH] perf: Fix oops when kthread execs user process Young Xiao 2019-05-28 12:31 ` Young Xiao 2019-05-28 12:41 ` Russell King - ARM Linux admin 2019-05-28 12:41 ` Russell King - ARM Linux admin 2019-05-28 14:01 ` Peter Zijlstra 2019-05-28 14:01 ` Peter Zijlstra 2019-05-28 15:32 ` Will Deacon 2019-05-28 15:32 ` Will Deacon 2019-05-28 16:12 ` Mark Rutland 2019-05-28 16:12 ` Mark Rutland 2019-05-28 17:32 ` Peter Zijlstra 2019-05-28 17:32 ` Peter Zijlstra 2019-05-29 9:17 ` Will Deacon 2019-05-29 9:17 ` Will Deacon 2019-05-29 10:10 ` Peter Zijlstra 2019-05-29 10:10 ` Peter Zijlstra 2019-05-29 10:20 ` Will Deacon 2019-05-29 10:20 ` Will Deacon 2019-05-29 12:55 ` Peter Zijlstra 2019-05-29 12:55 ` Peter Zijlstra 2019-05-29 13:05 ` Will Deacon 2019-05-29 13:05 ` Will Deacon 2019-05-29 13:25 ` Peter Zijlstra 2019-05-29 13:25 ` Peter Zijlstra 2019-05-29 14:35 ` Will Deacon 2019-05-29 14:35 ` Will Deacon 2019-05-29 16:19 ` Peter Zijlstra 2019-05-29 16:19 ` Peter Zijlstra 2019-05-29 16:24 ` Mark Rutland 2019-05-29 16:24 ` Mark Rutland 2019-05-29 16:38 ` Mark Rutland 2019-05-29 16:38 ` Mark Rutland 2019-05-29 17:03 ` Peter Zijlstra 2019-05-29 17:03 ` Peter Zijlstra 2019-05-30 10:35 ` Mark Rutland 2019-05-30 10:35 ` Mark Rutland 2019-05-29 16:25 ` Will Deacon 2019-05-29 16:25 ` Will Deacon 2019-05-29 16:44 ` Peter Zijlstra [this message] 2019-05-29 16:44 ` Peter Zijlstra 2019-05-30 7:28 ` Will Deacon 2019-05-30 7:28 ` Will Deacon 2019-05-30 8:38 ` Ravi Bangoria 2019-05-30 8:38 ` Ravi Bangoria 2019-05-30 10:27 ` Ravi Bangoria 2019-05-30 10:27 ` Ravi Bangoria 2019-05-31 15:37 ` Will Deacon 2019-05-31 15:37 ` Will Deacon 2019-06-03 11:23 ` Will Deacon 2019-06-03 11:23 ` Will Deacon 2019-06-03 11:48 ` Peter Zijlstra 2019-06-03 11:48 ` Peter Zijlstra 2019-06-03 13:30 ` Michael Ellerman 2019-06-03 13:30 ` Michael Ellerman 2019-05-29 10:11 ` Mark Rutland 2019-05-29 10:11 ` Mark Rutland 2019-05-29 4:21 ` Michael Ellerman 2019-05-29 4:21 ` Michael Ellerman 2019-05-29 1:44 ` Michael Ellerman 2019-05-29 1:44 ` Michael Ellerman
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20190529164407.GA2623@hirez.programming.kicks-ass.net \ --to=peterz@infradead.org \ --cc=92siuyang@gmail.com \ --cc=acme@redhat.com \ --cc=bp@alien8.de \ --cc=eranian@google.com \ --cc=fweisbec@gmail.com \ --cc=hpa@zytor.com \ --cc=jolsa@redhat.com \ --cc=kan.liang@linux.intel.com \ --cc=linux-arm-kernel@lists.infradead.org \ --cc=linux-kernel@vger.kernel.org \ --cc=linux@armlinux.org.uk \ --cc=mark.rutland@arm.com \ --cc=mingo@redhat.com \ --cc=mpe@ellerman.id.au \ --cc=ravi.bangoria@linux.vnet.ibm.com \ --cc=will.deacon@arm.com \ --cc=x86@kernel.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.