From: Baoquan He <bhe@redhat.com> To: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org, x86@kernel.org, kexec@lists.infradead.org Subject: Re: kdump kernel randomly hang with tick_periodic call trace on bare metal system Date: Sat, 14 Jan 2023 10:08:53 +0800 [thread overview] Message-ID: <Y8IOtY2x8XQT+Owd@fedora> (raw) In-Reply-To: <87mt6rjrra.ffs@tglx> On 01/09/23 at 10:57pm, Thomas Gleixner wrote: > On Tue, Dec 20 2022 at 13:41, Baoquan He wrote: > > On one intel bare metal system, I can randomly reproduce the kdump hang > > as below with tick_periodic call trace. Attach the kernel config for > > reference. > > This has absolutely nothing to do with x2apic IPI shorthands > > > [ 0.045980] Spurious LAPIC timer interrupt on cpu 0 > > So here the CPU receives a spurious Local APIC timer interrupt, but > that's a red herring. > > > [ 1.152690] BUG: kernel NULL pointer dereference, address: 0000000000000088 > > [ 1.159634] #PF: supervisor read access in kernel mode > > [ 1.164757] #PF: error_code(0x0000) - not-present page > > [ 1.169882] PGD 0 P4D 0 > > [ 1.172407] Oops: 0000 [#1] PREEMPT SMP PTI > > [ 1.176578] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 6.0.11-300.fc37.x86_64 #1 > > [ 1.183870] Hardware name: Dell Inc. PowerEdge R410/0N051F, BIOS 1.11.0 07/20/2012 > > [ 1.191420] RIP: 0010:tick_periodic+0x23/0x80 > > I'm willing to bet that this is caused by the following line in tick_periodic(): > > update_process_times(user_mode(get_irq_regs())); > > user_mode() is invoked with a NULL pointer. user_mode() accesses > regs->cs. CS is at offset 0x88.... > > The reason for this is here: > > > [ 1.280648] tick_handle_periodic+0x1f/0x70 > > [ 1.284821] timer_interrupt+0x14/0x20 > > [ 1.288561] __handle_irq_event_percpu+0x46/0x190 > > [ 1.293253] handle_irq_event+0x34/0x70 > > [ 1.297080] handle_level_irq+0xa8/0x180 > > [ 1.300993] resend_irqs+0x5d/0x70 > > [ 1.304386] tasklet_action_common.constprop.0+0xab/0xe0 > > [ 1.309686] __do_softirq+0xfb/0x319 > > [ 1.313254] __irq_exit_rcu+0xd7/0x140 > > [ 1.316993] common_interrupt+0xb9/0xd0 > > For some reason the timer interrupt is resent in software. I assume it is > the HPET interrupt because that's what just got initialized. > > > [ 1.143537] clocksource: hpet: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 133484882848 ns > > and the callchain below just confirms that: > > > [ 1.393315] _raw_spin_unlock_irqrestore+0x19/0x40 > > [ 1.398093] __setup_irq+0x443/0x6d0 > > [ 1.401659] request_threaded_irq+0x109/0x170 > > [ 1.406005] hpet_time_init+0x2d/0x4b > > [ 1.409661] x86_late_time_init+0x17/0x34 > > [ 1.413658] start_kernel+0x8cf/0x97f > > The software resend code does not go through the regular interrupt entry > path which explains why get_irq_regs() returns NULL. > > That software resend is bogus especially since the timer interrupt is > a level interrupt. As dmesg does not say anything about the APIC > delivery mode I assume this goes through i8259, which fails to set the > IRQ_LEVEL flag on all interrupt lines. > > The below should fix this. Sorry for late reply, just notice this mail. Thanks a lot for checking this and providing a fix, I will reserve the lab machine and give it a shot, will feedback once finished. > --- > --- a/arch/x86/kernel/i8259.c > +++ b/arch/x86/kernel/i8259.c > @@ -114,6 +114,7 @@ static void make_8259A_irq(unsigned int > disable_irq_nosync(irq); > io_apic_irqs &= ~(1<<irq); > irq_set_chip_and_handler(irq, &i8259A_chip, handle_level_irq); > + irq_set_status_flags(irq, IRQ_LEVEL); > enable_irq(irq); > lapic_assign_legacy_vector(irq, true); > } > --- a/arch/x86/kernel/irqinit.c > +++ b/arch/x86/kernel/irqinit.c > @@ -65,8 +65,10 @@ void __init init_ISA_irqs(void) > > legacy_pic->init(0); > > - for (i = 0; i < nr_legacy_irqs(); i++) > + for (i = 0; i < nr_legacy_irqs(); i++) { > irq_set_chip_and_handler(i, chip, handle_level_irq); > + irq_set_status_flags(i, IRQ_LEVEL); > + } > } > > void __init init_IRQ(void) >
WARNING: multiple messages have this Message-ID (diff)
From: Baoquan He <bhe@redhat.com> To: Thomas Gleixner <tglx@linutronix.de> Cc: linux-kernel@vger.kernel.org, x86@kernel.org, kexec@lists.infradead.org Subject: Re: kdump kernel randomly hang with tick_periodic call trace on bare metal system Date: Sat, 14 Jan 2023 10:08:53 +0800 [thread overview] Message-ID: <Y8IOtY2x8XQT+Owd@fedora> (raw) In-Reply-To: <87mt6rjrra.ffs@tglx> On 01/09/23 at 10:57pm, Thomas Gleixner wrote: > On Tue, Dec 20 2022 at 13:41, Baoquan He wrote: > > On one intel bare metal system, I can randomly reproduce the kdump hang > > as below with tick_periodic call trace. Attach the kernel config for > > reference. > > This has absolutely nothing to do with x2apic IPI shorthands > > > [ 0.045980] Spurious LAPIC timer interrupt on cpu 0 > > So here the CPU receives a spurious Local APIC timer interrupt, but > that's a red herring. > > > [ 1.152690] BUG: kernel NULL pointer dereference, address: 0000000000000088 > > [ 1.159634] #PF: supervisor read access in kernel mode > > [ 1.164757] #PF: error_code(0x0000) - not-present page > > [ 1.169882] PGD 0 P4D 0 > > [ 1.172407] Oops: 0000 [#1] PREEMPT SMP PTI > > [ 1.176578] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 6.0.11-300.fc37.x86_64 #1 > > [ 1.183870] Hardware name: Dell Inc. PowerEdge R410/0N051F, BIOS 1.11.0 07/20/2012 > > [ 1.191420] RIP: 0010:tick_periodic+0x23/0x80 > > I'm willing to bet that this is caused by the following line in tick_periodic(): > > update_process_times(user_mode(get_irq_regs())); > > user_mode() is invoked with a NULL pointer. user_mode() accesses > regs->cs. CS is at offset 0x88.... > > The reason for this is here: > > > [ 1.280648] tick_handle_periodic+0x1f/0x70 > > [ 1.284821] timer_interrupt+0x14/0x20 > > [ 1.288561] __handle_irq_event_percpu+0x46/0x190 > > [ 1.293253] handle_irq_event+0x34/0x70 > > [ 1.297080] handle_level_irq+0xa8/0x180 > > [ 1.300993] resend_irqs+0x5d/0x70 > > [ 1.304386] tasklet_action_common.constprop.0+0xab/0xe0 > > [ 1.309686] __do_softirq+0xfb/0x319 > > [ 1.313254] __irq_exit_rcu+0xd7/0x140 > > [ 1.316993] common_interrupt+0xb9/0xd0 > > For some reason the timer interrupt is resent in software. I assume it is > the HPET interrupt because that's what just got initialized. > > > [ 1.143537] clocksource: hpet: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 133484882848 ns > > and the callchain below just confirms that: > > > [ 1.393315] _raw_spin_unlock_irqrestore+0x19/0x40 > > [ 1.398093] __setup_irq+0x443/0x6d0 > > [ 1.401659] request_threaded_irq+0x109/0x170 > > [ 1.406005] hpet_time_init+0x2d/0x4b > > [ 1.409661] x86_late_time_init+0x17/0x34 > > [ 1.413658] start_kernel+0x8cf/0x97f > > The software resend code does not go through the regular interrupt entry > path which explains why get_irq_regs() returns NULL. > > That software resend is bogus especially since the timer interrupt is > a level interrupt. As dmesg does not say anything about the APIC > delivery mode I assume this goes through i8259, which fails to set the > IRQ_LEVEL flag on all interrupt lines. > > The below should fix this. Sorry for late reply, just notice this mail. Thanks a lot for checking this and providing a fix, I will reserve the lab machine and give it a shot, will feedback once finished. > --- > --- a/arch/x86/kernel/i8259.c > +++ b/arch/x86/kernel/i8259.c > @@ -114,6 +114,7 @@ static void make_8259A_irq(unsigned int > disable_irq_nosync(irq); > io_apic_irqs &= ~(1<<irq); > irq_set_chip_and_handler(irq, &i8259A_chip, handle_level_irq); > + irq_set_status_flags(irq, IRQ_LEVEL); > enable_irq(irq); > lapic_assign_legacy_vector(irq, true); > } > --- a/arch/x86/kernel/irqinit.c > +++ b/arch/x86/kernel/irqinit.c > @@ -65,8 +65,10 @@ void __init init_ISA_irqs(void) > > legacy_pic->init(0); > > - for (i = 0; i < nr_legacy_irqs(); i++) > + for (i = 0; i < nr_legacy_irqs(); i++) { > irq_set_chip_and_handler(i, chip, handle_level_irq); > + irq_set_status_flags(i, IRQ_LEVEL); > + } > } > > void __init init_IRQ(void) > _______________________________________________ kexec mailing list kexec@lists.infradead.org http://lists.infradead.org/mailman/listinfo/kexec
next prev parent reply other threads:[~2023-01-14 2:09 UTC|newest] Thread overview: 29+ messages / expand[flat|nested] mbox.gz Atom feed top 2022-12-20 5:34 [PATCH] Revert "x86/apic/x2apic: Implement IPI shorthands support" Baoquan He 2022-12-20 5:34 ` Baoquan He 2022-12-20 5:41 ` kdump kernel randomly hang with tick_periodic call trace on bare metal system Baoquan He 2022-12-20 5:41 ` Baoquan He 2022-12-20 5:51 ` Baoquan He 2022-12-20 5:51 ` Baoquan He 2022-12-21 15:46 ` Guilherme G. Piccoli 2022-12-21 15:46 ` Guilherme G. Piccoli 2022-12-22 4:09 ` Baoquan He 2022-12-22 4:09 ` Baoquan He 2023-01-09 21:57 ` Thomas Gleixner 2023-01-09 21:57 ` Thomas Gleixner 2023-01-14 2:08 ` Baoquan He [this message] 2023-01-14 2:08 ` Baoquan He 2023-01-16 9:08 ` Baoquan He 2023-01-16 9:08 ` Baoquan He 2023-01-16 16:27 ` [tip: x86/urgent] x86/i8259: Mark legacy PIC interrupts with IRQ_LEVEL tip-bot2 for Thomas Gleixner 2022-12-20 11:38 ` [PATCH] Revert "x86/apic/x2apic: Implement IPI shorthands support" Peter Zijlstra 2022-12-20 11:38 ` Peter Zijlstra 2022-12-20 12:38 ` Baoquan He 2022-12-20 12:38 ` Baoquan He 2023-01-04 15:18 ` Dr. David Alan Gilbert 2023-01-04 15:18 ` Dr. David Alan Gilbert 2023-01-09 21:59 ` Thomas Gleixner 2023-01-09 21:59 ` Thomas Gleixner 2023-01-10 2:24 ` Baoquan He 2023-01-10 2:24 ` Baoquan He 2023-01-17 8:27 ` Baoquan He 2023-01-17 8:27 ` Baoquan He
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=Y8IOtY2x8XQT+Owd@fedora \ --to=bhe@redhat.com \ --cc=kexec@lists.infradead.org \ --cc=linux-kernel@vger.kernel.org \ --cc=tglx@linutronix.de \ --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.