Message-ID: <b6c0bd81-3b8d-465d-a0eb-faa5323a6b05@amazon.com>
Date: Tue, 26 Mar 2024 14:34:45 +0200
From: "Margolin, Michael" <mrgolin@...zon.com>
To: Tao Liu <ltao@...hat.com>, Gal Pressman <gal.pressman@...ux.dev>
CC: <sleybo@...zon.com>, <jgg@...pe.ca>, <leon@...nel.org>,
	<kexec@...ts.infradead.org>, <linux-kernel@...r.kernel.org>,
	<linux-rdma@...r.kernel.org>
Subject: Re: Implementing .shutdown method for efa module

Hi Tao,

Thanks for bringing this up.

I've unsuccessfully tried to reproduce this kernel panic using 
production Red Hat 9.3 AMI (5.14.0-362.18.1.el9_3.aarch64).

Are there any related changes in the kernel you are testing?

Anyways we do need to handle shutdown properly, please let know if 
calling to efa_remove solves your issue.

Michael

On 3/26/2024 3:38 AM, Tao Liu wrote:
> Hi Gal,
>
> On Mon, Mar 25, 2024 at 4:06 PM Gal Pressman <gal.pressman@...ux.dev> wrote:
>> On 25/03/2024 4:10, Tao Liu wrote:
>>> Hi,
>>>
>>> Recently I experienced a kernel panic which is related to efa module
>>> when testing kexec -l && kexec -e to switch to a new kernel on AWS
>>> i4g.16xlarge instance.
>>>
>>> Here is the dmesg log:
>>>
>>> [    6.379918] systemd[1]: Mounting FUSE Control File System...
>>> [    6.381984] systemd[1]: Mounting Kernel Configuration File System...
>>> [    6.383918] systemd[1]: Starting Apply Kernel Variables...
>>> [    6.385430] systemd[1]: Started Journal Service.
>>> [    6.394221] ACPI: bus type drm_connector registered
>>> [    6.421408] systemd-journald[1263]: Received client request to
>>> flush runtime journal.
>>> [    7.262543] efa 0000:00:1b.0: enabling device (0010 -> 0012)
>>> [    7.432420] efa 0000:00:1b.0: Setup irq:191 name:efa-mgmnt@pci:0000:00:1b.0
>>> [    7.435581] efa 0000:00:1b.0 efa_0: IB device registered
>>> [    7.885564] random: crng init done
>>> [    8.139857] XFS (nvme0n1p2): Mounting V5 Filesystem
>>> d7003ecc-db6f-4bfb-bf92-60376b6a6563
>>> [    8.265233] XFS (nvme0n1p2): Ending clean mount
>>> [   10.555612] IPv6: ADDRCONF(NETDEV_CHANGE): eth0: link becomes ready
>>>
>>> Red Hat Enterprise Linux 9.4 Beta (Plow)
>>> Kernel 5.14.0-425.el9.aarch64 on an aarch64
>>>
>>> ip-10-0-27-226 login: [   29.940381] kexec_core: Starting new kernel
>>> [   30.079279] psci: CPU1 killed (polled 0 ms)
>>> [   30.119222] psci: CPU2 killed (polled 0 ms)
>>> [   30.199293] psci: CPU3 killed (polled 0 ms)
>>> [   30.309214] psci: CPU4 killed (polled 0 ms)
>>> [   30.379221] psci: CPU5 killed (polled 0 ms)
>>> [   30.419210] psci: CPU6 killed (polled 0 ms)
>>> [   30.489207] IRQ 191: no longer affine to CPU7
>>> [   30.489667] psci: CPU7 killed (polled 0 ms)
>>> ..snip...
>>> [   33.849123] psci: CPU63 killed (polled 0 ms)
>>> [   33.849943] Bye!
>>> [    0.000000] Booting Linux on physical CPU 0x0000000000 [0x413fd0c1]
>>> [    0.000000] Linux version 5.14.0-417.el9.aarch64
>>> (mockbuild@...64-025.build.eng.bos.redhat.com) (gcc (GCC) 11.4.1
>>> 20231218 (Red Hat 11.4.1-3), GNU ld version 2.35.2-42.el9) #1 SMP
>>> PREEMPT_DYNAMIC Thu Feb 1 21:23:03 EST 2024
>>> ...snip...
>>> [    1.012692] Freeing unused kernel memory: 6016K
>>> [    2.370947] Checked W+X mappings: passed, no W+X pages found
>>> [    2.370980] Run /init as init process
>>> [    2.370982]   with arguments:
>>> [    2.370983]     /init
>>> [    2.370984]   with environment:
>>> [    2.370984]     HOME=/
>>> [    2.370985]     TERM=linux
>>> [    2.373257] Kernel panic - not syncing: Attempted to kill init!
>>> exitcode=0x0000000b
>>> [    2.373259] CPU: 1 PID: 1 Comm: init Not tainted 5.14.0-417.el9.aarch64 #1
>>> [    2.382240] Hardware name: Amazon EC2 i4g.16xlarge/, BIOS 1.0 11/1/2018
>>> [    2.383814] Call trace:
>>> [    2.384410]  dump_backtrace+0xa8/0x120
>>> [    2.385318]  show_stack+0x1c/0x30
>>> [    2.386124]  dump_stack_lvl+0x74/0x8c
>>> [    2.387011]  dump_stack+0x14/0x24
>>> [    2.387810]  panic+0x158/0x368
>>> [    2.388553]  do_exit+0x3a8/0x3b0
>>> [    2.389333]  do_group_exit+0x38/0xa4
>>> [    2.390195]  get_signal+0x7a4/0x810
>>> [    2.391044]  do_signal+0x1bc/0x260
>>> [    2.391870]  do_notify_resume+0x108/0x210
>>> [    2.392839]  el0_da+0x154/0x160
>>> [    2.393603]  el0t_64_sync_handler+0xdc/0x150
>>> [    2.394628]  el0t_64_sync+0x17c/0x180
>>> [    2.395513] SMP: stopping secondary CPUs
>>> [    2.396483] Kernel Offset: 0x586f04e00000 from 0xffff800008000000
>>> [    2.397934] PHYS_OFFSET: 0x40000000
>>> [    2.398774] CPU features: 0x0,00000101,70020143,10417a0b
>>> [    2.400042] Memory Limit: none
>>> [    2.400783] ---[ end Kernel panic - not syncing: Attempted to kill
>>> init! exitcode=0x0000000b ]---
>>>
>>> In the dmesg log, I found "[   30.489207] IRQ 191: no longer affine to
>>> CPU7" is suspicious, which is related to efa module. After blacklist
>>> efa module from automatic loading when bootup, the kernel panic issue
>>> doesn't appear again.
>>>
>>> It looks to me it is due to the efa being not properly shutdown during
>>> kexec, so the ongoing DMA/interrupts etc overwrite the memory range.
>>>
>>> Though the issue is reproduced on rhel's kernel, the upstream kernel
>>> [1] doesn't have the .shutdown method implemented either. Since I'm
>>> not very familiar with the efa driver, could you please implement the
>>> .shutdown method in drivers/infiniband/hw/efa/efa_main.c? Thanks in
>>> advance!
>> Did you try to reproduce it on upstream kernel?
>>
> Thanks for your comments! No I haven't, I will give it a try.
>>> [1]: https://github.com/torvalds/linux/blob/master/drivers/infiniband/hw/efa/efa_main.c#L674
>>>
>>> Thanks,
>>> Tao Liu
>>>
>> Try assigning efa_remove as the shutdown callback:
>>      .shutdown = efa_remove,
>>
>> Does it fix it?
> Thanks, I will also try the code, and I will post the testing results.
>
> Thanks,
> Tao Liu
>
>

Powered by blists - more mailing lists