Re: [Bug][5.17-rc0] Between commits daadb3bd0e8d and 455e73a07f6e, the kernel stops loading on my devices.

[Heikki Krogerus]

Hi Mikhail,

On Tue, Jan 18, 2022 at 04:26:27PM +0500, Mikhail Gavrilov wrote:
> On Mon, 17 Jan 2022 at 17:21, Heikki Krogerus
> <heikki.krogerus@xxxxxxxxxxxxxxx> wrote:
> >
> > Have you modified the file, or something else that you have not
> > committed yet?
> >
> >         % git status
> >
> > Checkout the file, and then try to apply the patch:
> >
> >         % git checkout drivers/usb/typec/port-mapper.c
> >         % git apply -v 0001-usb-typec-Test-fix.patch
> >
> 
> I understood why the patch was not applied. The "Window line endings"
> in the attached file were.
> 
> $ dos2unix 0001-usb-typec-Test-fix.patch
> dos2unix: converting file 0001-usb-typec-Test-fix.patch to Unix format...
> $ git apply -v 0001-usb-typec-Test-fix.patch
> Checking patch drivers/usb/typec/port-mapper.c...
> Applied patch drivers/usb/typec/port-mapper.c cleanly.

Oh, I'm sorry. Something corrupted the patch, which is a bit alarming,
but good that you managed to test it in any case.

> Unfortunately the attached patch didn't fix the issue.
> 
>  BUG: kernel NULL pointer dereference, address: 0000000000000008
>  #PF: supervisor read access in kernel mode
>  #PF: error_code(0x0000) - not-present page
>  PGD 0 P4D 0
>  Oops: 0000 [#1] PREEMPT SMP NOPTI
>  CPU: 10 PID: 459 Comm: kworker/10:2 Not tainted
> 5.16.0-patched-455e73a07f6e+ #26
>  Hardware name: ASUSTeK COMPUTER INC. ROG Strix G513QY_G513QY/G513QY,
> BIOS G513QY.316 11/29/2021
>  Workqueue: events_long ucsi_init_work [typec_ucsi]
>  RIP: 0010:component_master_add_with_match+0x14/0x110
>  Code: 72 d7 48 89 ef 5b 5d e9 1a 5c 9f ff 66 2e 0f 1f 84 00 00 00 00
> 00 0f 1f 44 00 00 41 55 49 89 fd 41 54 49 89 f4 55 53 48 89 d3 <48> 8b
> 72 08 48 3b 32 74 10 48 89 d7 e8 4b fd ff ff 85 c0 0f 85 bb
>  RSP: 0018:ffffba4581837d90 EFLAGS: 00010282
>  RAX: 0000000000000000 RBX: 0000000000000000 RCX: 0000000000000000
>  RDX: 0000000000000000 RSI: ffffffffc04f58d0 RDI: ffff9c5afe43a808
>  RBP: ffff9c5a81076358 R08: 0000000000000001 R09: 0000000000000001
>  R10: 0000000000000001 R11: 0000000000000000 R12: ffffffffc04f58d0
>  R13: ffff9c5afe43a808 R14: ffff9c5afe43a808 R15: ffff9c5a832c4010
>  FS:  0000000000000000(0000) GS:ffff9c5d88a00000(0000) knlGS:0000000000000000
>  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>  CR2: 0000000000000008 CR3: 00000002e5028000 CR4: 0000000000750ee0
>  PKRU: 55555554
>  Call Trace:
>   <TASK>
>   typec_link_ports+0x5d/0x70 [typec]
>   typec_register_port+0x1e6/0x2e0 [typec]
>   ucsi_init_work+0x463/0x840 [typec_ucsi]
>   process_one_work+0x2bb/0x600
>   worker_thread+0x55/0x3b0
>   ? process_one_work+0x600/0x600
>   kthread+0x17c/0x1a0
>   ? set_kthread_struct+0x40/0x40
>   ret_from_fork+0x22/0x30
>   </TASK>

This is not the same issue, this is another bug :-(. The original
issue does seem to be fixed.

I can reproduce this one by simply not creating the component list in
the code. That function - component_master_add_with_match() - can't
handle situation where the list is empty. I'll prepare the fix.

Thanks,

-- 
heikki