[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <6eac5c37-a5a8-4ccf-aef6-62a4a0bfcea0@jevklidu.cz>
Date: Sat, 17 Aug 2024 19:57:24 +0200
From: Petr Valenta <petr@...klidu.cz>
To: "Rafael J. Wysocki" <rafael@...nel.org>, Jiri Slaby <jirislaby@...nel.org>
Cc: Len Brown <lenb@...nel.org>,
"linux-acpi@...r.kernel.org" <linux-acpi@...r.kernel.org>,
Linux kernel mailing list <linux-kernel@...r.kernel.org>,
Linux regressions mailing list <regressions@...ts.linux.dev>
Subject: Re: ACPI IRQ storm with 6.10
Dne 16. 08. 24 v 20:29 Rafael J. Wysocki napsal(a):
> On Wed, Aug 14, 2024 at 8:48 AM Jiri Slaby <jirislaby@...nel.org> wrote:
>>
>> On 14. 08. 24, 7:22, Jiri Slaby wrote:
>>> Hi,
>>>
>>> one openSUSE's user reported that with 6.10, he sees one CPU under an
>>> IRQ storm from ACPI (sci_interrupt):
>>> 9: 20220768 ... IR-IO-APIC 9-fasteoi acpi
>>>
>>> At:
>>> https://bugzilla.suse.com/show_bug.cgi?id=1229085
>>>
>>> 6.9 was OK.
>>>
>>> With acpi.debug_level=0x08000000 acpi.debug_layer=0xffffffff, there is a
>>> repeated load of:
>>>> evgpe-0673 ev_detect_gpe : Read registers for GPE 6D:
>>>> Status=20, Enable=00, RunEnable=4A, WakeEnable=00
>>
>> 0x6d seems to count excessively (10 snapshots every 1 second):
>>> /sys/firmware/acpi/interrupts/gpe6D: 82066 EN STS enabled unmasked
>>> /sys/firmware/acpi/interrupts/gpe6D: 86536 EN STS enabled unmasked
>>> /sys/firmware/acpi/interrupts/gpe6D: 90990 STS enabled unmasked
>>> /sys/firmware/acpi/interrupts/gpe6D: 95468 EN STS enabled unmasked
>>> /sys/firmware/acpi/interrupts/gpe6D: 100282 EN STS enabled unmasked
>>> /sys/firmware/acpi/interrupts/gpe6D: 105187 STS enabled unmasked
>>> /sys/firmware/acpi/interrupts/gpe6D: 110014 STS enabled unmasked
>>> /sys/firmware/acpi/interrupts/gpe6D: 114852 STS enabled unmasked
>>> /sys/firmware/acpi/interrupts/gpe6D: 119682 STS enabled unmasked
>>> /sys/firmware/acpi/interrupts/gpe6D: 124194 STS enabled unmasked
>>> /sys/firmware/acpi/interrupts/gpe6D: 128641 EN STS enabled unmasked
>>
>> acpidump:
>> https://bugzilla.suse.com/attachment.cgi?id=876677
>>
>> DSDT:
>> https://bugzilla.suse.com/attachment.cgi?id=876678
>>
>>> Any ideas?
>
> GPE 6D is listed in _PRW for some devices, so maybe one of them
> continues to trigger wakeup events?
>
Disabling powertop service (which calls /usr/sbin/powertop --auto-tune)
solves problem completely. After some search I have found this is the cause:
# causes IRQ storm on 6.10.x
# kernel 6.9.9 is immune
echo 'auto' > /sys/bus/pci/devices/0000:00:1f.6/power/control
lspci | grep 1f.6
00:1f.6 Ethernet controller: Intel Corporation Device 550b (rev 20)
journalctl -b | grep 1f.6
srp 17 19:44:17 e14 kernel: pci 0000:00:1f.6: [8086:550b] type 00 class
0x020000 conventional PCI endpoint
srp 17 19:44:17 e14 kernel: pci 0000:00:1f.6: BAR 0 [mem
0x9c300000-0x9c31ffff]
srp 17 19:44:17 e14 kernel: pci 0000:00:1f.6: PME# supported from D0
D3hot D3cold
srp 17 19:44:17 e14 kernel: pci 0000:00:1f.6: Adding to iommu group 12
srp 17 19:44:19 e14 kernel: e1000e 0000:00:1f.6: Interrupt Throttling
Rate (ints/sec) set to dynamic conservative mode
srp 17 19:44:19 e14 kernel: e1000e 0000:00:1f.6 0000:00:1f.6
(uninitialized): registered PHC clock
srp 17 19:44:20 e14 kernel: e1000e 0000:00:1f.6 eth0: (PCI
Express:2.5GT/s:Width x1) fc:5c:ee:b0:13:74
srp 17 19:44:20 e14 kernel: e1000e 0000:00:1f.6 eth0: Intel(R) PRO/1000
Network Connection
srp 17 19:44:20 e14 kernel: e1000e 0000:00:1f.6 eth0: MAC: 16, PHY: 12,
PBA No: FFFFFF-0FF
srp 17 19:44:20 e14 kernel: e1000e 0000:00:1f.6 enp0s31f6: renamed from eth0
srp 17 19:44:24 e14 ModemManager[1434]: <info> [base-manager] couldn't
check support for device '/sys/devices/pci0000:00/0000:00:1f.6': not
supported by any plugin
> You can ask the reporter to mask that GPE via "echo mask >
> /sys/firmware/acpi/interrupts/gpe6D" and see if the storm goes away
> then.
>
> The only ACPI core issue introduced between 6.9 and 6.10 I'm aware of
> is the one addressed by this series
>
> https://lore.kernel.org/linux-acpi/22385894.EfDdHjke4D@rjwysocki.net/
>
> but this is about the EC and the problem here doesn't appear to be
> EC-related. It may be worth trying anyway, though.
>
Powered by blists - more mailing lists