Add a patch for printing the AMD Zen CPU reset reason#514
Add a patch for printing the AMD Zen CPU reset reason#514saiarcot895 merged 2 commits intosonic-net:masterfrom
Conversation
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
How can these events be triggered? |
| From: Yazen Ghannam <yazen.ghannam@amd.com> | ||
| Date: Tue, 22 Apr 2025 18:48:30 -0500 | ||
| Subject: [PATCH 1/2] x86/CPU/AMD: Print the reason for the last reset | ||
|
|
There was a problem hiding this comment.
Please add the upstream commit hash as done for stable series commits.
There was a problem hiding this comment.
Thanks @paulmenzel , please take a look at 46ca756 to see if that is what you had in mind.
As I understand it, writing 0x6 to 0xcf9 is a standard way of reboting an x86 CPU. I can trigger this with Triggering the FCH watchdog on SONiC with an AMD Zen3 CPU, I can do by enabling the watchdog and never petting it. |
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
nate-nexthop
force-pushed
the
lotus.amd-log
branch
from
6800534 to
4ac95b1
Compare
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
nate-nexthop
force-pushed
the
lotus.amd-log
branch
from
4ac95b1 to
98cbc97
Compare
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
If I intentionally trigger a CPU soft reset I see this: ``` admin@gold208-dut:~$ sudo dmesg | grep -i reason [ 0.635233] x86/amd: Previous system reset reason [0x00080800]: software wrote 0x6 to reset control register 0xCF9 ``` If I intentionally trigger the CPU FCH Watchdog, I see this: ``` admin@gold208-dut:~$ sudo dmesg | grep reason [ 0.632563] x86/amd: Previous system reset reason [0x02000800]: hardware watchdog timer expired ``` Upstream from here: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/commit/?id=ab8131028710d009ab93d6bffd2a2749ade909b0 The patch had to be adapted to v6.1 we're using, that was basically adding the entire contents (5 constants) of `fch.h` as the file didn't exist in v6.1, and updating the patch for `amd.c` for context. Signed-off-by: Nate White <nate@nexthop.ai>
nate-nexthop
force-pushed
the
lotus.amd-log
branch
from
98cbc97 to
e9ddb34
Compare
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
I've adapted the backported patches to the current (6.12) kernel sources, and can confirm the functionality still works: |
|
@saiarcot895 could you take a look please? |
|
@lotus-nexthop Could you merge in the master branch into your branch? |
|
/azp run |
|
Azure Pipelines successfully started running 1 pipeline(s). |
|
@saiarcot895 Updated the branch with the latest master, looks like pipelines have started. |
5 participants
Upstream commits:
The patch had to be adapted to v6.1 we're using, that was basically adding the entire contents (5 constants) of
fch.has the file didn't exist in v6.1, and updating the patch foramd.cfor context.Testing
If we intentionally trigger a CPU soft reset (with
sudo reboot -f) I see this:If we intentionally trigger the CPU FCH Watchdog, I see this:
To enable watchdog we create a
/etc/systemd/system.conf.d/override.confwith the contents:
To trigger the watchdog:
sudo tee /dev/watchdog1and enter just one character and let the device be for a minute or so.