在 2025/7/17 06:25, Bjorn Helgaas 写道:

[+cc Ilpo, Jonathan (should have been included since the patch has his
Reviewed-by)]

Thanks.

Thanks for the ping; I noticed quite a bit of discussion but didn't
follow it myself, so didn't know it was basically all resolved.

On Mon, May 12, 2025 at 09:38:39AM +0800, Shuai Xue wrote:

quoted

Hotplug events are critical indicators for analyzing hardware health,
particularly in AI supercomputers where surprise link downs can
significantly impact system performance and reliability.

I dropped the "particularly in AI supercomputers" part because I think
this is relevant in general.

quoted

To this end, define a new TRACING_SYSTEM named pci, add a generic RAS
tracepoint for hotplug event to help healthy check, and generate
tracepoints for pcie hotplug event.

I'm not quite clear on the difference between "add generic RAS
tracepoint for hotplug event" and "generate tracepoints for pcie
hotplug event."  Are these two different things?

The purpose of this patch is to address the lack of tracepoints for PCIe
hotplug events in our production environment. In the initial RFC
version, I defined tracepoints such as "Link Up" and "Link Down"
specifically for PCIe hotplug. Later, Lukas suggested that these
tracepoints could be made more generic so that other PCI hotplug drivers
could also use them.

That’s why, when defining the event, I used a "generic" pci_hotplug_event
instead of a pcie_hotplug_event. If you're interested in more details
about this discussion, please refer to this link[1].

[1]https://erol.kernel.org/linux-pci/git/0/commit/?id=0ffd56f572f25bcd6c2265a1863848a18dce0e29

However, currently only PCIe hotplug is using these tracepoints, which
is why the CREATE_TRACE_POINTS macro is placed in
drivers/pci/hotplug/pciehp_ctrl.c.

I see the new TRACE_EVENT(pci_hp_event, ...) definition.  Is that what
you mean by the "generic RAS tracepoint"?

Yes.

And the five new trace_pci_hp_event() calls that use the TRACE_EVENT
are the "tracepoints for PCIe hotplug event"?

Actually, the tracepoints are generic, although right now they are only
used for PCIe hotplug.

quoted

Add enum pci_hotplug_event in
include/uapi/linux/pci.h so applications like rasdaemon can register
tracepoint event handlers for it.

The output like below:

$ echo 1 > /sys/kernel/debug/tracing/events/pci/pci_hp_event/enable
$ cat /sys/kernel/debug/tracing/trace_pipe
     <...>-206     [001] .....    40.373870: pci_hp_event: 0000:00:02.0 slot:10, event:Link Down

     <...>-206     [001] .....    40.374871: pci_hp_event: 0000:00:02.0 slot:10, event:Card not present

quoted

+#define PCI_HOTPLUG_EVENT					\
+	EM(PCI_HOTPLUG_LINK_UP,			"Link Up")	\
+	EM(PCI_HOTPLUG_LINK_DOWN,		"Link Down")	\
+	EM(PCI_HOTPLUG_CARD_PRESENT,		"Card present")	\
+	EMe(PCI_HOTPLUG_CARD_NOT_PRESENT,	"Card not present")

Running this:

   $ git grep -E "\<(EM|EMe)\("

I notice that these new events don't look like the others, which
mostly look like "word" or "event-type" or "VERB object".

I'm OK with this, but just giving you a chance to consider what will
be the least surprise to users and easiest for grep and shell
scripting.

I think this is also common. For example, MF_PAGE_TYPE for
memory_failure_event uses a similar format:

#define MF_PAGE_TYPE \
	EM ( MF_MSG_KERNEL, "reserved kernel page" ) \
	EM ( MF_MSG_KERNEL_HIGH_ORDER, "high-order kernel page" )


and aer_uncorrectable_errors for aer_event:

#define aer_uncorrectable_errors				\
	{PCI_ERR_UNC_UND,	"Undefined"},			\
	{PCI_ERR_UNC_DLP,	"Data Link Protocol Error"},	\
	{PCI_ERR_UNC_SURPDN,	"Surprise Down Error"},		\
	{PCI_ERR_UNC_POISON_TLP,"Poisoned TLP"},

I also noticed capitalization of "Up" and "Down", but not "present"
and "not present".

Aha, this is a bit tricky:)

The original kernel log messages are not consistent either:

ctrl_info(ctrl, "Slot(%s): Link Down\n",
ctrl_info(ctrl, "Slot(%s): Card not present\n",

I tried to keep the output as close as possible to the existing log
messages. If you prefer a more consistent capitalization style, I can
send another patch to fix that.

"Card" is only used occasionally and informally in the PCIe spec, and
not at all in the context of hotplug of Slot Status (Presence Detect
State refers to "adapter in the slot"), but it does match the pciehp
dmesg text, so it probably makes sense to use that.

Anyway, I applied this on pci/trace for v6.17.  If there's anything
you want to tweak in the commit log or event text, we can still do
that.

   https://git.kernel.org/pub/scm/linux/kernel/git/pci/pci.git/commit/?h=trace

Bjorn

Thank you again for applying this to pci/trace for v6.17. If there’s
anything more to tweak in the commit log or event text, please let me
know.

Best regards,

Shuai