Thread (8 messages) 8 messages, 5 authors, 2026-03-25

Re: [PATCH] Documentation: PCI: Document decoding of TLP Header in AER messages

From: <hidden>
Date: 2026-03-25 05:18:36
Also in: linux-pci

Thanks Lukas for the suggestion and Ilpo for the caveat on
Link Status 2.

I'll add auto-detection of the (Flit) suffix (7e077e6707b3,
v6.15+) to --aer mode so mixed flit / non-flit TLPs in the same
log are each parsed with the correct framing -- no --flit needed.
For --lspci I'll also pick up Flit+ from LnkSta2: per device.

The --flit flag stays as a global override for inputs without
auto-detection markers.  This will be a patch release (v0.5.1) --
fully backward compatible.

Maciej



On Tuesday, March 24th, 2026 at 4:18 AM, Ilpo Järvinen [off-list ref] wrote:
On Tue, 24 Mar 2026, mx2pg@pm.me wrote:
quoted
  One thing worth calling out: starting with PCIe 6.0, Flit Mode is
  mandatory at 64.0 GT/s and supported at all PCIe link speeds, so a
  Flit-capable PCIe 6.x link may operate below 64.0 GT/s and still be
  in Flit Mode.  The raw TLP Header bytes do not encode the framing —
  the same four bytes decode to entirely different packet types in
  non-Flit vs Flit framing.  The negotiated mode can be read from the
  Flit Mode Status bit in Link Status 2, or via lspci -vv on a recent
  pciutils build.
There's one caveat in using Link Status 2 Flit Mode Status bit, it can
only be used as the indicator when the Link is Up, which may come into
picture in troubleshooting scenarios.

The kernel code tries to hide that by indicating the Flit mode explicitly
in the log message it prints out.

Sadly, TLP Logging on DPC side was botched in the PCIe spec so it doesn't
indicate the Flit/non-Flit mode information explicitly (in contrast to AER
that has a flag that tells in which mode the TLP Log was captured). To
workaround that limitation, kernel has to save of the Link Status 2
contents and hope the information is not stale when DPC has brought the
Link Down (it seems relatively likely to remain valid but it's still
fundamentally racy way to get the Flit/non-Flit information).

--
 i.
quoted
  tlp-tool defaults to non-Flit, which is correct for the vast majority
  of hardware deployed today.  That will change: as PCIe 6.x adoption
  grows, a significant share of TLP debugging will involve Flit Mode
  links, and this is already a concern among switch and device vendors
  working through the transition.  Users on Flit Mode links must pass
  --flit:

    # non-Flit link (default, most common today)
    curl -L https://git.kernel.org/linus/2ca1c94ce0b6 | rtlp-tool --aer

    # Flit Mode link
    curl -L https://git.kernel.org/linus/2ca1c94ce0b6 | rtlp-tool --aer --flit

  It may be worth a one-liner in the Documentation patch:

    For PCIe 6.x links with Flit Mode negotiated (check Flit Mode Status
    in Link Status 2, or lspci -vv), pass --flit to rtlp-tool.

  Maciej



On Monday, March 23rd, 2026 at 9:50 AM, Bjorn Helgaas [off-list ref] wrote:
quoted
On Mon, Mar 23, 2026 at 07:52:39AM +0100, Lukas Wunner wrote:
quoted
The prefix/header of the TLP that caused an error is recorded by the Root
Complex and emitted to the kernel log in raw hex format.  Document the
existence and usage of tlp-tool, which allows decoding the TLP Header
into human-readable form.

The TLP Header hints at the root cause of an error, yet is often ignored
because of its seeming opaqueness.  Instead, PCIe errors are frequently
worked around by a change in the kernel without fully understanding the
actual source of the problem.  With more documentation on available tools
we'll hopefully come up with better solutions.

There are also wireshark dissectors for TLPs, but it seems they expect a
complete TLP, not just the header, and they cannot grok the hex format
emitted by the kernel directly.  tlp-tool appears to be the most cut and
dried solution out there.

Signed-off-by: Lukas Wunner <lukas@wunner.de>
Cc: Maciej Grochowski <redacted>
Applied to pci/for-linus for v7.0, thanks!

I tweaked the commit log to note that the Header Log is in the AER
Capability, which may be in any PCIe function.
quoted
---
We could also go one step further and point users to this tool
in a printk_once() message when the first error occurs.
For now, just amending the documentation is probably sufficient.

 Documentation/PCI/pcieaer-howto.rst | 10 ++++++++++
 1 file changed, 10 insertions(+)
diff --git a/Documentation/PCI/pcieaer-howto.rst b/Documentation/PCI/pcieaer-howto.rst
index 3210c47..90fdfdd 100644
--- a/Documentation/PCI/pcieaer-howto.rst
+++ b/Documentation/PCI/pcieaer-howto.rst
@@ -85,6 +85,16 @@ In the example, 'Requester ID' means the ID of the device that sent
 the error message to the Root Port. Please refer to PCIe specs for other
 fields.

+The 'TLP Header' is the prefix/header of the TLP that caused the error
+in raw hex format. To decode the TLP Header into human-readable form
+one may use tlp-tool:
+
+https://github.com/mmpg-x86/tlp-tool
+
+Example usage::
+
+  curl -L https://git.kernel.org/linus/2ca1c94ce0b6 | rtlp-tool --aer
+
 AER Ratelimits
 --------------

--
2.51.0
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help