Re: [PATCH 0/3] PCI: Add support for logging Flit Mode TLPs (PCIe6)

[Ilpo Järvinen]

On Tue, 17 Dec 2024, Yazen Ghannam wrote:

> On Fri, Jun 14, 2024 at 06:09:18PM +0300, Ilpo Järvinen wrote:
> > This series adds support for Flit Mode (PCIe6). The series is built on
> > top of the TLP Logging refactoring series:
> > 
> >   https://lore.kernel.org/linux-pci/20240514113109.6690-1-ilpo.jarvinen@xxxxxxxxxxxxxxx/
> > 
> > Important note to maintainer: The series carries
> > pcie_update_link_speed() refactoring change that is almost identical
> > with a patch in the PCIe BW controller series. The patch itself is
> > basically the same but the context has minor difference. This will need
> > to be considered if applying both series within the same kernel cycle.
> > 
> > Ilpo Järvinen (3):
> >   PCI: Refactor pcie_update_link_speed()
> >   PCI: Track Flit Mode Status & print it with link status
> >   PCI: Handle TLP Log in Flit mode
> > 
> >  drivers/pci/hotplug/pciehp_hpc.c |  5 +--
> >  drivers/pci/pci.c                | 12 ++++---
> >  drivers/pci/pci.h                | 13 ++++++--
> >  drivers/pci/pcie/aer.c           |  4 ++-
> >  drivers/pci/pcie/dpc.c           | 23 ++++++++++---
> >  drivers/pci/pcie/tlp.c           | 57 ++++++++++++++++++++++----------
> >  drivers/pci/probe.c              | 13 +++++---
> >  include/linux/aer.h              | 13 ++++++--
> >  include/linux/pci.h              |  1 +
> >  include/ras/ras_event.h          | 12 +++----
> >  include/uapi/linux/pci_regs.h    |  6 +++-
> >  11 files changed, 112 insertions(+), 47 deletions(-)
>
> Is there any new development on this feature? Or is it on hold while any
> spec oversights are worked out in the PCI-SIG?

As far as I know, the series is not on hold. It just tends to take time 
from Bjorn to get patches applied (and I don't want to pressure 
maintainers with frequent pings). But I think it might help if you would 
kindly review the patches. :-)

And of course this series depends on the TLP cleanup series that has to 
be applied first (but hopefully that happens soon enough).

When it comes to the spec oversight, we concluded with Lukas Wunner that 
even if DPC capability would eventually get a flag to indicate in which 
mode the TLP was logged we cannot assume the flags is always there. Thus, 
this link Flit mode status tracking has to be done anyway. I know it's not 
ideal because at least in theory the state kept by the kernel could be 
stale but there seems to no way around that given how the spec is.

-- 
 i.