Re: [PATCH v4 0/2] PCI: Fix the issue of failed speed limit lifting

[Bjorn Helgaas <helgaas@kernel.org>]

[+cc Andrew, Matthew, Sathy]

On Thu, Jan 23, 2025 at 01:51:53PM +0800, Jiwei Sun wrote:
> From: Jiwei Sun <sunjw10@lenovo.com>
> 
> Since commit de9a6c8d5dbf ("PCI/bwctrl: Add pcie_set_target_speed() to set
> PCIe Link Speed"), there are two potential issues in the function
> pcie_failed_link_retrain().
> 
> (1) The macro PCIE_LNKCTL2_TLS2SPEED() and PCIE_LNKCAP_SLS2SPEED() just
> use the link speed field of the registers. However, there are many other
> different function fields in the Link Control 2 Register or the Link
> Capabilities Register. If the register value is directly used by the two
> macros, it may cause getting an error link speed value (PCI_SPEED_UNKNOWN).
> 
> (2) In the pcie_failed_link_retrain(), the local variable lnkctl2 is not
> changed after reading from PCI_EXP_LNKCTL2. It might cause that the
> removing 2.5GT/s downstream link speed restriction codes are not executed.
> 
> In order to avoid the above-mentioned potential issues, only keep link
> speed field of the two registers before using and reread the Link Control 2
> Register before using.
> 
> This series focuses on the first patch of the original series [1]. The
> second one of the original series will submitted via the other single
> patch.
> 
> [1] https://lore.kernel.org/linux-pci/tencent_DD9CBE5B44210B43A04EF8DAF52506A08509@qq.com/
> ---
> v4 changes:
>  - rename the variable name in the macro
> 
> v3 changes:
>  - add fix tag in the commit messages of first patch
>  - add an empty line after the local variable definition in the macro
>  - adjust the position of reading the Link Control 2 register in the code
> 
> v2 changes:
>  - divide the two issues into different patches
>  - get fixed inside the macros
> 
> Jiwei Sun (2):
>   PCI: Fix the wrong reading of register fields
>   PCI: Adjust the position of reading the Link Control 2 register
> 
>  drivers/pci/pci.h    | 32 +++++++++++++++++++-------------
>  drivers/pci/quirks.c |  6 ++++--
>  2 files changed, 23 insertions(+), 15 deletions(-)

Sorry, this totally slipped through the cracks.  I applied both of
these to pci/enumeration for v6.17.

Andrew reported tripping over this issue fixed by the first patch, and
Lukas also posted a similar patch [1] to fix it, so I updated the
commit log as below to include details of Andrew's report.

As Lukas did, I added a stable tag but made it for v6.13+ (not v6.12+)
because I think the actual problem showed up with de9a6c8d5dbf
("PCI/bwctrl: Add pcie_set_target_speed() to set PCIe Link Speed"),
not with f68dea13405c ("PCI: Revert to the original speed after PCIe
failed link retraining").

[1] https://lore.kernel.org/r/1c92ef6bcb314ee6977839b46b393282e4f52e74.1750684771.git.lukas@wunner.de


    PCI: Fix link speed calculation on retrain failure

    When pcie_failed_link_retrain() fails to retrain, it tries to revert to the
    previous link speed.  However it calculates that speed from the Link
    Control 2 register without masking out non-speed bits first.

    PCIE_LNKCTL2_TLS2SPEED() converts such incorrect values to
    PCI_SPEED_UNKNOWN (0xff), which in turn causes a WARN splat in
    pcie_set_target_speed():

      pci 0000:00:01.1: [1022:14ed] type 01 class 0x060400 PCIe Root Port
      pci 0000:00:01.1: broken device, retraining non-functional downstream link at 2.5GT/s
      pci 0000:00:01.1: retraining failed
      WARNING: CPU: 1 PID: 1 at drivers/pci/pcie/bwctrl.c:168 pcie_set_target_speed
      RDX: 0000000000000001 RSI: 00000000000000ff RDI: ffff9acd82efa000
      pcie_failed_link_retrain
      pci_device_add
      pci_scan_single_device

    Mask out the non-speed bits in PCIE_LNKCTL2_TLS2SPEED() and
    PCIE_LNKCAP_SLS2SPEED() so they don't incorrectly return PCI_SPEED_UNKNOWN.