NAK/Cmt: [SRU][J][PATCH 0/1] UBUNTU SAUCE: spi: zynqmp-gqspi: fix broken driver

[Stewart Hore]

On 1/7/26 18:47, Stewart Hore wrote:
> BugLink: https://bugs.launchpad.net/bugs/2137603
>
> SRU Justification:
>
> [Impact]
>
> The FPGA userspace firmware update tool is broken due to a change in the
> spi-zynqmp-gqspi driver.
>
> ```
> ubuntu at kria:~$ sudo image_update -i BOOT-k26-smk-v1.04-20250410071436.bin
> [sudo] password for ubuntu:
> Marking last booted image as bootable
> Reading Image file
> Updating Image A
> Marking target image as non bootable
> Writing Image to Image A partition
> [ 7772.972367] zynqmp-qspi ff0f0000.spi: Chip select timed out
> [ 7773.976364] zynqmp-qspi ff0f0000.spi: Chip select timed out
> [ 7774.980366] zynqmp-qspi ff0f0000.spi: Chip select timed out
> ```
>
> With the commit reverted the tool works again:
> ```
> $ sudo image_update -i BOOT-k26-smk-v1.04-20250410071436.bin
> [sudo] password for ubuntu:
> Marking last booted image as bootable
> Reading Image file
> Updating Image A
> Marking target image as non bootable
> Writing Image to Image A partition
> Marking target image as non bootable and requested image
> Clearing multiboot register value
> BOOT-k26-smk-v1.04-20250410071436.bin successfully updated to Image A partition
> ```
>
>
> [Fix]
>
> A bisect identified commit `90b759a09f30d612844be245ff7e80c67eb22be5`
> changed the IRQ handler which broke the spi-zynqmp-gqspi driver.
> Reverting this commit fixes the issue.
>
>
> [Test Plan]
>
> The revert commit has been tested locally on KV260 Rev 1 #202405-34037
> in the linux-xilinx-zynqmp kernel. The test cases above show how to test
> the changes.
>
> [Where problems could occur]
>
> This fixes a regression. Further regressions can be considered low. The
> offending commit was to correct behaviour in the driver in the 6.x
> kernel family.
>
> [Other Info]
>
> This can be considered low priority because the change has been applied
> to the linux-xilinx-zynqmp kernel already to resolve the issue on
> hardware. This change fixes the problem in the parent kernel as well.
>
> I have not investigated if adding further upstream commits from 6.x
> kernel back to 5.15 resolves the issue without reverting the commit. I
> ran a bisect to identify the offending commit; it is isolated to
> spi-zynqmp-gqspi driver changes and can be easily reverted.
>
> Stewart Hore (1):
>   UBUNTU: SAUCE: Revert "spi: zynqmp-gqspi: Always acknowledge
>     interrupts"
>
>  drivers/spi/spi-zynqmp-gqspi.c | 20 ++++++++++++--------
>  1 file changed, 12 insertions(+), 8 deletions(-)


I was a bit too eager with a revert commit for generic. The breaking commit is in upstream 5.15, which means a revert commit in Ubuntu generic isn't the right fix. Either upstream is broken or there are other commits we missed in our kernel that need to go with the breaking commit. Fixing either of those cases should be the way to go instead of just reverting this commit in Ubuntu generic 5.15.


> --
> 2.51.0
>