On 11/21/24 02:55, Paolo Abeni wrote:
>
> On 11/18/24 00:48, dullfire@xxxxxxxxx wrote:
>> From: Jonathan Currier <dullfire@xxxxxxxxx>
>>
>> Commit 7d5ec3d36123 ("PCI/MSI: Mask all unused MSI-X entries")
>> introduces a readl() from ENTRY_VECTOR_CTRL before the writel() to
>> ENTRY_DATA. This is correct, however some hardware, like the Sun Neptune
>> chips, the niu module, will cause an error and/or fatal trap if any MSIX
>> table entry is read before the corresponding ENTRY_DATA field is written
>> to. This patch adds an optional early writel() in msix_prepare_msi_desc().
> Why the issue can't be addressed into the relevant device driver? It
> looks like an H/W bug, a driver specific fix looks IMHO more fitting.
I considered this approach, and thus asked about it in the mailing lists here:
https://lore.kernel.org/sparclinux/7de14cca-e2fa-49f7-b83e-5f8322cc9e56@xxxxxxxxx/T/
It sounds like you are suggesting approach 1 (add the MSIX register writes to niu).
I did do a quick and dirty test of that. However it ended up requiring
msix_map_region(), and pci_msix_desc_addr(), both of are internal to the
msi code, and not exported or in pubic headers. The msix_table_size()
macro was also needed. I could either export those functions, or reproduce
their code in the niu driver. However, as Thomas' suggestion seemed very
simple and elegant to me, I decided to got with it.
If it is actually preferable to either copy those msix functions into niu,
they are not very large, or export them (GPL I would assume?) let me know
and I can make that change.
>
> A cross subsystem series, like this one, gives some extra complication
> to maintainers.
>
> Thanks,
>
> Paolo
Sincerely,
Jonathan Currier
