[PATCH v2 1/2] PCI: Make pci_msi_setup_pci_dev() non-static for use by arch code

Guilherme G. Piccoli gpiccoli at linux.vnet.ibm.com
Mon Aug 24 22:18:34 AEST 2015


On 08/24/2015 04:37 AM, Michael Ellerman wrote:
>> more: I was testing driver issues on kernel 2.6.32 (RHEL 6.6), and when
>> I tried the mainline kernel, the driver wasn't able to enable MSI-X
>> capabilities. Interestingly, on kernel 4.1 this behavior doesn't happen
>> and the driver can use MSI-X interrupts.
>>
>> So, I figured that something was wrong and found the problem described
>> on the patches. I tried the proposed solution (calling manually the
>> function that is not reachable anymore) and it works.
>>
>> Regarding the bnx2x driver, below are two dmesg outputs:
>>
>> 1) With kernel 4.2-rc7
>> bnx2x 0000:01:00.0: no msix capability found
>
> OK. This is because the initialisation of dev->msix_cap was lost due to commit
> 1851617cd2da.
>
>> 2) With kernel 4.1
>> bnx2x 0000:01:00.0: msix capability found
>> bnx2x 0000:01:00.0 eth2: using MSI-X  IRQs: sp 24  fp[0] 26 ... fp[7] 33
>
> OK. And I assume with these patches you see the above output again.

Exactly. With the proposed patches applied, dev->msix_cap is initialized 
normally, so the driver is able to do its job as usual.


>>> Or anywhere after the first '---', which means the version commentary is
>>> discarded in the final commit.
>>
>> I used scissors, but there's no problem in stop using it in this list.
>
> Thanks, but my scripts don't grok scissors. So I prefer the commentary after
> the '---'.

Thanks for the info.

Cheers



More information about the Linuxppc-dev mailing list