[PATCH 1/2] powerpc/powernv: reduce multi-hit of iommu_add_device()

Wei Yang weiyang at linux.vnet.ibm.com
Wed Apr 30 13:28:03 EST 2014


On Wed, Apr 30, 2014 at 10:28:12AM +1000, Gavin Shan wrote:
>
>It seems that we have 2 problems here:
>
>- For non-SRIOV case, pcibios_setup_device() is called for towice. That
>  seems incorrect. We could simply remove pcibios_setup_bus_devices()
>  from pcibios_fixup_bus().

I have thought about this solution before, but I guess this would have some
side effect on other platforms.

Well, just did a test by removing this line in pcibios_fixup_bus(), the result
is:
1. system up, thanks god.
2. no one invoke the pcibios_setup_device() at bootup time.

The reason for no one invoke the pcibios_setup_device() is: in
pcibios_add_device() it will check whether the bus is added before calling
pcibios_setup_device(). And at this time, the bus is not added.

Still wierd, why the system could be up. But one thing for sure is, no one
invoke the pcibios_setup_device() at bootup stage. So this solution may not
work.

>
>- It's too early to register IOMMU group/device in pnv_pci_ioda_dma_dev_setup()
>  because the sysfs entries of the PCI device aren't finalized yet. So we could
>  remove all logic we have in pnv_pci_ioda_dma_dev_setup() and just purely rely
>  on "tce_iommu_bus_notifier".

This would be another solution. 

One concern:

    If we want to do like this, we need to retrieve the pe information and
    get the tce32_table base in tce_iommu_bus_notifier. Hmm... looks a little
    not that nice.

>
>By the way, I never tried EEH on SRIOV PF/VFs. However, I never hit similar
>issue in non-SRIOV cases.

I have test this case on a PF with no VF enabled.

I just make the mlx4_pci_err_detected() return PCI_ERS_RESULT_NONE and do
nothing. So this will force the eeh to do a hotplug recovery.

You could have a try on your machine too.

>
>Thanks,
>Gavin

-- 
Richard Yang
Help you, Help me



More information about the Linuxppc-dev mailing list