Re: [PATCH V2] PCI/MSI: Set device flag indicating only 32-bit MSI support

[Vidya Sagar <vidyas@nvidia.com>]

On 12/4/2020 1:24 AM, Bjorn Helgaas wrote:
> External email: Use caution opening links or attachments
> 
> 
> On Fri, Dec 04, 2020 at 12:33:45AM +0530, Vidya Sagar wrote:
>> On 12/3/2020 11:54 PM, Bjorn Helgaas wrote:
>>> On Tue, Nov 24, 2020 at 04:20:35PM +0530, Vidya Sagar wrote:
>>>> There are devices (Ex:- Marvell SATA controller) that don't support
>>>> 64-bit MSIs and the same is advertised through their MSI capability
>>>> register. Set no_64bit_msi flag explicitly for such devices in the
>>>> MSI setup code so that the msi_verify_entries() API would catch
>>>> if the MSI arch code tries to use 64-bit MSI.
>>>
>>> This seems good to me.  I'll post a possible revision to set
>>> dev->no_64bit_msi in the device enumeration path instead of in the IRQ
>>> allocation path, since it's really a property of the device, not of
>>> the msi_desc.
>>>
>>> I like the extra checking this gives us.  Was this prompted by
>>> tripping over something, or is it something you noticed by code
>>> reading?  If the former, a hint about what was wrong and how it's
>>> being fixed would be useful.
>> I observed functionality issue with Marvell SATA controller (1b4b:9171) when
>> the allocated MSI target address was a 64-bit address. I mentioned the
>> Marvell SATA controller as an example in the commit message.
> 
> I know you mentioned the Marvell controller, but apparently that
> device is working perfectly correctly: it does not support 64-bit MSI,
> and it does not advertise support for 64-bit MSI.
> 
> So if there's a functionality issue, that means something is wrong in
> Linux that caused us to assign a 64-bit MSI address to it.  *That*
> issue is what I want to know about.  Your patch only warns about the
> issue; it doesn't fix it.
The issue is in the DWC code. I pushed a generic patch to fix that issue 
by specifying the limit of MSI target address to 32-bit region. Patch is 
up for review at 
http://patchwork.ozlabs.org/project/linux-pci/patch/20201117165312.25847-1-vidyas@nvidia.com/

> 
> I don't think there's any point in specifically mentioning the Marvell
> device if it is working correctly, because the same issue should
> affect *any* device that doesn't support 64-bit MSI.  But if there's
> some arch code that incorrectly assigns a 64-bit address, it would
> definitely be useful to specify the arch.  And hopefully there's a fix
> for that arch code, too.
> 
>>>> Signed-off-by: Vidya Sagar <vidyas@nvidia.com>
>>>> ---
>>>> V2:
>>>> * Addressed Bjorn's comment and changed the error message
>>>>
>>>>    drivers/pci/msi.c | 11 +++++++----
>>>>    1 file changed, 7 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
>>>> index d52d118979a6..8de5ba6b4a59 100644
>>>> --- a/drivers/pci/msi.c
>>>> +++ b/drivers/pci/msi.c
>>>> @@ -581,10 +581,12 @@ msi_setup_entry(struct pci_dev *dev, int nvec, struct irq_affinity *affd)
>>>>         entry->msi_attrib.multi_cap     = (control & PCI_MSI_FLAGS_QMASK) >> 1;
>>>>         entry->msi_attrib.multiple      = ilog2(__roundup_pow_of_two(nvec));
>>>>
>>>> -     if (control & PCI_MSI_FLAGS_64BIT)
>>>> +     if (control & PCI_MSI_FLAGS_64BIT) {
>>>>                 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_64;
>>>> -     else
>>>> +     } else {
>>>>                 entry->mask_pos = dev->msi_cap + PCI_MSI_MASK_32;
>>>> +             dev->no_64bit_msi = 1;
>>>> +     }
>>>>
>>>>         /* Save the initial mask status */
>>>>         if (entry->msi_attrib.maskbit)
>>>> @@ -602,8 +604,9 @@ static int msi_verify_entries(struct pci_dev *dev)
>>>>         for_each_pci_msi_entry(entry, dev) {
>>>>                 if (!dev->no_64bit_msi || !entry->msg.address_hi)
>>>>                         continue;
>>>> -             pci_err(dev, "Device has broken 64-bit MSI but arch"
>>>> -                     " tried to assign one above 4G\n");
>>>> +             pci_err(dev, "Device has either broken 64-bit MSI or "
>>>> +                     "only 32-bit MSI support but "
>>>> +                     "arch tried to assign one above 4G\n");
>>>>                 return -EIO;
>>>>         }
>>>>         return 0;
>>>> --
>>>> 2.17.1
>>>>