pci_alloc_irq_vectors fails ENOSPC for XPS 13 9310

[Thomas Gleixner]

On Tue, Nov 03 2020 at 10:08, Bjorn Helgaas wrote:
> On Tue, Nov 03, 2020 at 08:49:06AM +0200, Kalle Valo wrote:
>> Bjorn Helgaas <helgaas at kernel.org> writes:
>> > On Mon, Nov 02, 2020 at 08:49:51PM +0200, Kalle Valo wrote:
>> >> Thomas Krause <thomaskrause at posteo.de> writes:
>> >> 
>> >> >> I had the same problem as well back in the days, for me enabling
>> >> >> CONFIG_IRQ_REMAP helped. If it helps for you also I wonder if we should
>> >> >> mention that in the ath11k warning above :)

Interrupt remapping only helps when the device supports only MSI (not
MSI-X) because x86 (kernel) does not support multiple MSI interrupts
without remapping.

So if only MSI is available then you get exactly _one_ MSI vector
without remapping.

>> >> > CONFIG_IRQ_REMAP did not do the trick.

The config alone does not help. The hardware has to support it and the
BIOS has to enable it.

Check the BIOS for a switch which is named 'VT-d' or such. It might
depend on 'Intel Virtualization Technology' or such.

>> >   00:1c.0 PCI bridge: Intel Corporation Device a0b8 (rev 20)
>> > 	Bus: primary=00, secondary=56, subordinate=56, sec-latency=0
>> > 	Memory behind bridge: 8c300000-8c3fffff [size=1M]
>> >   56:00.0 Network controller: Qualcomm Device 1101 (rev 01)
>> >      Region 0: Memory at 8c300000 (64-bit, non-prefetchable) [size=1M]

So I grabbed the PCI info from the link and it has:

     Capabilities: [50] MSI: Enable- Count=1/32 Maskable+ 64bit-

So no MSI-X, ergo only one MSI interrupt without remapping.
 
>> >> To summarise: Thomas is reporting[1] a problem with ath11k on QCA6390
>> >> PCI device where he is not having enough MSI vectors. ath11k needs 32
>> >> vectors but pci_alloc_irq_vectors() returns -ENOSPC. PCI support is new
>> >> for ath11k and introduced in v5.10-rc1. The irq allocation code is in
>> >> drivers/net/wireless/ath/ath11k/pci.c. [2]
>
>> > But it seems a little greedy if the device can't operate at all unless
>> > it gets 32 vectors.  Are you sure that's a hard requirement?  Most
>> > devices can work with fewer vectors, even if it reduces performance.

Right, even most high end network cards work with one interrupt.

>> This was my first reaction as well when I saw the code for the first
>> time. And the reply I got is that the firmware needs all 32 vectors, it
>> won't work with less.

Great design.

> I do see a couple other drivers that are completely inflexible (they
> request min==max).  But I don't know the system constraint you're
> hitting.  CC'd Thomas & Christoph in case they have time to give us a
> hint.

Can I have a full dmesg please?

Please enable CONFIG_IRQ_REMAP and CONFIG_INTEL_IOMMU (not strictly
required, but it's a Dell BIOS after all). Also set
CONFIG_INTEL_IOMMU_DEFAULT_ON.

Or simply try a distro kernel.

Thanks,

        tglx