Dropped/disabled NVME support in qemu

Dropped/disabled NVME support in qemu

[Guenter Roeck]

Hi all,

since qemu v5.1, nvme support has been dropped from several architectures.
Well, it was not exactly dropped, but disabled.

On those architectures, trying to boot from an nvme drive now results
in the following or a similar error message.

qemu-system-alpha: -device nvme,serial=foo,drive=d0: MSI-X is not supported by interrupt controller

This is the result of the following two commits.

fbf2e5375e ("hw/block/nvme: Verify msix_vector_use() returned value")
1c0c2163aa ("hw/block/nvme: verify msix_init_exclusive_bar() return value")

Affected architectures are alpha, parisc, ppc, sh, and sparc, and every other
architecture/platform where the interrupt controller does not support MSI-X.
After reverting above patches, I can boot from nvme again.

This does not really matter for me - I run my own versions of qemu anyway,
and don't mind carrying reverts if I have to. However, for my education,
I would like to understand why nvme support was disabled, especially since
it seemed to work for me just fine for years. Can someone please explain ?

Thanks,
Guenter

Re: Dropped/disabled NVME support in qemu

[Philippe Mathieu-Daudé]

Cc'ing NVMe emulation maintainers.

On 12/21/20 6:12 AM, Guenter Roeck wrote:
> Hi all,
> 
> since qemu v5.1, nvme support has been dropped from several architectures.
> Well, it was not exactly dropped, but disabled.
> 
> On those architectures, trying to boot from an nvme drive now results
> in the following or a similar error message.
> 
> qemu-system-alpha: -device nvme,serial=foo,drive=d0: MSI-X is not supported by interrupt controller
> 
> This is the result of the following two commits.
> 
> fbf2e5375e ("hw/block/nvme: Verify msix_vector_use() returned value")
> 1c0c2163aa ("hw/block/nvme: verify msix_init_exclusive_bar() return value")
> 
> Affected architectures are alpha, parisc, ppc, sh, and sparc, and every other
> architecture/platform where the interrupt controller does not support MSI-X.
> After reverting above patches, I can boot from nvme again.
> 
> This does not really matter for me - I run my own versions of qemu anyway,
> and don't mind carrying reverts if I have to. However, for my education,
> I would like to understand why nvme support was disabled, especially since
> it seemed to work for me just fine for years. Can someone please explain ?
> 
> Thanks,
> Guenter
> 

Re: Dropped/disabled NVME support in qemu

[Klaus Jensen]

[-- Attachment #1: Type: text/plain, Size: 1661 bytes --]

On Dec 21 15:11, Philippe Mathieu-Daudé wrote:
> Cc'ing NVMe emulation maintainers.
> 
> On 12/21/20 6:12 AM, Guenter Roeck wrote:
> > Hi all,
> > 
> > since qemu v5.1, nvme support has been dropped from several architectures.
> > Well, it was not exactly dropped, but disabled.
> > 
> > On those architectures, trying to boot from an nvme drive now results
> > in the following or a similar error message.
> > 
> > qemu-system-alpha: -device nvme,serial=foo,drive=d0: MSI-X is not supported by interrupt controller
> > 
> > This is the result of the following two commits.
> > 
> > fbf2e5375e ("hw/block/nvme: Verify msix_vector_use() returned value")
> > 1c0c2163aa ("hw/block/nvme: verify msix_init_exclusive_bar() return value")
> > 
> > Affected architectures are alpha, parisc, ppc, sh, and sparc, and every other
> > architecture/platform where the interrupt controller does not support MSI-X.
> > After reverting above patches, I can boot from nvme again.
> > 
> > This does not really matter for me - I run my own versions of qemu anyway,
> > and don't mind carrying reverts if I have to. However, for my education,
> > I would like to understand why nvme support was disabled, especially since
> > it seemed to work for me just fine for years. Can someone please explain ?
> > 
> > Thanks,
> > Guenter
> > 
> 

Hi Guenter,

Thanks for reporting this. My apologies - this is an unintended
regression. I'll get a revert staged.

I'll of course also look into why this was not picked up doing testing.
There is an nvme QTest that should have picked this up, but it obviously
was not run for these architectures.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: Dropped/disabled NVME support in qemu

[Guenter Roeck]

[-- Attachment #1.1: Type: text/plain, Size: 1907 bytes --]

Hi Klaus,

On 12/21/20 7:17 AM, Klaus Jensen wrote:
> On Dec 21 15:11, Philippe Mathieu-Daudé wrote:
>> Cc'ing NVMe emulation maintainers.
>>
>> On 12/21/20 6:12 AM, Guenter Roeck wrote:
>>> Hi all,
>>>
>>> since qemu v5.1, nvme support has been dropped from several architectures.
>>> Well, it was not exactly dropped, but disabled.
>>>
>>> On those architectures, trying to boot from an nvme drive now results
>>> in the following or a similar error message.
>>>
>>> qemu-system-alpha: -device nvme,serial=foo,drive=d0: MSI-X is not supported by interrupt controller
>>>
>>> This is the result of the following two commits.
>>>
>>> fbf2e5375e ("hw/block/nvme: Verify msix_vector_use() returned value")
>>> 1c0c2163aa ("hw/block/nvme: verify msix_init_exclusive_bar() return value")
>>>
>>> Affected architectures are alpha, parisc, ppc, sh, and sparc, and every other
>>> architecture/platform where the interrupt controller does not support MSI-X.
>>> After reverting above patches, I can boot from nvme again.
>>>
>>> This does not really matter for me - I run my own versions of qemu anyway,
>>> and don't mind carrying reverts if I have to. However, for my education,
>>> I would like to understand why nvme support was disabled, especially since
>>> it seemed to work for me just fine for years. Can someone please explain ?
>>>
>>> Thanks,
>>> Guenter
>>>
>>
> 
> Hi Guenter,
> 
> Thanks for reporting this. My apologies - this is an unintended
> regression. I'll get a revert staged.
> 
> I'll of course also look into why this was not picked up doing testing.
> There is an nvme QTest that should have picked this up, but it obviously
> was not run for these architectures.
> 

Ah, that explains a lot. Just a thought: instead of reverting the patches,
it might possibly make sense to handle lack of MSI-X support explicitly.

Thanks,
Guenter


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 833 bytes --]