mbox series

[SRU,bionic:linux-azure-4.15,0/1] UBUNTU: SAUCE: Call trace during nvme storage performance test in bionic/linux-azure 4.15.0-1154

Message ID 20221115170031.13179-1-john.cabaj@canonical.com
Headers show
Series UBUNTU: SAUCE: Call trace during nvme storage performance test in bionic/linux-azure 4.15.0-1154 | expand

Message

John Cabaj Nov. 15, 2022, 5 p.m. UTC
[Impact]

Hypervisor only allocates interrupts to happen on a single physical GPU, which can lead to CPU soft locks when using PCI NVME under heavy load.

[Test Case]

Microsoft tested

[Where things could go wrong]

Hypervisor PCI driver may not load correctly. Patch did not apply cleanly and could lead to merge conflicts once the upstream patch is pulled in.

[Other Info]

SF: #00347669

John Cabaj (1):
  PCI: hv: Only reuse existing IRTE allocation for Multi-MSI

 drivers/pci/host/pci-hyperv.c | 68 +++++++++++++++++++++++++++--------
 1 file changed, 53 insertions(+), 15 deletions(-)

Comments

Tim Gardner Nov. 15, 2022, 6:09 p.m. UTC | #1
On 11/15/22 10:00 AM, John Cabaj wrote:
> [Impact]
> 
> Hypervisor only allocates interrupts to happen on a single physical GPU, which can lead to CPU soft locks when using PCI NVME under heavy load.
> 
> [Test Case]
> 
> Microsoft tested
> 
> [Where things could go wrong]
> 
> Hypervisor PCI driver may not load correctly. Patch did not apply cleanly and could lead to merge conflicts once the upstream patch is pulled in.
> 
> [Other Info]
> 
> SF: #00347669
> 
> John Cabaj (1):
>    PCI: hv: Only reuse existing IRTE allocation for Multi-MSI
> 
>   drivers/pci/host/pci-hyperv.c | 68 +++++++++++++++++++++++++++--------
>   1 file changed, 53 insertions(+), 15 deletions(-)
> 
Acked-by: Tim Gardner <tim.gardner@canonical.com>

Good test results
Tim Gardner Nov. 16, 2022, 2:05 p.m. UTC | #2
On 11/15/22 11:09 AM, Tim Gardner wrote:
> On 11/15/22 10:00 AM, John Cabaj wrote:
>> [Impact]
>>
>> Hypervisor only allocates interrupts to happen on a single physical 
>> GPU, which can lead to CPU soft locks when using PCI NVME under heavy 
>> load.
>>
>> [Test Case]
>>
>> Microsoft tested
>>
>> [Where things could go wrong]
>>
>> Hypervisor PCI driver may not load correctly. Patch did not apply 
>> cleanly and could lead to merge conflicts once the upstream patch is 
>> pulled in.
>>
>> [Other Info]
>>
>> SF: #00347669
>>
>> John Cabaj (1):
>>    PCI: hv: Only reuse existing IRTE allocation for Multi-MSI
>>
>>   drivers/pci/host/pci-hyperv.c | 68 +++++++++++++++++++++++++++--------
>>   1 file changed, 53 insertions(+), 15 deletions(-)
>>
> Acked-by: Tim Gardner <tim.gardner@canonical.com>
> 
> Good test results

I did forget to mention that this patch needs additional provenance 
(which I'll fix at application time). Its now in linux-next: commit 
c234ba8042920fa83635808dc5673f36869ca280 ("PCI: hv: Only reuse existing 
IRTE allocation for Multi-MSI").

rtg
Tim Gardner Nov. 16, 2022, 2:31 p.m. UTC | #3
On 11/15/22 10:00 AM, John Cabaj wrote:
> [Impact]
> 
> Hypervisor only allocates interrupts to happen on a single physical GPU, which can lead to CPU soft locks when using PCI NVME under heavy load.
> 
> [Test Case]
> 
> Microsoft tested
> 
> [Where things could go wrong]
> 
> Hypervisor PCI driver may not load correctly. Patch did not apply cleanly and could lead to merge conflicts once the upstream patch is pulled in.
> 
> [Other Info]
> 
> SF: #00347669
> 
> John Cabaj (1):
>    PCI: hv: Only reuse existing IRTE allocation for Multi-MSI
> 
>   drivers/pci/host/pci-hyperv.c | 68 +++++++++++++++++++++++++++--------
>   1 file changed, 53 insertions(+), 15 deletions(-)
> 
Applied to bionic:linux-azure-4.15. Thanks.

I updated the commit message to reflect proper provenance.

-rtg