On Tue, Jan 18, 2022 at 11:12:43AM +0800, Zhu, Lingshan wrote:
> 
> 
> On 1/18/2022 11:01 AM, Jason Wang wrote:
> 
>     On Mon, Jan 17, 2022 at 3:57 PM Michael S. Tsirkin <[email protected]> 
> wrote:
> 
>         On Mon, Jan 17, 2022 at 02:15:55PM +0800, Jason Wang wrote:
> 
>             在 2022/1/14 下午9:39, Michael S. Tsirkin 写道:
> 
>                 The spec says (v1.1 4.1.4.5 ISR status capability):
> 
>                 The VIRTIO_PCI_CAP_ISR_CFG capability refers to at least a 
> single byte, which contains the 8­bit ISR
>                 status field to be used for INT#x interrupt handling.
> 
>                 and
> 
>                 to avoid an extra access, simply reading this register resets 
> it to 0 and causes the device to de­assert the
>                 interrupt.
>                 In this way, driver read of ISR status causes the device to 
> de­assert an interrupt.
> 
>                 See sections 4.1.5.3 and 4.1.5.4 for how this is used.
> 
>                 and in 4.1.5.4 Notification of Device Configuration Changes
> 
>                 it says:
> 
>                 • If MSI­X capability is disabled:
>                 1. Set the second lower bit of the ISR Status field for the 
> device.
>                 2. Send the appropriate PCI interrupt for the device.
> 
>                 If MSI­X capability is enabled:
>                 1. If config_msix_vector is not NO_VECTOR, request the 
> appropriate MSI­X interrupt message for
>                 the device, config_msix_vector sets the MSI­X Table entry 
> number.
> 
>                 all of the above make it looks like VIRTIO_PCI_CAP_ISR_CFG 
> capability is
>                 unused with MSIX.
> 
>                 This was actually the way the spec was understood by
>                 Zhu Lingshan from Intel (Cc'd).
> 
>                 However, looking at the conformance statements, one finds out 
> this is
>                 not the case:
> 
>                 4.1.4.5.1 Device Requirements: ISR status capability
> 
> 
>                 The device MUST present at least one VIRTIO_PCI_CAP_ISR_CFG 
> capability.
>                 The device MUST set the Device Configuration Interrupt bit in 
> ISR status before sending a device configu­
>                 ration change notification to the driver.
>                 If MSI­X capability is disabled, the device MUST set the 
> Queue Interrupt bit in ISR status before sending a
>                 virtqueue notification to the driver.
> 
>                 which implies that the Device Configuration Interrupt bit is 
> set unconditionally.
> 
> 
> 
>                 It is unfortunate that it does not copy this requirement in 
> more places,
>                 and that the non-conformance text is incomplete and does not
>                 mention the MSI-X usage at all.
> 
>                 I propose to extend 4.1.4.5 ISR status capability and
>                 4.1.5.4 Notification of Device Configuration Changes
>                 to mention the MSI use.
> 
> I agree to expand ISR cap usage to MSI(MSIX) usage with clear descriptions,
> and remove the limitations to MSIX. E.g,:
> 4.1.4.5.2 Driver Requirements: ISR status capability
> If MSI-X capability is enabled, the driver SHOULD NOT access ISR status upon
> detecting a Queue Interrupt.
> 
> Thanks
> 

No, we actually need that for bypass. Only config interrupts set ISR.

I will need to send a patch, people are still confused.


>             I wonder do we want
> 
>             1) mandate ISR bit
> 
>             or
> 
>             2) remove the ISR bit set for MSI mode?
> 
>             1) is the current Qemu behavior but seems a little bit contradict 
> with the
>             goal of MSI.
> 
>             Thanks
> 
> 
> 
>         I think we want the ISR bit
> 
>     This means there's no chance to use fast irq path but since it's a
>     less frequent operation. It should be fine.
> 
>     Thanks
> 
> 
>         since otherwise we need to go read
>         a ton of config space fields to check whether anything changed.
> 
>         --
>         MST
> 
> 
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to