Now we regenerate vconfig for all the BARs via vfio_bar_fixup(), every time
any offset of any of them are read. Though BARs aren't re-read regularly,
the regeneration can be avoid if no BARs had been written since they were
last read, in which case the vdev->bardirty is false.
Let's predicate the vfio_bar_fixup() on the bardirty so that it can return
immediately if !bardirty.
Suggested-by: Alex Williamson <[email protected]>
Signed-off-by: Zenghui Yu <[email protected]>
---
* From v1:
- Per Alex's suggestion, let vfio_bar_fixup() test vdev->bardirty to
avoid doing work if bardirty is false, instead of removing it entirely.
- Rewrite the commit message.
drivers/vfio/pci/vfio_pci_config.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/vfio/pci/vfio_pci_config.c b/drivers/vfio/pci/vfio_pci_config.c
index d98843feddce..5e02ba07e8e8 100644
--- a/drivers/vfio/pci/vfio_pci_config.c
+++ b/drivers/vfio/pci/vfio_pci_config.c
@@ -467,6 +467,9 @@ static void vfio_bar_fixup(struct vfio_pci_device *vdev)
__le32 *vbar;
u64 mask;
+ if (!vdev->bardirty)
+ return;
+
vbar = (__le32 *)&vdev->vconfig[PCI_BASE_ADDRESS_0];
for (i = 0; i < PCI_STD_NUM_BARS; i++, vbar++) {
--
2.19.1
On Mon, 21 Sep 2020 12:51:16 +0800
Zenghui Yu <[email protected]> wrote:
> Now we regenerate vconfig for all the BARs via vfio_bar_fixup(), every time
> any offset of any of them are read. Though BARs aren't re-read regularly,
> the regeneration can be avoid if no BARs had been written since they were
s/avoid/avoided/
> last read, in which case the vdev->bardirty is false.
s/the//
>
> Let's predicate the vfio_bar_fixup() on the bardirty so that it can return
> immediately if !bardirty.
Maybe
"Let's return immediately in vfio_bar_fixup() if bardirty is false." ?
>
> Suggested-by: Alex Williamson <[email protected]>
> Signed-off-by: Zenghui Yu <[email protected]>
> ---
> * From v1:
> - Per Alex's suggestion, let vfio_bar_fixup() test vdev->bardirty to
> avoid doing work if bardirty is false, instead of removing it entirely.
> - Rewrite the commit message.
>
> drivers/vfio/pci/vfio_pci_config.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/drivers/vfio/pci/vfio_pci_config.c b/drivers/vfio/pci/vfio_pci_config.c
> index d98843feddce..5e02ba07e8e8 100644
> --- a/drivers/vfio/pci/vfio_pci_config.c
> +++ b/drivers/vfio/pci/vfio_pci_config.c
> @@ -467,6 +467,9 @@ static void vfio_bar_fixup(struct vfio_pci_device *vdev)
> __le32 *vbar;
> u64 mask;
>
> + if (!vdev->bardirty)
Finally, bardirty can actually affect something :)
> + return;
> +
> vbar = (__le32 *)&vdev->vconfig[PCI_BASE_ADDRESS_0];
>
> for (i = 0; i < PCI_STD_NUM_BARS; i++, vbar++) {
Reviewed-by: Cornelia Huck <[email protected]>
Hi Cornelia,
On 2020/9/21 18:21, Cornelia Huck wrote:
> On Mon, 21 Sep 2020 12:51:16 +0800
> Zenghui Yu <[email protected]> wrote:
>
>> Now we regenerate vconfig for all the BARs via vfio_bar_fixup(), every time
>> any offset of any of them are read. Though BARs aren't re-read regularly,
>> the regeneration can be avoid if no BARs had been written since they were
>
> s/avoid/avoided/
>
>> last read, in which case the vdev->bardirty is false.
>
> s/the//
>
>>
>> Let's predicate the vfio_bar_fixup() on the bardirty so that it can return
>> immediately if !bardirty.
>
> Maybe
>
> "Let's return immediately in vfio_bar_fixup() if bardirty is false." ?
Yes.
>>
>> Suggested-by: Alex Williamson <[email protected]>
>> Signed-off-by: Zenghui Yu <[email protected]>
>> ---
>> * From v1:
>> - Per Alex's suggestion, let vfio_bar_fixup() test vdev->bardirty to
>> avoid doing work if bardirty is false, instead of removing it entirely.
>> - Rewrite the commit message.
>>
>> drivers/vfio/pci/vfio_pci_config.c | 3 +++
>> 1 file changed, 3 insertions(+)
>>
>> diff --git a/drivers/vfio/pci/vfio_pci_config.c b/drivers/vfio/pci/vfio_pci_config.c
>> index d98843feddce..5e02ba07e8e8 100644
>> --- a/drivers/vfio/pci/vfio_pci_config.c
>> +++ b/drivers/vfio/pci/vfio_pci_config.c
>> @@ -467,6 +467,9 @@ static void vfio_bar_fixup(struct vfio_pci_device *vdev)
>> __le32 *vbar;
>> u64 mask;
>>
>> + if (!vdev->bardirty)
>
> Finally, bardirty can actually affect something :)
>
>> + return;
>> +
>> vbar = (__le32 *)&vdev->vconfig[PCI_BASE_ADDRESS_0];
>>
>> for (i = 0; i < PCI_STD_NUM_BARS; i++, vbar++) {
>
> Reviewed-by: Cornelia Huck <[email protected]>
Thanks for you review! I think Alex can help fix the commit message when
applying? Otherwise I can send a v3.
Thanks,
Zenghui