Hi All,
Previously I used virtio-iommu as a non-s390x test vehicle[0] for the
single queue flushing scheme introduced by my s390x DMA API conversion
series[1]. For this I modified virtio-iommu to a) use .iotlb_sync_map
and b) enable IOMMU_CAP_DEFERRED_FLUSH. It turned out that deferred
flush and even just the introduction of ops->iotlb_sync_map yield
performance uplift[2] even with per-CPU queues. So here is a small
series of these two changes.
The code is also available on the b4/viommu-deferred-flush branch of my
kernel.org git repository[3].
Note on testing: I tested this series on my AMD Ryzen 3900X workstation
using QEMU 8.1.2 a pass-through NVMe and Intel 82599 NIC VFs. For the
NVMe I saw an increase of about 10% in IOPS and 30% in read bandwidth
compared with v6.7-rc2. One odd thing though is that QEMU seemed to make
the entire guest resident/pinned once I passed-through a PCI device.
I seem to remember this wasn't the case with my last version but not
sure which QEMU version I used back then.
@Jean-Philippe: I didn't include your R-b's as I changed back to the
nr_endpoints check and this is like 30% of the patches.
Thanks,
Niklas
[0] https://lore.kernel.org/lkml/[email protected]/
[1] https://lore.kernel.org/lkml/[email protected]/
[2] https://lore.kernel.org/lkml/20230802123612.GA6142@myrica/
Signed-off-by: Niklas Schnelle <[email protected]>
---
Changes in v3:
- Removed NULL check from viommu_sync_req() (Jason)
- Went back to checking for 0 endpoints in IOTLB ops (Robin)
- Rebased on v6.7-rc2 which includes necessary iommu-dma changes
- Link to v2: https://lore.kernel.org/r/[email protected]
Changes in v2:
- Check for viommu == NULL in viommu_sync_req() instead of for
0 endpoints in ops (Jean-Philippe)
- Added comment where viommu can be NULL (me)
- Link to v1: https://lore.kernel.org/r/[email protected]
To: Jean-Philippe Brucker <[email protected]>
To: Joerg Roedel <[email protected]>
To: Will Deacon <[email protected]>
To: Jason Gunthorpe <[email protected]>
To: Robin Murphy <[email protected]>
Cc: [email protected],
Cc: [email protected]
Cc: [email protected],
Cc: Niklas Schnelle <[email protected]>
---
Niklas Schnelle (2):
iommu/virtio: Make use of ops->iotlb_sync_map
iommu/virtio: Add ops->flush_iotlb_all and enable deferred flush
drivers/iommu/virtio-iommu.c | 33 ++++++++++++++++++++++++++++++++-
1 file changed, 32 insertions(+), 1 deletion(-)
---
base-commit: 98b1cc82c4affc16f5598d4fa14b1858671b2263
change-id: 20230825-viommu-sync-map-1bf0cc4fdc15
Best regards,
--
Niklas Schnelle
Add ops->flush_iotlb_all operation to enable virtio-iommu for the
dma-iommu deferred flush scheme. This results in a significant increase
in performance in exchange for a window in which devices can still
access previously IOMMU mapped memory when running with
CONFIG_IOMMU_DEFAULT_DMA_LAZY. The previous strict behavior can be
achieved with iommu.strict=1 on the kernel command line or
CONFIG_IOMMU_DEFAULT_DMA_STRICT.
Link: https://lore.kernel.org/lkml/20230802123612.GA6142@myrica/
Signed-off-by: Niklas Schnelle <[email protected]>
---
drivers/iommu/virtio-iommu.c | 16 ++++++++++++++++
1 file changed, 16 insertions(+)
diff --git a/drivers/iommu/virtio-iommu.c b/drivers/iommu/virtio-iommu.c
index 3f18f69d70d8..ef6a87445105 100644
--- a/drivers/iommu/virtio-iommu.c
+++ b/drivers/iommu/virtio-iommu.c
@@ -926,6 +926,19 @@ static int viommu_iotlb_sync_map(struct iommu_domain *domain,
return viommu_sync_req(vdomain->viommu);
}
+static void viommu_flush_iotlb_all(struct iommu_domain *domain)
+{
+ struct viommu_domain *vdomain = to_viommu_domain(domain);
+
+ /*
+ * May be called before the viommu is initialized including
+ * while creating direct mapping
+ */
+ if (!vdomain->nr_endpoints)
+ return;
+ viommu_sync_req(vdomain->viommu);
+}
+
static void viommu_get_resv_regions(struct device *dev, struct list_head *head)
{
struct iommu_resv_region *entry, *new_entry, *msi = NULL;
@@ -1051,6 +1064,8 @@ static bool viommu_capable(struct device *dev, enum iommu_cap cap)
switch (cap) {
case IOMMU_CAP_CACHE_COHERENCY:
return true;
+ case IOMMU_CAP_DEFERRED_FLUSH:
+ return true;
default:
return false;
}
@@ -1071,6 +1086,7 @@ static struct iommu_ops viommu_ops = {
.map_pages = viommu_map_pages,
.unmap_pages = viommu_unmap_pages,
.iova_to_phys = viommu_iova_to_phys,
+ .flush_iotlb_all = viommu_flush_iotlb_all,
.iotlb_sync = viommu_iotlb_sync,
.iotlb_sync_map = viommu_iotlb_sync_map,
.free = viommu_domain_free,
--
2.39.2
Hi Niklas,
On Mon, Nov 20, 2023 at 03:51:55PM +0100, Niklas Schnelle wrote:
> Hi All,
>
> Previously I used virtio-iommu as a non-s390x test vehicle[0] for the
> single queue flushing scheme introduced by my s390x DMA API conversion
> series[1]. For this I modified virtio-iommu to a) use .iotlb_sync_map
> and b) enable IOMMU_CAP_DEFERRED_FLUSH. It turned out that deferred
> flush and even just the introduction of ops->iotlb_sync_map yield
> performance uplift[2] even with per-CPU queues. So here is a small
> series of these two changes.
>
> The code is also available on the b4/viommu-deferred-flush branch of my
> kernel.org git repository[3].
>
> Note on testing: I tested this series on my AMD Ryzen 3900X workstation
> using QEMU 8.1.2 a pass-through NVMe and Intel 82599 NIC VFs. For the
> NVMe I saw an increase of about 10% in IOPS and 30% in read bandwidth
> compared with v6.7-rc2. One odd thing though is that QEMU seemed to make
> the entire guest resident/pinned once I passed-through a PCI device.
> I seem to remember this wasn't the case with my last version but not
> sure which QEMU version I used back then.
That's probably expected, now that boot-bypass is enabled by default: on
VM boot, endpoints are able to do DMA to the entire guest-physical address
space, until a virtio-iommu driver disables global bypass in the config
space (at which point the pinned memory is hopefully reclaimed by the
host). QEMU enables it by default to mimic other IOMMU implementations,
and to allow running firmware or OS that don't support virtio-iommu. It
can be disabled with boot-bypass=off
> @Jean-Philippe: I didn't include your R-b's as I changed back to the
> nr_endpoints check and this is like 30% of the patches.
Thank you for the patches. For the series:
Reviewed-by: Jean-Philippe Brucker <[email protected]>
On Mon, Nov 20, 2023 at 03:51:55PM +0100, Niklas Schnelle wrote:
> Niklas Schnelle (2):
> iommu/virtio: Make use of ops->iotlb_sync_map
> iommu/virtio: Add ops->flush_iotlb_all and enable deferred flush
>
> drivers/iommu/virtio-iommu.c | 33 ++++++++++++++++++++++++++++++++-
> 1 file changed, 32 insertions(+), 1 deletion(-)
Applied, thanks.