Intent of this post:
Seek reviews from Intel reviewers and anyone else in the list
interested in IO performance in confidential VMs. Need some acked-by
reviewed-by tags before I can add swiotlb maintainers to "to/cc" lists
and ask for a review from them.
Changes from v1 to v2:
- rebase to the latest dma-mapping tree.
- drop the duplicate patch for mitigating lock contention
- re-collect perf data
swiotlb is now widely used by confidential VMs. This series optimizes
swiotlb to reduce cache misses and lock contention during bounce buffer
allocation/free and memory bouncing to improve IO workload performance in
confidential VMs.
Here are some FIO tests we did to demonstrate the improvement.
Test setup
----------
A normal VM with 8vCPU and 32G memory, swiotlb is enabled by swiotlb=force.
FIO block size is 4K and iodepth is 256. Note that a normal VM is used so
that others lack of necessary hardware to host confidential VMs can reproduce
results below.
Results
-------
1 FIO job read/write IOPS (k)
vanilla read 216
write 251
optimized read 250
write 270
1-job FIO sequential read/write perf increase by 19% and 8% respectively.
Chao Gao (2):
swiotlb: use bitmap to track free slots
swiotlb: Allocate memory in a cache-friendly way
include/linux/swiotlb.h | 8 ++-
kernel/dma/swiotlb.c | 127 +++++++++++++++++-----------------------
2 files changed, 60 insertions(+), 75 deletions(-)
--
2.25.1
Ping. Intel reviewers, comments and suggestions are welcome.
On Mon, Jul 18, 2022 at 09:28:16AM +0800, Chao Gao wrote:
>Intent of this post:
> Seek reviews from Intel reviewers and anyone else in the list
> interested in IO performance in confidential VMs. Need some acked-by
> reviewed-by tags before I can add swiotlb maintainers to "to/cc" lists
> and ask for a review from them.
>
>Changes from v1 to v2:
>- rebase to the latest dma-mapping tree.
>- drop the duplicate patch for mitigating lock contention
>- re-collect perf data
>
>swiotlb is now widely used by confidential VMs. This series optimizes
>swiotlb to reduce cache misses and lock contention during bounce buffer
>allocation/free and memory bouncing to improve IO workload performance in
>confidential VMs.
>
>Here are some FIO tests we did to demonstrate the improvement.
>
>Test setup
>----------
>
>A normal VM with 8vCPU and 32G memory, swiotlb is enabled by swiotlb=force.
>FIO block size is 4K and iodepth is 256. Note that a normal VM is used so
>that others lack of necessary hardware to host confidential VMs can reproduce
>results below.
>
>Results
>-------
>
>1 FIO job read/write IOPS (k)
>vanilla read 216
> write 251
>optimized read 250
> write 270
>
>1-job FIO sequential read/write perf increase by 19% and 8% respectively.
>
>Chao Gao (2):
> swiotlb: use bitmap to track free slots
> swiotlb: Allocate memory in a cache-friendly way
>
> include/linux/swiotlb.h | 8 ++-
> kernel/dma/swiotlb.c | 127 +++++++++++++++++-----------------------
> 2 files changed, 60 insertions(+), 75 deletions(-)
>
>--
>2.25.1
>