Re: [PATCH v5 0/5] vfio/type1: optimize vfio_pin_pages_remote() and vfio_unpin_pages_remote()

[Alex Williamson <alex.williamson@redhat.com>]

On Thu, 14 Aug 2025 14:47:09 +0800
lizhe.67@bytedance.com wrote:

> From: Li Zhe <lizhe.67@bytedance.com>
> 
> This patchset is an integration of the two previous patchsets[1][2].
> 
> When vfio_pin_pages_remote() is called with a range of addresses that
> includes large folios, the function currently performs individual
> statistics counting operations for each page. This can lead to significant
> performance overheads, especially when dealing with large ranges of pages.
> 
> The function vfio_unpin_pages_remote() has a similar issue, where executing
> put_pfn() for each pfn brings considerable consumption.
> 
> This patchset primarily optimizes the performance of the relevant functions
> by batching the less efficient operations mentioned before.
> 
> The first two patch optimizes the performance of the function
> vfio_pin_pages_remote(), while the remaining patches optimize the
> performance of the function vfio_unpin_pages_remote().
> 
> The performance test results, based on v6.16, for completing the 16G
> VFIO MAP/UNMAP DMA, obtained through unit test[3] with slight
> modifications[4], are as follows.
> 
> Base(6.16):
> ------- AVERAGE (MADV_HUGEPAGE) --------
> VFIO MAP DMA in 0.049 s (328.5 GB/s)
> VFIO UNMAP DMA in 0.141 s (113.7 GB/s)
> ------- AVERAGE (MAP_POPULATE) --------
> VFIO MAP DMA in 0.268 s (59.6 GB/s)
> VFIO UNMAP DMA in 0.307 s (52.2 GB/s)
> ------- AVERAGE (HUGETLBFS) --------
> VFIO MAP DMA in 0.051 s (310.9 GB/s)
> VFIO UNMAP DMA in 0.135 s (118.6 GB/s)
> 
> With this patchset:
> ------- AVERAGE (MADV_HUGEPAGE) --------
> VFIO MAP DMA in 0.025 s (633.1 GB/s)
> VFIO UNMAP DMA in 0.044 s (363.2 GB/s)
> ------- AVERAGE (MAP_POPULATE) --------
> VFIO MAP DMA in 0.249 s (64.2 GB/s)
> VFIO UNMAP DMA in 0.289 s (55.3 GB/s)
> ------- AVERAGE (HUGETLBFS) --------
> VFIO MAP DMA in 0.030 s (533.2 GB/s)
> VFIO UNMAP DMA in 0.044 s (361.3 GB/s)
> 
> For large folio, we achieve an over 40% performance improvement for VFIO
> MAP DMA and an over 67% performance improvement for VFIO DMA UNMAP. For
> small folios, the performance test results show a slight improvement with
> the performance before optimization.
> 
> [1]: https://lore.kernel.org/all/20250529064947.38433-1-lizhe.67@bytedance.com/
> [2]: https://lore.kernel.org/all/20250620032344.13382-1-lizhe.67@bytedance.com/#t
> [3]: https://github.com/awilliam/tests/blob/vfio-pci-mem-dma-map/vfio-pci-mem-dma-map.c
> [4]: https://lore.kernel.org/all/20250610031013.98556-1-lizhe.67@bytedance.com/
> 
> Li Zhe (5):
>   mm: introduce num_pages_contiguous()
>   vfio/type1: optimize vfio_pin_pages_remote()
>   vfio/type1: batch vfio_find_vpfn() in function
>     vfio_unpin_pages_remote()
>   vfio/type1: introduce a new member has_rsvd for struct vfio_dma
>   vfio/type1: optimize vfio_unpin_pages_remote()
> 
>  drivers/vfio/vfio_iommu_type1.c | 112 ++++++++++++++++++++++++++------
>  include/linux/mm.h              |   7 +-
>  include/linux/mm_inline.h       |  35 ++++++++++
>  3 files changed, 132 insertions(+), 22 deletions(-)

I've added this to the vfio next branch, please test.  As described
previously, barring objections I'll try to get this into the current
merge window since it almost made v6.17, but was then dropped due to
disagreements in the mm space, then blocked by merge conflicts.  Thanks,

Alex