Re: [PATCH] iommu/vt-d: Introduce a rb_tree for looking up device

[Baolu Lu <baolu.lu@linux.intel.com>]

On 2023/8/22 0:52, Jason Gunthorpe wrote:
> On Mon, Aug 21, 2023 at 12:16:59AM -0700, Huang Jiaqing wrote:
>> The existing IO page fault handler locates the PCI device by calling
>> pci_get_domain_bus_and_slot(), which searches the list of all PCI
>> devices until the desired PCI device is found. This is inefficient
>> because the algorithm efficiency of searching a list is O(n). In the
>> critical path of handling an IO page fault, this can cause a significant
>> performance bottleneck.
>>
>> To improve the performance of the IO page fault handler, replace
>> pci_get_domain_bus_and_slot() with a local red-black tree. A red-black
>> tree is a self-balancing binary search tree, which means that the
>> average time complexity of searching a red-black tree is O(log(n)). This
>> is significantly faster than O(n), so it can significantly improve the
>> performance of the IO page fault handler.
>>
>> In addition, we can only insert the affected devices (those that have IO
>> page fault enabled) into the red-black tree. This can further improve
>> the performance of the IO page fault handler.
>>
>> Signed-off-by: Huang Jiaqing<jiaqing.huang@intel.com>
>> ---
>>   drivers/iommu/intel/iommu.c | 68 +++++++++++++++++++++++++++++++++++++
>>   drivers/iommu/intel/iommu.h |  8 +++++
>>   drivers/iommu/intel/svm.c   | 13 +++----
>>   3 files changed, 81 insertions(+), 8 deletions(-)
> I feel like this should be a helper library provided by the core
> code, doesn't every PRI driver basically need the same thing?

It seems to be. pci_get_domain_bus_and_slot() is also used in the amd
driver. And the risc-v iommu driver under discussion is also proposing
this.

Best regards,
baolu