On 6/1/23 6:46 PM, Jason Gunthorpe wrote: > On Thu, Jun 01, 2023 at 03:11:44PM -0400, Dennis Dalessandro wrote: >> On 6/1/23 2:58 PM, Jason Gunthorpe wrote: >>> On Thu, Jun 01, 2023 at 02:15:59PM -0400, Dennis Dalessandro wrote: >>>> On 6/1/23 1:45 PM, Jason Gunthorpe wrote: >>>>> On Fri, May 19, 2023 at 12:54:30PM -0400, Dennis Dalessandro wrote: >>>>>> From: Patrick Kelsey <pat.kelsey@xxxxxxxxxxxxxxxxxxxx> >>>>>> >>>>>> In order to handle user SDMA requests where the packet payload is to be >>>>>> constructed from memory other than system memory, do user SDMA >>>>>> page-pinning through the page-pinning interface. The page-pinning >>>>>> interface lets the user SDMA code operate on user_sdma_iovec objects >>>>>> without caring which type of memory that iovec's iov.iov_base points to. >>>>> >>>>> What is "other than system memory" memory?? >>>> >>>> For instance dmabuff, or something new in the future. This is pre-req work to >>>> make it more abstract and general purpose and design it in a way it probably >>>> should have been in the first place. >>> >>> But why is there uapi components to this? >> >> We added a new sdma req opcode and associated meminfo structure for the sdma >> request. >> >>> And HFI doesn't support DMABUF? >> >> It will through the psm2/libfabric interface soon. > > I don't want to add uapi stuff that has no use.. It has a use, psm2 and libfabric both use this interface. I don't see what the issue is.
