On Fri, Jun 13, 2025 at 02:50:48AM +0800, Edgecombe, Rick P wrote: > On Thu, 2025-06-12 at 15:19 +0800, Yan Zhao wrote: > > > TDX isn't setting PFERR_WRITE_MASK or PFERR_PRESENT_MASK in the error_code > > > passed into the fault handler. So page_fault_can_be_fast() should return > > > false > > > for that reason for private/mirror faults. > > Hmm, TDX does set PFERR_WRITE_MASK in the error_code when fault->prefetch is > > false (since exit_qual is set to EPT_VIOLATION_ACC_WRITE in > > tdx_handle_ept_violation()). > > > > PFERR_PRESENT_MASK is always unset. > > > > page_fault_can_be_fast() does always return false for private mirror faults > > though, due to the reason in > > https://lore.kernel.org/kvm/aEp6pDQgbjsfrg2h@xxxxxxxxxxxxxxxxxxxxxxxxx :) > > Seems cleanup worthy to me, but not a bug. I think we should follow up, > depending on the scope of Sean's cleanup. Ok. There was an explicit disallowing of fast page fault for mirror [1]. Maybe we can add it back after Sean's cleanup. [1] https://lore.kernel.org/kvm/af70ce8626cb7366d9b86a41c5d731f8ebd144ff.1708933498.git.isaku.yamahata@xxxxxxxxx/
