On Thu, Sep 17, 2020 at 08:32:21AM -0300, Jason Gunthorpe wrote: > On Thu, Sep 17, 2020 at 11:28:19AM +0100, Lorenzo Pieralisi wrote: > > Unfortunately if we merge this patch we _do_ know from this thread > > that userspace will suffer from a perf regression on TX2. > > > > Either we ignore it or we write some code to prevent it > > (ie first step make arch_can_pci_mmap_wc() return 0 on TX2 - > > possibly using the arm64 errata detection mechanism). > > If anyone complains they can send the change to arch_can_pci_mmap_wc() > - I'm pretty sure nobody will complain due to mlx5 If that's the case we can go ahead and merge this patch with a reworded commit log - this thread was very useful nonetheless for my own information (and others hopefully) so thank you. For VFIO WC + "message push ioremap" - to be continued. Clint, please reword the commit and resend, not sure we can hit v5.10 but we shall try. Thanks, Lorenzo
