On Fri, Sep 12, 2014 at 12:19:37PM -0700, Andrew Morton wrote: > On Fri, 12 Sep 2014 20:47:39 +0200 Joerg Roedel <jroedel@xxxxxxx> wrote: > > > thanks for your review, I tried to answer your questions below. > > You'd be amazed how helpful that was ;) > > > Fair enough, I hope I clarified a few things with my explanations > > above. I will also update the description of the patch-set when I > > re-send. > > Sounds good, thanks. > > > How does HMM play into all of this? Would HMM make this patchset > obsolete, or could HMM be evolved to do so? HMM should be consider as distinc from this. The hardware TLB we are talking with this patchset can be flush by the CPU from inside an atomic context (ie while holding cpu page table spinlock for instance). HMM on the other hand deals with hardware that have there own page table ie they do not necessarily walk the cpu page table. Flushing the TLB for this kind of hardware means scheduling some job on the hardware and this can not be done from kernel atomic context as this job might take a long time to complete (imagine preempting thousand of threads on a gpu). Still HMM can be use in a mixed environement where the IOMMUv2 is use for memory that reside into system ram while HMM only handle memory that have been migrated to the device memory. So while HMM intend to provide more features than IOMMUv2 hardware allow, it does not intend to replace it. On contrary hope is that both can work at same time. Cheers, Jérôme > _______________________________________________ > iommu mailing list > iommu@xxxxxxxxxxxxxxxxxxxxxxxxxx > https://lists.linuxfoundation.org/mailman/listinfo/iommu -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>