Hi, On Mon, Sep 16, 2019 at 8:57 PM Jean-Philippe Brucker <jean-philippe@xxxxxxxxxx> wrote: > On 13/09/2019 09:13, Guo Ren wrote: > > Another idea is seperate remote TLB invalidate into two instructions: > > > > - sfence.vma.b.asyc > > - sfence.vma.b.barrier // wait all async TLB invalidate operations > > finished for all harts. > > It's not clear to me how this helps, but I probably don't have the whole > picture. If you have a place where it is safe to wait for the barrier to > complete, why not do the whole invalidate there? > > > (I remember who mentioned me separate them into two instructions after > > session. Anup? Is the idea right ?) Forget it, I still use irq signal in my formal proposal [1]. I also couldn't image the whole picture :P > > To solve the problem, we could define a async mode in sfence.vma.b to > > slove the problem and finished with per_cpu_irq/exception. > > The solution I had to this problem is pinning the ASID [1] used by the > IOMMU, to prevent the CPU from recycling the ASID on rollover. This way > the CPU doesn't have to wait for IOMMU invalidations to complete, when > scheduling a task that might not even have anything to do with the IOMMU. > > In the Arm SMMU, ASID and IOASID (PASID) are separate identifiers. IOASID > indexes an entry in the context descriptor table, which contains the ASID. > So with unpinned shared ASID you don't need to invalidate the ATC on > rollover, since the IOASID doesn't change, but you do need to modify the > context descriptor and invalidate cached versions of it. The terminology confused me a lot. I perfer use PASID for IOMMU and ASID is for CPU. Arm's entry of the context descriptor table contains a "IOASID" IOASID != ASID for CPU_TLB and IOMMU_TLB. When you say "since the IOASID doesn't change",Is it PASID or my IOASID ? -_*! PASID in PCI-sig was used to determine transfer address space. For intel, the entry which is indexed by PASID also contain S1/S2.PGD and DID(VMID). For arm, the entry which is indexed by PASID only contain S1.PGD and IOASID. Compare to Intel Vt-d Scalable mode, arm's design can't support PCI Virtual Function. > > Once you have pinned ASIDs, you could also declare that IOASID = ASID. I > don't remember finding an argument to strictly forbid it, even though ASID > and IOASID have different sizes on Arm (respectively 8/16 and 20 bits). ASID and IOASID are hard to keep the same between CPU system and IOMMU system. So I introduce S1/S2.PGD.PPN as a bridge between CPUs and IOMMUs. See my proposal [1] 1: https://lore.kernel.org/linux-csky/1568896556-28769-1-git-send-email-guoren@xxxxxxxxxx/T/#u -- Best Regards Guo Ren ML: https://lore.kernel.org/linux-csky/ _______________________________________________ kvmarm mailing list kvmarm@xxxxxxxxxxxxxxxxxxxxx https://lists.cs.columbia.edu/mailman/listinfo/kvmarm
