Alexey Kardashevskiy <aik@xxxxxxxxx> writes: > On 17/12/2019 10:07, Thiago Jung Bauermann wrote: >> >> Alexey Kardashevskiy <aik@xxxxxxxxx> writes: >> >>> By default a pseries guest supports a H_PUT_TCE hypercall which maps >>> a single IOMMU page in a DMA window. Additionally the hypervisor may >>> support H_PUT_TCE_INDIRECT/H_STUFF_TCE which update multiple TCEs at once; >>> this is advertised via the device tree /rtas/ibm,hypertas-functions >>> property which Linux converts to FW_FEATURE_MULTITCE. >>> >>> FW_FEATURE_MULTITCE is checked when dma_iommu_ops is used; however >>> the code managing the huge DMA window (DDW) ignores it and calls >>> H_PUT_TCE_INDIRECT even if it is explicitly disabled via >>> the "multitce=off" kernel command line parameter. >>> >>> This adds FW_FEATURE_MULTITCE checking to the DDW code path. >>> >>> This changes tce_build_pSeriesLP to take liobn and page size as >>> the huge window does not have iommu_table descriptor which usually >>> the place to store these numbers. >>> >>> Fixes: 4e8b0cf46b25 ("powerpc/pseries: Add support for dynamic dma windows") >>> Signed-off-by: Alexey Kardashevskiy <aik@xxxxxxxxx> >> >> Reviewed-by: Thiago Jung Bauermann <bauerman@xxxxxxxxxxxxx> >> Tested-by: Thiago Jung Bauermann <bauerman@xxxxxxxxxxxxx> >> >> Some minor nits below. Feel free to ignore. >> >>> @@ -146,25 +146,25 @@ static int tce_build_pSeriesLP(struct iommu_table *tbl, long tcenum, >>> int ret = 0; >>> long tcenum_start = tcenum, npages_start = npages; >>> >>> - rpn = __pa(uaddr) >> TCE_SHIFT; >>> + rpn = __pa(uaddr) >> tceshift; >>> proto_tce = TCE_PCI_READ; >>> if (direction != DMA_TO_DEVICE) >>> proto_tce |= TCE_PCI_WRITE; >>> >>> while (npages--) { >>> - tce = proto_tce | (rpn & TCE_RPN_MASK) << TCE_RPN_SHIFT; >>> - rc = plpar_tce_put((u64)tbl->it_index, (u64)tcenum << 12, tce); >>> + tce = proto_tce | (rpn & TCE_RPN_MASK) << tceshift; >>> + rc = plpar_tce_put((u64)liobn, (u64)tcenum << tceshift, tce); >> >> Is it necessary to cast to u64 here? plpar_tce_put() takes unsigned long >> for both arguments. > > Looked as an unrelated change. Small but still unrelated. Ah, I hadn't noticed that the cast was already in the original code. >>> @@ -400,6 +402,20 @@ static int tce_setrange_multi_pSeriesLP(unsigned long start_pfn, >>> u64 rc = 0; >>> long l, limit; >>> >>> + if (!firmware_has_feature(FW_FEATURE_MULTITCE)) { >>> + unsigned long tceshift = be32_to_cpu(maprange->tce_shift); >>> + unsigned long dmastart = (start_pfn << PAGE_SHIFT) + >>> + be64_to_cpu(maprange->dma_base); >>> + unsigned long tcenum = dmastart >> tceshift; >>> + unsigned long npages = num_pfn << PAGE_SHIFT >> >>> + be32_to_cpu(maprange->tce_shift); >> >> Could use the tceshift variable here. > > > True, overlooked. > Thanks for the reviews! Thank you for the patches! -- Thiago Jung Bauermann IBM Linux Technology Center