Re: [RFC 6/7] iommu/vt-d: convert the intel iommu driver to the dma-iommu ops api
Hi, On 5/4/19 9:23 PM, Tom Murphy wrote: static int intel_iommu_add_device(struct device *dev) { + struct dmar_domain *dmar_domain; + struct iommu_domain *domain; struct intel_iommu *iommu; struct iommu_group *group; - struct iommu_domain *domain; + dma_addr_t base; u8 bus, devfn; iommu = device_to_iommu(dev, , ); @@ -4871,9 +4514,12 @@ static int intel_iommu_add_device(struct device *dev) if (IS_ERR(group)) return PTR_ERR(group); + base = IOVA_START_PFN << VTD_PAGE_SHIFT; domain = iommu_get_domain_for_dev(dev); + dmar_domain = to_dmar_domain(domain); if (domain->type == IOMMU_DOMAIN_DMA) - dev->dma_ops = _dma_ops; + iommu_setup_dma_ops(dev, base, + __DOMAIN_MAX_ADDR(dmar_domain->gaw) - base); I didn't find the implementation of iommu_setup_dma_ops() in this series. Will the iova resource be initialized in this function? If so, will this block iommu_group_create_direct_mappings() which reserves and maps the reserved iova ranges. iommu_group_put(group); return 0; @@ -5002,19 +4648,6 @@ int intel_iommu_enable_pasid(struct intel_iommu *iommu, struct intel_svm_dev *sd return ret; } -static void intel_iommu_apply_resv_region(struct device *dev, - struct iommu_domain *domain, - struct iommu_resv_region *region) -{ - struct dmar_domain *dmar_domain = to_dmar_domain(domain); - unsigned long start, end; - - start = IOVA_PFN(region->start); - end = IOVA_PFN(region->start + region->length - 1); - - WARN_ON_ONCE(!reserve_iova(_domain->iovad, start, end)); -} - struct intel_iommu *intel_svm_device_to_iommu(struct device *dev) { struct intel_iommu *iommu; @@ -5050,13 +4683,13 @@ const struct iommu_ops intel_iommu_ops = { .detach_dev = intel_iommu_detach_device, .map= intel_iommu_map, .unmap = intel_iommu_unmap, + .flush_iotlb_all= iommu_flush_iova, .flush_iotlb_range = intel_iommu_flush_iotlb_range, .iova_to_phys = intel_iommu_iova_to_phys, .add_device = intel_iommu_add_device, .remove_device = intel_iommu_remove_device, .get_resv_regions = intel_iommu_get_resv_regions, .put_resv_regions = intel_iommu_put_resv_regions, - .apply_resv_region = intel_iommu_apply_resv_region, With this removed, how will iommu_group_create_direct_mappings() work? Best regards, Lu Baolu ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [RFC 2/7] iommu/vt-d: Remove iova handling code from non-dma ops path
Hi, On 5/5/19 9:19 AM, Lu Baolu wrote: Hi, On 5/4/19 9:23 PM, Tom Murphy via iommu wrote: @@ -4181,58 +4168,37 @@ static int intel_iommu_memory_notifier(struct notifier_block *nb, unsigned long val, void *v) { struct memory_notify *mhp = v; - unsigned long long start, end; - unsigned long start_vpfn, last_vpfn; + unsigned long start_vpfn = mm_to_dma_pfn(mhp->start_pfn); + unsigned long last_vpfn = mm_to_dma_pfn(mhp->start_pfn + + mhp->nr_pages - 1); switch (val) { case MEM_GOING_ONLINE: - start = mhp->start_pfn << PAGE_SHIFT; - end = ((mhp->start_pfn + mhp->nr_pages) << PAGE_SHIFT) - 1; - if (iommu_domain_identity_map(si_domain, start, end)) { - pr_warn("Failed to build identity map for [%llx-%llx]\n", - start, end); + if (iommu_domain_identity_map(si_domain, start_vpfn, + last_vpfn)) { + pr_warn("Failed to build identity map for [%lx-%lx]\n", + start_vpfn, last_vpfn); return NOTIFY_BAD; } break; Actually we don't need to update the si_domain if iommu hardware supports pass-through mode. This should be made in a separated patch anyway. Oh! please ignore it. This callback is only registered when hardware doesn't support pass through mode. if (si_domain && !hw_pass_through) register_memory_notifier(_iommu_memory_nb); Best regards, Lu Baolu ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
Re: [RFC 2/7] iommu/vt-d: Remove iova handling code from non-dma ops path
Hi, On 5/4/19 9:23 PM, Tom Murphy via iommu wrote: @@ -4181,58 +4168,37 @@ static int intel_iommu_memory_notifier(struct notifier_block *nb, unsigned long val, void *v) { struct memory_notify *mhp = v; - unsigned long long start, end; - unsigned long start_vpfn, last_vpfn; + unsigned long start_vpfn = mm_to_dma_pfn(mhp->start_pfn); + unsigned long last_vpfn = mm_to_dma_pfn(mhp->start_pfn + + mhp->nr_pages - 1); switch (val) { case MEM_GOING_ONLINE: - start = mhp->start_pfn << PAGE_SHIFT; - end = ((mhp->start_pfn + mhp->nr_pages) << PAGE_SHIFT) - 1; - if (iommu_domain_identity_map(si_domain, start, end)) { - pr_warn("Failed to build identity map for [%llx-%llx]\n", - start, end); + if (iommu_domain_identity_map(si_domain, start_vpfn, + last_vpfn)) { + pr_warn("Failed to build identity map for [%lx-%lx]\n", + start_vpfn, last_vpfn); return NOTIFY_BAD; } break; Actually we don't need to update the si_domain if iommu hardware supports pass-through mode. This should be made in a separated patch anyway. Best regards, Lu Baolu ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
[RFC 5/7] iommu/dma-iommu: add wrapper for iommu_dma_free_cpu_cached_iovas
Add a wrapper for iommu_dma_free_cpu_cached_iovas in the dma-iommu api path to help with the intel-iommu driver conversion to the dma-iommu api path Signed-off-by: Tom Murphy --- drivers/iommu/dma-iommu.c | 9 + include/linux/dma-iommu.h | 3 +++ 2 files changed, 12 insertions(+) diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c index 82ba500886b4..1415b6f068c1 100644 --- a/drivers/iommu/dma-iommu.c +++ b/drivers/iommu/dma-iommu.c @@ -49,6 +49,15 @@ struct iommu_dma_cookie { struct iommu_domain *fq_domain; }; +void iommu_dma_free_cpu_cached_iovas(unsigned int cpu, + struct iommu_domain *domain) +{ + struct iommu_dma_cookie *cookie = domain->iova_cookie; + struct iova_domain *iovad = >iovad; + + free_cpu_cached_iovas(cpu, iovad); +} + static void iommu_dma_entry_dtor(unsigned long data) { struct page *freelist = (struct page *)data; diff --git a/include/linux/dma-iommu.h b/include/linux/dma-iommu.h index 3fc76918e9bf..1e5bee193feb 100644 --- a/include/linux/dma-iommu.h +++ b/include/linux/dma-iommu.h @@ -25,6 +25,9 @@ void iommu_setup_dma_ops(struct device *dev, u64 dma_base, u64 size); void iommu_dma_map_msi_msg(int irq, struct msi_msg *msg); void iommu_dma_get_resv_regions(struct device *dev, struct list_head *list); +void iommu_dma_free_cpu_cached_iovas(unsigned int cpu, + struct iommu_domain *domain); + #else /* CONFIG_IOMMU_DMA */ struct iommu_domain; -- 2.17.1 ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
[RFC 6/7] iommu/vt-d: convert the intel iommu driver to the dma-iommu ops api
Convert the intel iommu driver to the dma-iommu api to allow us to remove the iova handling code and the reserved region code Signed-off-by: Tom Murphy --- drivers/iommu/Kconfig | 1 + drivers/iommu/intel-iommu.c | 405 ++-- include/linux/intel-iommu.h | 1 - 3 files changed, 20 insertions(+), 387 deletions(-) diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig index 6f07f3b21816..dfed97f55b6e 100644 --- a/drivers/iommu/Kconfig +++ b/drivers/iommu/Kconfig @@ -180,6 +180,7 @@ config INTEL_IOMMU select IOMMU_IOVA select NEED_DMA_MAP_STATE select DMAR_TABLE + select IOMMU_DMA help DMA remapping (DMAR) devices support enables independent address translations for Direct Memory Access (DMA) from devices. diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c index 87622a28b854..980fc4816d72 100644 --- a/drivers/iommu/intel-iommu.c +++ b/drivers/iommu/intel-iommu.c @@ -39,6 +39,7 @@ #include #include #include +#include #include #include #include @@ -1180,13 +1181,6 @@ static void dma_free_pagelist(struct page *freelist) } } -static void iova_entry_free(unsigned long data) -{ - struct page *freelist = (struct page *)data; - - dma_free_pagelist(freelist); -} - /* iommu handling */ static int iommu_alloc_root_entry(struct intel_iommu *iommu) { @@ -1530,16 +1524,14 @@ static inline void __mapping_notify_one(struct intel_iommu *iommu, iommu_flush_write_buffer(iommu); } -static void iommu_flush_iova(struct iova_domain *iovad) +static void iommu_flush_iova(struct iommu_domain *domain) { - struct dmar_domain *domain; + struct dmar_domain *dmar_domain = to_dmar_domain(domain); int idx; - domain = container_of(iovad, struct dmar_domain, iovad); - - for_each_domain_iommu(idx, domain) { + for_each_domain_iommu(idx, dmar_domain) { struct intel_iommu *iommu = g_iommus[idx]; - u16 did = domain->iommu_did[iommu->seq_id]; + u16 did = dmar_domain->iommu_did[iommu->seq_id]; iommu->flush.flush_iotlb(iommu, did, 0, 0, DMA_TLB_DSI_FLUSH); @@ -1804,48 +1796,6 @@ static int domain_detach_iommu(struct dmar_domain *domain, return count; } -static struct iova_domain reserved_iova_list; -static struct lock_class_key reserved_rbtree_key; - -static int dmar_init_reserved_ranges(void) -{ - struct pci_dev *pdev = NULL; - struct iova *iova; - int i; - - init_iova_domain(_iova_list, VTD_PAGE_SIZE, IOVA_START_PFN); - - lockdep_set_class(_iova_list.iova_rbtree_lock, - _rbtree_key); - - /* IOAPIC ranges shouldn't be accessed by DMA */ - iova = reserve_iova(_iova_list, IOVA_PFN(IOAPIC_RANGE_START), - IOVA_PFN(IOAPIC_RANGE_END)); - if (!iova) { - pr_err("Reserve IOAPIC range failed\n"); - return -ENODEV; - } - - /* Reserve all PCI MMIO to avoid peer-to-peer access */ - for_each_pci_dev(pdev) { - struct resource *r; - - for (i = 0; i < PCI_NUM_RESOURCES; i++) { - r = >resource[i]; - if (!r->flags || !(r->flags & IORESOURCE_MEM)) - continue; - iova = reserve_iova(_iova_list, - IOVA_PFN(r->start), - IOVA_PFN(r->end)); - if (!iova) { - pci_err(pdev, "Reserve iova for %pR failed\n", r); - return -ENODEV; - } - } - } - return 0; -} - static inline int guestwidth_to_adjustwidth(int gaw) { int agaw; @@ -1871,7 +1821,7 @@ static void domain_exit(struct dmar_domain *domain) /* destroy iovas */ if (domain->domain.type == IOMMU_DOMAIN_DMA) - put_iova_domain(>iovad); + iommu_put_dma_cookie(>domain); freelist = domain_unmap(domain, 0, DOMAIN_MAX_PFN(domain->gaw), NULL); @@ -3213,296 +3163,6 @@ static int __init init_dmars(void) return ret; } -/* This takes a number of _MM_ pages, not VTD pages */ -static unsigned long intel_alloc_iova(struct device *dev, -struct dmar_domain *domain, -unsigned long nrpages, uint64_t dma_mask) -{ - unsigned long iova_pfn; - - /* Restrict dma_mask to the width that the iommu can handle */ - dma_mask = min_t(uint64_t, DOMAIN_MAX_ADDR(domain->gaw), dma_mask); - /* Ensure we reserve the whole size-aligned region */ - nrpages = __roundup_pow_of_two(nrpages); - - if (!dmar_forcedac && dma_mask > DMA_BIT_MASK(32)) { - /* -* First try to allocate an io virtual address in -
[RFC 7/7] iommu/vt-d: Always set DMA_PTE_READ if the iommu doens't support zero length reads
To match the dma-ops api path the DMA_PTE_READ should be set if ZLR isn't supported in the iommu Signed-off-by: Tom Murphy --- drivers/iommu/intel-iommu.c | 13 - 1 file changed, 12 insertions(+), 1 deletion(-) diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c index 980fc4816d72..e78b056d 100644 --- a/drivers/iommu/intel-iommu.c +++ b/drivers/iommu/intel-iommu.c @@ -4378,6 +4378,17 @@ static void intel_iommu_detach_device(struct iommu_domain *domain, dmar_remove_one_dev_info(dev); } +static bool supports_zlr(struct dmar_domain *domain) +{ + int i; + + for_each_domain_iommu(i, domain) { + if (cap_zlr(g_iommus[i]->cap)) + return true; + } + return false; +} + static int intel_iommu_map(struct iommu_domain *domain, unsigned long iova, phys_addr_t hpa, size_t size, int iommu_prot) @@ -4391,7 +4402,7 @@ static int intel_iommu_map(struct iommu_domain *domain, if (dmar_domain == si_domain && hw_pass_through) return 0; - if (iommu_prot & IOMMU_READ) + if (iommu_prot & IOMMU_READ || !supports_zlr(dmar_domain)) prot |= DMA_PTE_READ; if (iommu_prot & IOMMU_WRITE) prot |= DMA_PTE_WRITE; -- 2.17.1 ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
[RFC 2/7] iommu/vt-d: Remove iova handling code from non-dma ops path
There is no reason to keep track of the iovas in the non-dma ops path. All this code seems to be pointless and can be removed. Signed-off-by: Tom Murphy --- drivers/iommu/intel-iommu.c | 94 + 1 file changed, 33 insertions(+), 61 deletions(-) diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c index 2db1dc47e7e4..77895cd89f29 100644 --- a/drivers/iommu/intel-iommu.c +++ b/drivers/iommu/intel-iommu.c @@ -1846,11 +1846,6 @@ static int dmar_init_reserved_ranges(void) return 0; } -static void domain_reserve_special_ranges(struct dmar_domain *domain) -{ - copy_reserved_iova(_iova_list, >iovad); -} - static inline int guestwidth_to_adjustwidth(int gaw) { int agaw; @@ -1875,7 +1870,8 @@ static void domain_exit(struct dmar_domain *domain) rcu_read_unlock(); /* destroy iovas */ - put_iova_domain(>iovad); + if (domain->domain.type == IOMMU_DOMAIN_DMA) + put_iova_domain(>iovad); freelist = domain_unmap(domain, 0, DOMAIN_MAX_PFN(domain->gaw)); @@ -2554,19 +2550,9 @@ static struct dmar_domain *dmar_insert_one_dev_info(struct intel_iommu *iommu, } static int iommu_domain_identity_map(struct dmar_domain *domain, -unsigned long long start, -unsigned long long end) +unsigned long first_vpfn, +unsigned long last_vpfn) { - unsigned long first_vpfn = start >> VTD_PAGE_SHIFT; - unsigned long last_vpfn = end >> VTD_PAGE_SHIFT; - - if (!reserve_iova(>iovad, dma_to_mm_pfn(first_vpfn), - dma_to_mm_pfn(last_vpfn))) { - pr_err("Reserving iova failed\n"); - return -ENOMEM; - } - - pr_debug("Mapping reserved region %llx-%llx\n", start, end); /* * RMRR range might have overlap with physical memory range, * clear it first @@ -2613,7 +2599,8 @@ static int __init si_domain_init(int hw) for_each_mem_pfn_range(i, nid, _pfn, _pfn, NULL) { ret = iommu_domain_identity_map(si_domain, - PFN_PHYS(start_pfn), PFN_PHYS(end_pfn)); + mm_to_dma_pfn(start_pfn), + mm_to_dma_pfn(end_pfn)); if (ret) return ret; } @@ -4181,58 +4168,37 @@ static int intel_iommu_memory_notifier(struct notifier_block *nb, unsigned long val, void *v) { struct memory_notify *mhp = v; - unsigned long long start, end; - unsigned long start_vpfn, last_vpfn; + unsigned long start_vpfn = mm_to_dma_pfn(mhp->start_pfn); + unsigned long last_vpfn = mm_to_dma_pfn(mhp->start_pfn + + mhp->nr_pages - 1); switch (val) { case MEM_GOING_ONLINE: - start = mhp->start_pfn << PAGE_SHIFT; - end = ((mhp->start_pfn + mhp->nr_pages) << PAGE_SHIFT) - 1; - if (iommu_domain_identity_map(si_domain, start, end)) { - pr_warn("Failed to build identity map for [%llx-%llx]\n", - start, end); + if (iommu_domain_identity_map(si_domain, start_vpfn, + last_vpfn)) { + pr_warn("Failed to build identity map for [%lx-%lx]\n", + start_vpfn, last_vpfn); return NOTIFY_BAD; } break; case MEM_OFFLINE: case MEM_CANCEL_ONLINE: - start_vpfn = mm_to_dma_pfn(mhp->start_pfn); - last_vpfn = mm_to_dma_pfn(mhp->start_pfn + mhp->nr_pages - 1); - while (start_vpfn <= last_vpfn) { - struct iova *iova; + { struct dmar_drhd_unit *drhd; struct intel_iommu *iommu; struct page *freelist; - iova = find_iova(_domain->iovad, start_vpfn); - if (iova == NULL) { - pr_debug("Failed get IOVA for PFN %lx\n", -start_vpfn); - break; - } - - iova = split_and_remove_iova(_domain->iovad, iova, -start_vpfn, last_vpfn); - if (iova == NULL) { - pr_warn("Failed to split IOVA PFN [%lx-%lx]\n", - start_vpfn, last_vpfn); - return NOTIFY_BAD; - } - - freelist = domain_unmap(si_domain, iova->pfn_lo, -
[RFC 4/7] iommu/dma-iommu: Handle freelists in the dma-iommu api path
Currently the iova flush queue implementation in the dma-iommu api path doesn't handle freelists. Change the unmap_fast code to allow it to return any freelists which need to be handled. Signed-off-by: Tom Murphy --- drivers/iommu/dma-iommu.c | 39 +++-- drivers/iommu/iommu.c | 10 + drivers/vfio/vfio_iommu_type1.c | 2 +- include/linux/iommu.h | 3 ++- 4 files changed, 36 insertions(+), 18 deletions(-) diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c index fa5713a4f6f8..82ba500886b4 100644 --- a/drivers/iommu/dma-iommu.c +++ b/drivers/iommu/dma-iommu.c @@ -49,6 +49,18 @@ struct iommu_dma_cookie { struct iommu_domain *fq_domain; }; +static void iommu_dma_entry_dtor(unsigned long data) +{ + struct page *freelist = (struct page *)data; + + while (freelist != NULL) { + unsigned long p = (unsigned long)page_address(freelist); + + freelist = freelist->freelist; + free_page(p); + } +} + static inline size_t cookie_msi_granule(struct iommu_dma_cookie *cookie) { if (cookie->type == IOMMU_DMA_IOVA_COOKIE) @@ -313,7 +325,8 @@ static int iommu_dma_init_domain(struct iommu_domain *domain, dma_addr_t base, if (!cookie->fq_domain && !iommu_domain_get_attr(domain, DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE, ) && attr) { cookie->fq_domain = domain; - init_iova_flush_queue(iovad, iommu_dma_flush_iotlb_all, NULL); + init_iova_flush_queue(iovad, iommu_dma_flush_iotlb_all, + iommu_dma_entry_dtor); } if (!dev) @@ -393,7 +406,7 @@ static dma_addr_t iommu_dma_alloc_iova(struct iommu_domain *domain, } static void iommu_dma_free_iova(struct iommu_dma_cookie *cookie, - dma_addr_t iova, size_t size) + dma_addr_t iova, size_t size, struct page *freelist) { struct iova_domain *iovad = >iovad; @@ -402,7 +415,8 @@ static void iommu_dma_free_iova(struct iommu_dma_cookie *cookie, cookie->msi_iova -= size; else if (cookie->fq_domain) /* non-strict mode */ queue_iova(iovad, iova_pfn(iovad, iova), - size >> iova_shift(iovad), 0); + size >> iova_shift(iovad), + (unsigned long) freelist); else free_iova_fast(iovad, iova_pfn(iovad, iova), size >> iova_shift(iovad)); @@ -414,14 +428,15 @@ static void __iommu_dma_unmap(struct iommu_domain *domain, dma_addr_t dma_addr, struct iommu_dma_cookie *cookie = domain->iova_cookie; struct iova_domain *iovad = >iovad; size_t iova_off = iova_offset(iovad, dma_addr); + struct page *freelist; dma_addr -= iova_off; size = iova_align(iovad, size + iova_off); - WARN_ON(iommu_unmap_fast(domain, dma_addr, size) != size); + WARN_ON(iommu_unmap_fast(domain, dma_addr, size, ) != size); if (!cookie->fq_domain) - iommu_tlb_sync(domain); - iommu_dma_free_iova(cookie, dma_addr, size); + iommu_flush_iotlb_range(domain, dma_addr, size, freelist); + iommu_dma_free_iova(cookie, dma_addr, size, freelist); } static dma_addr_t __iommu_dma_map(struct device *dev, phys_addr_t phys, @@ -441,7 +456,7 @@ static dma_addr_t __iommu_dma_map(struct device *dev, phys_addr_t phys, return DMA_MAPPING_ERROR; if (iommu_map(domain, iova, phys - iova_off, size, prot)) { - iommu_dma_free_iova(cookie, iova, size); + iommu_dma_free_iova(cookie, iova, size, NULL); return DMA_MAPPING_ERROR; } return iova + iova_off; @@ -600,7 +615,7 @@ static void *iommu_dma_alloc_remap(struct device *dev, size_t size, struct iova_domain *iovad = >iovad; bool coherent = dev_is_dma_coherent(dev); int ioprot = dma_info_to_prot(DMA_BIDIRECTIONAL, coherent, attrs); - pgprot_t prot = arch_dma_mmap_pgprot(dev, PAGE_KERNEL, attrs); + pgprot_t prot = pgprot_noncached(PAGE_KERNEL); unsigned int count, min_size, alloc_sizes = domain->pgsize_bitmap; struct page **pages; struct sg_table sgt; @@ -659,7 +674,7 @@ static void *iommu_dma_alloc_remap(struct device *dev, size_t size, out_free_sg: sg_free_table(); out_free_iova: - iommu_dma_free_iova(cookie, iova, size); + iommu_dma_free_iova(cookie, iova, size, NULL); out_free_pages: __iommu_dma_free_pages(pages, count); return NULL; @@ -668,7 +683,7 @@ static void *iommu_dma_alloc_remap(struct device *dev, size_t size, static void *iommu_dma_alloc_contiguous_remap(struct device *dev, size_t size, dma_addr_t *dma_handle, gfp_t gfp, unsigned long attrs) { - pgprot_t prot =
[RFC 0/7] Convert the Intel iommu driver to the dma-ops api
Convert the intel iommu driver to the dma-ops api so that we can remove a bunch of repeated code. This patchset depends on the "iommu/vt-d: Delegate DMA domain to generic iommu" and "iommu/amd: Convert the AMD iommu driver to the dma-iommu api" patch sets which haven't yet merged so this is just a RFC to get some feedback before I do more testing. Tom Murphy (7): iommu/vt-d: Set the dma_ops per device so we can remove the iommu_no_mapping code iommu/vt-d: Remove iova handling code from non-dma ops path iommu: improve iommu iotlb flushing iommu/dma-iommu: Handle freelists in the dma-iommu api path iommu/dma-iommu: add wrapper for iommu_dma_free_cpu_cached_iovas iommu/vt-d: convert the intel iommu driver to the dma-iommu ops api iommu/vt-d: Always set DMA_PTE_READ if the iommu doens't support zero length reads drivers/iommu/Kconfig | 1 + drivers/iommu/amd_iommu.c | 14 +- drivers/iommu/arm-smmu-v3.c | 3 +- drivers/iommu/arm-smmu.c| 2 +- drivers/iommu/dma-iommu.c | 48 ++- drivers/iommu/exynos-iommu.c| 3 +- drivers/iommu/intel-iommu.c | 605 +--- drivers/iommu/iommu.c | 21 +- drivers/iommu/ipmmu-vmsa.c | 2 +- drivers/iommu/msm_iommu.c | 2 +- drivers/iommu/mtk_iommu.c | 3 +- drivers/iommu/mtk_iommu_v1.c| 3 +- drivers/iommu/omap-iommu.c | 2 +- drivers/iommu/qcom_iommu.c | 2 +- drivers/iommu/rockchip-iommu.c | 2 +- drivers/iommu/s390-iommu.c | 3 +- drivers/iommu/tegra-gart.c | 2 +- drivers/iommu/tegra-smmu.c | 2 +- drivers/vfio/vfio_iommu_type1.c | 3 +- include/linux/dma-iommu.h | 3 + include/linux/intel-iommu.h | 1 - include/linux/iommu.h | 24 +- 22 files changed, 175 insertions(+), 576 deletions(-) -- 2.17.1 ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu
[RFC 1/7] iommu/vt-d: Set the dma_ops per device so we can remove the iommu_no_mapping code
Set the dma_ops per device so we can remove the iommu_no_mapping code. Signed-off-by: Tom Murphy --- drivers/iommu/intel-iommu.c | 85 +++-- 1 file changed, 6 insertions(+), 79 deletions(-) diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c index eace915602f0..2db1dc47e7e4 100644 --- a/drivers/iommu/intel-iommu.c +++ b/drivers/iommu/intel-iommu.c @@ -2622,17 +2622,6 @@ static int __init si_domain_init(int hw) return 0; } -static int identity_mapping(struct device *dev) -{ - struct device_domain_info *info; - - info = dev->archdata.iommu; - if (info && info != DUMMY_DEVICE_DOMAIN_INFO) - return (info->domain == si_domain); - - return 0; -} - static int domain_add_dev_info(struct dmar_domain *domain, struct device *dev) { struct dmar_domain *ndomain; @@ -3270,43 +3259,6 @@ static unsigned long intel_alloc_iova(struct device *dev, return iova_pfn; } -/* Check if the dev needs to go through non-identity map and unmap process.*/ -static int iommu_no_mapping(struct device *dev) -{ - int found; - - if (iommu_dummy(dev)) - return 1; - - found = identity_mapping(dev); - if (found) { - /* -* If the device's dma_mask is less than the system's memory -* size then this is not a candidate for identity mapping. -*/ - u64 dma_mask = *dev->dma_mask; - - if (dev->coherent_dma_mask && - dev->coherent_dma_mask < dma_mask) - dma_mask = dev->coherent_dma_mask; - - if (dma_mask < dma_get_required_mask(dev)) { - /* -* 32 bit DMA is removed from si_domain and fall back -* to non-identity mapping. -*/ - dmar_remove_one_dev_info(dev); - dev_warn(dev, "32bit DMA uses non-identity mapping\n"); - - return 0; - } - - return 1; - } - - return 0; -} - static dma_addr_t __intel_map_single(struct device *dev, phys_addr_t paddr, size_t size, int dir, u64 dma_mask) { @@ -3320,9 +3272,6 @@ static dma_addr_t __intel_map_single(struct device *dev, phys_addr_t paddr, BUG_ON(dir == DMA_NONE); - if (iommu_no_mapping(dev)) - return paddr; - domain = find_domain(dev); if (!domain) return DMA_MAPPING_ERROR; @@ -3391,9 +3340,6 @@ static void intel_unmap(struct device *dev, dma_addr_t dev_addr, size_t size) struct intel_iommu *iommu; struct page *freelist; - if (iommu_no_mapping(dev)) - return; - domain = find_domain(dev); BUG_ON(!domain); @@ -3442,9 +3388,7 @@ static void *intel_alloc_coherent(struct device *dev, size_t size, size = PAGE_ALIGN(size); order = get_order(size); - if (!iommu_no_mapping(dev)) - flags &= ~(GFP_DMA | GFP_DMA32); - else if (dev->coherent_dma_mask < dma_get_required_mask(dev)) { + if (dev->coherent_dma_mask < dma_get_required_mask(dev)) { if (dev->coherent_dma_mask < DMA_BIT_MASK(32)) flags |= GFP_DMA; else @@ -3456,11 +3400,6 @@ static void *intel_alloc_coherent(struct device *dev, size_t size, page = dma_alloc_from_contiguous(dev, count, order, flags & __GFP_NOWARN); - if (page && iommu_no_mapping(dev) && - page_to_phys(page) + size > dev->coherent_dma_mask) { - dma_release_from_contiguous(dev, page, count); - page = NULL; - } } if (!page) @@ -3510,20 +3449,6 @@ static void intel_unmap_sg(struct device *dev, struct scatterlist *sglist, intel_unmap(dev, startaddr, nrpages << VTD_PAGE_SHIFT); } -static int intel_nontranslate_map_sg(struct device *hddev, - struct scatterlist *sglist, int nelems, int dir) -{ - int i; - struct scatterlist *sg; - - for_each_sg(sglist, sg, nelems, i) { - BUG_ON(!sg_page(sg)); - sg->dma_address = sg_phys(sg); - sg->dma_length = sg->length; - } - return nelems; -} - static int intel_map_sg(struct device *dev, struct scatterlist *sglist, int nelems, enum dma_data_direction dir, unsigned long attrs) { @@ -3538,8 +3463,6 @@ static int intel_map_sg(struct device *dev, struct scatterlist *sglist, int nele struct intel_iommu *iommu; BUG_ON(dir == DMA_NONE); - if (iommu_no_mapping(dev)) - return intel_nontranslate_map_sg(dev, sglist, nelems, dir); domain = find_domain(dev); if (!domain) @@ -4570,7
Re: [PATCH v6 0/1] iommu: enhance IOMMU dma mode build options
Hi all, Can anybody review or comment? On 2019/4/18 21:57, Zhen Lei wrote: > v5 --> v6: > 1. give up adding boot option iommu.dma_mode > > v4 --> v5: > As Hanjun and Thomas Gleixner's suggestion: > 1. Keep the old ARCH specific boot options no change. > 2. Keep build option CONFIG_IOMMU_DEFAULT_PASSTHROUGH no change. > > v4: > As Robin Murphy's suggestion: > "It's also not necessarily obvious to the user how this interacts with > IOMMU_DEFAULT_PASSTHROUGH, so if we really do go down this route, maybe it > would be better to refactor the whole lot into a single selection of something > like IOMMU_DEFAULT_MODE anyway." > > In this version, I tried to normalize the IOMMU dma mode boot options for all > ARCHs. When IOMMU is enabled, there are 3 dma modes: paasthrough(bypass), > lazy(mapping but defer the IOTLB invalidation), strict. But currently each > ARCHs defined their private boot options, different with each other. For > example, to enable/disable "passthrough", ARM64 use iommu.passthrough=1/0, > X86 use iommu=pt/nopt, PPC/POWERNV use iommu=nobypass. > > Zhen Lei (1): > iommu: enhance IOMMU dma mode build options > > arch/ia64/kernel/pci-dma.c| 2 +- > arch/powerpc/platforms/powernv/pci-ioda.c | 3 ++- > arch/s390/pci/pci_dma.c | 2 +- > arch/x86/kernel/pci-dma.c | 7 ++--- > drivers/iommu/Kconfig | 44 > ++- > drivers/iommu/amd_iommu_init.c| 3 ++- > drivers/iommu/intel-iommu.c | 2 +- > drivers/iommu/iommu.c | 3 ++- > 8 files changed, 48 insertions(+), 18 deletions(-) > -- Thanks! BestRegards ___ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu