paravirtual IOMMU interface
v6 cover paravirtual IOMMU interface
The idea of a paravirtual IOMMU interface was last discussed on xen-devel several years ago and narrowed down on a draft specification [1]. There was also an RFC patch series posted with an implementation, however this was never followed through.
In this patch series I have tried to simplify the interface and therefore have moved away from the draft specification. There is not yet any new specification but hopefully the interface in the introduced iommu_op header file will be understandable without such a specification.
[1] https://lists.xenproject.org/archives/html/xen-devel/2016-02/msg01428.html
Paul Durrant (14):
iommu: introduce the concept of BFN...
iommu: make use of type-safe BFN and MFN in exported functions
iommu: push use of type-safe BFN and MFN into iommu_ops
iommu: don't domain_crash() inside iommu_map/unmap_page() public / x86: introduce __HYPERCALL_iommu_op
iommu: track reserved ranges using a rangeset
x86: add iommu_op to query reserved ranges
vtd: add lookup_page method to iommu_ops
mm / iommu: include need_iommu() test in iommu_use_hap_pt() mm / iommu: split need_iommu() into has_iommu_pt() and
need_iommu_pt_sync()
x86: add iommu_op to enable modification of IOMMU mappings
memory: add get_paged_gfn() as a wrapper...
x86: add iommu_ops to modify and flush IOMMU mappings
x86: extend the map and unmap iommu_ops to support grant references
tools/flask/policy/modules/xen.if | 1 + xen/arch/arm/p2m.c | 9 +- xen/arch/x86/hvm/emulate.c | 32 +- xen/arch/x86/hvm/hvm.c | 16 +- xen/arch/x86/hvm/hypercall.c | 1 + xen/arch/x86/hvm/mtrr.c | 5 +- xen/arch/x86/hypercall.c | 1 + xen/arch/x86/mm.c | 15 +- xen/arch/x86/mm/mem_sharing.c | 2 +- xen/arch/x86/mm/p2m-ept.c | 19 +- xen/arch/x86/mm/p2m-pt.c | 52 ++- xen/arch/x86/mm/p2m.c | 42 +- xen/arch/x86/mm/paging.c | 2 +- xen/arch/x86/pv/hypercall.c | 1 + xen/arch/x86/x86_64/mm.c | 8 +- xen/common/Makefile | 1 + xen/common/grant_table.c | 197 +++++++-- xen/common/iommu_op.c | 598 ++++++++++++++++++++++++++ xen/common/memory.c | 69 ++- xen/common/vm_event.c | 2 +- xen/drivers/passthrough/amd/iommu_cmd.c | 18 +- xen/drivers/passthrough/amd/iommu_map.c | 86 ++-- xen/drivers/passthrough/amd/pci_amd_iommu.c | 6 +- xen/drivers/passthrough/arm/smmu.c | 20 +- xen/drivers/passthrough/device_tree.c | 21 +- xen/drivers/passthrough/iommu.c | 106 +++-- xen/drivers/passthrough/pci.c | 6 +- xen/drivers/passthrough/vtd/iommu.c | 90 ++-- xen/drivers/passthrough/vtd/iommu.h | 3 + xen/drivers/passthrough/vtd/x86/vtd.c | 18 +- xen/drivers/passthrough/x86/iommu.c | 8 +- xen/include/Makefile | 2 + xen/include/asm-arm/grant_table.h | 2 +- xen/include/asm-arm/iommu.h | 2 +- xen/include/asm-arm/p2m.h | 3 + xen/include/asm-x86/grant_table.h | 2 +- xen/include/asm-x86/hvm/svm/amd-iommu-proto.h | 8 +- xen/include/asm-x86/iommu.h | 5 +- xen/include/asm-x86/p2m.h | 2 + xen/include/public/iommu_op.h | 254 +++++++++++ xen/include/public/xen.h | 1 + xen/include/xen/grant_table.h | 7 + xen/include/xen/hypercall.h | 12 + xen/include/xen/iommu.h | 83 +++- xen/include/xen/mm.h | 5 + xen/include/xen/sched.h | 14 +- xen/include/xlat.lst | 8 + xen/include/xsm/dummy.h | 6 + xen/include/xsm/xsm.h | 6 + xen/xsm/dummy.c | 1 + xen/xsm/flask/hooks.c | 6 + xen/xsm/flask/policy/access_vectors | 2 + 52 files changed, 1576 insertions(+), 310 deletions(-) create mode 100644 xen/common/iommu_op.c create mode 100644 xen/include/public/iommu_op.h
---
Cc: Andrew Cooper <[email protected]>
Cc: Brian Woods <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Razvan Cojocaru <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>
Cc: Tamas K Lengyel <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 1/14 iommu: introduce the concept of BFN...
...meaning 'bus frame number' i.e. a frame number mapped in the IOMMU rather than the MMU.
This patch is a largely cosmetic change that substitutes the terms 'gfn' and 'gaddr' for 'bfn' and 'baddr' in all the places where the frame number or address relate to the IOMMU rather than the MMU.
The parts that are not purely cosmetic are:
- the introduction of a type-safe declaration of bfn_t and definition of INVALID_BFN to make the substitution of gfn_x(INVALID_GFN) mechanical.
- the introduction of __bfn_to_baddr and __baddr_to_bfn (and type-safe variants without the leading __) with some use of the former.
Subsequent patches will convert code to make use of type-safe BFNs.
---
Cc: Suravee Suthikulpanit <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Kevin Tian <[email protected]>
v6:
- Dropped changes to 'mfn' section in xen/mm.h as suggested by Kevin.
v3:
- Get rid of intermediate 'frame' variables again.
v2:
- Addressed comments from Jan.
--- xen/drivers/passthrough/amd/iommu_cmd.c | 18 +++---- xen/drivers/passthrough/amd/iommu_map.c | 76 ++++++++++++++--------------- xen/drivers/passthrough/amd/pci_amd_iommu.c | 2 +- xen/drivers/passthrough/arm/smmu.c | 16 +++--- xen/drivers/passthrough/iommu.c | 28 +++++------ xen/drivers/passthrough/vtd/iommu.c | 30 ++++++------ xen/include/xen/iommu.h | 38 ++++++++++++--- xen/include/xen/mm.h | 5 ++ 8 files changed, 122 insertions(+), 91 deletions(-)
xen/drivers/passthrough/amd/iommu_cmd.c +9 -9
@@ -284,7 +284,7 @@ void invalidate_iommu_all(struct amd_iommu *iommu) }void amd_iommu_flush_iotlb (u8 devfn,const struct pci_dev *pdev, -uint64_t gaddr,unsigned int order) +baddr_t baddr,unsigned int order) {unsigned long flags;struct amd_iommu *iommu;
@@ -315,12 +315,12 @@ void amd_iommu_flush_iotlb(u8 devfn, const struct pci_dev *pdev,/* send INVALIDATE_IOTLB_PAGES command */ spin_lock_irqsave (&iommu->lock, flags); -invalidate_iotlb_pages (iommu, maxpend,0 , queueid, gaddr, req_id, order); +invalidate_iotlb_pages (iommu, maxpend,0 , queueid, baddr, req_id, order);flush_command_buffer (iommu);spin_unlock_irqrestore (&iommu->lock, flags); } -static void amd_iommu_flush_all_iotlbs (struct domain *d,uint64_t gaddr, +static void amd_iommu_flush_all_iotlbs (struct domain *d,baddr_t baddr,unsigned int order) {struct pci_dev *pdev;
@@ -333,7 +333,7 @@ static void amd_iommu_flush_all_iotlbs(struct domain *d, uint64_t gaddr,u8 devfn = pdev->devfn;do { -amd_iommu_flush_iotlb (devfn, pdev, gaddr, order); +amd_iommu_flush_iotlb (devfn, pdev, baddr, order); devfn += pdev->phantom_stride; }while ( devfn != pdev->devfn &&PCI_SLOT (devfn) ==PCI_SLOT (pdev->devfn) );
@@ -342,7 +342,7 @@ static void amd_iommu_flush_all_iotlbs(struct domain *d, uint64_t gaddr,/* Flush iommu cache after p2m changes. */ static void _amd_iommu_flush_pages (struct domain *d, -uint64_t gaddr,unsigned int order) +baddr_t baddr,unsigned int order) {unsigned long flags;struct amd_iommu *iommu;
@@ -352,13 +352,13 @@ static void _amd_iommu_flush_pages(struct domain *d,for_each_amd_iommu ( iommu ) {spin_lock_irqsave (&iommu->lock, flags); -invalidate_iommu_pages (iommu, gaddr, dom_id, order); +invalidate_iommu_pages (iommu, baddr, dom_id, order);flush_command_buffer (iommu);spin_unlock_irqrestore (&iommu->lock, flags); }if ( ats_enabled ) -amd_iommu_flush_all_iotlbs (d, gaddr, order); +amd_iommu_flush_all_iotlbs (d, baddr, order); }void amd_iommu_flush_all_pages (struct domain *d)
@@ -367,9 +367,9 @@ void amd_iommu_flush_all_pages(struct domain *d) }void amd_iommu_flush_pages (struct domain *d, -unsigned long gfn,unsigned int order) +unsigned long bfn,unsigned int order) { -_amd_iommu_flush_pages (d, (uint64_t ) gfn <<PAGE_SHIFT , order); +_amd_iommu_flush_pages (d,__bfn_to_baddr (bfn), order); }void amd_iommu_flush_device (struct amd_iommu *iommu,uint16_t bdf)
xen/drivers/passthrough/amd/iommu_map.c +38 -38
251 lines not on this page · raw message
xen/drivers/passthrough/amd/pci_amd_iommu.c +1 -1
@@ -578,7 +578,7 @@ static void amd_dump_p2m_table_level(struct page_info* pg, int level,maddr_to_page (next_table_maddr), next_level, address, indent + 1); else - printk("%*sgfn: %08lx mfn: %08lx\n", + printk("%*sbfn: %08lx mfn: %08lx\n", indent,"" , (unsigned long )PFN_DOWN (address), (unsigned long)PFN_DOWN(next_table_maddr));
xen/drivers/passthrough/arm/smmu.c +8 -8
@@ -2551,7 +2551,7 @@ static int __must_check arm_smmu_iotlb_flush_all(struct domain *d) }static int __must_check arm_smmu_iotlb_flush (struct domain *d, -unsigned long gfn, +unsigned long bfn,unsigned int page_count) {/* ARM SMMU v1 doesn't have flush by VMA and VMID */
@@ -2737,7 +2737,7 @@ static void arm_smmu_iommu_domain_teardown(struct domain *d)xfree (xen_domain); } -static int __must_check arm_smmu_map_page (struct domain *d,unsigned long gfn, +static int __must_check arm_smmu_map_page (struct domain *d,unsigned long bfn,unsigned long mfn,unsigned int flags) {p2m_type_t t;
@@ -2748,10 +2748,10 @@ static int __must_check arm_smmu_map_page(struct domain *d, unsigned long gfn,* protected by an IOMMU, Xen needs to add a 1:1 mapping in the domain * p2m to allow DMA request to work. * This is only valid when the domain is directed mapped. Hence this -* function should only be used by gnttab code with gfn == mfn. +* function should only be used by gnttab code with gfn == mfn == bfn. */ BUG_ON (!is_domain_direct_mapped (d)); -BUG_ON (mfn != gfn); +BUG_ON (mfn != bfn);/* We only support readable and writable flags */ if (!(flags & (IOMMUF_readable | IOMMUF_writable)))
@@ -2763,19 +2763,19 @@ static int __must_check arm_smmu_map_page(struct domain *d, unsigned long gfn,* The function guest_physmap_add_entry replaces the current mapping * if there is already one... */ -return guest_physmap_add_entry (d,_gfn (gfn),_mfn (mfn),0 , t); +return guest_physmap_add_entry (d,_gfn (bfn),_mfn (bfn),0 , t); } -static int __must_check arm_smmu_unmap_page (struct domain *d,unsigned long gfn) +static int __must_check arm_smmu_unmap_page (struct domain *d,unsigned long bfn) {/* * This function should only be used by gnttab code when the domain -* is direct mapped +* is direct mapped (i.e. gfn == mfn == bfn). */ if ( !is_domain_direct_mapped (d) )return -EINVAL ; -return guest_physmap_remove_page (d,_gfn (gfn),_mfn (gfn),0 ); +return guest_physmap_remove_page (d,_gfn (bfn),_mfn (bfn),0 ); }static const struct iommu_ops arm_smmu_iommu_ops = {
xen/drivers/passthrough/iommu.c +14 -14
@@ -185,7 +185,7 @@ void __hwdom_init iommu_hwdom_init(struct domain *d) page_list_for_each ( page, &d->page_list ) {unsigned long mfn =mfn_x (page_to_mfn (page)); -unsigned long gfn =mfn_to_gmfn (d, mfn); +unsigned long bfn =mfn_to_gmfn (d, mfn);unsigned int mapping = IOMMUF_readable;int ret;
@@ -194,7 +194,7 @@ void __hwdom_init iommu_hwdom_init(struct domain *d) == PGT_writable_page) ) mapping |= IOMMUF_writable; - ret = hd->platform_ops->map_page (d, gfn, mfn, mapping); + ret = hd->platform_ops->map_page (d, bfn, mfn, mapping);if ( !rc ) rc = ret;
@@ -255,7 +255,7 @@ void iommu_domain_destroy(struct domain *d)arch_iommu_domain_destroy (d); } -int iommu_map_page (struct domain *d,unsigned long gfn,unsigned long mfn, +int iommu_map_page (struct domain *d,unsigned long bfn,unsigned long mfn,unsigned int flags) {const struct domain_iommu *hd =dom_iommu (d);
@@ -264,13 +264,13 @@ int iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,if ( !iommu_enabled || !hd->platform_ops )return 0 ; - rc = hd->platform_ops->map_page (d, gfn, mfn, flags); + rc = hd->platform_ops->map_page (d, bfn, mfn, flags);if (unlikely (rc) ) {if ( !d->is_shutting_down &&printk_ratelimit () )printk (XENLOG_ERR -"d%d: IOMMU mapping gfn %#lx to mfn %#lx failed: %d\n" , - d->domain_id, gfn, mfn, rc); +"d%d: IOMMU mapping bfn %#lx to mfn %#lx failed: %d\n" , + d->domain_id, bfn, mfn, rc);if ( !is_hardware_domain (d) )domain_crash (d);
@@ -279,7 +279,7 @@ int iommu_map_page(struct domain *d, unsigned long gfn, unsigned long mfn,return rc; } -int iommu_unmap_page (struct domain *d,unsigned long gfn) +int iommu_unmap_page (struct domain *d,unsigned long bfn) {const struct domain_iommu *hd =dom_iommu (d);int rc;
@@ -287,13 +287,13 @@ int iommu_unmap_page(struct domain *d, unsigned long gfn)if ( !iommu_enabled || !hd->platform_ops )return 0 ; - rc = hd->platform_ops->unmap_page (d, gfn); + rc = hd->platform_ops->unmap_page (d, bfn);if (unlikely (rc) ) {if ( !d->is_shutting_down &&printk_ratelimit () )printk (XENLOG_ERR -"d%d: IOMMU unmapping gfn %#lx failed: %d\n" , - d->domain_id, gfn, rc); +"d%d: IOMMU unmapping bfn %#lx failed: %d\n" , + d->domain_id, bfn, rc);if ( !is_hardware_domain (d) )domain_crash (d);
@@ -319,7 +319,7 @@ static void iommu_free_pagetables(unsigned long unused) cpumask_cycle(smp_processor_id(), &cpu_online_map)); } -int iommu_iotlb_flush (struct domain *d,unsigned long gfn, +int iommu_iotlb_flush (struct domain *d,unsigned long bfn,unsigned int page_count) {const struct domain_iommu *hd =dom_iommu (d);
@@ -328,13 +328,13 @@ int iommu_iotlb_flush(struct domain *d, unsigned long gfn,if ( !iommu_enabled || !hd->platform_ops || !hd->platform_ops->iotlb_flush )return 0 ; - rc = hd->platform_ops->iotlb_flush (d, gfn, page_count); + rc = hd->platform_ops->iotlb_flush (d, bfn, page_count);if (unlikely (rc) ) {if ( !d->is_shutting_down &&printk_ratelimit () )printk (XENLOG_ERR -"d%d: IOMMU IOTLB flush failed: %d, gfn %#lx, page count %u\n" , - d->domain_id, rc, gfn, page_count); +"d%d: IOMMU IOTLB flush failed: %d, bfn %#lx, page count %u\n" , + d->domain_id, rc, bfn, page_count);if ( !is_hardware_domain (d) )domain_crash (d);
xen/drivers/passthrough/vtd/iommu.c +15 -15
@@ -585,7 +585,7 @@ static int __must_check iommu_flush_all(void) }static int __must_check iommu_flush_iotlb (struct domain *d, -unsigned long gfn, +unsigned long bfn,bool_t dma_old_pte_present,unsigned int page_count) {
@@ -612,12 +612,12 @@ static int __must_check iommu_flush_iotlb(struct domain *d,if ( iommu_domid ==-1 )continue ; -if ( page_count !=1 || gfn ==gfn_x (INVALID_GFN ) ) +if ( page_count !=1 || bfn ==bfn_x (INVALID_BFN ) ) rc =iommu_flush_iotlb_dsi (iommu, iommu_domid,0 , flush_dev_iotlb);else rc =iommu_flush_iotlb_psi (iommu, iommu_domid, - (paddr_t )gfn <<PAGE_SHIFT_4K , +__bfn_to_baddr (bfn),PAGE_ORDER_4K , !dma_old_pte_present, flush_dev_iotlb);
@@ -633,15 +633,15 @@ static int __must_check iommu_flush_iotlb(struct domain *d, }static int __must_check iommu_flush_iotlb_pages (struct domain *d, -unsigned long gfn, +unsigned long bfn,unsigned int page_count) { -return iommu_flush_iotlb (d, gfn,1 , page_count); +return iommu_flush_iotlb (d, bfn,1 , page_count); }static int __must_check iommu_flush_iotlb_all (struct domain *d) { -return iommu_flush_iotlb (d,gfn_x (INVALID_GFN ),0 ,0 ); +return iommu_flush_iotlb (d,bfn_x (INVALID_BFN ),0 ,0 ); }/* clear one page's page table */
@@ -1767,7 +1767,7 @@ static void iommu_domain_teardown(struct domain *d) }static int __must_check intel_iommu_map_page (struct domain *d, -unsigned long gfn, +unsigned long bfn,unsigned long mfn,unsigned int flags) {
@@ -1786,14 +1786,14 @@ static int __must_check intel_iommu_map_page(struct domain *d,spin_lock (&hd->arch.mapping_lock); - pg_maddr =addr_to_dma_page_maddr (d, (paddr_t )gfn <<PAGE_SHIFT_4K ,1 ); + pg_maddr =addr_to_dma_page_maddr (d,__bfn_to_baddr (bfn),1 );if ( pg_maddr ==0 ) {spin_unlock (&hd->arch.mapping_lock);return -ENOMEM ; } page = (struct dma_pte *)map_vtd_domain_page (pg_maddr); - pte = page + (gfn &LEVEL_MASK ); + pte = page + (bfn &LEVEL_MASK ); old = *pte;dma_set_pte_addr (new, (paddr_t )mfn <<PAGE_SHIFT_4K ); dma_set_pte_prot(new,
@@ -1817,22 +1817,22 @@ static int __must_check intel_iommu_map_page(struct domain *d,unmap_vtd_domain_page (page);if ( !this_cpu (iommu_dont_flush_iotlb) ) - rc =iommu_flush_iotlb (d, gfn,dma_pte_present (old),1 ); + rc =iommu_flush_iotlb (d, bfn,dma_pte_present (old),1 );return rc; }static int __must_check intel_iommu_unmap_page (struct domain *d, -unsigned long gfn) +unsigned long bfn) {/* Do nothing if hardware domain and iommu supports pass thru. */ if ( iommu_passthrough &&is_hardware_domain (d) )return 0 ; -return dma_pte_clear_one (d, (paddr_t )gfn <<PAGE_SHIFT_4K ); +return dma_pte_clear_one (d,__bfn_to_baddr (bfn)); } -int iommu_pte_flush (struct domain *d,u64 gfn,u64 *pte, +int iommu_pte_flush (struct domain *d,uint64_t bfn,uint64_t *pte,int order,int present) {struct acpi_drhd_unit *drhd;
@@ -1856,7 +1856,7 @@ int iommu_pte_flush(struct domain *d, u64 gfn, u64 *pte,continue ; rc =iommu_flush_iotlb_psi (iommu, iommu_domid, - (paddr_t )gfn <<PAGE_SHIFT_4K , +__bfn_to_baddr (bfn), order, !present, flush_dev_iotlb);if ( rc >0 ) {
@@ -2626,7 +2626,7 @@ static void vtd_dump_p2m_table_level(paddr_t pt_maddr, int level, paddr_t gpa,vtd_dump_p2m_table_level (dma_pte_addr (*pte), next_level, address, indent +1 ); else - printk("%*sgfn: %08lx mfn: %08lx\n", + printk("%*sbfn: %08lx mfn: %08lx\n", indent,"" , (unsigned long )(address >>PAGE_SHIFT_4K ), (unsigned long)(dma_pte_addr(*pte) >> PAGE_SHIFT_4K));
xen/include/xen/iommu.h +32 -6
@@ -23,11 +23,37 @@#include <xen/page-defs.h> #include <xen/spinlock.h> #include <xen/pci.h> +#include <xen/typesafe.h> #include <public/hvm/ioreq.h> #include <public/domctl.h> #include <asm/device.h> #include <asm/iommu.h> +TYPE_SAFE (uint64_t, bfn); +#define PRI_bfn PRIx64 +#define INVALID_BFN _bfn(~0UL) + +#ifndef bfn_t +#define bfn_t/* Grep fodder: bfn_t, _bfn() and bfn_x() are defined above */ +#define _bfn +#define bfn_x +#undef bfn_t +#undef _bfn +#undef bfn_x +#endif + +#define IOMMU_PAGE_SHIFT 12 +#define IOMMU_PAGE_SIZE (_AC(1,L) << IOMMU_PAGE_SHIFT) +#define IOMMU_PAGE_MASK (~(IOMMU_PAGE_SIZE - 1)) + +typedef uint64_t baddr_t ; + +#define __bfn_to_baddr (bfn) ((baddr_t)(bfn) << IOMMU_PAGE_SHIFT) +#define __baddr_to_bfn (baddr) ((uint64_t)(baddr >> IOMMU_PAGE_SHIFT)) + +#define bfn_to_baddr (bfn) __bfn_to_baddr(bfn_x(bfn)) +#define baddr_to_bfn (baddr) _bfn(__baddr_to_bfn(baddr)) +extern bool_t iommu_enable, iommu_enabled;extern bool_t force_iommu, iommu_dom0_strict, iommu_verbose;extern bool_t iommu_workaround_bios_bug, iommu_igfx, iommu_passthrough;
@@ -60,9 +86,9 @@ void iommu_teardown(struct domain *d);#define IOMMUF_readable (1u<<_IOMMUF_readable)#define _IOMMUF_writable 1#define IOMMUF_writable (1u<<_IOMMUF_writable) -int __must_check iommu_map_page (struct domain *d,unsigned long gfn, +int __must_check iommu_map_page (struct domain *d,unsigned long bfn,unsigned long mfn,unsigned int flags); -int __must_check iommu_unmap_page (struct domain *d,unsigned long gfn); +int __must_check iommu_unmap_page (struct domain *d,unsigned long bfn);enum iommu_feature {
@@ -150,9 +176,9 @@ struct iommu_ops {#endif /* HAS_PCI */ void (*teardown)(struct domain *d); -int __must_check (*map_page)(struct domain *d,unsigned long gfn, +int __must_check (*map_page)(struct domain *d,unsigned long bfn,unsigned long mfn,unsigned int flags); -int __must_check (*unmap_page)(struct domain *d,unsigned long gfn); +int __must_check (*unmap_page)(struct domain *d,unsigned long bfn);void (*free_page_table)(struct page_info *);#ifdef CONFIG_X86 void (*update_ire_from_apic)(unsigned int apic,unsigned int reg,unsigned int value);
@@ -163,7 +189,7 @@ struct iommu_ops {void (*resume)(void );void (*share_p2m)(struct domain *d);void (*crash_shutdown)(void ); -int __must_check (*iotlb_flush)(struct domain *d,unsigned long gfn, +int __must_check (*iotlb_flush)(struct domain *d,unsigned long bfn,unsigned int page_count);int __must_check (*iotlb_flush_all)(struct domain *d);int (*get_reserved_device_memory)(iommu_grdm_t *,void *);
@@ -185,7 +211,7 @@ int iommu_do_pci_domctl(struct xen_domctl *, struct domain *d,int iommu_do_domctl (struct xen_domctl *,struct domain *d,XEN_GUEST_HANDLE_PARAM (xen_domctl_t )); -int __must_check iommu_iotlb_flush (struct domain *d,unsigned long gfn, +int __must_check iommu_iotlb_flush (struct domain *d,unsigned long bfn,unsigned int page_count);int __must_check iommu_iotlb_flush_all (struct domain *d);
xen/include/xen/mm.h +5 -0
@@ -26,6 +26,11 @@* A linear idea of a guest physical address space. For an auto-translated * guest, pfn == gfn while for a non-translated guest, pfn != gfn. * +* bfn: Bus Frame Number (definitions in include/xen/iommu.h) +* The linear frame numbers of IOMMU address space. All initiators for (i.e. +* all devices assigned to) a guest share a single IOMMU address space and, +* by default, Xen will ensure bfn == pfn. +* * WARNING: Some of these terms have changed over time while others have been * used inconsistently, meaning that a lot of existing code does not match the * definitions above. New code should use these terms as described here, and
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
xen/include/xen/mm.h line 32
Quoted 11 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:46, <[email protected]> wrote:
xen/include/xen/mm.h +4
-0@@ -26,6 +26,11 @@* A linear idea of a guest physical address space. For an auto-translated * guest, pfn == gfn while for a non-translated guest, pfn != gfn. * +* bfn: Bus Frame Number (definitions in include/xen/iommu.h) +* The linear frame numbers of IOMMU address space. All initiators for (i.e. +* all devices assigned to) a guest share a single IOMMU address space and, +* by default, Xen will ensure bfn == pfn.
The code changes are purely mechanical and hence fine, but I have to admit I continue to struggle with the "bus" part in the name here:
I don't think it is any less ambiguous than GFN, because which bus are you thinking about here? The (virtual) one as seen by the guest, aiui. The physical (host) one would be at least as natural to be indexed by such typed/named variables. I'd somehow like it to be made explicit in the name whose view these represent. GBFN? VBFN?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 24 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 30 August 2018 17:00
To: Paul Durrant <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Kevin Tian <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...xen/include/xen/mm.h line 30
Quoted 9 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:46, <[email protected]> wrote:
xen/include/xen/mm.h +2
-0@@ -26,6 +26,11 @@* A linear idea of a guest physical address space. For an auto-translated * guest, pfn == gfn while for a non-translated guest, pfn != gfn. * +* bfn: Bus Frame Number (definitions in include/xen/iommu.h) +* The linear frame numbers of IOMMU address space. All initiators for (i.e.
xen/include/xen/mm.h line 31
Quoted 1 line
xen/include/xen/mm.h +1
-0+* all devices assigned to) a guest share a single IOMMU address space and,
xen/include/xen/mm.h line 32
Quoted 1 line
xen/include/xen/mm.h +1
-0+* by default, Xen will ensure bfn == pfn.
Quoted 9 lines
The code changes are purely mechanical and hence fine, but I have to admit I continue to struggle with the "bus" part in the name here:
I don't think it is any less ambiguous than GFN, because which bus are you thinking about here? The (virtual) one as seen by the guest, aiui. The physical (host) one would be at least as natural to be indexed by such typed/named variables. I'd somehow like it to be made explicit in the name whose view these represent. GBFN? VBFN?
Well, it always refers to whatever physical bus is on the other side of the IOMMU from the core so it's the view of whatever peripheral devices are located on that bus. As Kevin said each device can have its own address space and the fact we always use a global per-VM space is an implementation detail so DFN for 'device frame number' or IOFN (since IOVA is reasonably widely used term) might be more future-proof?
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 27 lines
Quoted 18 lines
Quoted 1 line · On 03.09.18 at 10:23, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 30 August 2018 17:00xen/include/xen/mm.h line 30
Quoted 9 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:46, <[email protected]> wrote:
xen/include/xen/mm.h +2
-0@@ -26,6 +26,11 @@* A linear idea of a guest physical address space. For an auto-translated * guest, pfn == gfn while for a non-translated guest, pfn != gfn. * +* bfn: Bus Frame Number (definitions in include/xen/iommu.h) +* The linear frame numbers of IOMMU address space. All initiators for (i.e.
xen/include/xen/mm.h line 31
Quoted 1 line
xen/include/xen/mm.h +1
-0+* all devices assigned to) a guest share a single IOMMU address space and,
xen/include/xen/mm.h line 32
Quoted 1 line
xen/include/xen/mm.h +1
-0+* by default, Xen will ensure bfn == pfn. Quoted 8 lines
The code changes are purely mechanical and hence fine, but I have to admit I continue to struggle with the "bus" part in the name here:
I don't think it is any less ambiguous than GFN, because which bus are you thinking about here? The (virtual) one as seen by the guest, aiui. The physical (host) one would be at least as natural to be indexed by such typed/named variables. I'd somehow like it to be made explicit in the name whose view these represent. GBFN? VBFN?
Quoted 6 lines
Well, it always refers to whatever physical bus is on the other side of the IOMMU from the core so it's the view of whatever peripheral devices are located on that bus. As Kevin said each device can have its own address space and the fact we always use a global per-VM space is an implementation detail so DFN for 'device frame number' or IOFN (since IOVA is reasonably widely used term) might be more future-proof?
I don't think any name alone would make things future-proof. Considering the per-device address space case, a frame number would always need to come in a tuple paired with a device identifier. Nevertheless I think either of the names you suggest would be a slight improvement, perhaps DFN even more than IOFN, as that includes whose view this is. But I'd certainly appreciate opinions of others, so you don't again end up doing something that I've asked for,just to subsequently undo it again.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 43 lines
From: Jan Beulich [mailto:[email protected]]
Sent: Monday, September 3, 2018 7:47 PMQuoted 9 lines
Quoted 9 lines
Quoted 1 line · On 03.09.18 at 10:23, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 30 August 2018 17:00xen/include/xen/mm.h line 26
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:46, <[email protected]> wrote:
xen/include/xen/mm.h +0
-0@@ -26,6 +26,11 @@* A linear idea of a guest physical address space. For an auto- translated
Quoted 4 lines
Quoted 4 lines
xen/include/xen/mm.h
Quoted 4 lines
xen/include/xen/mm.h +2
-0* guest, pfn == gfn while for a non-translated guest, pfn != gfn. * +* bfn: Bus Frame Number (definitions in include/xen/iommu.h) +* The linear frame numbers of IOMMU address space. All initiators for
Quoted 2 lines
Quoted 2 lines
(i.e.
Quoted 1 line
xen/include/xen/mm.h +1
-0+* all devices assigned to) a guest share a single IOMMU address space
Quoted 12 lines
Quoted 11 lines
and,
xen/include/xen/mm.h line 32
Quoted 1 line
xen/include/xen/mm.h +1
-0+* by default, Xen will ensure bfn == pfn. The code changes are purely mechanical and hence fine, but I have to admit I continue to struggle with the "bus" part in the name here:
I don't think it is any less ambiguous than GFN, because which bus are you thinking about here? The (virtual) one as seen by the guest, aiui. The physical (host) one would be at least as natural to be indexed by such typed/named variables. I'd somehow like it to be made explicit in the name whose view these represent. GBFN? VBFN?Quoted 3 lines
Well, it always refers to whatever physical bus is on the other side of the IOMMU from the core so it's the view of whatever peripheral devices are located on that bus. As Kevin said each device can have its own address
space
Quoted 1 line
and the fact we always use a global per-VM space is an implementation
detail
Quoted 1 line
so DFN for 'device frame number' or IOFN (since IOVA is reasonably
widely
Quoted 1 line
used term) might be more future-proof?
Quoted 8 lines
I don't think any name alone would make things future-proof. Considering the per-device address space case, a frame number would always need to come in a tuple paired with a device identifier. Nevertheless I think either of the names you suggest would be a slight improvement, perhaps DFN even more than IOFN, as that includes whose view this is. But I'd certainly appreciate opinions of others, so you don't again end up doing something that I've asked for,just to subsequently undo it again.
bus address is commonly used along with physical/virtual address, to represent different views between devices and CPU. From that angle I think BFN is a clear term in this context. btw it is not necessary to differentiate GBFN and MBFN since there is only one BFN view per device.
DFN (device frame number) makes me think about something on the device. What does a device address mean? Even DMA frame number (thus DMA address) sounds clearer with same DFN notation.
IOFN/IOVA is less preferred to me, as in most cases it refers to virtual addresses explicitly managed by device driver to be used in I/O requests. it is not fit for the case where iofn=pfn.
so I prefer to staying with BFN. :-)
Thanks
Kevin
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 53 lines
Quoted 44 lines
Quoted 1 line · On 04.09.18 at 08:48, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Monday, September 3, 2018 7:47 PMQuoted 9 lines
Quoted 9 lines
Quoted 1 line · On 03.09.18 at 10:23, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 30 August 2018 17:00xen/include/xen/mm.h line 26
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:46, <[email protected]> wrote:
xen/include/xen/mm.h +0
-0@@ -26,6 +26,11 @@* A linear idea of a guest physical address space. For an auto- translated
Quoted 4 lines
Quoted 4 lines
xen/include/xen/mm.h
Quoted 4 lines
xen/include/xen/mm.h +2
-0* guest, pfn == gfn while for a non-translated guest, pfn != gfn. * +* bfn: Bus Frame Number (definitions in include/xen/iommu.h) +* The linear frame numbers of IOMMU address space. All initiators for
Quoted 2 lines
Quoted 2 lines
(i.e.
Quoted 1 line
xen/include/xen/mm.h +1
-0+* all devices assigned to) a guest share a single IOMMU address space
Quoted 12 lines
Quoted 11 lines
and,
xen/include/xen/mm.h line 32
Quoted 1 line
xen/include/xen/mm.h +1
-0+* by default, Xen will ensure bfn == pfn. The code changes are purely mechanical and hence fine, but I have to admit I continue to struggle with the "bus" part in the name here:
I don't think it is any less ambiguous than GFN, because which bus are you thinking about here? The (virtual) one as seen by the guest, aiui. The physical (host) one would be at least as natural to be indexed by such typed/named variables. I'd somehow like it to be made explicit in the name whose view these represent. GBFN? VBFN?Quoted 3 lines
Well, it always refers to whatever physical bus is on the other side of the IOMMU from the core so it's the view of whatever peripheral devices are located on that bus. As Kevin said each device can have its own address
space
Quoted 1 line
and the fact we always use a global per-VM space is an implementation
detail
Quoted 1 line
so DFN for 'device frame number' or IOFN (since IOVA is reasonably
widely
Quoted 1 line
used term) might be more future-proof?
Quoted 8 lines
I don't think any name alone would make things future-proof. Considering the per-device address space case, a frame number would always need to come in a tuple paired with a device identifier. Nevertheless I think either of the names you suggest would be a slight improvement, perhaps DFN even more than IOFN, as that includes whose view this is. But I'd certainly appreciate opinions of others, so you don't again end up doing something that I've asked for,just to subsequently undo it again.
Quoted 5 lines
bus address is commonly used along with physical/virtual address, to represent different views between devices and CPU. From that angle I think BFN is a clear term in this context. btw it is not necessary to differentiate GBFN and MBFN since there is only one BFN view per device.
Sure, but you neglect the presence of one or more IOMMUs when you say "between devices and CPU". There addresses prior to and after IOMMU translation are distinct, and while the one before the translation matches the device view, the one after translation does not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and the other representing the IOMMU (output) view.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 9 lines
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 6 lines
bus address is commonly used along with physical/virtual address, to represent different views between devices and CPU. From that angle I think BFN is a clear term in this context. btw it is not necessary to differentiate GBFN and MBFN since there is only one BFN view per device.
Quoted 8 lines
Sure, but you neglect the presence of one or more IOMMUs when you say "between devices and CPU". There addresses prior to and after IOMMU translation are distinct, and while the one before the translation matches the device view, the one after translation does not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and the other representing the IOMMU (output) view.
I didn't get. the output address from IOMMU is the one sent to memory controller, same as the one sent from CPU.
Thanks
Kevin
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 19 lines
Quoted 10 lines
Quoted 1 line · On 04.09.18 at 10:37, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 6 lines
bus address is commonly used along with physical/virtual address, to represent different views between devices and CPU. From that angle I think BFN is a clear term in this context. btw it is not necessary to differentiate GBFN and MBFN since there is only one BFN view per device.
Quoted 8 lines
Sure, but you neglect the presence of one or more IOMMUs when you say "between devices and CPU". There addresses prior to and after IOMMU translation are distinct, and while the one before the translation matches the device view, the one after translation does not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and the other representing the IOMMU (output) view.
Quoted 2 lines
I didn't get. the output address from IOMMU is the one sent to memory controller, same as the one sent from CPU.
That's on present x86 systems, but aiui not in the general case. The terminology to be used in Xen should fit the general case though.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 30 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 09:47
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Paul Durrant <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [Xen-devel] [PATCH v6 01/14] iommu: introduce the concept of BFN...Quoted 19 lines
Quoted 18 lines
Quoted 1 line · On 04.09.18 at 10:37, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 6 lines
bus address is commonly used along with physical/virtual address, to represent different views between devices and CPU. From that angle I think BFN is a clear term in this context. btw it is not necessary to differentiate GBFN and MBFN since there is only one BFN view per device.
Sure, but you neglect the presence of one or more IOMMUs when you say "between devices and CPU". There addresses prior to and after IOMMU translation are distinct, and while the one before the translation matches the device view, the one after translation does not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and the other representing the IOMMU (output) view.
Quoted 5 lines
Quoted 2 lines
I didn't get. the output address from IOMMU is the one sent to memory controller, same as the one sent from CPU.
That's on present x86 systems, but aiui not in the general case. The terminology to be used in Xen should fit the general case though.
So your concern is cascaded IOMMUs?
Paul
Quoted 3 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 31 lines
Quoted 31 lines
Quoted 1 line · On 04.09.18 at 10:49, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 09:47
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Paul Durrant <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [Xen-devel] [PATCH v6 01/14] iommu: introduce the concept of BFN...Quoted 19 lines
Quoted 18 lines
Quoted 1 line · On 04.09.18 at 10:37, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 6 lines
bus address is commonly used along with physical/virtual address, to represent different views between devices and CPU. From that angle I think BFN is a clear term in this context. btw it is not necessary to differentiate GBFN and MBFN since there is only one BFN view per device.
Sure, but you neglect the presence of one or more IOMMUs when you say "between devices and CPU". There addresses prior to and after IOMMU translation are distinct, and while the one before the translation matches the device view, the one after translation does not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and the other representing the IOMMU (output) view.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
I didn't get. the output address from IOMMU is the one sent to memory controller, same as the one sent from CPU.
That's on present x86 systems, but aiui not in the general case. The terminology to be used in Xen should fit the general case though.
So your concern is cascaded IOMMUs?
Not primarily. My concern are systems with an I/O address space (behind the IOMMU) distinct from the CPU address space. Iirc at least Alpha is/was that way.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 42 lines
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 5:08 PMQuoted 6 lines
Quoted 6 lines
Quoted 1 line · On 04.09.18 at 10:49, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 09:47
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; JulienGrall
Quoted 1 line
Quoted 1 line
<[email protected]>; Paul Durrant <[email protected]>;
Stefano
Quoted 3 lines
Quoted 3 lines
Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [Xen-devel] [PATCH v6 01/14] iommu: introduce the conceptof
Quoted 7 lines
Quoted 7 lines
BFN...
Quoted 5 lines
Quoted 5 lines
Quoted 1 line · On 04.09.18 at 10:37, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 2 lines
bus address is commonly used along with physical/virtual address,
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
represent different views between devices and CPU. From that
angle
Quoted 16 lines
Quoted 16 lines
Quoted 15 lines
Quoted 4 lines
Quoted 3 lines
I think BFN is a clear term in this context. btw it is not necessary to differentiate GBFN and MBFN since there is only one BFN view per device.
Quoted 8 lines
Sure, but you neglect the presence of one or more IOMMUs when you say "between devices and CPU". There addresses prior to and after IOMMU translation are distinct, and while the one before the translation matches the device view, the one after translation does not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and the other representing the IOMMU (output) view.
I didn't get. the output address from IOMMU is the one sent to memory controller, same as the one sent from CPU.
Quoted 9 lines
Quoted 4 lines
Quoted 2 lines
That's on present x86 systems, but aiui not in the general case. The terminology to be used in Xen should fit the general case though.
So your concern is cascaded IOMMUs?
Not primarily. My concern are systems with an I/O address space (behind the IOMMU) distinct from the CPU address space. Iirc at least Alpha is/was that way.
Then Paul please documents clearly that this bus address refers to the input side of IOMMU. :-)
Thanks
Kevin
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 48 lines
Quoted 48 lines
Quoted 1 line · On 05.09.18 at 02:42, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 5:08 PMQuoted 6 lines
Quoted 6 lines
Quoted 1 line · On 04.09.18 at 10:49, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 09:47
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; JulienGrall
Quoted 1 line
Quoted 1 line
<[email protected]>; Paul Durrant <[email protected]>;
Stefano
Quoted 3 lines
Quoted 3 lines
Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [Xen-devel] [PATCH v6 01/14] iommu: introduce the conceptof
Quoted 7 lines
Quoted 7 lines
BFN...
Quoted 5 lines
Quoted 5 lines
Quoted 1 line · On 04.09.18 at 10:37, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 2 lines
bus address is commonly used along with physical/virtual address,
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
represent different views between devices and CPU. From that
angle
Quoted 20 lines
Quoted 13 lines
Quoted 13 lines
Quoted 12 lines
Quoted 3 lines
I think BFN is a clear term in this context. btw it is not necessary to differentiate GBFN and MBFN since there is only one BFN view per device.
Sure, but you neglect the presence of one or more IOMMUs when you say "between devices and CPU". There addresses prior to and after IOMMU translation are distinct, and while the one before the translation matches the device view, the one after translation does not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and the other representing the IOMMU (output) view.
Quoted 5 lines
Quoted 2 lines
I didn't get. the output address from IOMMU is the one sent to memory controller, same as the one sent from CPU.
That's on present x86 systems, but aiui not in the general case. The terminology to be used in Xen should fit the general case though.
So your concern is cascaded IOMMUs?
Quoted 7 lines
Quoted 4 lines
Not primarily. My concern are systems with an I/O address space (behind the IOMMU) distinct from the CPU address space. Iirc at least Alpha is/was that way.
Then Paul please documents clearly that this bus address refers to the input side of IOMMU. :-)
But when reading code you can't always go back to look at the one place where its meaning is documented. Hence my desire for a name which properly conveys the meaning.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 60 lines
From: Jan Beulich [mailto:[email protected]]
Sent: Wednesday, September 5, 2018 2:49 PMQuoted 10 lines
Quoted 10 lines
Quoted 1 line · On 05.09.18 at 02:42, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 5:08 PMQuoted 6 lines
Quoted 6 lines
Quoted 1 line · On 04.09.18 at 10:49, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 09:47
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; JulienQuoted 6 lines
Quoted 6 lines
Grall
Quoted 1 line
Quoted 1 line
<[email protected]>; Paul Durrant <[email protected]>;
Stefano
Quoted 3 lines
Quoted 3 lines
Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [Xen-devel] [PATCH v6 01/14] iommu: introduce theconcept
Quoted 8 lines
Quoted 8 lines
of
Quoted 7 lines
Quoted 7 lines
BFN...
Quoted 5 lines
Quoted 5 lines
Quoted 1 line · On 04.09.18 at 10:37, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 2 lines
bus address is commonly used along with physical/virtual
address,
Quoted 4 lines
Quoted 4 lines
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
represent different views between devices and CPU. From that
angle
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
I think BFN is a clear term in this context. btw it is not necessary
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
differentiate GBFN and MBFN since there is only one BFN view
per
Quoted 24 lines
Quoted 23 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 10 lines
Quoted 1 line
device.
Sure, but you neglect the presence of one or more IOMMUs when you say "between devices and CPU". There addresses prior to and after IOMMU translation are distinct, and while the one before the translation matches the device view, the one after translation does not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and the other representing the IOMMU (output) view.
Quoted 7 lines
Quoted 5 lines
Quoted 2 lines
I didn't get. the output address from IOMMU is the one sent to memory controller, same as the one sent from CPU.
That's on present x86 systems, but aiui not in the general case. The terminology to be used in Xen should fit the general case though.
So your concern is cascaded IOMMUs?
Not primarily. My concern are systems with an I/O address space (behind the IOMMU) distinct from the CPU address space. Iirc at least Alpha is/was that way.
Quoted 7 lines
Quoted 2 lines
Then Paul please documents clearly that this bus address refers to the input side of IOMMU. :-)
But when reading code you can't always go back to look at the one place where its meaning is documented. Hence my desire for a name which properly conveys the meaning.
Then possibly go back to DFN, but take 'D' as DMA instead of device?
Thanks
Kevin
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 64 lines
Quoted 64 lines
Quoted 1 line · On 05.09.18 at 08:56, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Wednesday, September 5, 2018 2:49 PMQuoted 10 lines
Quoted 10 lines
Quoted 1 line · On 05.09.18 at 02:42, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 5:08 PMQuoted 6 lines
Quoted 6 lines
Quoted 1 line · On 04.09.18 at 10:49, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 09:47
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; JulienQuoted 6 lines
Quoted 6 lines
Grall
Quoted 1 line
Quoted 1 line
<[email protected]>; Paul Durrant <[email protected]>;
Stefano
Quoted 3 lines
Quoted 3 lines
Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [Xen-devel] [PATCH v6 01/14] iommu: introduce theconcept
Quoted 8 lines
Quoted 8 lines
of
Quoted 7 lines
Quoted 7 lines
BFN...
Quoted 5 lines
Quoted 5 lines
Quoted 1 line · On 04.09.18 at 10:37, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 2 lines
bus address is commonly used along with physical/virtual
address,
Quoted 4 lines
Quoted 4 lines
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
represent different views between devices and CPU. From that
angle
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
I think BFN is a clear term in this context. btw it is not necessary
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
differentiate GBFN and MBFN since there is only one BFN view
per
Quoted 26 lines
Quoted 14 lines
Quoted 14 lines
Quoted 14 lines
Quoted 13 lines
Quoted 10 lines
Quoted 1 line
device.
Sure, but you neglect the presence of one or more IOMMUs when you say "between devices and CPU". There addresses prior to and after IOMMU translation are distinct, and while the one before the translation matches the device view, the one after translation does not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and the other representing the IOMMU (output) view.
I didn't get. the output address from IOMMU is the one sent to memory controller, same as the one sent from CPU.
Quoted 9 lines
Quoted 4 lines
Quoted 2 lines
That's on present x86 systems, but aiui not in the general case. The terminology to be used in Xen should fit the general case though.
So your concern is cascaded IOMMUs?
Not primarily. My concern are systems with an I/O address space (behind the IOMMU) distinct from the CPU address space. Iirc at least Alpha is/was that way.
Then Paul please documents clearly that this bus address refers to the input side of IOMMU. :-)
Quoted 6 lines
Quoted 4 lines
But when reading code you can't always go back to look at the one place where its meaning is documented. Hence my desire for a name which properly conveys the meaning.
Then possibly go back to DFN, but take 'D' as DMA instead of device?
How would "DMA" be any better than "bus"? Whose view it is then still is unclear.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 88 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 08:12
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Paul Durrant <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the concept of BFN...Quoted 8 lines
Quoted 8 lines
Quoted 1 line · On 05.09.18 at 08:56, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Wednesday, September 5, 2018 2:49 PMQuoted 4 lines
Quoted 4 lines
Quoted 1 line · On 05.09.18 at 02:42, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 5:08 PMQuoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 1 line · On 04.09.18 at 10:49, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 09:47
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>;Julien
Quoted 10 lines
Quoted 10 lines
Quoted 6 lines
Quoted 6 lines
Grall
Quoted 1 line
Quoted 1 line
<[email protected]>; Paul Durrant <[email protected]>;
Stefano
Quoted 3 lines
Quoted 3 lines
Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [Xen-devel] [PATCH v6 01/14] iommu: introduce theconcept
Quoted 3 lines
Quoted 3 lines
of
Quoted 2 lines
Quoted 2 lines
BFN...
Quoted 8 lines
Quoted 8 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 1 line · On 04.09.18 at 10:37, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 2 lines
bus address is commonly used along with physical/virtual
address,
Quoted 2 lines
Quoted 2 lines
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
represent different views between devices and CPU. From
that
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
angle
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
I think BFN is a clear term in this context. btw it is not
necessary
Quoted 6 lines
Quoted 6 lines
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
differentiate GBFN and MBFN since there is only one BFN view
per
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
device.
Sure, but you neglect the presence of one or more IOMMUs
when
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
you say "between devices and CPU". There addresses prior to
and
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
after IOMMU translation are distinct, and while the one before
the
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
translation matches the device view, the one after translation
does
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and
the
Quoted 24 lines
Quoted 23 lines
Quoted 11 lines
Quoted 11 lines
Quoted 10 lines
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
other representing the IOMMU (output) view.
I didn't get. the output address from IOMMU is the one sent to memory controller, same as the one sent from CPU.
That's on present x86 systems, but aiui not in the general case. The terminology to be used in Xen should fit the general case though.
So your concern is cascaded IOMMUs?
Quoted 7 lines
Quoted 4 lines
Not primarily. My concern are systems with an I/O address space (behind the IOMMU) distinct from the CPU address space. Iirc at least Alpha is/was that way.
Then Paul please documents clearly that this bus address refers to the input side of IOMMU. :-)
But when reading code you can't always go back to look at the one place where its meaning is documented. Hence my desire for a name which properly conveys the meaning.
Quoted 5 lines
Quoted 1 line
Then possibly go back to DFN, but take 'D' as DMA instead of device?
How would "DMA" be any better than "bus"? Whose view it is then still is unclear.
Personally I think 'bus address' is commonly enough used term for addresses used by devices for DMA. Indeed we have already 'dev_bus_addr' in the grant map and unmap hypercalls. So baddr and bfn seem like ok terms to me. It's also not impossible to rename these later if they prove problematic.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 91 lines
Quoted 91 lines
Quoted 1 line · On 05.09.18 at 11:13, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 08:12
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Paul Durrant <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the concept of BFN...Quoted 8 lines
Quoted 8 lines
Quoted 1 line · On 05.09.18 at 08:56, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Wednesday, September 5, 2018 2:49 PMQuoted 4 lines
Quoted 4 lines
Quoted 1 line · On 05.09.18 at 02:42, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 5:08 PMQuoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 1 line · On 04.09.18 at 10:49, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 09:47
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>;Julien
Quoted 10 lines
Quoted 10 lines
Quoted 6 lines
Quoted 6 lines
Grall
Quoted 1 line
Quoted 1 line
<[email protected]>; Paul Durrant <[email protected]>;
Stefano
Quoted 3 lines
Quoted 3 lines
Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [Xen-devel] [PATCH v6 01/14] iommu: introduce theconcept
Quoted 3 lines
Quoted 3 lines
of
Quoted 2 lines
Quoted 2 lines
BFN...
Quoted 8 lines
Quoted 8 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 1 line · On 04.09.18 at 10:37, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 2 lines
bus address is commonly used along with physical/virtual
address,
Quoted 2 lines
Quoted 2 lines
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
represent different views between devices and CPU. From
that
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
angle
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
I think BFN is a clear term in this context. btw it is not
necessary
Quoted 6 lines
Quoted 6 lines
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
differentiate GBFN and MBFN since there is only one BFN view
per
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
device.
Sure, but you neglect the presence of one or more IOMMUs
when
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
you say "between devices and CPU". There addresses prior to
and
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
after IOMMU translation are distinct, and while the one before
the
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
translation matches the device view, the one after translation
does
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
not necessarily match the CPU view. Hence there are two "bus" frame numbers here - one representing the device view, and
the
Quoted 25 lines
Quoted 16 lines
Quoted 16 lines
Quoted 15 lines
Quoted 10 lines
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines
other representing the IOMMU (output) view.
I didn't get. the output address from IOMMU is the one sent to memory controller, same as the one sent from CPU.
That's on present x86 systems, but aiui not in the general case. The terminology to be used in Xen should fit the general case though.
So your concern is cascaded IOMMUs?
Not primarily. My concern are systems with an I/O address space (behind the IOMMU) distinct from the CPU address space. Iirc at least Alpha is/was that way.
Quoted 7 lines
Quoted 2 lines
Then Paul please documents clearly that this bus address refers to the input side of IOMMU. :-)
But when reading code you can't always go back to look at the one place where its meaning is documented. Hence my desire for a name which properly conveys the meaning.
Then possibly go back to DFN, but take 'D' as DMA instead of device?
Quoted 8 lines
Quoted 3 lines
How would "DMA" be any better than "bus"? Whose view it is then still is unclear.
Personally I think 'bus address' is commonly enough used term for addresses used by devices for DMA. Indeed we have already '
dev_bus_addr' in the grant map and unmap hypercalls. So baddr and bfn seem like ok terms to me. It's also not impossible to rename these later if they prove problematic.
But that's the point - the names are problematic (to me): I permanently have to remind myself that they do _not_ refer to the addresses as seen when accessing memory, but the ones going _into_ the IOMMU. The confusion (on my part) arises every time I see a mixture of gfn, bfn, and mfn in the same patch, perhaps including some 1:1-ness assumptions between pairs of them.
Take these two hunks as example (mixing in some pfn as well):
@@ -436,7 +436,7 @@ static int iommu_merge_pages(struct domain *d, unsigned long pt_mfn,
* {Re, un}mapping super page frames causes re-allocation of io
* page tables.
*/
-static int iommu_pde_from_gfn(struct domain *d, unsigned long pfn, +static int iommu_pde_from_bfn(struct domain *d, unsigned long pfn,
unsigned long pt_mfn[])
{
u64 *pde, *next_table_vaddr;
@@ -477,11 +477,11 @@ static int iommu_pde_from_gfn(struct domain *d, unsigned long pfn,
next_table_mfn != 0 )
{
int i;
- unsigned long mfn, gfn;
+ unsigned long mfn, bfn;
unsigned int page_sz;
page_sz = 1 << (PTE_PER_TABLE_SHIFT * (next_level - 1));
- gfn = pfn & ~((1 << (PTE_PER_TABLE_SHIFT * next_level)) - 1);
+ bfn = pfn & ~((1 << (PTE_PER_TABLE_SHIFT * next_level)) - 1);
mfn = next_table_mfn;
/* allocate lower level page table */Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 118 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 10:39
To: Paul Durrant <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Kevin Tian <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the concept of BFN...Quoted 7 lines
Quoted 7 lines
Quoted 1 line · On 05.09.18 at 11:13, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 08:12
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Paul Durrant <[email protected]>;Stefano
Quoted 3 lines
Quoted 3 lines
Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the conceptof
Quoted 11 lines
Quoted 11 lines
BFN...
Quoted 9 lines
Quoted 9 lines
Quoted 1 line · On 05.09.18 at 08:56, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Wednesday, September 5, 2018 2:49 PMQuoted 5 lines
Quoted 5 lines
Quoted 1 line · On 05.09.18 at 02:42, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 5:08 PMQuoted 1 line
Quoted 1 line
Quoted 1 line · On 04.09.18 at 10:49, <[email protected]> wrote:
Quoted 8 lines
Quoted 8 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 09:47
To: Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>;Julien
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Grall
Quoted 1 line
Quoted 1 line
<[email protected]>; Paul Durrant
Quoted 14 lines
Quoted 14 lines
Quoted 14 lines
Quoted 14 lines
Quoted 4 lines
Quoted 4 lines
Stefano
Quoted 3 lines
Quoted 3 lines
Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [Xen-devel] [PATCH v6 01/14] iommu: introduce theconcept
Quoted 8 lines
Quoted 8 lines
of
Quoted 7 lines
Quoted 7 lines
BFN...
Quoted 5 lines
Quoted 5 lines
Quoted 1 line · On 04.09.18 at 10:37, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 4, 2018 4:33 PMQuoted 2 lines
bus address is commonly used along with physical/virtual
address,
Quoted 8 lines
Quoted 8 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
represent different views between devices and CPU. From
that
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
angle
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
I think BFN is a clear term in this context. btw it is not
necessary
Quoted 2 lines
Quoted 2 lines
to
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
differentiate GBFN and MBFN since there is only one BFN
view
Quoted 6 lines
Quoted 6 lines
Quoted 4 lines
Quoted 4 lines
per
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
device.
Sure, but you neglect the presence of one or more IOMMUs
when
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
you say "between devices and CPU". There addresses prior
to
Quoted 2 lines
Quoted 2 lines
and
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
after IOMMU translation are distinct, and while the one
before
Quoted 4 lines
Quoted 4 lines
the
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
translation matches the device view, the one after translation
does
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
not necessarily match the CPU view. Hence there are two
"bus"
Quoted 6 lines
Quoted 6 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
frame numbers here - one representing the device view, and
the
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
other representing the IOMMU (output) view.
I didn't get. the output address from IOMMU is the one sent
to
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
memory controller, same as the one sent from CPU.
That's on present x86 systems, but aiui not in the general case.
The
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
terminology to be used in Xen should fit the general case
though.
Quoted 16 lines
Quoted 16 lines
Quoted 16 lines
Quoted 15 lines
Quoted 10 lines
Quoted 7 lines
Quoted 2 lines
So your concern is cascaded IOMMUs?
Not primarily. My concern are systems with an I/O address space (behind the IOMMU) distinct from the CPU address space. Iirc at least Alpha is/was that way.
Then Paul please documents clearly that this bus address refers to the input side of IOMMU. :-)
But when reading code you can't always go back to look at the one place where its meaning is documented. Hence my desire for a name which properly conveys the meaning.
Quoted 7 lines
Quoted 5 lines
Quoted 1 line
Then possibly go back to DFN, but take 'D' as DMA instead of device?
How would "DMA" be any better than "bus"? Whose view it is then still is unclear.
Personally I think 'bus address' is commonly enough used term for
addresses
Quoted 8 lines
Quoted 1 line
used by devices for DMA. Indeed we have already '
dev_bus_addr' in thegrant
Quoted 2 lines
map and unmap hypercalls. So baddr and bfn seem like ok terms to me. It's also not impossible to rename these later if they prove problematic.
But that's the point - the names are problematic (to me): I permanently have to remind myself that they do _not_ refer to the addresses as seen when accessing memory, but the ones going _into_ the IOMMU.
Ok. Could we agree on 'IOFN' then? I think 'iova' and 'io address' are also reasonably widely used terms to refer to address from a device's PoV. I'd really like to unblock these early patches.
Quoted 18 lines
The confusion (on my part) arises every time I see a mixture of gfn, bfn, and mfn in the same patch, perhaps including some 1:1-ness assumptions between pairs of them.
Take these two hunks as example (mixing in some pfn as well):
(quoted diff) +1
-1@@ -436,7 +436,7 @@ static int iommu_merge_pages(struct domain *d, unsigned long pt_mfn, * {Re, un}mapping super page frames causes re-allocation of io * page tables. */ -static int iommu_pde_from_gfn(struct domain *d, unsigned long pfn, +static int iommu_pde_from_bfn(struct domain *d, unsigned long pfn, unsigned long pt_mfn[]) {u64 *pde, *next_table_vaddr;(quoted diff) +0
-0@@ -477,11 +477,11 @@ static int iommu_pde_from_gfn(struct domain *d, unsigned long pfn,
Review static int iommu_pde_from_gfn(struct domain *d, · line 485
Quoted 10 lines
(quoted diff) +2
-2next_table_mfn != 0 ) { int i; - unsigned long mfn, gfn; + unsigned long mfn, bfn; unsigned int page_sz; page_sz = 1 << (PTE_PER_TABLE_SHIFT * (next_level - 1)); - gfn = pfn & ~((1 << (PTE_PER_TABLE_SHIFT * next_level)) - 1); + bfn = pfn & ~((1 << (PTE_PER_TABLE_SHIFT * next_level)) - 1);
This is not wonderful code, agreed. In this particular case it looks like I may be able to just rename the pfn argument to iofn (assuming we go with that name) and lose the stack variable, if that helps.
Paul
Quoted 5 lines
(quoted diff) +0
-0mfn = next_table_mfn;/* allocate lower level page table */Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 12 lines
Quoted 12 lines
Quoted 1 line · On 06.09.18 at 12:36, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 10:39Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 05.09.18 at 11:13, <[email protected]> wrote:
Personally I think 'bus address' is commonly enough used term for
addresses
Quoted 1 line
used by devices for DMA. Indeed we have already '
dev_bus_addr' in thegrant
Quoted 2 lines
map and unmap hypercalls. So baddr and bfn seem like ok terms to me. It's also not impossible to rename these later if they prove problematic.
Quoted 7 lines
Quoted 3 lines
But that's the point - the names are problematic (to me): I permanently have to remind myself that they do _not_ refer to the addresses as seen when accessing memory, but the ones going _into_ the IOMMU.
Ok. Could we agree on 'IOFN' then? I think 'iova' and 'io address' are also reasonably widely used terms to refer to address from a device's PoV. I'd really like to unblock these early patches.
Hmm, earlier I had indicated I'd prefer DFN (as this make clear whose view we are talking about). Kevin seemed to prefer DFN too, just with a different association for D (which, as said, I consider unhelpful). So is there a particular reason you're now suggesting IOFN nevertheless?
Quoted 25 lines
Quoted 25 lines
The confusion (on my part) arises every time I see a mixture of gfn, bfn, and mfn in the same patch, perhaps including some 1:1-ness assumptions between pairs of them.
Take these two hunks as example (mixing in some pfn as well):
(quoted diff) +1
-1@@ -436,7 +436,7 @@ static int iommu_merge_pages(struct domain *d, unsigned long pt_mfn, * {Re, un}mapping super page frames causes re-allocation of io * page tables. */ -static int iommu_pde_from_gfn(struct domain *d, unsigned long pfn, +static int iommu_pde_from_bfn(struct domain *d, unsigned long pfn, unsigned long pt_mfn[]) {u64 *pde, *next_table_vaddr;(quoted diff) +1
-1@@ -477,11 +477,11 @@ static int iommu_pde_from_gfn(struct domain *d, unsigned long pfn, next_table_mfn != 0 ) { int i; - unsigned long mfn, gfn; + unsigned long mfn, bfn; unsigned int page_sz;
Quoted 7 lines
Review static int iommu_pde_from_gfn(struct domain *d, · line 485
Quoted 3 lines
(quoted diff) +1
-1page_sz = 1 << (PTE_PER_TABLE_SHIFT * (next_level - 1)); - gfn = pfn & ~((1 << (PTE_PER_TABLE_SHIFT * next_level)) - 1); + bfn = pfn & ~((1 << (PTE_PER_TABLE_SHIFT * next_level)) - 1);This is not wonderful code, agreed. In this particular case it looks like I may be able to just rename the pfn argument to iofn (assuming we go with that name) and lose the stack variable, if that helps.
Renaming the parameter will likely help, I agree. Getting rid of the local variable, otoh, I'm not sure is going to work (you need to retain the function parameter's original value for the next iteration of the outer loop).
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 29 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 06 September 2018 14:13
To: Paul Durrant <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Kevin Tian <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the concept of BFN...Quoted 8 lines
Quoted 8 lines
Quoted 1 line · On 06.09.18 at 12:36, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 10:39Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 05.09.18 at 11:13, <[email protected]> wrote:
Personally I think 'bus address' is commonly enough used term for
addresses
Quoted 1 line
used by devices for DMA. Indeed we have already '
dev_bus_addr' inthe
Quoted 2 lines
Quoted 2 lines
grant
Quoted 1 line
map and unmap hypercalls. So baddr and bfn seem like ok terms to me.
It's
Quoted 6 lines
Quoted 5 lines
Quoted 1 line
also not impossible to rename these later if they prove problematic.
But that's the point - the names are problematic (to me): I permanently have to remind myself that they do _not_ refer to the addresses as seen when accessing memory, but the ones going _into_ the IOMMU.
Quoted 8 lines
Quoted 3 lines
Ok. Could we agree on 'IOFN' then? I think 'iova' and 'io address' are also reasonably widely used terms to refer to address from a device's PoV. I'd really like to unblock these early patches.
Hmm, earlier I had indicated I'd prefer DFN (as this make clear whose view we are talking about). Kevin seemed to prefer DFN too, just with a different association for D (which, as said, I consider unhelpful). So is there a particular reason you're now suggesting IOFN nevertheless?
It was the ambiguity and lack of agreement over the 'D' that made me think that the other alternative would be better. Kevin, would you be ok with 'IOFN'?
Quoted 33 lines
Quoted 7 lines
Quoted 7 lines
The confusion (on my part) arises every time I see a mixture of gfn, bfn, and mfn in the same patch, perhaps including some 1:1-ness assumptions between pairs of them.
Take these two hunks as example (mixing in some pfn as well):
Quoted 11 lines
Review static int iommu_pde_from_gfn(struct domain
Quoted 11 lines
(quoted diff) +1
-1@@ -436,7 +436,7 @@ static int iommu_merge_pages(struct domain *d, unsigned long pt_mfn, * {Re, un}mapping super page frames causes re-allocation of io * page tables. */ -static int iommu_pde_from_gfn(struct domain *d, unsigned long pfn, +static int iommu_pde_from_bfn(struct domain *d, unsigned long pfn, unsigned long pt_mfn[]) {u64 *pde, *next_table_vaddr;(quoted diff) +0
-0@@ -477,11 +477,11 @@ static int iommu_pde_from_gfn(struct domain*d,
Quoted 8 lines
Quoted 8 lines
(quoted diff) +1
-1unsigned long pfn, next_table_mfn != 0 ) { int i; - unsigned long mfn, gfn; + unsigned long mfn, bfn; unsigned int page_sz;Quoted 6 lines
Review static int iommu_pde_from_gfn(struct domain · line 485
Quoted 3 lines
(quoted diff) +1
-1page_sz = 1 << (PTE_PER_TABLE_SHIFT * (next_level - 1)); - gfn = pfn & ~((1 << (PTE_PER_TABLE_SHIFT * next_level)) - 1); + bfn = pfn & ~((1 << (PTE_PER_TABLE_SHIFT * next_level)) - 1);This is not wonderful code, agreed. In this particular case it looks like I may be able to just rename the pfn argument to iofn (assuming we go with
Quoted 7 lines
that
Quoted 1 line
name) and lose the stack variable, if that helps.
Renaming the parameter will likely help, I agree. Getting rid of the local variable, otoh, I'm not sure is going to work (you need to retain the function parameter's original value for the next iteration of the outer loop).
Oh, yes I missed that. I'll see what I can do once we have agreement on a name.
Paul
Quoted 1 line
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 41 lines
From: Paul Durrant [mailto:[email protected]]
Sent: Thursday, September 6, 2018 10:54 PMQuoted 9 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 06 September 2018 14:13
To: Paul Durrant <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Kevin Tian <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the conceptof
Quoted 13 lines
BFN...
Quoted 8 lines
Quoted 8 lines
Quoted 1 line · On 06.09.18 at 12:36, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 10:39Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 05.09.18 at 11:13, <[email protected]> wrote:
Personally I think 'bus address' is commonly enough used term for
addresses
Quoted 1 line
used by devices for DMA. Indeed we have already '
dev_bus_addr' inthe
Quoted 2 lines
Quoted 2 lines
grant
Quoted 1 line
map and unmap hypercalls. So baddr and bfn seem like ok terms to
me.
Quoted 4 lines
It's
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
also not impossible to rename these later if they prove problematic.
But that's the point - the names are problematic (to me): I
permanently
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
have to remind myself that they do _not_ refer to the addresses as seen when accessing memory, but the ones going _into_ the IOMMU.
Ok. Could we agree on 'IOFN' then? I think 'iova' and 'io address' are
also
Quoted 1 line
Quoted 1 line
reasonably widely used terms to refer to address from a device's PoV.
I'd
Quoted 2 lines
Quoted 1 line
really like to unblock these early patches.
Quoted 9 lines
Quoted 4 lines
Hmm, earlier I had indicated I'd prefer DFN (as this make clear whose view we are talking about). Kevin seemed to prefer DFN too, just with a different association for D (which, as said, I consider unhelpful). So is there a particular reason you're now suggesting IOFN nevertheless?
It was the ambiguity and lack of agreement over the 'D' that made me think that the other alternative would be better.
Kevin, would you be ok with 'IOFN'?
My problem with DFN is when combining D with address then "device address" is not very clear to me while interpreting D as DMA is also not that clear from Jan's point.
I didn't see a perfect candidate without causing any ambiguity - at this point I feel IOFN/IOVA might not be the best one but it wins to me when considering the fact that it is widely used term in other places (e.g. in vtd spec, in Linux vfio/iommu driver, etc.).
Thanks
Kevin
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 47 lines
Quoted 47 lines
Quoted 1 line · On 07.09.18 at 03:47, <[email protected]> wrote:
From: Paul Durrant [mailto:[email protected]]
Sent: Thursday, September 6, 2018 10:54 PMQuoted 9 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 06 September 2018 14:13
To: Paul Durrant <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Kevin Tian <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the conceptof
Quoted 13 lines
BFN...
Quoted 8 lines
Quoted 8 lines
Quoted 1 line · On 06.09.18 at 12:36, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 10:39Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 05.09.18 at 11:13, <[email protected]> wrote:
Personally I think 'bus address' is commonly enough used term for
addresses
Quoted 1 line
used by devices for DMA. Indeed we have already '
dev_bus_addr' inthe
Quoted 2 lines
Quoted 2 lines
grant
Quoted 1 line
map and unmap hypercalls. So baddr and bfn seem like ok terms to
me.
Quoted 4 lines
It's
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
also not impossible to rename these later if they prove problematic.
But that's the point - the names are problematic (to me): I
permanently
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
have to remind myself that they do _not_ refer to the addresses as seen when accessing memory, but the ones going _into_ the IOMMU.
Ok. Could we agree on 'IOFN' then? I think 'iova' and 'io address' are
also
Quoted 1 line
Quoted 1 line
reasonably widely used terms to refer to address from a device's PoV.
I'd
Quoted 6 lines
Quoted 1 line
really like to unblock these early patches.
Hmm, earlier I had indicated I'd prefer DFN (as this make clear whose view we are talking about). Kevin seemed to prefer DFN too, just with a different association for D (which, as said, I consider unhelpful). So is there a particular reason you're now suggesting IOFN nevertheless?
Quoted 8 lines
Quoted 4 lines
It was the ambiguity and lack of agreement over the 'D' that made me think that the other alternative would be better.
Kevin, would you be ok with 'IOFN'?My problem with DFN is when combining D with address then "device address" is not very clear to me while interpreting D as DMA is also not that clear from Jan's point.
What about making its description mention both possible interpretations?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 69 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 07:24
To: Paul Durrant <[email protected]>; Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Stefano Stabellini <[email protected]>; xen- devel <[email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the concept of BFN...Quoted 9 lines
Quoted 9 lines
Quoted 1 line · On 07.09.18 at 03:47, <[email protected]> wrote:
From: Paul Durrant [mailto:[email protected]]
Sent: Thursday, September 6, 2018 10:54 PMQuoted 5 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 06 September 2018 14:13
To: Paul Durrant <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; JulienGrall
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
<[email protected]>; Kevin Tian <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce theconcept
Quoted 11 lines
Quoted 11 lines
of
Quoted 10 lines
BFN...
Quoted 8 lines
Quoted 8 lines
Quoted 1 line · On 06.09.18 at 12:36, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 10:39Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 05.09.18 at 11:13, <[email protected]> wrote:
Personally I think 'bus address' is commonly enough used term for
addresses
Quoted 1 line
used by devices for DMA. Indeed we have already '
dev_bus_addr'in
Quoted 6 lines
Quoted 6 lines
Quoted 3 lines
the
Quoted 2 lines
Quoted 2 lines
grant
Quoted 1 line
map and unmap hypercalls. So baddr and bfn seem like ok terms to
me.
Quoted 2 lines
It's
Quoted 1 line
Quoted 1 line
Quoted 1 line
also not impossible to rename these later if they prove
problematic.
Quoted 5 lines
Quoted 5 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
But that's the point - the names are problematic (to me): I
permanently
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
have to remind myself that they do _not_ refer to the addresses as seen when accessing memory, but the ones going _into_ the
IOMMU.
Quoted 5 lines
Quoted 5 lines
Quoted 2 lines
Quoted 2 lines
Ok. Could we agree on 'IOFN' then? I think 'iova' and 'io address' are
also
Quoted 1 line
Quoted 1 line
reasonably widely used terms to refer to address from a device's PoV.
I'd
Quoted 8 lines
Quoted 8 lines
Quoted 6 lines
Quoted 1 line
really like to unblock these early patches.
Hmm, earlier I had indicated I'd prefer DFN (as this make clear whose view we are talking about). Kevin seemed to prefer DFN too, just with a different association for D (which, as said, I consider unhelpful). So is there a particular reason you're now suggesting IOFN nevertheless?
It was the ambiguity and lack of agreement over the 'D' that made me
think
Quoted 4 lines
Quoted 3 lines
that the other alternative would be better.
Kevin, would you be ok with 'IOFN'?
Quoted 6 lines
Quoted 3 lines
My problem with DFN is when combining D with address then "device address" is not very clear to me while interpreting D as DMA is also not that clear from Jan's point.
What about making its description mention both possible interpretations?
I'm ok with DFN plus supporting text. Kevin, are you ok with that?
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 80 lines
From: Paul Durrant [mailto:[email protected]]
Sent: Friday, September 7, 2018 4:13 PMQuoted 9 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 07:24
To: Paul Durrant <[email protected]>; Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Stefano Stabellini <[email protected]>; xen- devel <[email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the conceptof
Quoted 11 lines
BFN...
Quoted 9 lines
Quoted 9 lines
Quoted 1 line · On 07.09.18 at 03:47, <[email protected]> wrote:
From: Paul Durrant [mailto:[email protected]]
Sent: Thursday, September 6, 2018 10:54 PMQuoted 5 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 06 September 2018 14:13
To: Paul Durrant <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; JulienQuoted 15 lines
Grall
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
<[email protected]>; Kevin Tian <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce theconcept
Quoted 9 lines
Quoted 9 lines
of
Quoted 8 lines
BFN...
Quoted 6 lines
Quoted 6 lines
Quoted 1 line · On 06.09.18 at 12:36, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 10:39Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 05.09.18 at 11:13, <[email protected]> wrote:
Personally I think 'bus address' is commonly enough used term
for
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
addresses
Quoted 1 line
used by devices for DMA. Indeed we have already
'
dev_bus_addr'Quoted 4 lines
in
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
the
Quoted 2 lines
Quoted 2 lines
grant
Quoted 1 line
map and unmap hypercalls. So baddr and bfn seem like ok
terms to
Quoted 8 lines
Quoted 3 lines
Quoted 3 lines
me.
Quoted 2 lines
It's
Quoted 1 line
Quoted 1 line
Quoted 1 line
also not impossible to rename these later if they prove
problematic.
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
But that's the point - the names are problematic (to me): I
permanently
Quoted 1 line
Quoted 1 line
Quoted 1 line
have to remind myself that they do _not_ refer to the addresses
as
Quoted 4 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
seen when accessing memory, but the ones going _into_ the
IOMMU.
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Ok. Could we agree on 'IOFN' then? I think 'iova' and 'io address'
are
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
also
Quoted 1 line
Quoted 1 line
reasonably widely used terms to refer to address from a device's
PoV.
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
I'd
Quoted 3 lines
Quoted 1 line
really like to unblock these early patches.
Hmm, earlier I had indicated I'd prefer DFN (as this make clear
whose
Quoted 10 lines
Quoted 5 lines
Quoted 5 lines
Quoted 3 lines
view we are talking about). Kevin seemed to prefer DFN too, just with a different association for D (which, as said, I consider unhelpful). So is there a particular reason you're now suggesting IOFN nevertheless?
It was the ambiguity and lack of agreement over the 'D' that made me
think
Quoted 4 lines
Quoted 3 lines
that the other alternative would be better.
Kevin, would you be ok with 'IOFN'?
Quoted 9 lines
Quoted 6 lines
Quoted 3 lines
My problem with DFN is when combining D with address then "device address" is not very clear to me while interpreting D as DMA is also not that clear from Jan's point.
What about making its description mention both possible interpretations?
I'm ok with DFN plus supporting text. Kevin, are you ok with that?
sure
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 01/14] iommu: introduce the concept of BFN...
Quoted 99 lines
-----Original Message-----
From: Tian, Kevin [mailto:[email protected]]
Sent: 07 September 2018 09:17
To: Paul Durrant <[email protected]>; 'Jan Beulich' <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Stefano Stabellini <[email protected]>; xen- devel <[email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the concept of BFN...Quoted 10 lines
From: Paul Durrant [mailto:[email protected]]
Sent: Friday, September 7, 2018 4:13 PMQuoted 7 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 07:24
To: Paul Durrant <[email protected]>; Kevin Tian <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Stefano Stabellini <[email protected]>;xen-
Quoted 16 lines
Quoted 2 lines
devel <[email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce the conceptof
Quoted 13 lines
BFN...
Quoted 9 lines
Quoted 9 lines
Quoted 1 line · On 07.09.18 at 03:47, <[email protected]> wrote:
From: Paul Durrant [mailto:[email protected]]
Sent: Thursday, September 6, 2018 10:54 PMQuoted 5 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 06 September 2018 14:13
To: Paul Durrant <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; JulienGrall
Quoted 1 line
Quoted 1 line
Quoted 1 line
<[email protected]>; Kevin Tian <[email protected]>;
Stefano
Quoted 36 lines
Quoted 13 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: RE: [Xen-devel] [PATCH v6 01/14] iommu: introduce theconcept
Quoted 9 lines
Quoted 9 lines
of
Quoted 8 lines
BFN...
Quoted 6 lines
Quoted 6 lines
Quoted 1 line · On 06.09.18 at 12:36, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 05 September 2018 10:39Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 05.09.18 at 11:13, <[email protected]> wrote:
Personally I think 'bus address' is commonly enough used term
for
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
addresses
Quoted 1 line
used by devices for DMA. Indeed we have already
'
dev_bus_addr'Quoted 4 lines
in
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
the
Quoted 2 lines
Quoted 2 lines
grant
Quoted 1 line
map and unmap hypercalls. So baddr and bfn seem like ok
terms to
Quoted 8 lines
Quoted 3 lines
Quoted 3 lines
me.
Quoted 2 lines
It's
Quoted 1 line
Quoted 1 line
Quoted 1 line
also not impossible to rename these later if they prove
problematic.
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
But that's the point - the names are problematic (to me): I
permanently
Quoted 1 line
Quoted 1 line
Quoted 1 line
have to remind myself that they do _not_ refer to the addresses
as
Quoted 4 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
seen when accessing memory, but the ones going _into_ the
IOMMU.
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Ok. Could we agree on 'IOFN' then? I think 'iova' and 'io address'
are
Quoted 9 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
also
Quoted 1 line
Quoted 1 line
reasonably widely used terms to refer to address from a device's
PoV.
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
I'd
Quoted 3 lines
Quoted 1 line
really like to unblock these early patches.
Hmm, earlier I had indicated I'd prefer DFN (as this make clear
whose
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
view we are talking about). Kevin seemed to prefer DFN too, just
with
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
a different association for D (which, as said, I consider unhelpful). So is there a particular reason you're now suggesting IOFN
nevertheless?
Quoted 11 lines
Quoted 11 lines
Quoted 2 lines
Quoted 2 lines
It was the ambiguity and lack of agreement over the 'D' that made me
think
Quoted 7 lines
Quoted 3 lines
that the other alternative would be better.
Kevin, would you be ok with 'IOFN'?My problem with DFN is when combining D with address then "device address" is not very clear to me while interpreting D as DMA is also not that clear from Jan's point.
Quoted 8 lines
Quoted 1 line
Quoted 1 line
What about making its description mention both possible
interpretations?
Quoted 4 lines
Quoted 1 line
I'm ok with DFN plus supporting text. Kevin, are you ok with that?
sure
Ok. Decision made. I will re-work the patches.
Paul
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 2/14 iommu: make use of type-safe BFN and MFN in exported functions
In reply to Paul DurrantThis patch modifies the declaration of the entry points to the IOMMU sub-system to use bfn_t and mfn_t in place of unsigned long. A subsequent patch will similarly modify the methods in the iommu_ops structure.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: George Dunlap <[email protected]>
v6:
- Re-base.
v3:
- Removed most of the uses of an intermediate 'frame' variable.
v2:
- Addressed comments from Jan.
- Use intermediate 'frame' variable to avoid directly encapsulating mfn or gfn values as bfns.
--- xen/arch/arm/p2m.c | 3 ++- xen/arch/x86/mm.c | 10 ++++---- xen/arch/x86/mm/p2m-ept.c | 10 +++++--- xen/arch/x86/mm/p2m-pt.c | 45 ++++++++++++++++++++--------------- xen/arch/x86/mm/p2m.c | 16 ++++++++----- xen/arch/x86/x86_64/mm.c | 5 ++-- xen/common/grant_table.c | 12 +++++----- xen/common/memory.c | 4 ++-- xen/drivers/passthrough/iommu.c | 25 ++++++++++--------- xen/drivers/passthrough/vtd/x86/vtd.c | 3 ++- xen/include/xen/iommu.h | 14 +++++++---- 11 files changed, 85 insertions(+), 62 deletions(-)
xen/arch/arm/p2m.c +2 -1
@@ -957,7 +957,8 @@ static int __p2m_set_entry(struct p2m_domain *p2m,if (need_iommu (p2m->domain) && (lpae_is_valid (orig_pte) ||lpae_is_valid (*entry)) ) - rc =iommu_iotlb_flush (p2m->domain,gfn_x (sgfn),1UL << page_order); + rc =iommu_iotlb_flush (p2m->domain,_bfn (gfn_x (sgfn)), +1UL << page_order);else rc =0 ;
xen/arch/x86/mm.c +5 -5
@@ -2752,14 +2752,14 @@ static int _get_page_type(struct page_info *page, unsigned long type,struct domain *d =page_get_owner (page);if ( d &&is_pv_domain (d) &&unlikely (need_iommu (d)) ) { -gfn_t gfn =_gfn (mfn_to_gmfn (d,mfn_x (page_to_mfn (page)))); +mfn_t mfn =page_to_mfn (page);if ( (x & PGT_type_mask) == PGT_writable_page ) - iommu_ret =iommu_unmap_page (d,gfn_x (gfn)); + iommu_ret =iommu_unmap_page (d,_bfn (mfn_x (mfn)));else if ( type == PGT_writable_page ) - iommu_ret =iommu_map_page (d,gfn_x (gfn), -mfn_x (page_to_mfn (page)), - IOMMUF_readable|IOMMUF_writable); + iommu_ret =iommu_map_page (d,_bfn (mfn_x (mfn)), mfn, + IOMMUF_readable | + IOMMUF_writable); } }
xen/arch/x86/mm/p2m-ept.c +7 -3
@@ -870,15 +870,19 @@ out: rc =iommu_pte_flush (d, gfn, &ept_entry->epte, order, vtd_pte_present); else { +bfn_t bfn =_bfn (gfn); +if ( iommu_flags )for ( i =0 ; i < (1 << order); i++ ) { - rc =iommu_map_page (d, gfn + i,mfn_x (mfn) + i, iommu_flags); + rc =iommu_map_page (d,bfn_add (bfn, i), +mfn_add (mfn, i), iommu_flags);if (unlikely (rc) ) {while ( i-- )/* If statement to satisfy __must_check. */ -if (iommu_unmap_page (p2m->domain, gfn + i) ) +if (iommu_unmap_page (p2m->domain, +bfn_add (bfn, i)) )continue ;break ;
@@ -887,7 +891,7 @@ out:else for ( i =0 ; i < (1 << order); i++ ) { - ret =iommu_unmap_page (d, gfn + i); + ret =iommu_unmap_page (d,bfn_add (bfn, i));if ( !rc ) rc = ret; }
xen/arch/x86/mm/p2m-pt.c +26 -19
55 lines not on this page · raw message
xen/arch/x86/mm/p2m.c +10 -6
@@ -715,9 +715,11 @@ p2m_remove_page(struct p2m_domain *p2m, unsigned long gfn_l, unsigned long mfn,if (need_iommu (p2m->domain) ) { +bfn_t bfn =_bfn (mfn); +for ( i =0 ; i < (1 << page_order); i++ ) { -int ret =iommu_unmap_page (p2m->domain, mfn + i); +int ret =iommu_unmap_page (p2m->domain,bfn_add (bfn, i));if ( !rc ) rc = ret;
@@ -774,16 +776,17 @@ guest_physmap_add_entry(struct domain *d, gfn_t gfn, mfn_t mfn, {if (need_iommu (d) && t == p2m_ram_rw ) { +bfn_t bfn =_bfn (mfn_x (mfn)); +for ( i =0 ; i < (1 << page_order); i++ ) { - rc =iommu_map_page (d,mfn_x (mfn_add (mfn, i)), -mfn_x (mfn_add (mfn, i)), + rc =iommu_map_page (d,bfn_add (bfn, i),mfn_add (mfn, i), IOMMUF_readable|IOMMUF_writable);if ( rc !=0 ) {while ( i-- >0 )/* If statement to satisfy __must_check. */ -if (iommu_unmap_page (d,mfn_x (mfn_add (mfn, i))) ) +if (iommu_unmap_page (d,bfn_add (bfn, i)) )continue ;return rc;
@@ -1158,7 +1161,8 @@ int set_identity_p2m_entry(struct domain *d, unsigned long gfn_l, {if ( !need_iommu (d) )return 0 ; -return iommu_map_page (d, gfn_l, gfn_l, IOMMUF_readable|IOMMUF_writable); +return iommu_map_page (d,_bfn (gfn_l),_mfn (gfn_l), + IOMMUF_readable | IOMMUF_writable); }gfn_lock (p2m, gfn,0 );
@@ -1248,7 +1252,7 @@ int clear_identity_p2m_entry(struct domain *d, unsigned long gfn_l) {if ( !need_iommu (d) )return 0 ; -return iommu_unmap_page (d, gfn_l); +return iommu_unmap_page (d,_bfn (gfn_l)); }gfn_lock (p2m, gfn,0 );
xen/arch/x86/x86_64/mm.c +3 -2
@@ -1429,13 +1429,14 @@ int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm)if ( iommu_enabled && !iommu_passthrough && !need_iommu (hardware_domain) ) {for ( i = spfn; i < epfn; i++ ) -if (iommu_map_page (hardware_domain, i, i, IOMMUF_readable|IOMMUF_writable) ) +if (iommu_map_page (hardware_domain,_bfn (i),_mfn (i), + IOMMUF_readable | IOMMUF_writable) )break ;if ( i != epfn ) {while (i-- > old_max)/* If statement to satisfy __must_check. */ -if (iommu_unmap_page (hardware_domain, i) ) +if (iommu_unmap_page (hardware_domain,_bfn (i)) )continue ; goto destroy_m2p;
xen/common/grant_table.c +6 -6
@@ -1143,13 +1143,13 @@ map_grant_ref( !(old_pin & (GNTPIN_hstw_mask|GNTPIN_devw_mask)) ) {if ( !(kind &MAPKIND_WRITE ) ) - err =iommu_map_page (ld,mfn_x (mfn),mfn_x (mfn), - IOMMUF_readable|IOMMUF_writable); + err =iommu_map_page (ld,_bfn (mfn_x (mfn)), mfn, + IOMMUF_readable | IOMMUF_writable); } else if ( act_pin && !old_pin ) {if ( !kind ) - err =iommu_map_page (ld,mfn_x (mfn),mfn_x (mfn), + err =iommu_map_page (ld,_bfn (mfn_x (mfn)), mfn, IOMMUF_readable); }if ( err )
@@ -1398,10 +1398,10 @@ unmap_common( kind =mapkind (lgt, rd, op->mfn);if ( !kind ) - err =iommu_unmap_page (ld,mfn_x (op->mfn)); + err =iommu_unmap_page (ld,_bfn (mfn_x (op->mfn)));else if ( !(kind &MAPKIND_WRITE ) ) - err =iommu_map_page (ld,mfn_x (op->mfn), -mfn_x (op->mfn), IOMMUF_readable); + err =iommu_map_page (ld,_bfn (mfn_x (op->mfn)), op->mfn, + IOMMUF_readable);double_gt_unlock (lgt, rgt);
xen/common/memory.c +2 -2
@@ -834,11 +834,11 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,this_cpu (iommu_dont_flush_iotlb) =0 ; - ret =iommu_iotlb_flush (d, xatp->idx - done, done); + ret =iommu_iotlb_flush (d,_bfn (xatp->idx - done), done);if (unlikely (ret) && rc >=0 ) rc = ret; - ret =iommu_iotlb_flush (d, xatp->gpfn - done, done); + ret =iommu_iotlb_flush (d,_bfn (xatp->gpfn - done), done);if (unlikely (ret) && rc >=0 ) rc = ret; }
xen/drivers/passthrough/iommu.c +12 -13
@@ -255,7 +255,7 @@ void iommu_domain_destroy(struct domain *d)arch_iommu_domain_destroy (d); } -int iommu_map_page (struct domain *d,unsigned long bfn,unsigned long mfn, +int iommu_map_page (struct domain *d,bfn_t bfn,mfn_t mfn,unsigned int flags) {const struct domain_iommu *hd =dom_iommu (d);
@@ -264,13 +264,13 @@ int iommu_map_page(struct domain *d, unsigned long bfn, unsigned long mfn,if ( !iommu_enabled || !hd->platform_ops )return 0 ; - rc = hd->platform_ops->map_page (d, bfn, mfn, flags); + rc = hd->platform_ops->map_page (d,bfn_x (bfn),mfn_x (mfn), flags);if (unlikely (rc) ) {if ( !d->is_shutting_down &&printk_ratelimit () )printk (XENLOG_ERR -"d%d: IOMMU mapping bfn %#lx to mfn %#lx failed: %d\n" , - d->domain_id, bfn, mfn, rc); +"d%d: IOMMU mapping bfn %" PRI_bfn" to mfn %" PRI_mfn" failed: %d\n" , + d->domain_id,bfn_x (bfn),mfn_x (mfn), rc);if ( !is_hardware_domain (d) )domain_crash (d);
@@ -279,7 +279,7 @@ int iommu_map_page(struct domain *d, unsigned long bfn, unsigned long mfn,return rc; } -int iommu_unmap_page (struct domain *d,unsigned long bfn) +int iommu_unmap_page (struct domain *d,bfn_t bfn) {const struct domain_iommu *hd =dom_iommu (d);int rc;
@@ -287,13 +287,13 @@ int iommu_unmap_page(struct domain *d, unsigned long bfn)if ( !iommu_enabled || !hd->platform_ops )return 0 ; - rc = hd->platform_ops->unmap_page (d, bfn); + rc = hd->platform_ops->unmap_page (d,bfn_x (bfn));if (unlikely (rc) ) {if ( !d->is_shutting_down &&printk_ratelimit () )printk (XENLOG_ERR -"d%d: IOMMU unmapping bfn %#lx failed: %d\n" , - d->domain_id, bfn, rc); +"d%d: IOMMU unmapping bfn %" PRI_bfn" failed: %d\n" , + d->domain_id,bfn_x (bfn), rc);if ( !is_hardware_domain (d) )domain_crash (d);
@@ -319,8 +319,7 @@ static void iommu_free_pagetables(unsigned long unused) cpumask_cycle(smp_processor_id(), &cpu_online_map)); } -int iommu_iotlb_flush (struct domain *d,unsigned long bfn, -unsigned int page_count) +int iommu_iotlb_flush (struct domain *d,bfn_t bfn,unsigned int page_count) {const struct domain_iommu *hd =dom_iommu (d);int rc;
@@ -328,13 +327,13 @@ int iommu_iotlb_flush(struct domain *d, unsigned long bfn,if ( !iommu_enabled || !hd->platform_ops || !hd->platform_ops->iotlb_flush )return 0 ; - rc = hd->platform_ops->iotlb_flush (d, bfn, page_count); + rc = hd->platform_ops->iotlb_flush (d,bfn_x (bfn), page_count);if (unlikely (rc) ) {if ( !d->is_shutting_down &&printk_ratelimit () )printk (XENLOG_ERR -"d%d: IOMMU IOTLB flush failed: %d, bfn %#lx, page count %u\n" , - d->domain_id, rc, bfn, page_count); +"d%d: IOMMU IOTLB flush failed: %d, bfn %" PRI_bfn", page count %u\n" , + d->domain_id, rc,bfn_x (bfn), page_count);if ( !is_hardware_domain (d) )domain_crash (d);
xen/drivers/passthrough/vtd/x86/vtd.c +2 -1
@@ -152,7 +152,8 @@ void __hwdom_init vtd_set_hwdom_mapping(struct domain *d) page_is_ram_type(pfn, RAM_TYPE_CONVENTIONAL) )continue ; - rc =iommu_map_page (d, pfn, pfn, IOMMUF_readable|IOMMUF_writable); + rc =iommu_map_page (d,_bfn (pfn),_mfn (pfn), + IOMMUF_readable | IOMMUF_writable);if ( rc ) printk(XENLOG_WARNING VTDPREFIX " d%d: IOMMU mapping failed: %d\n", d->domain_id, rc);
xen/include/xen/iommu.h +10 -4
@@ -24,6 +24,7 @@#include <xen/spinlock.h> #include <xen/pci.h> #include <xen/typesafe.h> +#include <xen/mm.h> #include <public/hvm/ioreq.h> #include <public/domctl.h> #include <asm/device.h>
@@ -42,6 +43,11 @@ TYPE_SAFE(uint64_t, bfn);#undef bfn_x#endif +static inline bfn_t bfn_add (bfn_t bfn,unsigned long i) +{ +return _bfn (bfn_x (bfn) + i); +} +#define IOMMU_PAGE_SHIFT 12#define IOMMU_PAGE_SIZE (_AC(1,L) << IOMMU_PAGE_SHIFT)#define IOMMU_PAGE_MASK (~(IOMMU_PAGE_SIZE - 1))
@@ -86,9 +92,9 @@ void iommu_teardown(struct domain *d);#define IOMMUF_readable (1u<<_IOMMUF_readable)#define _IOMMUF_writable 1#define IOMMUF_writable (1u<<_IOMMUF_writable) -int __must_check iommu_map_page (struct domain *d,unsigned long bfn, -unsigned long mfn,unsigned int flags); -int __must_check iommu_unmap_page (struct domain *d,unsigned long bfn); +int __must_check iommu_map_page (struct domain *d,bfn_t bfn, +mfn_t mfn,unsigned int flags); +int __must_check iommu_unmap_page (struct domain *d,bfn_t bfn);enum iommu_feature {
@@ -211,7 +217,7 @@ int iommu_do_pci_domctl(struct xen_domctl *, struct domain *d,int iommu_do_domctl (struct xen_domctl *,struct domain *d,XEN_GUEST_HANDLE_PARAM (xen_domctl_t )); -int __must_check iommu_iotlb_flush (struct domain *d,unsigned long bfn, +int __must_check iommu_iotlb_flush (struct domain *d,bfn_t bfn,unsigned int page_count);int __must_check iommu_iotlb_flush_all (struct domain *d);
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 02/14] iommu: make use of type-safe BFN and MFN in exported functions
Quoted 8 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:46, <[email protected]> wrote:
This patch modifies the declaration of the entry points to the IOMMU sub-system to use
bfn_tandmfn_tin place of unsigned long. A subsequent patch will similarly modify the methods in theiommu_opsstructure.Signed-off-by Paul Durrant <[email protected]>Reviewed-by Wei Liu <[email protected]>Reviewed-by Kevin Tian <[email protected]>
(obviously subject to possible renaming as per discussion on patch 1)
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 3/14 iommu: push use of type-safe BFN and MFN into iommu_ops
In reply to Paul DurrantThis patch modifies the methods in struct iommu_ops to use type-safe BFN and MFN. This follows on from the prior patch that modified the functions exported in xen/iommu.h.
---
Cc: Suravee Suthikulpanit <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
v6:
- Re-base.
v3:
- Remove some use of intermediate 'frame' variables.
v2:
- Addressed comments from Jan.
- Extend use of intermediate 'frame' variable to avoid directly encapsulating gfn values as bfns or vice versa.
--- xen/drivers/passthrough/amd/iommu_map.c | 46 ++++++++++++++++----------- xen/drivers/passthrough/amd/pci_amd_iommu.c | 2 +- xen/drivers/passthrough/arm/smmu.c | 16 +++++----- xen/drivers/passthrough/iommu.c | 9 +++--- xen/drivers/passthrough/vtd/iommu.c | 26 +++++++-------- xen/drivers/passthrough/x86/iommu.c | 2 +- xen/include/asm-x86/hvm/svm/amd-iommu-proto.h | 8 ++--- xen/include/xen/iommu.h | 13 +++++--- 8 files changed, 67 insertions(+), 55 deletions(-)
xen/drivers/passthrough/amd/iommu_map.c +27 -19
@@ -631,7 +631,7 @@ static int update_paging_mode(struct domain *d, unsigned long bfn)return 0 ; } -int amd_iommu_map_page (struct domain *d,unsigned long bfn,unsigned long mfn, +int amd_iommu_map_page (struct domain *d,bfn_t bfn,mfn_t mfn,unsigned int flags) {bool_t need_flush =0 ;
@@ -651,7 +651,8 @@ int amd_iommu_map_page(struct domain *d, unsigned long bfn, unsigned long mfn,if ( rc ) {spin_unlock (&hd->arch.mapping_lock); -AMD_IOMMU_DEBUG ("Root table alloc failed, bfn = %lx\n" , bfn); +AMD_IOMMU_DEBUG ("Root table alloc failed, bfn = %" PRI_bfn"\n" , +bfn_x (bfn));domain_crash (d);return rc; }
@@ -660,25 +661,27 @@ int amd_iommu_map_page(struct domain *d, unsigned long bfn, unsigned long mfn,* we might need a deeper page table for wider bfn now */ if (is_hvm_domain (d) ) { -if (update_paging_mode (d, bfn) ) +if (update_paging_mode (d,bfn_x (bfn)) ) {spin_unlock (&hd->arch.mapping_lock); -AMD_IOMMU_DEBUG ("Update page mode failed bfn = %lx\n" , bfn); +AMD_IOMMU_DEBUG ("Update page mode failed bfn = %" PRI_bfn"\n" , +bfn_x (bfn));domain_crash (d);return -EFAULT ; } } -if (iommu_pde_from_bfn (d, bfn, pt_mfn) || (pt_mfn[1 ] ==0 ) ) +if (iommu_pde_from_bfn (d,bfn_x (bfn), pt_mfn) || (pt_mfn[1 ] ==0 ) ) {spin_unlock (&hd->arch.mapping_lock); -AMD_IOMMU_DEBUG ("Invalid IO pagetable entry bfn = %lx\n" , bfn); +AMD_IOMMU_DEBUG ("Invalid IO pagetable entry bfn = %" PRI_bfn"\n" , +bfn_x (bfn));domain_crash (d);return -EFAULT ; }/* Install 4k mapping first */ - need_flush =set_iommu_pte_present (pt_mfn[1 ], bfn, mfn, + need_flush =set_iommu_pte_present (pt_mfn[1 ],bfn_x (bfn),mfn_x (mfn),IOMMU_PAGING_MODE_LEVEL_1 , !!(flags & IOMMUF_writable), !!(flags & IOMMUF_readable));
@@ -690,7 +693,7 @@ int amd_iommu_map_page(struct domain *d, unsigned long bfn, unsigned long mfn,/* 4K mapping for PV guests never changes, * no need to flush if we trust non-present bits */ if (is_hvm_domain (d) ) -amd_iommu_flush_pages (d, bfn,0 ); +amd_iommu_flush_pages (d,bfn_x (bfn),0 );for ( merge_level =IOMMU_PAGING_MODE_LEVEL_2 ; merge_level <= hd->arch.paging_mode; merge_level++ )
@@ -698,15 +701,16 @@ int amd_iommu_map_page(struct domain *d, unsigned long bfn, unsigned long mfn,if ( pt_mfn[merge_level] ==0 )break ;if ( !iommu_update_pde_count (d, pt_mfn[merge_level], - bfn, mfn, merge_level) ) +bfn_x (bfn),mfn_x (mfn), merge_level) )break ; -if (iommu_merge_pages (d, pt_mfn[merge_level], bfn, +if (iommu_merge_pages (d, pt_mfn[merge_level],bfn_x (bfn), flags, merge_level) ) {spin_unlock (&hd->arch.mapping_lock);AMD_IOMMU_DEBUG ("Merge iommu page failed at level %d, " -"bfn = %lx mfn = %lx\n" , merge_level, bfn, mfn); +"bfn = %" PRI_bfn" mfn = %" PRI_mfn"\n" , + merge_level,bfn_x (bfn),mfn_x (mfn));domain_crash (d);return -EFAULT ; }
@@ -720,7 +724,7 @@ out:return 0 ; } -int amd_iommu_unmap_page (struct domain *d,unsigned long bfn) +int amd_iommu_unmap_page (struct domain *d,bfn_t bfn) {unsigned long pt_mfn[7 ];struct domain_iommu *hd =dom_iommu (d);
@@ -742,31 +746,33 @@ int amd_iommu_unmap_page(struct domain *d, unsigned long bfn)* we might need a deeper page table for lager bfn now */ if (is_hvm_domain (d) ) { -int rc =update_paging_mode (d, bfn); +int rc =update_paging_mode (d,bfn_x (bfn));if ( rc ) {spin_unlock (&hd->arch.mapping_lock); -AMD_IOMMU_DEBUG ("Update page mode failed bfn = %lx\n" , bfn); +AMD_IOMMU_DEBUG ("Update page mode failed bfn = %" PRI_bfn"\n" , +bfn_x (bfn));if ( rc != -EADDRNOTAVAIL )domain_crash (d);return rc; } } -if (iommu_pde_from_bfn (d, bfn, pt_mfn) || (pt_mfn[1 ] ==0 ) ) +if (iommu_pde_from_bfn (d,bfn_x (bfn), pt_mfn) || (pt_mfn[1 ] ==0 ) ) {spin_unlock (&hd->arch.mapping_lock); -AMD_IOMMU_DEBUG ("Invalid IO pagetable entry bfn = %lx\n" , bfn); +AMD_IOMMU_DEBUG ("Invalid IO pagetable entry bfn = %" PRI_bfn"\n" , +bfn_x (bfn));domain_crash (d);return -EFAULT ; }/* mark PTE as 'page not present' */ -clear_iommu_pte_present (pt_mfn[1 ], bfn); +clear_iommu_pte_present (pt_mfn[1 ],bfn_x (bfn));spin_unlock (&hd->arch.mapping_lock); -amd_iommu_flush_pages (d, bfn,0 ); +amd_iommu_flush_pages (d,bfn_x (bfn),0 );return 0 ; }
@@ -787,7 +793,9 @@ int amd_iommu_reserve_domain_unity_map(struct domain *domain, gfn = phys_addr >>PAGE_SHIFT ;for ( i =0 ; i < npages; i++ ) { - rt =amd_iommu_map_page (domain, gfn +i, gfn +i, flags); +unsigned long frame = gfn + i; + + rt =amd_iommu_map_page (domain,_bfn (frame),_mfn (frame), flags);if ( rt !=0 )return rt; }
xen/drivers/passthrough/amd/pci_amd_iommu.c +1 -1
@@ -271,7 +271,7 @@ static void __hwdom_init amd_iommu_hwdom_init(struct domain *d)*/ if (mfn_valid (_mfn (pfn)) ) { -int ret =amd_iommu_map_page (d, pfn, pfn, +int ret =amd_iommu_map_page (d,_bfn (pfn),_mfn (pfn), IOMMUF_readable|IOMMUF_writable);if ( !rc )
xen/drivers/passthrough/arm/smmu.c +8 -8
@@ -2550,8 +2550,7 @@ static int __must_check arm_smmu_iotlb_flush_all(struct domain *d)return 0 ; } -static int __must_check arm_smmu_iotlb_flush (struct domain *d, -unsigned long bfn, +static int __must_check arm_smmu_iotlb_flush (struct domain *d,bfn_t bfn,unsigned int page_count) {/* ARM SMMU v1 doesn't have flush by VMA and VMID */
@@ -2737,8 +2736,8 @@ static void arm_smmu_iommu_domain_teardown(struct domain *d)xfree (xen_domain); } -static int __must_check arm_smmu_map_page (struct domain *d,unsigned long bfn, -unsigned long mfn,unsigned int flags) +static int __must_check arm_smmu_map_page (struct domain *d,bfn_t bfn, +mfn_t mfn,unsigned int flags) {p2m_type_t t;
@@ -2751,7 +2750,7 @@ static int __must_check arm_smmu_map_page(struct domain *d, unsigned long bfn,* function should only be used by gnttab code with gfn == mfn == bfn. */ BUG_ON (!is_domain_direct_mapped (d)); -BUG_ON (mfn != bfn); +BUG_ON (mfn_x (mfn) !=bfn_x (bfn));/* We only support readable and writable flags */ if (!(flags & (IOMMUF_readable | IOMMUF_writable)))
@@ -2763,10 +2762,11 @@ static int __must_check arm_smmu_map_page(struct domain *d, unsigned long bfn,* The function guest_physmap_add_entry replaces the current mapping * if there is already one... */ -return guest_physmap_add_entry (d,_gfn (bfn),_mfn (bfn),0 , t); +return guest_physmap_add_entry (d,_gfn (bfn_x (bfn)),_mfn (bfn_x (bfn)), +0 , t); } -static int __must_check arm_smmu_unmap_page (struct domain *d,unsigned long bfn) +static int __must_check arm_smmu_unmap_page (struct domain *d,bfn_t bfn) {/* * This function should only be used by gnttab code when the domain
@@ -2775,7 +2775,7 @@ static int __must_check arm_smmu_unmap_page(struct domain *d, unsigned long bfn)if ( !is_domain_direct_mapped (d) )return -EINVAL ; -return guest_physmap_remove_page (d,_gfn (bfn),_mfn (bfn),0 ); +return guest_physmap_remove_page (d,_gfn (bfn_x (bfn)),_mfn (bfn_x (bfn)),0 ); }static const struct iommu_ops arm_smmu_iommu_ops = {
xen/drivers/passthrough/iommu.c +5 -4
@@ -194,7 +194,8 @@ void __hwdom_init iommu_hwdom_init(struct domain *d) == PGT_writable_page) ) mapping |= IOMMUF_writable; - ret = hd->platform_ops->map_page (d, bfn, mfn, mapping); + ret = hd->platform_ops->map_page (d,_bfn (bfn),_mfn (mfn), + mapping);if ( !rc ) rc = ret;
@@ -264,7 +265,7 @@ int iommu_map_page(struct domain *d, bfn_t bfn, mfn_t mfn,if ( !iommu_enabled || !hd->platform_ops )return 0 ; - rc = hd->platform_ops->map_page (d,bfn_x (bfn),mfn_x (mfn), flags); + rc = hd->platform_ops->map_page (d, bfn, mfn, flags);if (unlikely (rc) ) {if ( !d->is_shutting_down &&printk_ratelimit () )
@@ -287,7 +288,7 @@ int iommu_unmap_page(struct domain *d, bfn_t bfn)if ( !iommu_enabled || !hd->platform_ops )return 0 ; - rc = hd->platform_ops->unmap_page (d,bfn_x (bfn)); + rc = hd->platform_ops->unmap_page (d, bfn);if (unlikely (rc) ) {if ( !d->is_shutting_down &&printk_ratelimit () )
@@ -327,7 +328,7 @@ int iommu_iotlb_flush(struct domain *d, bfn_t bfn, unsigned int page_count)if ( !iommu_enabled || !hd->platform_ops || !hd->platform_ops->iotlb_flush )return 0 ; - rc = hd->platform_ops->iotlb_flush (d,bfn_x (bfn), page_count); + rc = hd->platform_ops->iotlb_flush (d, bfn, page_count);if (unlikely (rc) ) {if ( !d->is_shutting_down &&printk_ratelimit () )
xen/drivers/passthrough/vtd/iommu.c +12 -14
@@ -584,8 +584,7 @@ static int __must_check iommu_flush_all(void)return rc; } -static int __must_check iommu_flush_iotlb (struct domain *d, -unsigned long bfn, +static int __must_check iommu_flush_iotlb (struct domain *d,bfn_t bfn,bool_t dma_old_pte_present,unsigned int page_count) {
@@ -612,12 +611,12 @@ static int __must_check iommu_flush_iotlb(struct domain *d,if ( iommu_domid ==-1 )continue ; -if ( page_count !=1 || bfn ==bfn_x (INVALID_BFN ) ) +if ( page_count !=1 ||bfn_eq (bfn,INVALID_BFN ) ) rc =iommu_flush_iotlb_dsi (iommu, iommu_domid,0 , flush_dev_iotlb);else rc =iommu_flush_iotlb_psi (iommu, iommu_domid, -__bfn_to_baddr (bfn), +bfn_to_baddr (bfn),PAGE_ORDER_4K , !dma_old_pte_present, flush_dev_iotlb);
@@ -633,7 +632,7 @@ static int __must_check iommu_flush_iotlb(struct domain *d, }static int __must_check iommu_flush_iotlb_pages (struct domain *d, -unsigned long bfn, +bfn_t bfn,unsigned int page_count) {return iommu_flush_iotlb (d, bfn,1 , page_count);
@@ -641,7 +640,7 @@ static int __must_check iommu_flush_iotlb_pages(struct domain *d,static int __must_check iommu_flush_iotlb_all (struct domain *d) { -return iommu_flush_iotlb (d,bfn_x (INVALID_BFN ),0 ,0 ); +return iommu_flush_iotlb (d,INVALID_BFN ,0 ,0 ); }/* clear one page's page table */
@@ -676,7 +675,7 @@ static int __must_check dma_pte_clear_one(struct domain *domain, u64 addr)iommu_flush_cache_entry (pte,sizeof (struct dma_pte ));if ( !this_cpu (iommu_dont_flush_iotlb) ) - rc =iommu_flush_iotlb_pages (domain, addr >>PAGE_SHIFT_4K ,1 ); + rc =iommu_flush_iotlb_pages (domain,baddr_to_bfn (addr),1 );unmap_vtd_domain_page (page);
@@ -1767,8 +1766,7 @@ static void iommu_domain_teardown(struct domain *d) }static int __must_check intel_iommu_map_page (struct domain *d, -unsigned long bfn, -unsigned long mfn, +bfn_t bfn,mfn_t mfn,unsigned int flags) {struct domain_iommu *hd =dom_iommu (d);
@@ -1786,16 +1784,16 @@ static int __must_check intel_iommu_map_page(struct domain *d,spin_lock (&hd->arch.mapping_lock); - pg_maddr =addr_to_dma_page_maddr (d,__bfn_to_baddr (bfn),1 ); + pg_maddr =addr_to_dma_page_maddr (d,bfn_to_baddr (bfn),1 );if ( pg_maddr ==0 ) {spin_unlock (&hd->arch.mapping_lock);return -ENOMEM ; } page = (struct dma_pte *)map_vtd_domain_page (pg_maddr); - pte = page + (bfn &LEVEL_MASK ); + pte = page + (bfn_x (bfn) &LEVEL_MASK ); old = *pte; -dma_set_pte_addr (new, (paddr_t )mfn <<PAGE_SHIFT_4K ); +dma_set_pte_addr (new,mfn_to_maddr (mfn));dma_set_pte_prot (new, ((flags & IOMMUF_readable) ?DMA_PTE_READ :0 ) | ((flags & IOMMUF_writable) ?DMA_PTE_WRITE :0 ));
@@ -1823,13 +1821,13 @@ static int __must_check intel_iommu_map_page(struct domain *d, }static int __must_check intel_iommu_unmap_page (struct domain *d, -unsigned long bfn) +bfn_t bfn) {/* Do nothing if hardware domain and iommu supports pass thru. */ if ( iommu_passthrough &&is_hardware_domain (d) )return 0 ; -return dma_pte_clear_one (d,__bfn_to_baddr (bfn)); +return dma_pte_clear_one (d,bfn_to_baddr (bfn)); } int iommu_pte_flush(struct domain *d, uint64_t bfn, uint64_t *pte,
xen/drivers/passthrough/x86/iommu.c +1 -1
@@ -65,7 +65,7 @@ int arch_iommu_populate_page_table(struct domain *d) {ASSERT (!(gfn >>DEFAULT_DOMAIN_ADDRESS_WIDTH ));BUG_ON (SHARED_M2P (gfn)); - rc = hd->platform_ops->map_page (d, gfn, mfn, + rc = hd->platform_ops->map_page (d,_bfn (gfn),_mfn (mfn), IOMMUF_readable | IOMMUF_writable); }
xen/include/asm-x86/hvm/svm/amd-iommu-proto.h +4 -4
@@ -52,9 +52,9 @@ int amd_iommu_init(void);int amd_iommu_update_ivrs_mapping_acpi (void );/* mapping functions */ -int __must_check amd_iommu_map_page (struct domain *d,unsigned long gfn, -unsigned long mfn,unsigned int flags); -int __must_check amd_iommu_unmap_page (struct domain *d,unsigned long gfn); +int __must_check amd_iommu_map_page (struct domain *d,bfn_t bfn, +mfn_t mfn,unsigned int flags); +int __must_check amd_iommu_unmap_page (struct domain *d,bfn_t bfn);u64 amd_iommu_get_next_table_from_pte (u32 *entry);int __must_check amd_iommu_alloc_root (struct domain_iommu *hd); int amd_iommu_reserve_domain_unity_map(struct domain *domain,
@@ -77,7 +77,7 @@ void iommu_dte_set_guest_cr3(u32 *dte, u16 dom_id, u64 gcr3,/* send cmd to iommu */ void amd_iommu_flush_all_pages (struct domain *d); -void amd_iommu_flush_pages (struct domain *d,unsigned long gfn, +void amd_iommu_flush_pages (struct domain *d,unsigned long bfn,unsigned int order);void amd_iommu_flush_iotlb (u8 devfn,const struct pci_dev *pdev,uint64_t gaddr,unsigned int order);
xen/include/xen/iommu.h +9 -4
@@ -48,6 +48,11 @@ static inline bfn_t bfn_add(bfn_t bfn, unsigned long i)return _bfn (bfn_x (bfn) + i); } +static inline bool_t bfn_eq (bfn_t x,bfn_t y) +{ +return bfn_x (x) ==bfn_x (y); +} +#define IOMMU_PAGE_SHIFT 12#define IOMMU_PAGE_SIZE (_AC(1,L) << IOMMU_PAGE_SHIFT)#define IOMMU_PAGE_MASK (~(IOMMU_PAGE_SIZE - 1))
@@ -182,9 +187,9 @@ struct iommu_ops {#endif /* HAS_PCI */ void (*teardown)(struct domain *d); -int __must_check (*map_page)(struct domain *d,unsigned long bfn, -unsigned long mfn,unsigned int flags); -int __must_check (*unmap_page)(struct domain *d,unsigned long bfn); +int __must_check (*map_page)(struct domain *d,bfn_t bfn,mfn_t mfn, +unsigned int flags); +int __must_check (*unmap_page)(struct domain *d,bfn_t bfn);void (*free_page_table)(struct page_info *);#ifdef CONFIG_X86 void (*update_ire_from_apic)(unsigned int apic,unsigned int reg,unsigned int value);
@@ -195,7 +200,7 @@ struct iommu_ops {void (*resume)(void );void (*share_p2m)(struct domain *d);void (*crash_shutdown)(void ); -int __must_check (*iotlb_flush)(struct domain *d,unsigned long bfn, +int __must_check (*iotlb_flush)(struct domain *d,bfn_t bfn,unsigned int page_count);int __must_check (*iotlb_flush_all)(struct domain *d);int (*get_reserved_device_memory)(iommu_grdm_t *,void *);
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 03/14] iommu: push use of type-safe BFN and MFN into iommu_ops
Quoted 8 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch modifies the methods in struct
iommu_opsto use type-safe BFN and MFN. This follows on from the prior patch that modified the functions exported inxen/iommu.h.Signed-off-by Paul Durrant <[email protected]>Reviewed-by Wei Liu <[email protected]>Reviewed-by Kevin Tian <[email protected]>
(same note as for patch 2)
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 4/14 iommu: don't domain_crash() inside iommu_map/unmap_page()
In reply to Paul DurrantThis patch removes the implicit domain_crash() from iommu_map(), unmap_page() and iommu_iotlb_flush() and turns them into straightforward wrappers that check the existence of the relevant iommu_op and call through to it. This makes them usable by PV IOMMU code to be delivered in future patches.
This patch adds a helper macro, domu_crash(), that will only invoke domain_crash() if the domain is not the hardware domain and modifies callers of iommu_map(), unmap_page() and iommu_iotlb_flush() to use this should an operation fail.
NOTE: This patch includes one bit of clean-up in set_identity_p2m_entry()
replacing use of p2m->domain with the domain pointer passed into the
function.---
Cc: Wei Liu <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>
Cc: George Dunlap <[email protected]>
v6:
- Introduce domu_crash() (idea suggested by Kevin, name suggested by Jan) to crash non-hardware domains.
- Dropped Wei's and George's R-b because of change.
v2:
- New in v2.
--- xen/arch/arm/p2m.c | 4 ++++ xen/arch/x86/mm.c | 3 +++ xen/arch/x86/mm/p2m-ept.c | 3 +++ xen/arch/x86/mm/p2m-pt.c | 3 +++ xen/arch/x86/mm/p2m.c | 22 ++++++++++++++++++---- xen/common/grant_table.c | 4 ++++ xen/common/memory.c | 3 +++ xen/drivers/passthrough/iommu.c | 12 ------------ xen/drivers/passthrough/x86/iommu.c | 4 ++++ xen/include/xen/sched.h | 5 +++++ 10 files changed, 47 insertions(+), 16 deletions(-)
xen/arch/arm/p2m.c +4 -0
@@ -957,8 +957,12 @@ static int __p2m_set_entry(struct p2m_domain *p2m,if (need_iommu (p2m->domain) && (lpae_is_valid (orig_pte) ||lpae_is_valid (*entry)) ) + { rc =iommu_iotlb_flush (p2m->domain,_bfn (gfn_x (sgfn)),1UL << page_order); +if (unlikely (rc) ) +domu_crash (p2m->domain); + }else rc =0 ;
xen/arch/x86/mm.c +3 -0
@@ -2760,6 +2760,9 @@ static int _get_page_type(struct page_info *page, unsigned long type, iommu_ret =iommu_map_page (d,_bfn (mfn_x (mfn)), mfn, IOMMUF_readable | IOMMUF_writable); + +if (unlikely (iommu_ret) ) +domu_crash (d); } }
xen/arch/x86/mm/p2m-ept.c +3 -0
@@ -895,6 +895,9 @@ out:if ( !rc ) rc = ret; } + +if (unlikely (rc) ) +domu_crash (d); } }
xen/arch/x86/mm/p2m-pt.c +3 -0
@@ -717,6 +717,9 @@ p2m_pt_set_entry(struct p2m_domain *p2m, gfn_t gfn_, mfn_t mfn, rc = ret; } } + +if (unlikely (rc) ) +domu_crash (p2m->domain); }/*
xen/arch/x86/mm/p2m.c +18 -4
@@ -724,6 +724,9 @@ p2m_remove_page(struct p2m_domain *p2m, unsigned long gfn_l, unsigned long mfn,if ( !rc ) rc = ret; } + +if (unlikely (rc) ) +domu_crash (p2m->domain); }return rc;
@@ -789,6 +792,7 @@ guest_physmap_add_entry(struct domain *d, gfn_t gfn, mfn_t mfn,if (iommu_unmap_page (d,bfn_add (bfn, i)) )continue ; +domu_crash (d);return rc; } }
@@ -1157,12 +1161,17 @@ int set_identity_p2m_entry(struct domain *d, unsigned long gfn_l,struct p2m_domain *p2m =p2m_get_hostp2m (d);int ret; -if ( !paging_mode_translate (p2m->domain) ) +if ( !paging_mode_translate (d) ) {if ( !need_iommu (d) )return 0 ; -return iommu_map_page (d,_bfn (gfn_l),_mfn (gfn_l), - IOMMUF_readable | IOMMUF_writable); + + ret =iommu_map_page (d,_bfn (gfn_l),_mfn (gfn_l), + IOMMUF_readable | IOMMUF_writable); +if (unlikely (ret) ) +domu_crash (d); + +return ret; }gfn_lock (p2m, gfn,0 );
@@ -1252,7 +1261,12 @@ int clear_identity_p2m_entry(struct domain *d, unsigned long gfn_l) {if ( !need_iommu (d) )return 0 ; -return iommu_unmap_page (d,_bfn (gfn_l)); + + ret =iommu_unmap_page (d,_bfn (gfn_l)); +if (unlikely (ret) ) +domu_crash (d); + +return ret; }gfn_lock (p2m, gfn,0 );
xen/common/grant_table.c +4 -0
@@ -1154,6 +1154,7 @@ map_grant_ref( }if ( err ) { +domu_crash (ld);double_gt_unlock (lgt, rgt); rc = GNTST_general_error; goto undo_out;
@@ -1406,7 +1407,10 @@ unmap_common(double_gt_unlock (lgt, rgt);if ( err ) + { +domu_crash (ld); rc = GNTST_general_error; + } }/* If just unmapped a writable mapping, mark as dirtied */
xen/common/memory.c +3 -0
@@ -841,6 +841,9 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, ret =iommu_iotlb_flush (d,_bfn (xatp->gpfn - done), done);if (unlikely (ret) && rc >=0 ) rc = ret; + +if (unlikely (rc <0 ) ) + domu_crash(d); }#endif
xen/drivers/passthrough/iommu.c +0 -12
@@ -272,9 +272,6 @@ int iommu_map_page(struct domain *d, bfn_t bfn, mfn_t mfn,printk (XENLOG_ERR "d%d: IOMMU mapping bfn %" PRI_bfn" to mfn %" PRI_mfn" failed: %d\n" , d->domain_id,bfn_x (bfn),mfn_x (mfn), rc); - -if ( !is_hardware_domain (d) ) -domain_crash (d); }return rc;
@@ -295,9 +292,6 @@ int iommu_unmap_page(struct domain *d, bfn_t bfn)printk (XENLOG_ERR "d%d: IOMMU unmapping bfn %" PRI_bfn" failed: %d\n" , d->domain_id,bfn_x (bfn), rc); - -if ( !is_hardware_domain (d) ) -domain_crash (d); }return rc;
@@ -335,9 +329,6 @@ int iommu_iotlb_flush(struct domain *d, bfn_t bfn, unsigned int page_count)printk (XENLOG_ERR "d%d: IOMMU IOTLB flush failed: %d, bfn %" PRI_bfn", page count %u\n" , d->domain_id, rc,bfn_x (bfn), page_count); - -if ( !is_hardware_domain (d) ) -domain_crash (d); }return rc;
@@ -358,9 +349,6 @@ int iommu_iotlb_flush_all(struct domain *d)printk (XENLOG_ERR "d%d: IOMMU IOTLB flush all failed: %d\n" , d->domain_id, rc); - -if ( !is_hardware_domain (d) ) -domain_crash (d); }return rc;
xen/drivers/passthrough/x86/iommu.c +4 -0
@@ -104,7 +104,11 @@ int arch_iommu_populate_page_table(struct domain *d)this_cpu (iommu_dont_flush_iotlb) =0 ;if ( !rc ) + { rc =iommu_iotlb_flush_all (d); +if (unlikely (rc) ) +domain_crash (d); + }if ( rc && rc != -ERESTART )iommu_teardown (d);
xen/include/xen/sched.h +5 -0
@@ -616,6 +616,11 @@ void __domain_crash(struct domain *d);__domain_crash (d); \ } while (0) +#define domu_crash (d) do { \ + if ( !is_hardware_domain(d) ) \ + domain_crash(d); \ +} while (false) +/* * Called from assembly code, with an optional address to help indicate why * the crash occured. If addr is 0, look up address from last extable
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 04/14] iommu: don't domain_crash() inside iommu_map/unmap_page()
xen/common/grant_table.c map_grant_ref( · line 1158
Quoted 9 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/grant_table.c +1
-0@@ -1154,6 +1154,7 @@ map_grant_ref( }if ( err ) { +domu_crash (ld);double_gt_unlock (lgt, rgt);
You crash the domain with both locks held here, but ...
xen/common/grant_table.c unmap_common( · line 1413
Quoted 8 lines
xen/common/grant_table.c +3
-0@@ -1406,7 +1407,10 @@ unmap_common(double_gt_unlock (lgt, rgt);if ( err ) + { +domu_crash (ld); rc = GNTST_general_error; + }
... outside of the locked region here. I think the latter is fine, and hence the former should be changed.
With that
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 04/14] iommu: don't domain_crash() inside iommu_map/unmap_page()
Quoted 29 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 11:38
To: Paul Durrant <[email protected]>
Cc: Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; Wei Liu <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Jun Nakajima <[email protected]>; Kevin Tian <[email protected]>; Stefano Stabellini <[email protected]>; xen- devel <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>
Subject: Re: [PATCH v6 04/14] iommu: don'tdomain_crash()inside iommu_map/unmap_page()xen/common/grant_table.c map_grant_ref( · line 1158
Quoted 9 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/grant_table.c +1
-0@@ -1154,6 +1154,7 @@ map_grant_ref( }if ( err ) { +domu_crash (ld);double_gt_unlock (lgt, rgt);You crash the domain with both locks held here, but ...
Quoted 3 lines
xen/common/grant_table.c +0
-0@@ -1406,7 +1407,10 @@ unmap_common(double_gt_unlock (lgt, rgt);
Quoted 8 lines
xen/common/grant_table.c unmap_common( · line 1413
Quoted 5 lines
xen/common/grant_table.c +3
-0if ( err ) + { +domu_crash (ld); rc = GNTST_general_error; + }... outside of the locked region here. I think the latter is fine, and hence the former should be changed.
Sure. Will do.
Quoted 4 lines
With that
Reviewed-by Jan Beulich <[email protected]>
Thanks,
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 5/14 public / x86: introduce __HYPERCALL_iommu_op
In reply to Paul DurrantThis patch introduces the boilerplate for a new hypercall to allow a domain to control IOMMU mappings for its own pages.
Whilst there is duplication of code between the native and compat entry points which appears ripe for some form of combination, I think it is better to maintain the separation as-is because the compat entry point will necessarily gain complexity in subsequent patches.
NOTE: This hypercall is only implemented for x86 and is currently
restricted by XSM to dom0. Its scope can be expanded in future if need be.
---
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
v6:
- Move iommu_op.c to xen/common as suggested by Kevin, but only build for x86 (as suggested by Jan).
v3:
- Push op code into XSM check.
v2:
- Get rid of the can_control_iommu() function, leaving this patch as pure boilerplate.
- Re-structure the hypercall to use a buffer array, similar to that used by __HYPERCALL_dm_op, to allow for future expansion of op structure without affecting binary compatibility.
- Drop use of __ in public header guard.
--- tools/flask/policy/modules/xen.if | 1 + xen/arch/x86/hvm/hypercall.c | 1 + xen/arch/x86/hypercall.c | 1 + xen/arch/x86/pv/hypercall.c | 1 + xen/common/Makefile | 1 + xen/common/iommu_op.c | 184 ++++++++++++++++++++++++++++++++++++ xen/include/Makefile | 2 + xen/include/public/iommu_op.h | 64 +++++++++++++ xen/include/public/xen.h | 1 + xen/include/xen/hypercall.h | 12 +++ xen/include/xlat.lst | 2 + xen/include/xsm/dummy.h | 6 ++ xen/include/xsm/xsm.h | 6 ++ xen/xsm/dummy.c | 1 + xen/xsm/flask/hooks.c | 6 ++ xen/xsm/flask/policy/access_vectors | 2 + 16 files changed, 291 insertions(+) create mode 100644 xen/common/iommu_op.c create mode 100644 xen/include/public/iommu_op.h
tools/flask/policy/modules/xen.if +1 -0
@@ -60,6 +60,7 @@ define(`create_domain_common', ` allow $1 $2:grant setup; allow $1 $2:hvm { getparam hvmctl sethvmc setparam nested altp2mhvm altp2mhvm_op dm }; + allow $1 $2:resource control_iommu; ') # create_domain(priv, target)
xen/arch/x86/hvm/hypercall.c +1 -0
@@ -137,6 +137,7 @@ static const hypercall_table_t hvm_hypercall_table[] = {COMPAT_CALL (mmuext_op),HYPERCALL (xenpmu_op),COMPAT_CALL (dm_op), +COMPAT_CALL (iommu_op), HYPERCALL(arch_1) };
xen/arch/x86/hypercall.c +1 -0
@@ -68,6 +68,7 @@ const hypercall_args_t hypercall_args_table[NR_hypercalls] =ARGS (xenpmu_op,2 ),ARGS (dm_op,3 ),ARGS (mca,1 ), +ARGS (iommu_op,2 ), ARGS(arch_1, 1), };
xen/arch/x86/pv/hypercall.c +1 -0
@@ -80,6 +80,7 @@ const hypercall_table_t pv_hypercall_table[] = {HYPERCALL (xenpmu_op),COMPAT_CALL (dm_op),HYPERCALL (mca), +COMPAT_CALL (iommu_op), HYPERCALL(arch_1), };
xen/common/Makefile +1 -0
@@ -13,6 +13,7 @@ obj-$(CONFIG_CRASH_DEBUG) += gdbstub.o obj-y += grant_table.o obj-y += guestcopy.o obj-bin-y += gunzip.init.o +obj-$(CONFIG_X86) += iommu_op.o obj-y += irq.o obj-y += kernel.o obj-y += keyhandler.o
xen/common/iommu_op.c new file +184 -0
184 lines not on this page · raw message
xen/include/Makefile +2 -0
@@ -11,6 +11,7 @@ headers-y := \ compat/features.h \ compat/grant_table.h \ compat/kexec.h \ + compat/iommu_op.h \ compat/memory.h \ compat/nmi.h \ compat/physdev.h \
@@ -29,6 +30,7 @@ headers-$(CONFIG_X86) += compat/arch-x86/xen-$(compat-arch-y).h headers-$(CONFIG_X86) += compat/hvm/dm_op.h headers-$(CONFIG_X86) += compat/hvm/hvm_op.h headers-$(CONFIG_X86) += compat/hvm/hvm_vcpu.h +headers-$(CONFIG_X86) += compat/iommu_op.h headers-y += compat/arch-$(compat-arch-y).h compat/pmu.h compat/xlat.h headers-$(CONFIG_FLASK) += compat/xsm/flask_op.h
xen/include/public/iommu_op.h new file +64 -0
64 lines not on this page · raw message
xen/include/public/xen.h +1 -0
@@ -121,6 +121,7 @@ DEFINE_XEN_GUEST_HANDLE(xen_ulong_t);#define __HYPERVISOR_xc_reserved_op 39/* reserved for XenClient */ #define __HYPERVISOR_xenpmu_op 40#define __HYPERVISOR_dm_op 41 +#define __HYPERVISOR_iommu_op 42/* Architecture-specific hypercall definitions. */ #define __HYPERVISOR_arch_0 48
xen/include/xen/hypercall.h +12 -0
@@ -16,6 +16,7 @@#include <public/version.h> #include <public/pmu.h> #include <public/hvm/dm_op.h> +#include <public/iommu_op.h> #include <asm/hypercall.h> #include <xsm/xsm.h>
@@ -148,6 +149,10 @@ do_dm_op( unsigned int nr_bufs, XEN_GUEST_HANDLE_PARAM(xen_dm_op_buf_t) bufs); +extern long +do_iommu_op (unsigned int nr_bufs, +XEN_GUEST_HANDLE_PARAM (xen_iommu_op_buf_t) bufs); +#ifdef CONFIG_COMPAT extern int
@@ -205,6 +210,13 @@ compat_dm_op( unsigned int nr_bufs, XEN_GUEST_HANDLE_PARAM(void) bufs); +#include <compat/iommu_op.h> + +DEFINE_XEN_GUEST_HANDLE (compat_iommu_op_buf_t); +extern int +compat_iommu_op (unsigned int nr_bufs, +XEN_GUEST_HANDLE_PARAM (compat_iommu_op_buf_t) bufs); +#endif void arch_get_xen_caps (xen_capabilities_info_t *info);
xen/include/xlat.lst +2 -0
@@ -77,6 +77,8 @@ ? vcpu_hvm_context hvm/hvm_vcpu.h ? vcpu_hvm_x86_32 hvm/hvm_vcpu.h ? vcpu_hvm_x86_64 hvm/hvm_vcpu.h +! iommu_op iommu_op.h +! iommu_op_buf iommu_op.h ? kexec_exec kexec.h ! kexec_image kexec.h ! kexec_range kexec.h
xen/include/xsm/dummy.h +6 -0
@@ -715,6 +715,12 @@ static XSM_INLINE int xsm_dm_op(XSM_DEFAULT_ARG struct domain *d)return xsm_default_action (action, current->domain, d); } +static XSM_INLINE int xsm_iommu_op(XSM_DEFAULT_ARG struct domain *d, unsigned int op) +{ +XSM_ASSERT_ACTION (XSM_PRIV ); +return xsm_default_action (action, current->domain, d); +} +#endif /* CONFIG_X86 */ #include <public/version.h>
xen/include/xsm/xsm.h +6 -0
@@ -178,6 +178,7 @@ struct xsm_operations {int (*ioport_mapping) (struct domain *d,uint32_t s,uint32_t e,uint8_t allow);int (*pmu_op) (struct domain *d,unsigned int op);int (*dm_op) (struct domain *d); +int (*iommu_op) (struct domain *d,unsigned int op);#endif int (*xen_version) (uint32_t cmd);int (*domain_resource_map) (struct domain *d);
@@ -686,6 +687,11 @@ static inline int xsm_dm_op(xsm_default_t def, struct domain *d) return xsm_ops->dm_op(d); } +static inline int xsm_iommu_op (xsm_default_t def,struct domain *d,unsigned int op) +{ +return xsm_ops->iommu_op (d, op); +} +#endif /* CONFIG_X86 */ static inline int xsm_xen_version (xsm_default_t def,uint32_t op)
xen/xsm/dummy.c +1 -0
@@ -155,6 +155,7 @@ void __init xsm_fixup_ops (struct xsm_operations *ops)set_to_dummy_if_null (ops, ioport_mapping);set_to_dummy_if_null (ops, pmu_op);set_to_dummy_if_null (ops, dm_op); +set_to_dummy_if_null (ops, iommu_op);#endif set_to_dummy_if_null (ops, xen_version);set_to_dummy_if_null (ops, domain_resource_map);
xen/xsm/flask/hooks.c +6 -0
@@ -1669,6 +1669,11 @@ static int flask_dm_op(struct domain *d) return current_has_perm(d, SECCLASS_HVM, HVM__DM); } +static int flask_iommu_op (struct domain *d) +{ +return current_has_perm (d,SECCLASS_RESOURCE ,RESOURCE__CONTROL_IOMMU ); +} +#endif /* CONFIG_X86 */ static int flask_xen_version (uint32_t op)
@@ -1847,6 +1852,7 @@ static struct xsm_operations flask_ops = { .ioport_mapping = flask_ioport_mapping, .pmu_op = flask_pmu_op, .dm_op = flask_dm_op, + .iommu_op = flask_iommu_op, #endif .xen_version = flask_xen_version, .domain_resource_map = flask_domain_resource_map,
xen/xsm/flask/policy/access_vectors +2 -0
@@ -471,6 +471,8 @@ class resource # checked for PHYSDEVOP_setup_gsi (target IRQ) # checked for PHYSDEVOP_pci_mmcfg_reserved (target xen_t) setup +# checked for IOMMU_OP + control_iommu } # Class security describes the FLASK security server itself; these operations
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 05/14] public / x86: introduce __HYPERCALL_iommu_op
xen/common/iommu_op.c line 2
Quoted 6 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/iommu_op.c new file +2
-0@@ -0,0 +1,184 @@ +/****************************************************************************** +* x86/iommu_op.c
Oops?
xen/common/iommu_op.c line 44
Quoted 9 lines
xen/common/iommu_op.c new file +9
-0+int do_one_iommu_op (xen_iommu_op_buf_t *buf) +{ +xen_iommu_op_t op; +int rc; + +if ( buf->size <sizeof (op) ) +return -EFAULT ; + +if (copy_from_guest ((void *)&op, buf->h,sizeof (op)) )
This cast could be avoided if you made ...
xen/common/iommu_op.c line 56
Quoted 12 lines
xen/common/iommu_op.c new file +12
-0+return -EFAULT ; + +if ( op.pad ) +return -EINVAL ; + + rc =xsm_iommu_op (XSM_PRIV , current->domain, op.op); +if ( rc ) +return rc; + +iommu_op (&op); + + if ( __copy_field_to_guest(guest_handle_cast(buf->h, xen_iommu_op_t),
... this cast the initializer of a local variable of suitable handle type (same on the compat path then).
Quoted 18 lines
xen/common/iommu_op.c new file +18
-0+int compat_one_iommu_op (compat_iommu_op_buf_t *buf) +{ +compat_iommu_op_t cmp; +xen_iommu_op_t nat; +int rc; + +if ( buf->size <sizeof (cmp) ) +return -EFAULT ; + +if (copy_from_compat ((void *)&cmp, buf->h,sizeof (cmp)) ) +return -EFAULT ; + +if ( cmp.pad ) +return -EINVAL ; + + rc =xsm_iommu_op (XSM_PRIV , current->domain, cmp.op); +if ( rc ) +return rc;
xen/common/iommu_op.c line 130
Quoted 10 lines
xen/common/iommu_op.c new file +10
-0+ +XLAT_iommu_op (&nat, &cmp); + +iommu_op (&nat); + +XLAT_iommu_op (&cmp, &nat); + +if (__copy_field_to_compat (compat_handle_cast (buf->h, + compat_iommu_op_t), + &cmp, status) )
Since you're only after the status field, perhaps better to avoid the full-blown reverse XLAT_iommu_op() and copy just that one field?
xen/include/Makefile headers-y := \ · line 17
Quoted 10 lines
xen/include/Makefile +1
-0@@ -11,6 +11,7 @@ headers-y := \ compat/features.h \ compat/grant_table.h \ compat/kexec.h \ + compat/iommu_op.h \ compat/memory.h \ compat/nmi.h \ compat/physdev.h \
I guess this is just an off-by-one wrt sorting?
xen/include/Makefile headers-$(CONFIG_X86) += compat/arch-x86/xen-$(compat-arch-y).h · line 33
Quoted 5 lines
xen/include/Makefile +1
-0@@ -29,6 +30,7 @@ headers-$(CONFIG_X86) += compat/arch-x86/xen-$(compat-arch-y).h headers-$(CONFIG_X86) += compat/hvm/dm_op.h headers-$(CONFIG_X86) += compat/hvm/hvm_op.h headers-$(CONFIG_X86) += compat/hvm/hvm_vcpu.h +headers-$(CONFIG_X86) += compat/iommu_op.h
Did you forget to remove this when adding the entry above?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 05/14] public / x86: introduce __HYPERCALL_iommu_op
Quoted 20 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 12:50
To: Paul Durrant <[email protected]>
Cc: Andrew Cooper <[email protected]>; Wei Liu <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>
Subject: Re: [PATCH v6 05/14] public / x86: introduce__HYPERCALL_iommu_opxen/common/iommu_op.c
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/iommu_op.c new file +0
-0@@ -0,0 +1,184 @@+/*********************************************************
*********************
Quoted 4 lines
xen/common/iommu_op.c line 2
Quoted 1 line
xen/common/iommu_op.c new file +1
-0+* x86/iommu_op.c Oops?
Yep. Missed that in the move to common.
Quoted 26 lines
xen/common/iommu_op.c line 44
Quoted 9 lines
xen/common/iommu_op.c new file +9
-0+int do_one_iommu_op (xen_iommu_op_buf_t *buf) +{ +xen_iommu_op_t op; +int rc; + +if ( buf->size <sizeof (op) ) +return -EFAULT ; + +if (copy_from_guest ((void *)&op, buf->h,sizeof (op)) )This cast could be avoided if you made ...
xen/common/iommu_op.c
Quoted 12 lines
xen/common/iommu_op.c new file +12
-0+return -EFAULT ; + +if ( op.pad ) +return -EINVAL ; + + rc =xsm_iommu_op (XSM_PRIV , current->domain, op.op); +if ( rc ) +return rc; + +iommu_op (&op); + + if ( __copy_field_to_guest(guest_handle_cast(buf->h,
xen_iommu_op_t),
Quoted 2 lines
... this cast the initializer of a local variable of suitable handle type (same on the compat path then).
Ok. I'll look at that.
Quoted 24 lines
Quoted 24 lines
xen/common/iommu_op.c new file +23
-0+int compat_one_iommu_op (compat_iommu_op_buf_t *buf) +{ +compat_iommu_op_t cmp; +xen_iommu_op_t nat; +int rc; + +if ( buf->size <sizeof (cmp) ) +return -EFAULT ; + +if (copy_from_compat ((void *)&cmp, buf->h,sizeof (cmp)) ) +return -EFAULT ; + +if ( cmp.pad ) +return -EINVAL ; + + rc =xsm_iommu_op (XSM_PRIV , current->domain, cmp.op); +if ( rc ) +return rc; + +XLAT_iommu_op (&nat, &cmp); + +iommu_op (&nat); +
Quoted 9 lines
xen/common/iommu_op.c line 130
Quoted 5 lines
xen/common/iommu_op.c new file +5
-0+XLAT_iommu_op (&cmp, &nat); + +if (__copy_field_to_compat (compat_handle_cast (buf->h, + compat_iommu_op_t), + &cmp, status) )Since you're only after the status field, perhaps better to avoid the full-blown reverse
XLAT_iommu_op()and copy just that one field?
I kind of like the fact that the two calls mirror each other so I'd prefer to keep it.
Quoted 4 lines
Quoted 4 lines
xen/include/Makefile +0
-0@@ -11,6 +11,7 @@ headers-y := \ compat/features.h \
Quoted 9 lines
xen/include/Makefile headers-y := \ · line 17
Quoted 6 lines
xen/include/Makefile +1
-0compat/grant_table.h \ compat/kexec.h \ + compat/iommu_op.h \ compat/memory.h \ compat/nmi.h \ compat/physdev.h \I guess this is just an off-by-one wrt sorting?
Yep. I'll move.
Quoted 9 lines
xen/include/Makefile headers-$(CONFIG_X86) += compat/arch-x86/xen-
Quoted 1 line
xen/include/Makefile +0
-0@@ -29,6 +30,7 @@ headers-$(CONFIG_X86) += compat/arch-x86/xen-$(compat-arch-y).h
xen/include/Makefile headers-$(CONFIG_X86) += compat/arch-x86/xen-$(compat-arch-y).h · line 33
Quoted 4 lines
xen/include/Makefile +1
-0headers-$(CONFIG_X86) += compat/hvm/dm_op.h headers-$(CONFIG_X86) += compat/hvm/hvm_op.h headers-$(CONFIG_X86) += compat/hvm/hvm_vcpu.h +headers-$(CONFIG_X86) += compat/iommu_op.hDid you forget to remove this when adding the entry above?
Yes, it should have gone.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 05/14] public / x86: introduce __HYPERCALL_iommu_op
Quoted 38 lines
Quoted 28 lines
Quoted 1 line · On 04.09.18 at 14:23, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 12:50Quoted 24 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/iommu_op.c new file +23
-0+int compat_one_iommu_op (compat_iommu_op_buf_t *buf) +{ +compat_iommu_op_t cmp; +xen_iommu_op_t nat; +int rc; + +if ( buf->size <sizeof (cmp) ) +return -EFAULT ; + +if (copy_from_compat ((void *)&cmp, buf->h,sizeof (cmp)) ) +return -EFAULT ; + +if ( cmp.pad ) +return -EINVAL ; + + rc =xsm_iommu_op (XSM_PRIV , current->domain, cmp.op); +if ( rc ) +return rc; + +XLAT_iommu_op (&nat, &cmp); + +iommu_op (&nat); +Quoted 9 lines
xen/common/iommu_op.c line 130
Quoted 5 lines
xen/common/iommu_op.c new file +5
-0+XLAT_iommu_op (&cmp, &nat); + +if (__copy_field_to_compat (compat_handle_cast (buf->h, + compat_iommu_op_t), + &cmp, status) )Since you're only after the status field, perhaps better to avoid the full-blown reverse
XLAT_iommu_op()and copy just that one field?
Quoted 2 lines
I kind of like the fact that the two calls mirror each other so I'd prefer to keep it.
Would you mind looking at the generated code (once you have a few sub-ops in place)? If the compiler manages to remove most of the cruft, I'd be fine keeping it as is. If, however, a whole lot of extra code gets generated, I'd really like to ask to use the shorter form.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 05/14] public / x86: introduce __HYPERCALL_iommu_op
Quoted 50 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 13:55
To: Paul Durrant <[email protected]>
Cc: Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Wei Liu <[email protected]>; Stefano Stabellini <[email protected]>; xen- devel <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>
Subject: RE: [PATCH v6 05/14] public / x86: introduce__HYPERCALL_iommu_opQuoted 38 lines
Quoted 37 lines
Quoted 1 line · On 04.09.18 at 14:23, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 04 September 2018 12:50Quoted 19 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/iommu_op.c new file +18
-0+int compat_one_iommu_op (compat_iommu_op_buf_t *buf) +{ +compat_iommu_op_t cmp; +xen_iommu_op_t nat; +int rc; + +if ( buf->size <sizeof (cmp) ) +return -EFAULT ; + +if (copy_from_compat ((void *)&cmp, buf->h,sizeof (cmp)) ) +return -EFAULT ; + +if ( cmp.pad ) +return -EINVAL ; + + rc =xsm_iommu_op (XSM_PRIV , current->domain, cmp.op); +if ( rc ) +return rc;xen/common/iommu_op.c line 130
Quoted 10 lines
xen/common/iommu_op.c new file +10
-0+ +XLAT_iommu_op (&nat, &cmp); + +iommu_op (&nat); + +XLAT_iommu_op (&cmp, &nat); + +if (__copy_field_to_compat (compat_handle_cast (buf->h, + compat_iommu_op_t), + &cmp, status) )Since you're only after the status field, perhaps better to avoid the full-blown reverse
XLAT_iommu_op()and copy just that one field?
Quoted 8 lines
Quoted 2 lines
I kind of like the fact that the two calls mirror each other so I'd prefer to keep it.
Would you mind looking at the generated code (once you have a few sub-ops in place)? If the compiler manages to remove most of the cruft, I'd be fine keeping it as is. If, however, a whole lot of extra code gets generated, I'd really like to ask to use the shorter form.
I checked... it's not wonderfully compact (because of the nested switches I guess), so I'll cherry-pick the status and add a comment.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 05/14] public / x86: introduce __HYPERCALL_iommu_op
In reply to Paul Durrantxen/common/Makefile obj-$(CONFIG_CRASH_DEBUG) += gdbstub.o · line 16
Quoted 8 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/Makefile +1
-0@@ -13,6 +13,7 @@ obj-$(CONFIG_CRASH_DEBUG) += gdbstub.o obj-y += grant_table.o obj-y += guestcopy.o obj-bin-y += gunzip.init.o +obj-$(CONFIG_X86) += iommu_op.o
Btw - irrespective of this I think you'd better ...
xen/common/Makefile obj-$(CONFIG_CRASH_DEBUG) += gdbstub.o · line 17
Quoted 1 line
xen/common/Makefile +1
-0+int compat_one_iommu_op(compat_iommu_op_buf_t *buf)
... have #ifdef CONFIG_COMPAT above here.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 6/14 iommu: track reserved ranges using a rangeset
In reply to Paul DurrantRanges that should be considered reserved in the IOMMU are not necessarily limited to RMRRs. If iommu_inclusive_mapping is set then any frame number falling within an E820 reserved region should also be considered as reserved in the IOMMU.
This patch adds a rangeset to the domain_iommu structure that is then used to track all reserved ranges. This will be needed by a subsequent patch to test whether it is safe to modify a particular IOMMU entry.
---
Cc: Jan Beulich <[email protected]>
Cc: Kevin Tian <[email protected]>
v6:
- Re-base.
v2:
- New in v2.
--- xen/drivers/passthrough/iommu.c | 10 +++++++++- xen/drivers/passthrough/vtd/iommu.c | 20 +++++++++++++------- xen/drivers/passthrough/vtd/x86/vtd.c | 15 ++++++++++++++- xen/include/xen/iommu.h | 6 ++++++ 4 files changed, 42 insertions(+), 9 deletions(-)
xen/drivers/passthrough/iommu.c +9 -1
@@ -147,6 +147,10 @@ int iommu_domain_init(struct domain *d)if ( !iommu_enabled )return 0 ; + hd->reserved_ranges =rangeset_new (d,NULL ,0 ); +if ( !hd->reserved_ranges ) +return -ENOMEM ; + hd->platform_ops =iommu_get_ops ();return hd->platform_ops->init (d); }
@@ -248,12 +252,16 @@ int iommu_construct(struct domain *d)void iommu_domain_destroy (struct domain *d) { -if ( !iommu_enabled || !dom_iommu (d)->platform_ops ) +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops )return ;iommu_teardown (d);arch_iommu_domain_destroy (d); + +rangeset_destroy (hd->reserved_ranges); } int iommu_map_page(struct domain *d, bfn_t bfn, mfn_t mfn,
xen/drivers/passthrough/vtd/iommu.c +13 -7
@@ -1910,6 +1910,7 @@ static int rmrr_identity_mapping(struct domain *d, bool_t map,unsigned long end_pfn =PAGE_ALIGN_4K (rmrr->end_address) >>PAGE_SHIFT_4K ;struct mapped_rmrr *mrmrr;struct domain_iommu *hd =dom_iommu (d); +int err =0 ;ASSERT (pcidevs_locked ());ASSERT (rmrr->base_address < rmrr->end_address);
@@ -1923,8 +1924,6 @@ static int rmrr_identity_mapping(struct domain *d, bool_t map,if ( mrmrr->base == rmrr->base_address && mrmrr->end == rmrr->end_address ) { -int ret =0 ; -if ( map ) { ++mrmrr->count;
@@ -1934,28 +1933,35 @@ static int rmrr_identity_mapping(struct domain *d, bool_t map,if ( --mrmrr->count )return 0 ; -while ( base_pfn < end_pfn ) + err =rangeset_remove_range (hd->reserved_ranges, + base_pfn, end_pfn); +while ( !err && base_pfn < end_pfn ) {if (clear_identity_p2m_entry (d, base_pfn) ) - ret = -ENXIO ; + err = -ENXIO ; + base_pfn++; }list_del (&mrmrr->list);xfree (mrmrr); -return ret; +return err; } }if ( !map )return -ENOENT ; + err =rangeset_add_range (hd->reserved_ranges, base_pfn, end_pfn); +if ( err ) +return err; +while ( base_pfn < end_pfn ) { -int err =set_identity_p2m_entry (d, base_pfn, p2m_access_rw, flag); - + err =set_identity_p2m_entry (d, base_pfn, p2m_access_rw, flag);if ( err )return err; + base_pfn++; }
xen/drivers/passthrough/vtd/x86/vtd.c +14 -1
@@ -154,8 +154,21 @@ void __hwdom_init vtd_set_hwdom_mapping(struct domain *d) rc =iommu_map_page (d,_bfn (pfn),_mfn (pfn), IOMMUF_readable | IOMMUF_writable); + +/* +* The only reason a reserved page would be mapped is that +* iommu_inclusive_mapping is set, in which case it needs to be +* marked as reserved in the IOMMU. +*/ +if ( !rc &&page_is_ram_type (pfn,RAM_TYPE_RESERVED ) ) + { +ASSERT (iommu_inclusive_mapping); + + rc =rangeset_add_singleton (dom_iommu (d)->reserved_ranges, pfn); + } +if ( rc ) - printk(XENLOG_WARNING VTDPREFIX " d%d: IOMMU mapping failed: %d\n", + printk(XENLOG_WARNING VTDPREFIX " d%d: IOMMU mapping/reservation failed: %d\n", d->domain_id, rc);if (!(i &0xfffff ))
xen/include/xen/iommu.h +6 -0
@@ -122,6 +122,12 @@ struct domain_iommu {/* Features supported by the IOMMU */ DECLARE_BITMAP (features, IOMMU_FEAT_count); + +/* +* BFN ranges that are reserved in the domain IOMMU mappings and +* must not be modified after initialization. +*/ +struct rangeset *reserved_ranges; };#define dom_iommu (d) (&(d)->iommu)
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 06/14] iommu: track reserved ranges using a rangeset
Quoted 6 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/drivers/passthrough/iommu.c +0
-1@@ -248,12 +252,16 @@ int iommu_construct(struct domain *d)void iommu_domain_destroy (struct domain *d) { -if ( !iommu_enabled || !dom_iommu (d)->platform_ops )
xen/drivers/passthrough/iommu.c int iommu_construct(struct domain *d) · line 264
Quoted 10 lines
xen/drivers/passthrough/iommu.c +5
-0+const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops )return ;iommu_teardown (d);arch_iommu_domain_destroy (d); + +rangeset_destroy (hd->reserved_ranges);
For idempotency reasons perhaps better to store NULL after the call?
Quoted 8 lines
xen/drivers/passthrough/vtd/x86/vtd.c +2
-0@@ -154,8 +154,21 @@ void __hwdom_init vtd_set_hwdom_mapping(struct domain *d) rc =iommu_map_page (d,_bfn (pfn),_mfn (pfn), IOMMUF_readable | IOMMUF_writable); + +/*
xen/drivers/passthrough/vtd/x86/vtd.c void __hwdom_init vtd_set_hwdom_mapping(struct domain *d) · line 168
Quoted 10 lines
xen/drivers/passthrough/vtd/x86/vtd.c +10
-0+* The only reason a reserved page would be mapped is that +* iommu_inclusive_mapping is set, in which case it needs to be +* marked as reserved in the IOMMU. +*/ +if ( !rc &&page_is_ram_type (pfn,RAM_TYPE_RESERVED ) ) + { +ASSERT (iommu_inclusive_mapping); + + rc =rangeset_add_singleton (dom_iommu (d)->reserved_ranges, pfn); + }
Why would this be restricted to the E820 reserved type? I think this should cover everything that gets mapped with iommu_inclusive_mapping set, but not mapped with the flag clear.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 06/14] iommu: track reserved ranges using a rangeset
Quoted 19 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 11:40
To: Paul Durrant <[email protected]>
Cc: Kevin Tian <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [PATCH v6 06/14] iommu: track reserved ranges using a rangesetQuoted 11 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/drivers/passthrough/iommu.c +3
-1@@ -248,12 +252,16 @@ int iommu_construct(struct domain *d)void iommu_domain_destroy (struct domain *d) { -if ( !iommu_enabled || !dom_iommu (d)->platform_ops ) +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops )return ;
Quoted 8 lines
xen/drivers/passthrough/iommu.c int iommu_construct(struct domain *d) · line 264
Quoted 5 lines
xen/drivers/passthrough/iommu.c +2
-0iommu_teardown (d);arch_iommu_domain_destroy (d); + +rangeset_destroy (hd->reserved_ranges);For idempotency reasons perhaps better to store NULL after the call?
Ok.
Quoted 20 lines
xen/drivers/passthrough/vtd/x86/vtd.c void __hwdom_init
Quoted 4 lines
xen/drivers/passthrough/vtd/x86/vtd.c +0
-0@@ -154,8 +154,21 @@ void __hwdom_init
vtd_set_hwdom_mapping(struct domain *d)Quoted 4 lines
xen/drivers/passthrough/vtd/x86/vtd.c +1
-0rc =iommu_map_page (d,_bfn (pfn),_mfn (pfn), IOMMUF_readable | IOMMUF_writable); +xen/drivers/passthrough/vtd/x86/vtd.c void __hwdom_init vtd_set_hwdom_mapping(struct domain *d)
Quoted 10 lines
xen/drivers/passthrough/vtd/x86/vtd.c +10
-0+/* +* The only reason a reserved page would be mapped is that +* iommu_inclusive_mapping is set, in which case it needs to be +* marked as reserved in the IOMMU. +*/ +if ( !rc &&page_is_ram_type (pfn,RAM_TYPE_RESERVED ) ) + { +ASSERT (iommu_inclusive_mapping); + + rc = rangeset_add_singleton(dom_iommu(d)->reserved_ranges,pfn);
Quoted 6 lines
xen/drivers/passthrough/vtd/x86/vtd.c void __hwdom_init vtd_set_hwdom_mapping(struct domain *d) · line 168
Quoted 1 line
xen/drivers/passthrough/vtd/x86/vtd.c +1
-0+ }Why would this be restricted to the E820 reserved type? I think this should cover everything that gets mapped with
iommu_inclusive_mappingset, but not mapped with the flag clear.
When PV-IOMMU is enabled it clearly needs to be modify the domain's memory mappings (otherwise what's the point of the API) so I could change this range to include everything mapped for the domain except ranges explicitly identified as RAM.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 7/14 x86: add iommu_op to query reserved ranges
In reply to Paul DurrantThis patch adds an iommu_op to allow the domain IOMMU reserved ranges to be queried by the guest.
NOTE: The number of reserved ranges is determined by system firmware, in
conjunction with Xen command line options, and is expected to be small. Thus, to avoid over-complicating the code, there is no pre-emption check within the operation.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
v4:
- Make xen_bfn_t strictly 64 bits wide and drop associated compat translation.
v3:
- Avoid speculation beyond array bounds check.
v2:
- Re-implemented for v2 based on new rangeset.
--- xen/common/iommu_op.c | 164 ++++++++++++++++++++++++++++++++++++++++-- xen/include/public/iommu_op.h | 39 ++++++++++ xen/include/xlat.lst | 2 + 3 files changed, 199 insertions(+), 6 deletions(-)
xen/common/iommu_op.c +158 -6
@@ -22,11 +22,70 @@#include <xen/event.h> #include <xen/guest_access.h> #include <xen/hypercall.h> +#include <xen/nospec.h> + +struct get_reserved_ctxt { +unsigned int max_entries; +unsigned int nr_entries; +XEN_GUEST_HANDLE (xen_iommu_reserved_range_t) ranges; +}; + +static int get_reserved (unsigned long s,unsigned long e,void *arg) +{ +struct get_reserved_ctxt *ctxt = arg; + +if ( ctxt->nr_entries < ctxt->max_entries ) + { +xen_iommu_reserved_range_t range = { + .start_bfn = s, + .nr_frames = e - s, + }; + +if (copy_to_guest_offset (ctxt->ranges, ctxt->nr_entries, &range, +1 ) ) +return -EFAULT ; + } + + ctxt->nr_entries++; +return 0 ; +} + +static int iommu_op_query_reserved (struct xen_iommu_op_query_reserved *op) +{ +struct domain *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); +struct get_reserved_ctxt ctxt = { + .max_entries = op->nr_entries, + .ranges = op->ranges, + }; +int rc; + +if ( op->pad ) +return -EINVAL ; + + rc =rangeset_report_ranges (iommu->reserved_ranges,0 , ~0ul , + get_reserved, &ctxt); +if ( rc ) +return rc; + +/* Pass back the actual number of reserved ranges */ + op->nr_entries = ctxt.nr_entries; + +if ( !guest_handle_is_null (ctxt.ranges) && + ctxt.nr_entries > ctxt.max_entries ) +return -ENOBUFS ; + +return 0 ; +}static void iommu_op (xen_iommu_op_t *op) {switch ( op->op ) { +case XEN_IOMMUOP_query_reserved: + op->status =iommu_op_query_reserved (&op->u.query_reserved); +break ; +default : op->status = -EOPNOTSUPP ;break ;
@@ -35,13 +94,20 @@ static void iommu_op(xen_iommu_op_t *op)int do_one_iommu_op (xen_iommu_op_buf_t *buf) { -xen_iommu_op_t op; +xen_iommu_op_t op = {}; +size_t offset; +static const size_t op_size[] = { + [XEN_IOMMUOP_query_reserved] =sizeof (struct xen_iommu_op_query_reserved ), + }; +size_t size;int rc; -if ( buf->size <sizeof (op) ) + offset = offsetof(struct xen_iommu_op , u); + +if ( buf->size < offset )return -EFAULT ; -if (copy_from_guest ((void *)&op, buf->h,sizeof (op)) ) +if (copy_from_guest ((void *)&op, buf->h, offset) )return -EFAULT ;if ( op.pad )
@@ -51,6 +117,16 @@ int do_one_iommu_op(xen_iommu_op_buf_t *buf)if ( rc )return rc; +if ( op.op >=ARRAY_SIZE (op_size) ) +return -EOPNOTSUPP ; + + size = op_size[array_index_nospec (op.op,ARRAY_SIZE (op_size))]; +if ( buf->size < offset + size ) +return -EFAULT ; + +if (copy_from_guest_offset ((void *)&op.u, buf->h, offset, size) ) +return -EFAULT ; +iommu_op (&op); if ( __copy_field_to_guest(guest_handle_cast(buf->h, xen_iommu_op_t),
@@ -100,16 +176,27 @@ long do_iommu_op(unsigned int nr_bufs,return rc; } +CHECK_iommu_reserved_range; +int compat_one_iommu_op (compat_iommu_op_buf_t *buf) { -compat_iommu_op_t cmp; +compat_iommu_op_t cmp = {}; +size_t offset; +static const size_t op_size[] = { + [XEN_IOMMUOP_query_reserved] =sizeof (struct compat_iommu_op_query_reserved ), + }; +size_t size;xen_iommu_op_t nat; +unsigned int u; +int32_t status;int rc; -if ( buf->size <sizeof (cmp) ) + offset = offsetof(struct compat_iommu_op , u); + +if ( buf->size < offset )return -EFAULT ; -if (copy_from_compat ((void *)&cmp, buf->h,sizeof (cmp)) ) +if (copy_from_compat ((void *)&cmp, buf->h, offset) )return -EFAULT ;if ( cmp.pad )
@@ -119,17 +206,82 @@ int compat_one_iommu_op(compat_iommu_op_buf_t *buf)if ( rc )return rc; +if ( cmp.op >=ARRAY_SIZE (op_size) ) +return -EOPNOTSUPP ; + + size = op_size[array_index_nospec (cmp.op,ARRAY_SIZE (op_size))]; +if ( buf->size < offset + size ) +return -EFAULT ; + +if (copy_from_compat_offset ((void *)&cmp.u, buf->h, offset, size) ) +return -EFAULT ; + +/* +* The xlat magic doesn't quite know how to handle the union so +* we need to fix things up here. +*/ +#define XLAT_iommu_op_u_query_reserved XEN_IOMMUOP_query_reserved + u = cmp.op; + +#define XLAT_iommu_op_query_reserved_HNDL_ranges (_d_, _s_) \ + do \ + { \ + if ( !compat_handle_is_null((_s_)->ranges) ) \ + { \ + unsigned int *nr_entries = COMPAT_ARG_XLAT_VIRT_BASE; \ + xen_iommu_reserved_range_t *ranges = \ + (void *)(nr_entries + 1); \ + \ + if ( sizeof(*nr_entries) + \ + (sizeof(*ranges) * (_s_)->nr_entries) > \ + COMPAT_ARG_XLAT_SIZE ) \ + return -E2BIG; \ + \ + *nr_entries = (_s_)->nr_entries; \ + set_xen_guest_handle((_d_)->ranges, ranges); \ + } \ + else \ + set_xen_guest_handle((_d_)->ranges, NULL); \ + } while (false) +XLAT_iommu_op (&nat, &cmp); +#undef XLAT_iommu_op_query_reserved_HNDL_ranges +iommu_op (&nat); + status = nat.status; + +#define XLAT_iommu_op_query_reserved_HNDL_ranges (_d_, _s_) \ + do \ + { \ + if ( !compat_handle_is_null((_d_)->ranges) ) \ + { \ + unsigned int *nr_entries = COMPAT_ARG_XLAT_VIRT_BASE; \ + compat_iommu_reserved_range_t *ranges = \ + (void *)(nr_entries + 1); \ + unsigned int nr = \ + min_t(unsigned int, (_d_)->nr_entries, *nr_entries); \ + \ + if ( __copy_to_compat_offset((_d_)->ranges, 0, ranges, nr) ) \ + status = -EFAULT; \ + } \ + } while (false) +XLAT_iommu_op (&cmp, &nat); +/* status will have been modified if __copy_to_compat_offset() failed */ + cmp.status = status; + +#undef XLAT_iommu_op_query_reserved_HNDL_ranges +if (__copy_field_to_compat (compat_handle_cast (buf->h, compat_iommu_op_t), &cmp, status) )return -EFAULT ; +#undef XLAT_iommu_op_u_query_reserved +return 0 ; }
xen/include/public/iommu_op.h +39 -0
@@ -25,11 +25,50 @@#include "xen.h" +typedef uint64_t xen_bfn_t ; + +/* Structure describing a single range reserved in the IOMMU */ +struct xen_iommu_reserved_range { +xen_bfn_t start_bfn; +unsigned int nr_frames; +unsigned int pad; +}; +typedef struct xen_iommu_reserved_range xen_iommu_reserved_range_t ; +DEFINE_XEN_GUEST_HANDLE (xen_iommu_reserved_range_t); + +/* +* XEN_IOMMUOP_query_reserved: Query ranges reserved in the IOMMU. +*/ +#define XEN_IOMMUOP_query_reserved 1 + +struct xen_iommu_op_query_reserved { +/* +* IN/OUT - On entry this is the number of entries available +* in the ranges array below. +* On exit this is the actual number of reserved ranges. +*/ +unsigned int nr_entries; +unsigned int pad; +/* +* OUT - This array is populated with reserved ranges. If it is +* not sufficiently large then available entries are populated, +* but the op status code will be set to -ENOBUFS. +* It is permissable to set this to NULL if nr_entries is also +* set to zero. In this case, on exit, nr_entries will still be +* set to the actual number of reserved ranges but the status +* code will be set to zero. +*/ +XEN_GUEST_HANDLE (xen_iommu_reserved_range_t) ranges; +}; +struct xen_iommu_op {uint16_t op;/* op type */ uint16_t pad;int32_t status;/* op completion status: */ /* 0 for success otherwise, negative errno */ +union { +struct xen_iommu_op_query_reserved query_reserved; + } u; };typedef struct xen_iommu_op xen_iommu_op_t ;DEFINE_XEN_GUEST_HANDLE (xen_iommu_op_t);
xen/include/xlat.lst +2 -0
@@ -79,6 +79,8 @@ ? vcpu_hvm_x86_64 hvm/hvm_vcpu.h ! iommu_op iommu_op.h ! iommu_op_buf iommu_op.h +! iommu_op_query_reserved iommu_op.h +? iommu_reserved_range iommu_op.h ? kexec_exec kexec.h ! kexec_image kexec.h ! kexec_range kexec.h
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 07/14] x86: add iommu_op to query reserved ranges
Quoted 8 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds an
iommu_opto allow the domain IOMMU reserved ranges to be queried by the guest.NOTE: The number of reserved ranges is determined by system firmware, in
conjunction with Xen command line options, and is expected to be small. Thus, to avoid over-complicating the code, there is no pre-emption check within the operation.
Hmm, RMRRs reportedly can cover a fair part of (the entire?) frame buffer of a graphics device.
Quoted 15 lines
xen/common/iommu_op.c +8
-1@@ -100,16 +176,27 @@ long do_iommu_op(unsigned int nr_bufs,return rc; } +CHECK_iommu_reserved_range; +int compat_one_iommu_op (compat_iommu_op_buf_t *buf) { -compat_iommu_op_t cmp; +compat_iommu_op_t cmp = {}; +size_t offset; +static const size_t op_size[] = { + [XEN_IOMMUOP_query_reserved] =sizeof (struct compat_iommu_op_query_reserved ), + }; +size_t size;
xen/common/iommu_op.c long do_iommu_op(unsigned int nr_bufs, · line 197
Quoted 10 lines
xen/common/iommu_op.c +5
-1xen_iommu_op_t nat; +unsigned int u; +int32_t status;int rc; -if ( buf->size <sizeof (cmp) ) + offset = offsetof(struct compat_iommu_op , u); + +if ( buf->size < offset )return -EFAULT ;
For some reason I notice this only now and here - -EFAULT isn't really appropriately describing the error condition here. -ENODATA or -ENOBUFS perhaps?
Quoted 17 lines
xen/common/iommu_op.c +13
-0@@ -119,17 +206,82 @@ int compat_one_iommu_op(compat_iommu_op_buf_t *buf)if ( rc )return rc; +if ( cmp.op >=ARRAY_SIZE (op_size) ) +return -EOPNOTSUPP ; + + size = op_size[array_index_nospec (cmp.op,ARRAY_SIZE (op_size))]; +if ( buf->size < offset + size ) +return -EFAULT ; + +if (copy_from_compat_offset ((void *)&cmp.u, buf->h, offset, size) ) +return -EFAULT ; + +/* +* The xlat magic doesn't quite know how to handle the union so +* we need to fix things up here.
xen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf) · line 231
Quoted 10 lines
xen/common/iommu_op.c +10
-0+*/ +#define XLAT_iommu_op_u_query_reserved XEN_IOMMUOP_query_reserved + u = cmp.op; + +#define XLAT_iommu_op_query_reserved_HNDL_ranges (_d_, _s_) \ + do \ + { \ + if ( !compat_handle_is_null((_s_)->ranges) ) \ + { \ + unsigned int *nr_entries = COMPAT_ARG_XLAT_VIRT_BASE; \
uint32_t (see below) or perhaps even better typeof().
Quoted 45 lines
xen/common/iommu_op.c +39
-0+xen_iommu_reserved_range_t *ranges = \ + (void *)(nr_entries +1 ); \ + \ +if (sizeof (*nr_entries) + \ + (sizeof (*ranges) * (_s_)->nr_entries) > \ +COMPAT_ARG_XLAT_SIZE ) \ +return -E2BIG ; \ + \ + *nr_entries = (_s_)->nr_entries; \ +set_xen_guest_handle ((_d_)->ranges, ranges); \ + } \ +else \ +set_xen_guest_handle ((_d_)->ranges,NULL ); \ + }while (false) +XLAT_iommu_op (&nat, &cmp); +#undef XLAT_iommu_op_query_reserved_HNDL_ranges +iommu_op (&nat); + status = nat.status; + +#define XLAT_iommu_op_query_reserved_HNDL_ranges (_d_, _s_) \ + do \ + { \ + if ( !compat_handle_is_null((_d_)->ranges) ) \ + { \ + unsigned int *nr_entries = COMPAT_ARG_XLAT_VIRT_BASE; \ + compat_iommu_reserved_range_t *ranges = \ + (void *)(nr_entries + 1); \ + unsigned int nr = \ + min_t(unsigned int, (_d_)->nr_entries, *nr_entries); \ + \ + if ( __copy_to_compat_offset((_d_)->ranges, 0, ranges, nr) ) \ + status = -EFAULT; \ + } \ + } while (false) +XLAT_iommu_op (&cmp, &nat); +/* status will have been modified if __copy_to_compat_offset() failed */ + cmp.status = status; + +#undef XLAT_iommu_op_query_reserved_HNDL_ranges
xen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf) · line 286
Quoted 10 lines
xen/common/iommu_op.c +3
-0+if (__copy_field_to_compat (compat_handle_cast (buf->h, compat_iommu_op_t), &cmp, status) )return -EFAULT ; +#undef XLAT_iommu_op_u_query_reserved +return 0 ; }
It's somehow more evident here, but I think even on the native path the nr_entries field doesn't get copied back despite it being an OUT.
xen/include/public/iommu_op.h line 34
Quoted 13 lines
xen/include/public/iommu_op.h +7
-0@@ -25,11 +25,50 @@#include "xen.h" +typedef uint64_t xen_bfn_t ; + +/* Structure describing a single range reserved in the IOMMU */ +struct xen_iommu_reserved_range { +xen_bfn_t start_bfn; +unsigned int nr_frames; +unsigned int pad;
uint32_t
xen/include/public/iommu_op.h line 41
Quoted 7 lines
xen/include/public/iommu_op.h +7
-0+}; +typedef struct xen_iommu_reserved_range xen_iommu_reserved_range_t ; +DEFINE_XEN_GUEST_HANDLE (xen_iommu_reserved_range_t); + +/* +* XEN_IOMMUOP_query_reserved: Query ranges reserved in the IOMMU. +*/
Single line comment please.
xen/include/public/iommu_op.h line 51
Quoted 10 lines
xen/include/public/iommu_op.h +10
-0+#define XEN_IOMMUOP_query_reserved 1 + +struct xen_iommu_op_query_reserved { +/* +* IN/OUT - On entry this is the number of entries available +* in the ranges array below. +* On exit this is the actual number of reserved ranges. +*/ +unsigned int nr_entries; +unsigned int pad;
uint32_t
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 07/14] x86: add iommu_op to query reserved ranges
Quoted 17 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:01
To: Paul Durrant <[email protected]>
Cc: Andrew Cooper <[email protected]>; Wei Liu <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>
Subject: Re: [PATCH v6 07/14] x86: addiommu_opto query reserved rangesQuoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds an
iommu_opto allow the domain IOMMU reservedranges to be
Quoted 2 lines
queried by the guest.
Quoted 8 lines
Quoted 1 line
NOTE: The number of reserved ranges is determined by system firmware,
in
Quoted 3 lines
conjunction with Xen command line options, and is expected to be small. Thus, to avoid over-complicating the code, there is no pre-emption check within the operation.
Hmm, RMRRs reportedly can cover a fair part of (the entire?) frame buffer of a graphics device.
That would still be phrased as a single range though, right?
Quoted 28 lines
Quoted 4 lines
xen/common/iommu_op.c +0
-0@@ -100,16 +176,27 @@ long do_iommu_op(unsigned int nr_bufs,return rc; }xen/common/iommu_op.c long do_iommu_op(unsigned int nr_bufs, · line 186
Quoted 10 lines
xen/common/iommu_op.c +6
-1+CHECK_iommu_reserved_range; + int compat_one_iommu_op(compat_iommu_op_buf_t *buf) { -compat_iommu_op_t cmp; + compat_iommu_op_t cmp = {}; + size_t offset; + static const size_t op_size[] = { + [XEN_IOMMUOP_query_reserved] = sizeof(struct
compat_iommu_op_query_reserved),xen/common/iommu_op.c long do_iommu_op(unsigned int nr_bufs, · line 197
Quoted 12 lines
xen/common/iommu_op.c +7
-1+ }; +size_t size;xen_iommu_op_t nat; +unsigned int u; +int32_t status;int rc; -if ( buf->size <sizeof (cmp) ) + offset = offsetof(struct compat_iommu_op , u); + +if ( buf->size < offset )return -EFAULT ;
Quoted 3 lines
For some reason I notice this only now and here -
-EFAULTisn't really appropriately describing the error condition here.-ENODATAor-ENOBUFSperhaps?
Ok. ENODATA seems best.
Quoted 26 lines
Review int
Quoted 2 lines
(quoted diff) +0
-0@@ -119,17 +206,82 @@ int
compat_one_iommu_op(compat_iommu_op_buf_t *buf)Quoted 8 lines
xen/common/iommu_op.c +5
-0if ( rc )return rc; +if ( cmp.op >=ARRAY_SIZE (op_size) ) +return -EOPNOTSUPP ; + + size = op_size[array_index_nospec (cmp.op,ARRAY_SIZE (op_size))]; +if ( buf->size < offset + size )xen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf)
Quoted 10 lines
xen/common/iommu_op.c +10
-0+return -EFAULT ; + +if (copy_from_compat_offset ((void *)&cmp.u, buf->h, offset, size) ) +return -EFAULT ; + +/* +* The xlat magic doesn't quite know how to handle the union so +* we need to fix things up here. +*/ +#define XLAT_iommu_op_u_query_reserved
XEN_IOMMUOP_query_reservedxen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf)
Quoted 3 lines
xen/common/iommu_op.c +3
-0+ u = cmp.op; + +#define XLAT_iommu_op_query_reserved_HNDL_ranges (_d_, _s_)\
Quoted 8 lines
xen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf) · line 231
Quoted 5 lines
xen/common/iommu_op.c +5
-0+ do \ + { \ + if ( !compat_handle_is_null((_s_)->ranges) ) \ + { \ + unsigned int *nr_entries = COMPAT_ARG_XLAT_VIRT_BASE; \
uint32_t(see below) or perhaps even bettertypeof().
Ok.
Quoted 53 lines
Quoted 14 lines
xen/common/iommu_op.c +14
-0+xen_iommu_reserved_range_t *ranges = \ + (void *)(nr_entries +1 ); \ + \ +if (sizeof (*nr_entries) + \ + (sizeof (*ranges) * (_s_)->nr_entries) > \ +COMPAT_ARG_XLAT_SIZE ) \ +return -E2BIG ; \ + \ + *nr_entries = (_s_)->nr_entries; \ +set_xen_guest_handle ((_d_)->ranges, ranges); \ + } \ +else \ +set_xen_guest_handle ((_d_)->ranges,NULL ); \ + }while (false)xen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf)
Quoted 10 lines
xen/common/iommu_op.c +6
-0+XLAT_iommu_op (&nat, &cmp); +#undef XLAT_iommu_op_query_reserved_HNDL_ranges +iommu_op (&nat); + status = nat.status; + +#define XLAT_iommu_op_query_reserved_HNDL_ranges (_d_, _s_)\
Quoted 8 lines
xen/common/iommu_op.c +8
-0+ do \ + { \ + if ( !compat_handle_is_null((_d_)->ranges) ) \ + { \ + unsigned int *nr_entries = COMPAT_ARG_XLAT_VIRT_BASE; \ + compat_iommu_reserved_range_t *ranges = \ + (void *)(nr_entries + 1); \ + unsigned int nr = \xen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf)
Quoted 10 lines
xen/common/iommu_op.c +8
-0+ min_t(unsigned int, (_d_)->nr_entries, *nr_entries); \ + \ +if (__copy_to_compat_offset ((_d_)->ranges,0 , ranges, nr) ) \ + status = -EFAULT ; \ + } \ + }while (false) +XLAT_iommu_op (&cmp, &nat); +/* status will have been modified if __copy_to_compat_offset() failed */
Quoted 9 lines
xen/common/iommu_op.c +4
-0+ cmp.status = status; + +#undef XLAT_iommu_op_query_reserved_HNDL_ranges +if (__copy_field_to_compat (compat_handle_cast (buf->h, compat_iommu_op_t), &cmp, status) )return -EFAULT ;
Quoted 7 lines
xen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf) · line 286
Quoted 4 lines
xen/common/iommu_op.c +2
-0+#undef XLAT_iommu_op_u_query_reserved +return 0 ; }It's somehow more evident here, but I think even on the native path the
nr_entriesfield doesn't get copied back despite it being an OUT.
Indeed. It would be so much simpler just to copy back the entire buf to avoid the need to special-case this for each op.
Quoted 7 lines
Quoted 7 lines
xen/include/public/iommu_op.h +0
-0@@ -25,11 +25,50 @@#include "xen.h"
Quoted 9 lines
xen/include/public/iommu_op.h line 34
Quoted 7 lines
xen/include/public/iommu_op.h +7
-0+typedef uint64_t xen_bfn_t ; + +/* Structure describing a single range reserved in the IOMMU */ +struct xen_iommu_reserved_range { +xen_bfn_t start_bfn; +unsigned int nr_frames; +unsigned int pad;
uint32_t
Ok.
Quoted 4 lines
xen/include/public/iommu_op.h line 36
Quoted 3 lines
xen/include/public/iommu_op.h +2
-0+}; +typedef struct xen_iommu_reserved_range
xen_iommu_reserved_range_t;
Quoted 9 lines
xen/include/public/iommu_op.h
Quoted 4 lines
xen/include/public/iommu_op.h +4
-0+DEFINE_XEN_GUEST_HANDLE (xen_iommu_reserved_range_t); + +/* +* XEN_IOMMUOP_query_reserved: Query ranges reserved in the IOMMU.
xen/include/public/iommu_op.h line 41
Quoted 1 line
xen/include/public/iommu_op.h +1
-0+*/ Single line comment please.
Ok.
Quoted 4 lines
Quoted 4 lines
xen/include/public/iommu_op.h +4
-0+#define XEN_IOMMUOP_query_reserved 1 + +struct xen_iommu_op_query_reserved { +/*
Quoted 9 lines
xen/include/public/iommu_op.h line 51
Quoted 6 lines
xen/include/public/iommu_op.h +6
-0+* IN/OUT - On entry this is the number of entries available +* in the ranges array below. +* On exit this is the actual number of reserved ranges. +*/ +unsigned int nr_entries; +unsigned int pad;
uint32_t
Ok.
Paul
Quoted 1 line
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 07/14] x86: add iommu_op to query reserved ranges
Quoted 11 lines
Quoted 11 lines
Quoted 1 line · On 11.09.18 at 11:34, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:01Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds an
iommu_opto allow the domain IOMMU reservedranges to be
Quoted 3 lines
queried by the guest.
NOTE: The number of reserved ranges is determined by system firmware,
in
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
conjunction with Xen command line options, and is expected to be small. Thus, to avoid over-complicating the code, there is no pre-emption check within the operation.
Hmm, RMRRs reportedly can cover a fair part of (the entire?) frame buffer of a graphics device.
That would still be phrased as a single range though, right?
Ah, yes, indeed.
Quoted 42 lines
Quoted 34 lines
xen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf)
Quoted 5 lines
iommu_op(&nat);xen/common/iommu_op.c +3
-0+ status = nat.status; + +#define XLAT_iommu_op_query_reserved_HNDL_ranges (_d_, _s_)\
Quoted 8 lines
xen/common/iommu_op.c +8
-0+ do \ + { \ + if ( !compat_handle_is_null((_d_)->ranges) ) \ + { \ + unsigned int *nr_entries = COMPAT_ARG_XLAT_VIRT_BASE; \ + compat_iommu_reserved_range_t *ranges = \ + (void *)(nr_entries + 1); \ + unsigned int nr = \xen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf)
Quoted 10 lines
xen/common/iommu_op.c +8
-0+ min_t(unsigned int, (_d_)->nr_entries, *nr_entries); \ + \ +if (__copy_to_compat_offset ((_d_)->ranges,0 , ranges, nr) ) \ + status = -EFAULT ; \ + } \ + }while (false) +XLAT_iommu_op (&cmp, &nat); +/* status will have been modified if __copy_to_compat_offset() failed */
Quoted 9 lines
xen/common/iommu_op.c +4
-0+ cmp.status = status; + +#undef XLAT_iommu_op_query_reserved_HNDL_ranges +if (__copy_field_to_compat (compat_handle_cast (buf->h, compat_iommu_op_t), &cmp, status) )return -EFAULT ;Quoted 7 lines
xen/common/iommu_op.c int compat_one_iommu_op(compat_iommu_op_buf_t *buf) · line 286
Quoted 4 lines
xen/common/iommu_op.c +2
-0+#undef XLAT_iommu_op_u_query_reserved +return 0 ; }It's somehow more evident here, but I think even on the native path the
nr_entriesfield doesn't get copied back despite it being an OUT.
Quoted 2 lines
Indeed. It would be so much simpler just to copy back the entire buf to avoid the need to special-case this for each op.
In case of such complications I view doing so as perfectly legitimate. It's just that in cases where copying a single field is obviously easy and sufficient that I suggest/request that only that one field be copied.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 07/14] x86: add iommu_op to query reserved ranges
In reply to Jan BeulichQuoted 8 lines
From: Jan Beulich
Sent: Friday, September 7, 2018 7:01 PMQuoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds an
iommu_opto allow the domain IOMMU reservedranges to be
Quoted 2 lines
queried by the guest.
Quoted 9 lines
Quoted 1 line
NOTE: The number of reserved ranges is determined by system firmware,
in
Quoted 3 lines
conjunction with Xen command line options, and is expected to be small. Thus, to avoid over-complicating the code, there is no pre-emption check within the operation.
Hmm, RMRRs reportedly can cover a fair part of (the entire?) frame buffer of a graphics device.
but it is still just one range
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 8/14 vtd: add lookup_page method to iommu_ops
In reply to Paul DurrantThis patch adds a new method to the VT-d IOMMU implementation to find the MFN currently mapped by the specified BFN along with a wrapper function in generic IOMMU code to call the implementation if it exists.
This functionality will be used by a subsequent patch.
---
Cc: Kevin Tian <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: George Dunlap <[email protected]>
v3:
- Addressed comments from George.
v2:
- Addressed some comments from Jan.
--- xen/drivers/passthrough/iommu.c | 11 +++++++++++ xen/drivers/passthrough/vtd/iommu.c | 34 ++++++++++++++++++++++++++++++++++ xen/drivers/passthrough/vtd/iommu.h | 3 +++ xen/include/xen/iommu.h | 4 ++++ 4 files changed, 52 insertions(+)
xen/drivers/passthrough/iommu.c +11 -0
@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain *d, bfn_t bfn)return rc; } +int iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +} +static void iommu_free_pagetables (unsigned long unused) { do {
xen/drivers/passthrough/vtd/iommu.c +34 -0
@@ -1830,6 +1830,39 @@ static int __must_check intel_iommu_unmap_page(struct domain *d,return dma_pte_clear_one (d,bfn_to_baddr (bfn)); } +static int intel_iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +struct domain_iommu *hd =dom_iommu (d); +struct dma_pte *page =NULL , *pte =NULL , val; +u64 pg_maddr; + +spin_lock (&hd->arch.mapping_lock); + + pg_maddr =addr_to_dma_page_maddr (d,bfn_to_baddr (bfn),0 ); +if ( pg_maddr ==0 ) + { +spin_unlock (&hd->arch.mapping_lock); +return -ENOMEM ; + } + + page =map_vtd_domain_page (pg_maddr); + pte = page + (bfn_x (bfn) &LEVEL_MASK ); + val = *pte; + +unmap_vtd_domain_page (page); +spin_unlock (&hd->arch.mapping_lock); + +if ( !dma_pte_present (val) ) +return -ENOENT ; + + *mfn =maddr_to_mfn (dma_pte_addr (val)); + *flags =dma_pte_read (val) ? IOMMUF_readable :0 ; + *flags |=dma_pte_write (val) ? IOMMUF_writable :0 ; + +return 0 ; +} +int iommu_pte_flush (struct domain *d,uint64_t bfn,uint64_t *pte,int order,int present) {
@@ -2661,6 +2694,7 @@ const struct iommu_ops intel_iommu_ops = { .teardown = iommu_domain_teardown, .map_page = intel_iommu_map_page, .unmap_page = intel_iommu_unmap_page, + .lookup_page = intel_iommu_lookup_page, .free_page_table = iommu_free_page_table, .reassign_device = reassign_device_ownership, .get_device_group_id = intel_iommu_group_id,
xen/drivers/passthrough/vtd/iommu.h +3 -0
@@ -272,6 +272,9 @@ struct dma_pte {#define dma_set_pte_prot (p, prot) do { \ (p).val = ((p).val & ~DMA_PTE_PROT) | ((prot) & DMA_PTE_PROT); \ } while (0) +#define dma_pte_prot (p) ((p).val & DMA_PTE_PROT) +#define dma_pte_read (p) (dma_pte_prot(p) & DMA_PTE_READ) +#define dma_pte_write (p) (dma_pte_prot(p) & DMA_PTE_WRITE)#define dma_pte_addr (p) ((p).val & PADDR_MASK & PAGE_MASK_4K)#define dma_set_pte_addr (p, addr) do {\ (p).val |= ((addr) & PAGE_MASK_4K); } while (0)
xen/include/xen/iommu.h +4 -0
@@ -100,6 +100,8 @@ void iommu_teardown(struct domain *d);int __must_check iommu_map_page (struct domain *d,bfn_t bfn,mfn_t mfn,unsigned int flags);int __must_check iommu_unmap_page (struct domain *d,bfn_t bfn); +int __must_check iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags);enum iommu_feature {
@@ -196,6 +198,8 @@ struct iommu_ops {int __must_check (*map_page)(struct domain *d,bfn_t bfn,mfn_t mfn,unsigned int flags);int __must_check (*unmap_page)(struct domain *d,bfn_t bfn); +int __must_check (*lookup_page)(struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags);void (*free_page_table)(struct page_info *);#ifdef CONFIG_X86 void (*update_ire_from_apic)(unsigned int apic,unsigned int reg,unsigned int value);
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation to find the MFN currently mapped by the specified BFN along with a wrapper function in generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be implemented.
Quoted 6 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain *d, bfn_t bfn)return rc; }
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t bfn) · line 317
Quoted 10 lines
xen/drivers/passthrough/iommu.c +10
-0+int iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +}
Shouldn't this be restricted to PV guests? HVM ones aren't supposed to know about MFNs.
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t bfn)
Quoted 5 lines
xen/drivers/passthrough/iommu.c +5
-0+static int intel_iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +struct domain_iommu *hd =dom_iommu (d); +struct dma_pte *page =NULL , *pte =NULL , val;
Pointless initializers. I also question the usefulness of "pte":
Quoted 4 lines
xen/drivers/passthrough/iommu.c +4
-0+u64 pg_maddr; + +spin_lock (&hd->arch.mapping_lock); +
xen/drivers/passthrough/iommu.c
Quoted 10 lines
xen/drivers/passthrough/iommu.c +10
-0+ pg_maddr =addr_to_dma_page_maddr (d,bfn_to_baddr (bfn),0 ); +if ( pg_maddr ==0 ) + { +spin_unlock (&hd->arch.mapping_lock); +return -ENOMEM ; + } + + page =map_vtd_domain_page (pg_maddr); + pte = page + (bfn_x (bfn) &LEVEL_MASK ); + val = *pte;
val = page[bfn_x(bfn) & LEVEL_MASK];
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 8 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: Re: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_ops
Quoted 8 lines
Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation to find
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper function
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be implemented.
Why? It can be done later. Nothing existing is going to break if it is not implemented.
Paul
Quoted 48 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t
Quoted 4 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain *d, bfn_tbfn)
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t bfn) · line 317
Quoted 13 lines
xen/drivers/passthrough/iommu.c +10
-0return rc; } +int iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +}Shouldn't this be restricted to PV guests? HVM ones aren't supposed to know about MFNs.
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t · line 318
Quoted 1 line
xen/drivers/passthrough/iommu.c +1
-0+static int intel_iommu_lookup_page(struct domain *d, bfn_t bfn, mfn_t*mfn,
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t
Quoted 4 lines
xen/drivers/passthrough/iommu.c +4
-0+ unsigned int *flags) +{ +struct domain_iommu *hd =dom_iommu (d); +struct dma_pte *page =NULL , *pte =NULL , val;Pointless initializers. I also question the usefulness of "pte":
Quoted 4 lines
xen/drivers/passthrough/iommu.c +4
-0+u64 pg_maddr; + +spin_lock (&hd->arch.mapping_lock); +xen/drivers/passthrough/iommu.c
Quoted 10 lines
xen/drivers/passthrough/iommu.c +10
-0+ pg_maddr =addr_to_dma_page_maddr (d,bfn_to_baddr (bfn),0 ); +if ( pg_maddr ==0 ) + { +spin_unlock (&hd->arch.mapping_lock); +return -ENOMEM ; + } + + page =map_vtd_domain_page (pg_maddr); + pte = page + (bfn_x (bfn) &LEVEL_MASK ); + val = *pte;val = page[bfn_x(bfn) & LEVEL_MASK];
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 12 lines
Quoted 12 lines
Quoted 1 line · On 07.09.18 at 14:36, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: Re: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation to find
the
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper function
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be implemented.
Why? It can be done later. Nothing existing is going to break if it is not implemented.
If it was something that's terribly difficult to implement, I'd probably agree. But introducing PV IOMMU for Intel only (and hence once again making AMD a second class citizen) I don't really like. Another thing would be if you had the implementation ready, but the maintainer(s) don't respond...
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 29 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 15:56
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 8 lines
Quoted 8 lines
Quoted 1 line · On 07.09.18 at 14:36, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: Re: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 2 lines
Quoted 2 lines · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation to
find
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper
function
Quoted 5 lines
Quoted 4 lines
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be implemented.
Quoted 9 lines
Quoted 2 lines
Why? It can be done later. Nothing existing is going to break if it is not implemented.
If it was something that's terribly difficult to implement, I'd probably agree. But introducing PV IOMMU for Intel only (and hence once again making AMD a second class citizen) I don't really like. Another thing would be if you had the implementation ready, but the maintainer(s) don't respond...
It's all time though. The fact is that, in XenServer, we've never had PV-IOMMU for AMD. It would be wonderful to have it for AMD too and indeed I may find the time to do it, but is that really a reason to block integration of these patches when no actual regression will be caused by of the lack of AMD support?
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 29 lines
Quoted 19 lines
Quoted 1 line · On 07.09.18 at 17:24, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 15:56Quoted 6 lines
Quoted 6 lines
Quoted 1 line · On 07.09.18 at 14:36, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation to
find
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper
function
Quoted 5 lines
Quoted 4 lines
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be implemented.
Quoted 9 lines
Quoted 2 lines
Why? It can be done later. Nothing existing is going to break if it is not implemented.
If it was something that's terribly difficult to implement, I'd probably agree. But introducing PV IOMMU for Intel only (and hence once again making AMD a second class citizen) I don't really like. Another thing would be if you had the implementation ready, but the maintainer(s) don't respond...
Quoted 5 lines
It's all time though. The fact is that, in XenServer, we've never had PV-IOMMU for AMD. It would be wonderful to have it for AMD too and indeed I may find the time to do it, but is that really a reason to block integration of these patches when no actual regression will be caused by of the lack of AMD support?
First of all - I don't mean to block anything. I deliberately said "I think it wants to be", not "It needs to be". But beyond that my second class citizen statement stands - it's no different from e.g. introducing VMX support, but not SVM one. That also is not resulting in any regression, but still is not very desirable.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
In reply to Jan BeulichQuoted 24 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: Re: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation to find
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper function
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be implemented.
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t
Quoted 3 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain *d, bfn_tbfn)
Quoted 3 lines
xen/drivers/passthrough/iommu.c +0
-0return rc; }
Quoted 13 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t bfn) · line 317
Quoted 10 lines
xen/drivers/passthrough/iommu.c +10
-0+int iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +}Shouldn't this be restricted to PV guests? HVM ones aren't supposed to know about MFNs.
Agreed, but I think this is the wrong level to be applying such a check: iommu_map_page() is supplied an MFN regardless of whether the domain is PV or HVM, so I think it is reasonable for a lookup function to work in terms of MFNs.
Quoted 10 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t · line 318
Quoted 2 lines
xen/drivers/passthrough/iommu.c +1
-0+static int intel_iommu_lookup_page(struct domain *d, bfn_t bfn, mfn_t*mfn,
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t
Quoted 4 lines
xen/drivers/passthrough/iommu.c +4
-0+ unsigned int *flags) +{ +struct domain_iommu *hd =dom_iommu (d); +struct dma_pte *page =NULL , *pte =NULL , val;Pointless initializers. I also question the usefulness of "pte":
Yes, I'll tidy up the initializers.
Quoted 8 lines
Quoted 8 lines
xen/drivers/passthrough/iommu.c +8
-0+u64 pg_maddr; + +spin_lock (&hd->arch.mapping_lock); + + pg_maddr =addr_to_dma_page_maddr (d,bfn_to_baddr (bfn),0 ); +if ( pg_maddr ==0 ) + { +spin_unlock (&hd->arch.mapping_lock);
Quoted 9 lines
xen/drivers/passthrough/iommu.c
Quoted 6 lines
xen/drivers/passthrough/iommu.c +6
-0+return -ENOMEM ; + } + + page =map_vtd_domain_page (pg_maddr); + pte = page + (bfn_x (bfn) &LEVEL_MASK ); + val = *pte;val = page[bfn_x(bfn) & LEVEL_MASK];
Yes, that's neater.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 34 lines
Quoted 20 lines
Quoted 1 line · On 12.09.18 at 10:31, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation to find
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper function
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be implemented.
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t
Quoted 3 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain *d, bfn_tbfn)
Quoted 3 lines
xen/drivers/passthrough/iommu.c +0
-0return rc; }Quoted 13 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t bfn) · line 317
Quoted 10 lines
xen/drivers/passthrough/iommu.c +10
-0+int iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +}Shouldn't this be restricted to PV guests? HVM ones aren't supposed to know about MFNs.
Quoted 4 lines
Agreed, but I think this is the wrong level to be applying such a check:
iommu_map_page()is supplied an MFN regardless of whether the domain is PV or HVM, so I think it is reasonable for a lookup function to work in terms of MFNs.
I don't mind much where the check sits, but ASSERT(!is_hvm_domain()), if placed here, should not trigger.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 45 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 09:44
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 6 lines
Quoted 6 lines
Quoted 1 line · On 12.09.18 at 10:31, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation to
find
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper
function
Quoted 8 lines
Quoted 8 lines
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be implemented.
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d,
Quoted 3 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain *d,
bfn_tQuoted 18 lines
Quoted 17 lines
bfn)
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t bfn) · line 317
Quoted 13 lines
xen/drivers/passthrough/iommu.c +10
-0return rc; } +int iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +}Shouldn't this be restricted to PV guests? HVM ones aren't supposed to know about MFNs.
Quoted 9 lines
Quoted 2 lines
Agreed, but I think this is the wrong level to be applying such a check:
iommu_map_page()is supplied an MFN regardless of whether the domainis PV or
Quoted 2 lines
HVM, so I think it is reasonable for a lookup function to work in terms of MFNs.
I don't mind much where the check sits, but ASSERT(!is_hvm_domain()), if placed here, should not trigger.
It will though. I'm going to need to use this function for HVM guests after having done a GFN lookup.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 64 lines
-----Original Message-----
From: Xen-devel [mailto:[email protected]] On Behalf Of Paul Durrant
Sent: 12 September 2018 09:45
To: 'Jan Beulich' <[email protected]>
Cc: xen-devel <[email protected]>; Kevin Tian <[email protected]>; George Dunlap <[email protected]>
Subject: Re: [Xen-devel] [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 7 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 09:44
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 2 lines · On 12.09.18 at 10:31, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11Quoted 10 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation to
find
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper
function
Quoted 4 lines
Quoted 4 lines
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be
implemented.
Quoted 25 lines
Quoted 4 lines
Quoted 4 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d,
Quoted 4 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain *d,
bfn_tQuoted 20 lines
Quoted 4 lines
bfn)
Quoted 3 lines
xen/drivers/passthrough/iommu.c +0
-0return rc; }Quoted 13 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t bfn) · line 317
Quoted 10 lines
xen/drivers/passthrough/iommu.c +10
-0+int iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +}Shouldn't this be restricted to PV guests? HVM ones aren't supposed to know about MFNs.
Agreed, but I think this is the wrong level to be applying such a check:
iommu_map_page()is supplied an MFN regardless of whether thedomain
Quoted 4 lines
is PV or
Quoted 2 lines
HVM, so I think it is reasonable for a lookup function to work in terms of MFNs.
Quoted 6 lines
Quoted 3 lines
I don't mind much where the check sits, but ASSERT(!is_hvm_domain()), if placed here, should not trigger.
It will though. I'm going to need to use this function for HVM guests after having done a GFN lookup.
Sorry... I'm getting confused myself now. It won't fire because in my case the domain here will always by PV (because it is the not the domain owning the GFN). I still think this is the wrong level for such a check though, but I'll put in the ASSERT.
Paul
Quoted 11 lines
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 80 lines
-----Original Message-----
From: Paul Durrant
Sent: 12 September 2018 09:52
To: Paul Durrant <[email protected]>; 'Jan Beulich' <[email protected]>
Cc: xen-devel <[email protected]>; Kevin Tian <[email protected]>; George Dunlap <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 2 lines
-----Original Message-----
From: Xen-devel [mailto:[email protected]] OnBehalf
Quoted 6 lines
Of Paul Durrant
Sent: 12 September 2018 09:45
To: 'Jan Beulich' <[email protected]>
Cc: xen-devel <[email protected]>; Kevin Tian <[email protected]>; George Dunlap <[email protected]>
Subject: Re: [Xen-devel] [PATCH v6 08/14] vtd: addlookup_pagemethodto
Quoted 17 lines
iommu_opsQuoted 7 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 09:44
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 2 lines · On 12.09.18 at 10:31, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation
to
Quoted 43 lines
Quoted 8 lines
find
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper
function
Quoted 4 lines
Quoted 4 lines
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be
implemented.
Quoted 25 lines
Quoted 4 lines
Quoted 4 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d,
Quoted 4 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain *d,
bfn_tQuoted 20 lines
Quoted 4 lines
bfn)
Quoted 3 lines
xen/drivers/passthrough/iommu.c +0
-0return rc; }Quoted 13 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t bfn) · line 317
Quoted 10 lines
xen/drivers/passthrough/iommu.c +10
-0+int iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +}Shouldn't this be restricted to PV guests? HVM ones aren't supposed to know about MFNs.
Agreed, but I think this is the wrong level to be applying such a check:
iommu_map_page()is supplied an MFN regardless of whether thedomain
Quoted 7 lines
is PV or
Quoted 2 lines
HVM, so I think it is reasonable for a lookup function to work in terms of MFNs.
I don't mind much where the check sits, but ASSERT(!is_hvm_domain()), if placed here, should not trigger.
Quoted 8 lines
Quoted 2 lines
It will though. I'm going to need to use this function for HVM guests after having done a GFN lookup.
Sorry... I'm getting confused myself now. It won't fire because in my case the domain here will always by PV (because it is the not the domain owning the GFN). I still think this is the wrong level for such a check though, but I'll put in the ASSERT.
Actually, no I still don't think the ASSERT is correct. Why should we rule out HVM guests being able to use PV-IOMMU?
Paul
Quoted 13 lines
Paul
Quoted 11 lines
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 62 lines
Quoted 62 lines
Quoted 1 line · On 12.09.18 at 10:53, <[email protected]> wrote:
From: Paul Durrant
Sent: 12 September 2018 09:52Quoted 1 line
From: Xen-devel [mailto:[email protected]] On
Behalf
Quoted 12 lines
Of Paul Durrant
Sent: 12 September 2018 09:45Quoted 9 lines
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 09:44Quoted 6 lines
Quoted 6 lines
Quoted 1 line · On 12.09.18 at 10:31, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation
to
Quoted 43 lines
Quoted 8 lines
find
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper
function
Quoted 4 lines
Quoted 4 lines
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be
implemented.
Quoted 25 lines
Quoted 4 lines
Quoted 4 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d,
Quoted 4 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain *d,
bfn_tQuoted 20 lines
Quoted 4 lines
bfn)
Quoted 3 lines
xen/drivers/passthrough/iommu.c +0
-0return rc; }Quoted 13 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, · line 317
Quoted 10 lines
xen/drivers/passthrough/iommu.c +10
-0+int iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +}Shouldn't this be restricted to PV guests? HVM ones aren't supposed to know about MFNs.
Agreed, but I think this is the wrong level to be applying such a check:
iommu_map_page()is supplied an MFN regardless of whether thedomain
Quoted 7 lines
is PV or
Quoted 2 lines
HVM, so I think it is reasonable for a lookup function to work in terms of MFNs.
I don't mind much where the check sits, but ASSERT(!is_hvm_domain()), if placed here, should not trigger.
Quoted 7 lines
Quoted 7 lines
Quoted 2 lines
It will though. I'm going to need to use this function for HVM guests after having done a GFN lookup.
Sorry... I'm getting confused myself now. It won't fire because in my case the domain here will always by PV (because it is the not the domain owning the GFN). I still think this is the wrong level for such a check though, but I'll put in the ASSERT.
And what would guarantee the ASSERT() to not trigger? As said, I'm fine with this being enforced at a different level, but it needs to be enforced somewhere.
Quoted 2 lines
Actually, no I still don't think the ASSERT is correct. Why should we rule out HVM guests being able to use PV-IOMMU?
A HVM guest using the PV IOMMU is quite fine, but it shouldn't talk to it in terms of MFNs.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 94 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 12 lines
Quoted 12 lines
Quoted 1 line · On 12.09.18 at 10:53, <[email protected]> wrote:
From: Paul Durrant
Sent: 12 September 2018 09:52Quoted 1 line
From: Xen-devel [mailto:[email protected]] On
Behalf
Quoted 6 lines
Of Paul Durrant
Sent: 12 September 2018 09:45Quoted 3 lines
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 09:44Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 1 line · On 12.09.18 at 10:31, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU
implementation
Quoted 4 lines
Quoted 4 lines
to
Quoted 3 lines
Quoted 3 lines
find
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a
wrapper
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 5 lines
function
Quoted 4 lines
Quoted 4 lines
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be
implemented.
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain
Quoted 4 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain*d,
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
bfn_tQuoted 5 lines
Quoted 5 lines
bfn)
Quoted 4 lines
xen/drivers/passthrough/iommu.c +1
-0return rc; } +int iommu_lookup_page(struct domain *d, bfn_t bfn, mfn_t*mfn,
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain · line 317
Quoted 9 lines
xen/drivers/passthrough/iommu.c +9
-0+ unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +}Shouldn't this be restricted to PV guests? HVM ones aren't
supposed
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
to know about MFNs.
Agreed, but I think this is the wrong level to be applying such a
check:
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 1 line
Quoted 1 line
iommu_map_page()is supplied an MFN regardless of whether thedomain
Quoted 2 lines
is PV or
Quoted 1 line
HVM, so I think it is reasonable for a lookup function to work in
terms of
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
MFNs.
I don't mind much where the check sits, but
ASSERT(!is_hvm_domain()),
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
if placed here, should not trigger.
It will though. I'm going to need to use this function for HVM guests
after
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
having done a GFN lookup.
Sorry... I'm getting confused myself now. It won't fire because in my case
the
Quoted 1 line
Quoted 1 line
domain here will always by PV (because it is the not the domain owning
the
Quoted 1 line
Quoted 1 line
GFN). I still think this is the wrong level for such a check though, but I'll put
in
Quoted 1 line
Quoted 1 line
the ASSERT.
And what would guarantee the
ASSERT()to not trigger? As said, I'm fine with this being enforced at a different level, but it needs to be enforced somewhere.
Quoted 6 lines
Quoted 2 lines
Actually, no I still don't think the ASSERT is correct. Why should we rule out HVM guests being able to use PV-IOMMU?
A HVM guest using the PV IOMMU is quite fine, but it shouldn't talk to it in terms of MFNs.
Well, it has to talk MFNs at some level, surely? The output of the IOMMU is not subject to EPT/NPT, right?
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 10 lines
Quoted 7 lines
Quoted 1 line · On 12.09.18 at 11:05, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03A HVM guest using the PV IOMMU is quite fine, but it shouldn't talk to it in terms of MFNs.
Well, it has to talk MFNs at some level, surely? The output of the IOMMU is not subject to EPT/NPT, right?
Yes to the second question, but no to the first: The GFN -> MFN translation should still be done inside Xen in the HVM case, imo (in the course of manufacturing the PTE).
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 16 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:13
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 8 lines
Quoted 7 lines
Quoted 1 line · On 12.09.18 at 11:05, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03A HVM guest using the PV IOMMU is quite fine, but it shouldn't talk to it in terms of MFNs.
Quoted 6 lines
Quoted 2 lines
Well, it has to talk MFNs at some level, surely? The output of the IOMMU is not subject to EPT/NPT, right?
Yes to the second question, but no to the first: The GFN -> MFN translation should still be done inside Xen in the HVM case, imo (in the course of manufacturing the PTE).
Indeed. This function is very much internal to Xen (it's simply an abstraction on top of a vendor implementation), so why should it not work in terms of MFNs? There's no hypercall that allows this to be blindly called by an HVM guest.
Paul
Quoted 3 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 20 lines
Quoted 20 lines
Quoted 1 line · On 12.09.18 at 11:15, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:13
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 10 lines
Quoted 7 lines
Quoted 1 line · On 12.09.18 at 11:05, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03A HVM guest using the PV IOMMU is quite fine, but it shouldn't talk to it in terms of MFNs.
Well, it has to talk MFNs at some level, surely? The output of the IOMMU is not subject to EPT/NPT, right?
Quoted 7 lines
Quoted 3 lines
Yes to the second question, but no to the first: The GFN -> MFN translation should still be done inside Xen in the HVM case, imo (in the course of manufacturing the PTE).
Indeed. This function is very much internal to Xen (it's simply an abstraction on top of a vendor implementation), so why should it not work in terms of MFNs?
Because "MFN" is a concept a HVM guest is not knowing about, or supposed to be knowing. The only time where (part of) it might legitimately (have to) know is when it comes to managing the host (including any guests), i.e. in the tool stack of a PVH Dom0.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 37 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:21
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 8 lines
Quoted 8 lines
Quoted 1 line · On 12.09.18 at 11:15, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:13
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 8 lines
Quoted 2 lines · On 12.09.18 at 11:05, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03A HVM guest using the PV IOMMU is quite fine, but it shouldn't talk to it in terms of MFNs.
Well, it has to talk MFNs at some level, surely? The output of the
IOMMU is
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
not subject to EPT/NPT, right?
Yes to the second question, but no to the first: The GFN -> MFN
translation
Quoted 5 lines
Quoted 2 lines
should still be done inside Xen in the HVM case, imo (in the course of manufacturing the PTE).
Indeed. This function is very much internal to Xen (it's simply an abstraction on top of a vendor implementation), so why should it not work
Quoted 7 lines
in
Quoted 1 line
terms of MFNs?
Because "MFN" is a concept a HVM guest is not knowing about, or supposed to be knowing. The only time where (part of) it might legitimately (have to) know is when it comes to managing the host (including any guests), i.e. in the tool stack of a PVH Dom0.
Ok. So consider validating a PV-IOMMU unmap request from an HVM guest. It passes in a DFN and a GFN belonging to itself. Now Xen needs to figure out whether that BFN actually maps to the GFN. It can look up the MFN backing the GFN (from the p2m). How does Xen now validate it if it cannot lookup what MFN is actually present in the PTE referenced by the DFN?
Paul
Quoted 3 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 46 lines
Quoted 38 lines
Quoted 1 line · On 12.09.18 at 11:30, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:21
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 8 lines
Quoted 8 lines
Quoted 1 line · On 12.09.18 at 11:15, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:13
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 8 lines
Quoted 2 lines · On 12.09.18 at 11:05, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03A HVM guest using the PV IOMMU is quite fine, but it shouldn't talk to it in terms of MFNs.
Well, it has to talk MFNs at some level, surely? The output of the
IOMMU is
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
not subject to EPT/NPT, right?
Yes to the second question, but no to the first: The GFN -> MFN
translation
Quoted 5 lines
Quoted 2 lines
should still be done inside Xen in the HVM case, imo (in the course of manufacturing the PTE).
Indeed. This function is very much internal to Xen (it's simply an abstraction on top of a vendor implementation), so why should it not work
Quoted 7 lines
in
Quoted 1 line
terms of MFNs?
Because "MFN" is a concept a HVM guest is not knowing about, or supposed to be knowing. The only time where (part of) it might legitimately (have to) know is when it comes to managing the host (including any guests), i.e. in the tool stack of a PVH Dom0.
Quoted 5 lines
Ok. So consider validating a PV-IOMMU unmap request from an HVM guest. It passes in a DFN and a GFN belonging to itself. Now Xen needs to figure out whether that BFN actually maps to the GFN. It can look up the MFN backing the GFN (from the p2m). How does Xen now validate it if it cannot lookup what MFN is actually present in the PTE referenced by the DFN?
I'm afraid I don't understand: The passed in GFN gets translated to an MFN using a p2m lookup. The passed in DFN (which aiui ought to match the GFN anyway on x86) gets translated to an MFN using an IOMMU page table lookup. The resulting two MFNs have to match for the request to be valid.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 63 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 11:08
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 8 lines
Quoted 8 lines
Quoted 1 line · On 12.09.18 at 11:30, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:21
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 2 lines · On 12.09.18 at 11:15, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:13
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toQuoted 7 lines
Quoted 7 lines
iommu_opsQuoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines · On 12.09.18 at 11:05, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03A HVM guest using the PV IOMMU is quite fine, but it shouldn't talk
to
Quoted 7 lines
Quoted 7 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
it in terms of MFNs.
Well, it has to talk MFNs at some level, surely? The output of the
IOMMU is
Quoted 2 lines
Quoted 2 lines
Quoted 1 line
not subject to EPT/NPT, right?
Quoted 7 lines
Quoted 7 lines
Quoted 1 line
Quoted 1 line
Yes to the second question, but no to the first: The GFN -> MFN
translation
Quoted 5 lines
Quoted 2 lines
should still be done inside Xen in the HVM case, imo (in the course of manufacturing the PTE).
Indeed. This function is very much internal to Xen (it's simply an abstraction on top of a vendor implementation), so why should it not
work
Quoted 9 lines
Quoted 7 lines
in
Quoted 1 line
terms of MFNs?
Because "MFN" is a concept a HVM guest is not knowing about, or supposed to be knowing. The only time where (part of) it might legitimately (have to) know is when it comes to managing the host (including any guests), i.e. in the tool stack of a PVH Dom0.
Ok. So consider validating a PV-IOMMU unmap request from an HVM
guest. It
Quoted 2 lines
passes in a DFN and a GFN belonging to itself. Now Xen needs to figure out whether that BFN actually maps to the GFN. It can look up the MFN backing
the
Quoted 1 line
GFN (from the p2m). How does Xen now validate it if it cannot lookup what
Quoted 9 lines
MFN
Quoted 1 line
is actually present in the PTE referenced by the DFN?
I'm afraid I don't understand: The passed in GFN gets translated to an MFN using a p2m lookup. The passed in DFN (which aiui ought to match the GFN anyway on x86) gets translated to an MFN using an IOMMU page table lookup. The resulting two MFNs have to match for the request to be valid.
Quite. So how does that work if iommu_lookup_page() is ASSERTing that the domain in question is not HVM?
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 74 lines
Quoted 64 lines
Quoted 1 line · On 12.09.18 at 12:09, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 11:08
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 8 lines
Quoted 8 lines
Quoted 1 line · On 12.09.18 at 11:30, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:21
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 2 lines · On 12.09.18 at 11:15, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:13
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toQuoted 7 lines
Quoted 7 lines
iommu_opsQuoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines · On 12.09.18 at 11:05, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03A HVM guest using the PV IOMMU is quite fine, but it shouldn't talk
to
Quoted 7 lines
Quoted 7 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
it in terms of MFNs.
Well, it has to talk MFNs at some level, surely? The output of the
IOMMU is
Quoted 2 lines
Quoted 2 lines
Quoted 1 line
not subject to EPT/NPT, right?
Quoted 7 lines
Quoted 7 lines
Quoted 1 line
Quoted 1 line
Yes to the second question, but no to the first: The GFN -> MFN
translation
Quoted 5 lines
Quoted 2 lines
should still be done inside Xen in the HVM case, imo (in the course of manufacturing the PTE).
Indeed. This function is very much internal to Xen (it's simply an abstraction on top of a vendor implementation), so why should it not
work
Quoted 9 lines
Quoted 7 lines
in
Quoted 1 line
terms of MFNs?
Because "MFN" is a concept a HVM guest is not knowing about, or supposed to be knowing. The only time where (part of) it might legitimately (have to) know is when it comes to managing the host (including any guests), i.e. in the tool stack of a PVH Dom0.
Ok. So consider validating a PV-IOMMU unmap request from an HVM
guest. It
Quoted 2 lines
passes in a DFN and a GFN belonging to itself. Now Xen needs to figure out whether that BFN actually maps to the GFN. It can look up the MFN backing
the
Quoted 1 line
GFN (from the p2m). How does Xen now validate it if it cannot lookup what
Quoted 9 lines
MFN
Quoted 1 line
is actually present in the PTE referenced by the DFN?
I'm afraid I don't understand: The passed in GFN gets translated to an MFN using a p2m lookup. The passed in DFN (which aiui ought to match the GFN anyway on x86) gets translated to an MFN using an IOMMU page table lookup. The resulting two MFNs have to match for the request to be valid.
Quoted 2 lines
Quite. So how does that work if
iommu_lookup_page()is ASSERTing that the domain in question is not HVM?
Well, as soon as the function doesn't hand back MFNs anymore to HVM callers, no such assertion would be needed anymore either.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 89 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 13:15
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 8 lines
Quoted 8 lines
Quoted 1 line · On 12.09.18 at 12:09, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 11:08
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 11 lines
Quoted 11 lines
Quoted 9 lines
Quoted 9 lines
Quoted 2 lines · On 12.09.18 at 11:30, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:21
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 1 line
Quoted 1 line
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 1 line · On 12.09.18 at 11:15, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:13
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <xen-Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Subject: RE: [PATCH v6 08/14] vtd: add
lookup_pagemethod to
iommu_opsQuoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines · On 12.09.18 at 11:05, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03A HVM guest using the PV IOMMU is quite fine, but it shouldn't
talk
Quoted 11 lines
Quoted 11 lines
to
Quoted 10 lines
Quoted 10 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
it in terms of MFNs.
Well, it has to talk MFNs at some level, surely? The output of the
IOMMU is
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
not subject to EPT/NPT, right?
Yes to the second question, but no to the first: The GFN -> MFN
translation
Quoted 1 line
Quoted 1 line
should still be done inside Xen in the HVM case, imo (in the course
of
Quoted 8 lines
Quoted 8 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 1 line
manufacturing the PTE).
Indeed. This function is very much internal to Xen (it's simply an abstraction on top of a vendor implementation), so why should it not
work
Quoted 3 lines
Quoted 3 lines
in
Quoted 1 line
terms of MFNs?
Quoted 8 lines
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Because "MFN" is a concept a HVM guest is not knowing about, or supposed to be knowing. The only time where (part of) it might legitimately (have to) know is when it comes to managing the host (including any guests), i.e. in the tool stack of a PVH Dom0.
Ok. So consider validating a PV-IOMMU unmap request from an HVM
guest. It
Quoted 1 line
passes in a DFN and a GFN belonging to itself. Now Xen needs to figure
out
Quoted 1 line
Quoted 1 line
Quoted 1 line
whether that BFN actually maps to the GFN. It can look up the MFN
backing
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
GFN (from the p2m). How does Xen now validate it if it cannot lookup
what
Quoted 10 lines
Quoted 9 lines
MFN
Quoted 1 line
is actually present in the PTE referenced by the DFN?
I'm afraid I don't understand: The passed in GFN gets translated to an MFN using a p2m lookup. The passed in DFN (which aiui ought to match the GFN anyway on x86) gets translated to an MFN using an IOMMU page table lookup. The resulting two MFNs have to match for the request to be valid.
Quoted 7 lines
Quoted 1 line
Quite. So how does that work if
iommu_lookup_page()is ASSERTing thatthe
Quoted 1 line
domain in question is not HVM?
Well, as soon as the function doesn't hand back MFNs anymore to HVM callers, no such assertion would be needed anymore either.
So you'd prefer I add an ASSERTion that I'm going to remove as soon as I add a caller of the function?
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 91 lines
Quoted 91 lines
Quoted 1 line · On 12.09.18 at 14:22, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 13:15
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 8 lines
Quoted 8 lines
Quoted 1 line · On 12.09.18 at 12:09, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 11:08
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 11 lines
Quoted 11 lines
Quoted 9 lines
Quoted 9 lines
Quoted 2 lines · On 12.09.18 at 11:30, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:21
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 1 line
Quoted 1 line
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 1 line · On 12.09.18 at 11:15, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:13
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <xen-Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Subject: RE: [PATCH v6 08/14] vtd: add
lookup_pagemethod to
iommu_opsQuoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines · On 12.09.18 at 11:05, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03A HVM guest using the PV IOMMU is quite fine, but it shouldn't
talk
Quoted 11 lines
Quoted 11 lines
to
Quoted 10 lines
Quoted 10 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
it in terms of MFNs.
Well, it has to talk MFNs at some level, surely? The output of the
IOMMU is
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
not subject to EPT/NPT, right?
Yes to the second question, but no to the first: The GFN -> MFN
translation
Quoted 1 line
Quoted 1 line
should still be done inside Xen in the HVM case, imo (in the course
of
Quoted 8 lines
Quoted 8 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 1 line
manufacturing the PTE).
Indeed. This function is very much internal to Xen (it's simply an abstraction on top of a vendor implementation), so why should it not
work
Quoted 3 lines
Quoted 3 lines
in
Quoted 1 line
terms of MFNs?
Quoted 8 lines
Quoted 8 lines
Quoted 6 lines
Quoted 4 lines
Because "MFN" is a concept a HVM guest is not knowing about, or supposed to be knowing. The only time where (part of) it might legitimately (have to) know is when it comes to managing the host (including any guests), i.e. in the tool stack of a PVH Dom0.
Ok. So consider validating a PV-IOMMU unmap request from an HVM
guest. It
Quoted 1 line
passes in a DFN and a GFN belonging to itself. Now Xen needs to figure
out
Quoted 1 line
Quoted 1 line
Quoted 1 line
whether that BFN actually maps to the GFN. It can look up the MFN
backing
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
GFN (from the p2m). How does Xen now validate it if it cannot lookup
what
Quoted 11 lines
Quoted 9 lines
MFN
Quoted 1 line
is actually present in the PTE referenced by the DFN?
I'm afraid I don't understand: The passed in GFN gets translated to an MFN using a p2m lookup. The passed in DFN (which aiui ought to match the GFN anyway on x86) gets translated to an MFN using an IOMMU page table lookup. The resulting two MFNs have to match for the request to be valid.
Quite. So how does that work if
iommu_lookup_page()is ASSERTing that
Quoted 8 lines
Quoted 5 lines
the
Quoted 1 line
domain in question is not HVM?
Well, as soon as the function doesn't hand back MFNs anymore to HVM callers, no such assertion would be needed anymore either.
So you'd prefer I add an ASSERTion that I'm going to remove as soon as I add a caller of the function?
No. I guess I'm increasingly confused: The function at present returns MFNs. Hence it must not be called by a HVM guest. Either you assert that the calling domain isn't HVM, or you make the function return GFNs for HVM domains (which then is a no-op due to gfn == dfn here, at least for now).
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 115 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 13:39
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 8 lines
Quoted 8 lines
Quoted 1 line · On 12.09.18 at 14:22, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 13:15
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 11 lines
Quoted 11 lines
Quoted 9 lines
Quoted 9 lines
Quoted 2 lines · On 12.09.18 at 12:09, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 11:08
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 1 line
Quoted 1 line
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
Quoted 1 line · On 12.09.18 at 11:30, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:21
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <xen-Quoted 11 lines
Quoted 11 lines
Quoted 10 lines
Quoted 10 lines
Quoted 1 line
Quoted 1 line
Subject: RE: [PATCH v6 08/14] vtd: add
lookup_pagemethod to
iommu_opsQuoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 2 lines · On 12.09.18 at 11:15, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:13
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <xen-Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 1 line
Quoted 1 line
Subject: RE: [PATCH v6 08/14] vtd: add
lookup_pagemethod to
iommu_opsQuoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines · On 12.09.18 at 11:05, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 10:03A HVM guest using the PV IOMMU is quite fine, but it
shouldn't
Quoted 6 lines
Quoted 6 lines
talk
Quoted 5 lines
Quoted 5 lines
to
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 2 lines
it in terms of MFNs.
Well, it has to talk MFNs at some level, surely? The output of
the
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
IOMMU is
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
not subject to EPT/NPT, right?
Yes to the second question, but no to the first: The GFN -> MFN
translation
Quoted 1 line
Quoted 1 line
should still be done inside Xen in the HVM case, imo (in the
course
Quoted 5 lines
Quoted 5 lines
of
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 1 line
manufacturing the PTE).
Indeed. This function is very much internal to Xen (it's simply an abstraction on top of a vendor implementation), so why should it
not
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
work
Quoted 9 lines
Quoted 7 lines
in
Quoted 1 line
terms of MFNs?
Because "MFN" is a concept a HVM guest is not knowing about, or supposed to be knowing. The only time where (part of) it might legitimately (have to) know is when it comes to managing the host (including any guests), i.e. in the tool stack of a PVH Dom0.
Ok. So consider validating a PV-IOMMU unmap request from an
HVM
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines
guest. It
Quoted 1 line
passes in a DFN and a GFN belonging to itself. Now Xen needs to
figure
Quoted 5 lines
Quoted 5 lines
out
Quoted 1 line
Quoted 1 line
Quoted 1 line
whether that BFN actually maps to the GFN. It can look up the MFN
backing
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
GFN (from the p2m). How does Xen now validate it if it cannot
lookup
Quoted 4 lines
Quoted 4 lines
what
Quoted 3 lines
Quoted 3 lines
MFN
Quoted 1 line
is actually present in the PTE referenced by the DFN?
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 6 lines
I'm afraid I don't understand: The passed in GFN gets translated to an MFN using a p2m lookup. The passed in DFN (which aiui ought to match the GFN anyway on x86) gets translated to an MFN using an IOMMU page table lookup. The resulting two MFNs have to match for the request to be valid.
Quite. So how does that work if
iommu_lookup_page()is ASSERTingthat
Quoted 6 lines
Quoted 5 lines
the
Quoted 1 line
domain in question is not HVM?
Well, as soon as the function doesn't hand back MFNs anymore to HVM callers, no such assertion would be needed anymore either.
Quoted 6 lines
Quoted 1 line
So you'd prefer I add an ASSERTion that I'm going to remove as soon as I
add
Quoted 1 line
a caller of the function?
No. I guess I'm increasingly confused: The function at present returns MFNs. Hence it must not be called by a HVM guest.
That's the part I don't get. What do you mean by 'called by a HVM guest'? I completely agree that MFN values must not be exposed to an HVM guest so there is no way the output of this function should ever be exposed through a hypercall and I'm not proposing that ever be done.
Quoted 5 lines
Either you assert
that the calling domain isn't HVM, or you make the function return GFNs for HVM domains (which then is a no-op due to gfn == dfn here, at least for now).
The function will never return its results to a guest, PV or HVM, so I really don't see the concern. It's a low level function, for Xen's internal use only. It's essentially the equivalent of a p2m lookup function and there's no way we'd ever expose the results of such a lookup to the guest either.
Paul
Quoted 1 line
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 6 lines
Quoted 1 line
Quoted 1 line · On 12.09.18 at 14:53, <[email protected]> wrote:
The function will never return its results to a guest, PV or HVM, so I really don't see the concern. It's a low level function, for Xen's internal use only. It's essentially the equivalent of a p2m lookup function and there's no way we'd ever expose the results of such a lookup to the guest either.
Oh, that was utter confusion on my part then, and I'm sorry for all the noise. I've got mislead by the titles of this patch and patches 13 and 14, all of which have "iommu_ops" as parts of them.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 15 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 14:20
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 6 lines
Quoted 1 line
Quoted 1 line · On 12.09.18 at 14:53, <[email protected]> wrote:
The function will never return its results to a guest, PV or HVM, so I really don't see the concern. It's a low level function, for Xen's internal use only. It's essentially the equivalent of a p2m lookup function and there's no way we'd ever expose the results of such a lookup to the guest either.
Quoted 4 lines
Oh, that was utter confusion on my part then, and I'm sorry for all the noise. I've got mislead by the titles of this patch and patches 13 and 14, all of which have "
iommu_ops" as parts of them.
I recently noticed this name overloading so sorry for leading you astray. I will seriously consider renaming the hypercall when I re-work the later patches.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 16 lines
Quoted 16 lines
Quoted 1 line · On 12.09.18 at 15:25, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 14:20
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 6 lines
Quoted 1 line
Quoted 1 line · On 12.09.18 at 14:53, <[email protected]> wrote:
The function will never return its results to a guest, PV or HVM, so I really don't see the concern. It's a low level function, for Xen's internal use only. It's essentially the equivalent of a p2m lookup function and there's no way we'd ever expose the results of such a lookup to the guest either.
Quoted 8 lines
Quoted 4 lines
Oh, that was utter confusion on my part then, and I'm sorry for all the noise. I've got mislead by the titles of this patch and patches 13 and 14, all of which have "
iommu_ops" as parts of them.I recently noticed this name overloading so sorry for leading you astray. I will seriously consider renaming the hypercall when I re-work the later patches.
The hypercall name is fine, I think. Perhaps the distinction could already be made more clear by using iommu_ops for the internal interface(s) (matching their structure name) and iommu_op / iommu_op-s for the hypercall based operations, seeing that the structure is names xen_iommu_op?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
Quoted 35 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 14:39
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 8 lines
Quoted 8 lines
Quoted 1 line · On 12.09.18 at 15:25, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 14:20
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod to
iommu_opsQuoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines · On 12.09.18 at 14:53, <[email protected]> wrote:
The function will never return its results to a guest, PV or HVM, so I really don't see the concern. It's a low level function, for Xen's internal use only. It's essentially the equivalent of a p2m lookup function and there's no way we'd ever expose the results of such a lookup to the
guest
Quoted 10 lines
Quoted 6 lines
Quoted 1 line
either.
Oh, that was utter confusion on my part then, and I'm sorry for all the noise. I've got mislead by the titles of this patch and patches 13 and 14, all of which have "
iommu_ops" as parts of them.I recently noticed this name overloading so sorry for leading you astray. I will seriously consider renaming the hypercall when I re-work the later patches.
Quoted 6 lines
The hypercall name is fine, I think. Perhaps the distinction could already be made more clear by using
iommu_opsfor the internal interface(s) (matching their structure name) andiommu_op/ iommu_op-s for the hypercall based operations, seeing that the structure is namesxen_iommu_op?
Ok. I'll leave the filenames as iommu_op.[ch] but I'll try to refer consistently to xen_iommu_op in comments and descriptions relating to the hypercalls and iommu_ops for the internal abstraction layer.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 08/14] vtd: add lookup_page method to iommu_ops
In reply to Paul DurrantQuoted 52 lines
Quoted 52 lines
Quoted 1 line · On 12.09.18 at 10:45, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 09:44
To: Paul Durrant <[email protected]>
Cc: George Dunlap <[email protected]>; Kevin Tian <[email protected]>; xen-devel <[email protected]>
Subject: RE: [PATCH v6 08/14] vtd: addlookup_pagemethod toiommu_opsQuoted 6 lines
Quoted 6 lines
Quoted 1 line · On 12.09.18 at 10:31, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:11Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds a new method to the VT-d IOMMU implementation to
find
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
MFN currently mapped by the specified BFN along with a wrapper
function
Quoted 8 lines
Quoted 8 lines
in
Quoted 1 line
generic IOMMU code to call the implementation if it exists.
For this to go in, I think the AMD side of it wants to also be implemented.
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d,
Quoted 3 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -305,6 +305,17 @@ int iommu_unmap_page(struct domain *d,
bfn_tQuoted 18 lines
Quoted 4 lines
bfn)
Quoted 3 lines
xen/drivers/passthrough/iommu.c +0
-0return rc; }Quoted 13 lines
xen/drivers/passthrough/iommu.c int iommu_unmap_page(struct domain *d, bfn_t bfn) · line 317
Quoted 10 lines
xen/drivers/passthrough/iommu.c +10
-0+int iommu_lookup_page (struct domain *d,bfn_t bfn,mfn_t *mfn, +unsigned int *flags) +{ +const struct domain_iommu *hd =dom_iommu (d); + +if ( !iommu_enabled || !hd->platform_ops ) +return -EOPNOTSUPP ; + +return hd->platform_ops->lookup_page (d, bfn, mfn, flags); +}Shouldn't this be restricted to PV guests? HVM ones aren't supposed to know about MFNs.
Quoted 2 lines
Agreed, but I think this is the wrong level to be applying such a check:
iommu_map_page()is supplied an MFN regardless of whether the domainis PV or
Quoted 2 lines
HVM, so I think it is reasonable for a lookup function to work in terms of MFNs.
Quoted 6 lines
Quoted 3 lines
I don't mind much where the check sits, but ASSERT(!is_hvm_domain()), if placed here, should not trigger.
It will though. I'm going to need to use this function for HVM guests after having done a GFN lookup.
That's the subject domain then, not the calling one.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 9/14 mm / iommu: include need_iommu() test in iommu_use_hap_pt()
In reply to Paul DurrantThe name 'iommu_use_hap_pt' suggests that that P2M table is in use as the domain's IOMMU pagetable which, prior to this patch, is not strictly true since the macro did not test whether the domain actually has IOMMU mappings.
---
Cc: Jun Nakajima <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Julien Grall <[email protected]>
v4:
- New in v4.
--- xen/arch/x86/mm/p2m-ept.c | 6 +++--- xen/arch/x86/mm/p2m-pt.c | 6 +++--- xen/arch/x86/mm/p2m.c | 2 +- xen/drivers/passthrough/iommu.c | 2 +- xen/include/asm-arm/iommu.h | 2 +- xen/include/asm-x86/iommu.h | 5 +++-- 6 files changed, 12 insertions(+), 11 deletions(-)
xen/arch/x86/mm/p2m-ept.c +3 -3
@@ -863,12 +863,12 @@ out:ept_sync_domain (p2m);/* For host p2m, may need to change VT-d page table.*/ -if ( rc ==0 &&p2m_is_hostp2m (p2m) &&need_iommu (d) && +if ( rc ==0 &&p2m_is_hostp2m (p2m) && need_modify_vtd_table ) { -if ( iommu_hap_pt_share ) +if (iommu_use_hap_pt (d) ) rc =iommu_pte_flush (d, gfn, &ept_entry->epte, order, vtd_pte_present); -else +else if (need_iommu (d) ) {bfn_t bfn =_bfn (gfn);
xen/arch/x86/mm/p2m-pt.c +3 -3
@@ -677,8 +677,8 @@ p2m_pt_set_entry(struct p2m_domain *p2m, gfn_t gfn_, mfn_t mfn, && (gfn + (1UL << page_order) - 1 > p2m->max_mapped_pfn) ) p2m->max_mapped_pfn = gfn + (1UL << page_order) - 1; -if ( iommu_enabled &&need_iommu (p2m->domain) && - (iommu_old_flags != iommu_pte_flags || old_mfn !=mfn_x (mfn)) ) +if ( iommu_enabled && (iommu_old_flags != iommu_pte_flags || + old_mfn !=mfn_x (mfn)) ) {ASSERT (rc ==0 );
@@ -687,7 +687,7 @@ p2m_pt_set_entry(struct p2m_domain *p2m, gfn_t gfn_, mfn_t mfn,if ( iommu_old_flags )amd_iommu_flush_pages (p2m->domain, gfn, page_order); } -else +else if (need_iommu (p2m->domain) ) {bfn_t bfn =_bfn (gfn);
xen/arch/x86/mm/p2m.c +1 -1
@@ -2090,7 +2090,7 @@ static unsigned int mmio_order(const struct domain *d,* - exclude PV guests, should execution reach this code for such. * So be careful when altering this. */ -if ( !need_iommu (d) || !iommu_use_hap_pt (d) || +if ( !iommu_use_hap_pt (d) || (start_fn & ((1UL <<PAGE_ORDER_2M ) -1 )) || !(nr >>PAGE_ORDER_2M ) )return PAGE_ORDER_4K ;
xen/drivers/passthrough/iommu.c +1 -1
@@ -453,7 +453,7 @@ int iommu_do_domctl(void iommu_share_p2m_table (struct domain * d) { -if ( iommu_enabled &&iommu_use_hap_pt (d) ) +if (iommu_use_hap_pt (d) )iommu_get_ops ()->share_p2m (d); }
xen/include/asm-arm/iommu.h +1 -1
@@ -21,7 +21,7 @@ struct arch_iommu };/* Always share P2M Table between the CPU and the IOMMU */ -#define iommu_use_hap_pt (d) (1) +#define iommu_use_hap_pt (d) (need_iommu(d))const struct iommu_ops *iommu_get_ops (void );void __init iommu_set_ops (const struct iommu_ops *ops);
xen/include/asm-x86/iommu.h +3 -2
@@ -77,8 +77,9 @@ static inline int iommu_hardware_setup(void)return -ENODEV ; } -/* Does this domain have a P2M table we can use as its IOMMU pagetable? */ -#define iommu_use_hap_pt (d) (hap_enabled(d) && iommu_hap_pt_share) +/* Are we using the domain P2M table as its IOMMU pagetable? */ +#define iommu_use_hap_pt (d) \ + (hap_enabled(d) && need_iommu(d) && iommu_hap_pt_share)void iommu_update_ire_from_apic (unsigned int apic,unsigned int reg,unsigned int value);unsigned int iommu_read_apic_from_ire (unsigned int apic,unsigned int reg);
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 09/14] mm / iommu: include need_iommu() test in iommu_use_hap_pt()
Quoted 5 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
The name '
iommu_use_hap_pt' suggests that that P2M table is in use as the domain's IOMMU pagetable which, prior to this patch, is not strictly true since the macro did not test whether the domain actually has IOMMU mappings.
Hmm, I would never have implied "has IOMMU mappings" from this variable name. To me it has always been "use HAP page tables for IOMMU if an IOMMU is in use". The code change looks sane, but I'm not sure it is a clear improvement. Hence I wonder whether you have a need for this change in subsequent patches which goes beyond what you say above.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 09/14] mm / iommu: include need_iommu() test in iommu_use_hap_pt()
Quoted 19 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:20
To: Paul Durrant <[email protected]>
Cc: Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Jun Nakajima <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [PATCH v6 09/14] mm / iommu: includeneed_iommu()test iniommu_use_hap_pt()Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
The name '
iommu_use_hap_pt' suggests that that P2M table is in use asthe
Quoted 3 lines
domain's IOMMU pagetable which, prior to this patch, is not strictly true since the macro did not test whether the domain actually has IOMMU mappings.
Quoted 7 lines
Hmm, I would never have implied "has IOMMU mappings" from this variable name. To me it has always been "use HAP page tables for IOMMU if an IOMMU is in use". The code change looks sane, but I'm not sure it is a clear improvement. Hence I wonder whether you have a need for this change in subsequent patches which goes beyond what you say above.
I could take it out but it would mean a non-trivial rebase, and to me - if true - the name sill implies the IOMMU is in use for the domain so I'd like to keep the change.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 09/14] mm / iommu: include need_iommu() test in iommu_use_hap_pt()
Quoted 28 lines
Quoted 20 lines
Quoted 1 line · On 11.09.18 at 11:39, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:20
To: Paul Durrant <[email protected]>
Cc: Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Jun Nakajima <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [PATCH v6 09/14] mm / iommu: includeneed_iommu()test iniommu_use_hap_pt()Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
The name '
iommu_use_hap_pt' suggests that that P2M table is in use asthe
Quoted 3 lines
domain's IOMMU pagetable which, prior to this patch, is not strictly true since the macro did not test whether the domain actually has IOMMU mappings.
Quoted 7 lines
Hmm, I would never have implied "has IOMMU mappings" from this variable name. To me it has always been "use HAP page tables for IOMMU if an IOMMU is in use". The code change looks sane, but I'm not sure it is a clear improvement. Hence I wonder whether you have a need for this change in subsequent patches which goes beyond what you say above.
Quoted 3 lines
I could take it out but it would mean a non-trivial rebase, and to me - if true - the name sill implies the IOMMU is in use for the domain so I'd like to keep the change.
Let's broaden the Cc list a little - perhaps we can get further opinions this way.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 09/14] mm / iommu: include need_iommu() test in iommu_use_hap_pt()
Quoted 34 lines
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 11, 2018 5:48 PMQuoted 11 lines
Quoted 11 lines
Quoted 1 line · On 11.09.18 at 11:39, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:20
To: Paul Durrant <[email protected]>
Cc: Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Jun Nakajima <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [PATCH v6 09/14] mm / iommu: includeneed_iommu()testin
Quoted 4 lines
Quoted 4 lines
iommu_use_hap_pt()Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
The name '
iommu_use_hap_pt' suggests that that P2M table is in useas
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
domain's IOMMU pagetable which, prior to this patch, is not strictly
true
Quoted 11 lines
Quoted 10 lines
Quoted 2 lines
since the macro did not test whether the domain actually has IOMMU mappings.
Hmm, I would never have implied "has IOMMU mappings" from this variable name. To me it has always been "use HAP page tables for IOMMU if an IOMMU is in use". The code change looks sane, but I'm not sure it is a clear improvement. Hence I wonder whether you have a need for this change in subsequent patches which goes beyond what you say above.
Quoted 8 lines
Quoted 2 lines
I could take it out but it would mean a non-trivial rebase, and to me - if true - the name sill implies the IOMMU is in use for the domain so I'd like
to
Quoted 1 line
keep the change.
Let's broaden the Cc list a little - perhaps we can get further opinions this way.
the previous definition checks hap_enabled:
-#define iommu_use_hap_pt(d) (hap_enabled(d) && iommu_hap_pt_share)
if following Jan's way it should be "use HAP page tables for IOMMU if an IOMMU is in use and HAP is enabled", then hap_enabled should be moved out too.
including iommu check in the macro looks more consistent.
Thanks
Kevin
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 09/14] mm / iommu: include need_iommu() test in iommu_use_hap_pt()
Quoted 59 lines
-----Original Message-----
From: Tian, Kevin [mailto:[email protected]]
Sent: 13 September 2018 07:24
To: Jan Beulich <[email protected]>; Paul Durrant <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Nakajima, Jun <[email protected]>; Stefano Stabellini <[email protected]>; xen- devel <[email protected]>
Subject: RE: [PATCH v6 09/14] mm / iommu: includeneed_iommu()test iniommu_use_hap_pt()Quoted 11 lines
From: Jan Beulich [mailto:[email protected]]
Sent: Tuesday, September 11, 2018 5:48 PMQuoted 8 lines
Quoted 8 lines
Quoted 1 line · On 11.09.18 at 11:39, <[email protected]> wrote:
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 07 September 2018 12:20
To: Paul Durrant <[email protected]>
Cc: Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Jun NakajimaQuoted 23 lines
Quoted 3 lines
Quoted 3 lines
Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>
Subject: Re: [PATCH v6 09/14] mm / iommu: includeneed_iommu()testin
Quoted 4 lines
Quoted 4 lines
iommu_use_hap_pt()Quoted 2 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
The name '
iommu_use_hap_pt' suggests that that P2M table is in useas
Quoted 2 lines
Quoted 2 lines
the
Quoted 1 line
domain's IOMMU pagetable which, prior to this patch, is not strictly
true
Quoted 11 lines
Quoted 10 lines
Quoted 2 lines
since the macro did not test whether the domain actually has IOMMU mappings.
Hmm, I would never have implied "has IOMMU mappings" from this variable name. To me it has always been "use HAP page tables for IOMMU if an IOMMU is in use". The code change looks sane, but I'm not sure it is a clear improvement. Hence I wonder whether you have a need for this change in subsequent patches which goes beyond what you say above.
Quoted 8 lines
Quoted 2 lines
I could take it out but it would mean a non-trivial rebase, and to me - if true - the name sill implies the IOMMU is in use for the domain so I'd like
to
Quoted 1 line
keep the change.
Let's broaden the Cc list a little - perhaps we can get further opinions this way.
the previous definition checks
hap_enabled:
Quoted 9 lines
-#define
iommu_use_hap_pt(d)(hap_enabled(d)&&iommu_hap_pt_share)if following Jan's way it should be "use HAP page tables for IOMMU if an IOMMU is in use and HAP is enabled", then
hap_enabledshould be moved out too.including iommu check in the macro looks more consistent.
Thanks. I'll keep this patch.
Paul
Quoted 2 lines
Thanks
Kevin
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 10/14 mm / iommu: split need_iommu() into has_iommu_pt() and need_iommu_pt_sync()
In reply to Paul DurrantThe name 'need_iommu()' is a little confusing as it suggests a domain needs to use the IOMMU but something might not be set up yet, when in fact it represents a tri-state value (not a boolean as might be expected) where -1 means 'IOMMU mappings being set up' and 1 means 'IOMMU mappings have been fully set up'.
Two different meanings are also inferred from the macro it in various places in the code:
- Some callers want to test whether a domain has IOMMU mappings at all
- Some callers want to test whether they need to synchronize the domain's P2M and IOMMU mappings
This patch replaces the 'need_iommu' tri-state value with a defined enumeration and adds a boolean flag 'need_sync' to separate these meanings, and places both of these in struct domain_iommu, rather than directly in struct domain.
This patch also creates two new boolean macros:
- 'has_iommu_pt()' evaluates to true if a domain has IOMMU mappings, even if they are still under construction.
- 'need_iommu_pt_sync()' evaluates to true if a domain requires explicit synchronization of the P2M and IOMMU mappings.
All callers of need_iommu() are then modified to use the macro appropriate to what they are trying to test.
NOTE: The test of need_iommu(d) in the AMD IOMMU initialization code has
been replaced with a test of iommu_dom0_strict since this more accurately reflects the meaning of the test and brings it into line with a similar test in the Intel VT-d code.
---
Cc: Stefano Stabellini <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Tamas K Lengyel <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>
Cc: Razvan Cojocaru <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>
Cc: Brian Woods <[email protected]>
v6:
- Deal with need_iommu being tri-state.
- Change the name of 'sync_iommu_pt' to 'need_iommu_pt_sync' and make sure it is set as soon as the mappings are under construction.
- Not adding Razvan's A-b because of change.
v4:
- New in v4.
--- xen/arch/arm/p2m.c | 2 +- xen/arch/x86/hvm/mtrr.c | 5 ++-- xen/arch/x86/mm.c | 2 +- xen/arch/x86/mm/mem_sharing.c | 2 +- xen/arch/x86/mm/p2m-ept.c | 2 +- xen/arch/x86/mm/p2m-pt.c | 2 +- xen/arch/x86/mm/p2m.c | 8 +++--- xen/arch/x86/mm/paging.c | 2 +- xen/arch/x86/x86_64/mm.c | 3 ++- xen/common/memory.c | 6 ++--- xen/common/vm_event.c | 2 +- xen/drivers/passthrough/amd/pci_amd_iommu.c | 2 +- xen/drivers/passthrough/device_tree.c | 21 ++++++++-------- xen/drivers/passthrough/iommu.c | 39 +++++++++++++++++++++-------- xen/drivers/passthrough/pci.c | 6 ++--- xen/drivers/passthrough/x86/iommu.c | 2 -- xen/include/asm-arm/grant_table.h | 2 +- xen/include/asm-arm/iommu.h | 2 +- xen/include/asm-x86/grant_table.h | 2 +- xen/include/asm-x86/iommu.h | 2 +- xen/include/xen/iommu.h | 17 +++++++++++++ xen/include/xen/sched.h | 9 +++---- 22 files changed, 87 insertions(+), 53 deletions(-)
xen/arch/arm/p2m.c +1 -1
@@ -955,7 +955,7 @@ static int __p2m_set_entry(struct p2m_domain *p2m,if (lpae_is_valid (orig_pte) && entry->p2m.base != orig_pte.p2m.base )p2m_free_entry (p2m, orig_pte, level); -if (need_iommu (p2m->domain) && +if (need_iommu_pt_sync (p2m->domain) && (lpae_is_valid (orig_pte) ||lpae_is_valid (*entry)) ) { rc = iommu_iotlb_flush(p2m->domain, _bfn(gfn_x(sgfn)),
xen/arch/x86/hvm/mtrr.c +3 -2
@@ -793,7 +793,8 @@ HVM_REGISTER_SAVE_RESTORE(MTRR, hvm_save_mtrr_msr, hvm_load_mtrr_msr,void memory_type_changed (struct domain *d) { -if (need_iommu (d) && d->vcpu && d->vcpu[0 ] ) +if ( (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d)) && + d->vcpu && d->vcpu[0 ] ) {p2m_memory_type_changed (d);flush_all (FLUSH_CACHE );
@@ -841,7 +842,7 @@ int epte_get_entry_emt(struct domain *d, unsigned long gfn, mfn_t mfn,return MTRR_TYPE_UNCACHABLE ; } -if ( !need_iommu (d) && !cache_flush_permitted (d) ) +if ( !has_iommu_pt (d) && !cache_flush_permitted (d) ) { *ipat =1 ;return MTRR_TYPE_WRBACK ;
xen/arch/x86/mm.c +1 -1
@@ -2750,7 +2750,7 @@ static int _get_page_type(struct page_info *page, unsigned long type, {/* Special pages should not be accessible from devices. */ struct domain *d =page_get_owner (page); -if ( d &&is_pv_domain (d) &&unlikely (need_iommu (d)) ) +if ( d &&is_pv_domain (d) &&unlikely (need_iommu_pt_sync (d)) ) {mfn_t mfn =page_to_mfn (page);
xen/arch/x86/mm/mem_sharing.c +1 -1
@@ -1610,7 +1610,7 @@ int mem_sharing_domctl(struct domain *d, struct xen_domctl_mem_sharing_op *mec)case XEN_DOMCTL_MEM_SHARING_CONTROL : { rc =0 ; -if (unlikely (need_iommu (d) && mec->u.enable) ) +if (unlikely (has_iommu_pt (d) && mec->u.enable) ) rc = -EXDEV ;else d->arch.hvm_domain.mem_sharing_enabled = mec->u.enable;
xen/arch/x86/mm/p2m-ept.c +1 -1
@@ -868,7 +868,7 @@ out: {if (iommu_use_hap_pt (d) ) rc =iommu_pte_flush (d, gfn, &ept_entry->epte, order, vtd_pte_present); -else if (need_iommu (d) ) +else if (need_iommu_pt_sync (d) ) {bfn_t bfn =_bfn (gfn);
xen/arch/x86/mm/p2m-pt.c +1 -1
@@ -687,7 +687,7 @@ p2m_pt_set_entry(struct p2m_domain *p2m, gfn_t gfn_, mfn_t mfn,if ( iommu_old_flags )amd_iommu_flush_pages (p2m->domain, gfn, page_order); } -else if (need_iommu (p2m->domain) ) +else if (need_iommu_pt_sync (p2m->domain) ) {bfn_t bfn =_bfn (gfn);
xen/arch/x86/mm/p2m.c +4 -4
@@ -713,7 +713,7 @@ p2m_remove_page(struct p2m_domain *p2m, unsigned long gfn_l, unsigned long mfn, {int rc =0 ; -if (need_iommu (p2m->domain) ) +if (need_iommu_pt_sync (p2m->domain) ) {bfn_t bfn =_bfn (mfn);
@@ -777,7 +777,7 @@ guest_physmap_add_entry(struct domain *d, gfn_t gfn, mfn_t mfn,if ( !paging_mode_translate (d) ) { -if (need_iommu (d) && t == p2m_ram_rw ) +if (need_iommu_pt_sync (d) && t == p2m_ram_rw ) {bfn_t bfn =_bfn (mfn_x (mfn));
@@ -1163,7 +1163,7 @@ int set_identity_p2m_entry(struct domain *d, unsigned long gfn_l,if ( !paging_mode_translate (d) ) { -if ( !need_iommu (d) ) +if ( !need_iommu_pt_sync (d) )return 0 ; ret = iommu_map_page(d, _bfn(gfn_l), _mfn(gfn_l),
@@ -1259,7 +1259,7 @@ int clear_identity_p2m_entry(struct domain *d, unsigned long gfn_l)if ( !paging_mode_translate (d) ) { -if ( !need_iommu (d) ) +if ( !need_iommu_pt_sync (d) )return 0 ; ret =iommu_unmap_page (d,_bfn (gfn_l));
xen/arch/x86/mm/paging.c +1 -1
@@ -213,7 +213,7 @@ int paging_log_dirty_enable(struct domain *d, bool_t log_global) {int ret; -if (need_iommu (d) && log_global ) +if (has_iommu_pt (d) && log_global ) {/* * Refuse to turn on global log-dirty mode
xen/arch/x86/x86_64/mm.c +2 -1
@@ -1426,7 +1426,8 @@ int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm)if ( ret ) goto destroy_m2p; -if ( iommu_enabled && !iommu_passthrough && !need_iommu (hardware_domain) ) +if ( iommu_enabled && !iommu_passthrough && + !need_iommu_pt_sync (hardware_domain) ) {for ( i = spfn; i < epfn; i++ ) if ( iommu_map_page(hardware_domain, _bfn(i), _mfn(i),
xen/common/memory.c +3 -3
@@ -805,8 +805,8 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, xatp->size -= start;#ifdef CONFIG_HAS_PASSTHROUGH -if (need_iommu (d) ) -this_cpu (iommu_dont_flush_iotlb) =1 ; +if (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d) ) +this_cpu (iommu_dont_flush_iotlb) =1 ;#endif while ( xatp->size > done )
@@ -828,7 +828,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, }#ifdef CONFIG_HAS_PASSTHROUGH -if (need_iommu (d) ) +if (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d) ) {int ret;
xen/common/vm_event.c +1 -1
@@ -644,7 +644,7 @@ int vm_event_domctl(struct domain *d, struct xen_domctl_vm_event_op *vec,/* No paging if iommu is used */ rc = -EMLINK ; -if (unlikely (need_iommu (d)) ) +if (unlikely (has_iommu_pt (d)) )break ; rc = -EXDEV ;
xen/drivers/passthrough/amd/pci_amd_iommu.c +1 -1
@@ -256,7 +256,7 @@ static void __hwdom_init amd_iommu_hwdom_init(struct domain *d)if (allocate_domain_resources (dom_iommu (d)) )BUG (); -if ( !iommu_passthrough && !need_iommu (d) ) +if ( !iommu_passthrough && !iommu_dom0_strict ) {int rc =0 ;
xen/drivers/passthrough/device_tree.c +10 -11
@@ -40,17 +40,16 @@ int iommu_assign_dt_device(struct domain *d, struct dt_device_node *dev)if ( !list_empty (&dev->domain_list) ) goto fail; -if (need_iommu (d) <=0 ) - { -/* -* The hwdom is forced to use IOMMU for protecting assigned -* device. Therefore the IOMMU data is already set up. -*/ -ASSERT (!is_hardware_domain (d)); - rc =iommu_construct (d); -if ( rc ) - goto fail; - } +/* +* The hwdom is forced to use IOMMU for protecting assigned +* device. Therefore the IOMMU data is already set up. +*/ +ASSERT (!is_hardware_domain (d) || + hd->status == IOMMU_STATUS_initialized); + + rc =iommu_construct (d); +if ( rc ) + goto fail;/* The flag field doesn't matter to DT device. */ rc = hd->platform_ops->assign_device (d,0 ,dt_to_dev (dev),0 );
xen/drivers/passthrough/iommu.c +29 -10
@@ -171,7 +171,7 @@ static void __hwdom_init check_hwdom_reqs(struct domain *d) void __hwdom_init iommu_hwdom_init(struct domain *d) { -const struct domain_iommu *hd =dom_iommu (d); +struct domain_iommu *hd =dom_iommu (d);check_hwdom_reqs (d);
@@ -179,8 +179,10 @@ void __hwdom_init iommu_hwdom_init(struct domain *d)return ;register_keyhandler ('o' , &iommu_dump_p2m_table,"dump iommu p2m table" ,0 ); - d->need_iommu = !!iommu_dom0_strict; -if (need_iommu (d) && !iommu_use_hap_pt (d) ) + + hd->status = IOMMU_STATUS_initializing; + hd->need_sync = !!iommu_dom0_strict && !iommu_use_hap_pt (d); +if (need_iommu_pt_sync (d) ) {struct page_info *page;unsigned int i =0 ;
@@ -213,35 +215,51 @@ void __hwdom_init iommu_hwdom_init(struct domain *d) } hd->platform_ops->hwdom_init (d); + + hd->status = IOMMU_STATUS_initialized; }void iommu_teardown (struct domain *d) { -const struct domain_iommu *hd =dom_iommu (d); +struct domain_iommu *hd =dom_iommu (d); - d->need_iommu =0 ; + hd->status = IOMMU_STATUS_disabled; hd->platform_ops->teardown (d);tasklet_schedule (&iommu_pt_cleanup_tasklet); }int iommu_construct (struct domain *d) { -if (need_iommu (d) >0 ) +struct domain_iommu *hd =dom_iommu (d); + +if ( hd->status == IOMMU_STATUS_initialized )return 0 ; -if ( !iommu_use_hap_pt (d) ) +if ( !iommu_use_hap_pt (d) && hd->status < IOMMU_STATUS_initialized ) {int rc; + hd->status = IOMMU_STATUS_initializing; + hd->need_sync = true; + rc =arch_iommu_populate_page_table (d);if ( rc ) + { +if ( rc != -ERESTART ) + { + hd->need_sync = false; + hd->status = IOMMU_STATUS_disabled; + } +return rc; + } } - d->need_iommu =1 ; + hd->status = IOMMU_STATUS_initialized; +/* * There may be dirty cache lines when a device is assigned -* and before need_iommu(d) becoming true, this will cause +* and before has_iommu_pt(d) becoming true, this will cause * memory_type_changed lose effect if memory type changes. * Call memory_type_changed here to amend this. */
@@ -500,7 +518,8 @@ static void iommu_dump_p2m_table(unsigned char key) ops =iommu_get_ops ();for_each_domain (d) { -if (is_hardware_domain (d) ||need_iommu (d) <=0 ) +if (is_hardware_domain (d) || +dom_iommu (d)->status < IOMMU_STATUS_initialized )continue ;if (iommu_use_hap_pt (d) )
xen/drivers/passthrough/pci.c +3 -3
@@ -1416,7 +1416,7 @@ static int assign_device(struct domain *d, u16 seg, u8 bus, u8 devfn, u32 flag)/* Prevent device assign if mem paging or mem sharing have been * enabled for this domain */ -if (unlikely (!need_iommu (d) && +if (unlikely (!has_iommu_pt (d) && (d->arch.hvm_domain.mem_sharing_enabled ||vm_event_check_ring (d->vm_event_paging) ||p2m_get_hostp2m (d)->global_logdirty)) )
@@ -1460,7 +1460,7 @@ static int assign_device(struct domain *d, u16 seg, u8 bus, u8 devfn, u32 flag) } done: -if ( !has_arch_pdevs (d) &&need_iommu (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) )iommu_teardown (d);pcidevs_unlock ();
@@ -1510,7 +1510,7 @@ int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn) pdev->fault.count =0 ; -if ( !has_arch_pdevs (d) &&need_iommu (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) )iommu_teardown (d);return ret;
xen/drivers/passthrough/x86/iommu.c +0 -2
@@ -45,8 +45,6 @@ int arch_iommu_populate_page_table(struct domain *d)struct page_info *page;int rc =0 , n =0 ; - d->need_iommu =-1 ; -this_cpu (iommu_dont_flush_iotlb) =1 ;spin_lock (&d->page_alloc_lock);
xen/include/asm-arm/grant_table.h +1 -1
@@ -92,7 +92,7 @@ static inline unsigned int gnttab_dom0_max(void) gfn_x(((i) >= nr_status_frames(t)) ? INVALID_GFN : (t)->arch.status_gfn[i])#define gnttab_need_iommu_mapping (d) \ - (is_domain_direct_mapped(d) && need_iommu(d)) + (is_domain_direct_mapped(d) && need_iommu_pt_sync(d))#endif /* __ASM_GRANT_TABLE_H__ */ /*
xen/include/asm-arm/iommu.h +1 -1
@@ -21,7 +21,7 @@ struct arch_iommu };/* Always share P2M Table between the CPU and the IOMMU */ -#define iommu_use_hap_pt (d) (need_iommu(d)) +#define iommu_use_hap_pt (d) (has_iommu_pt(d))const struct iommu_ops *iommu_get_ops (void );void __init iommu_set_ops (const struct iommu_ops *ops);
xen/include/asm-x86/grant_table.h +1 -1
@@ -99,6 +99,6 @@ static inline void gnttab_clear_flag(unsigned int nr, uint16_t *st)#define gnttab_release_host_mappings (domain) ( paging_mode_external(domain) )#define gnttab_need_iommu_mapping (d) \ - (!paging_mode_translate(d) && need_iommu(d)) + (!paging_mode_translate(d) && need_iommu_pt_sync(d))#endif /* __ASM_GRANT_TABLE_H__ */
xen/include/asm-x86/iommu.h +1 -1
@@ -79,7 +79,7 @@ static inline int iommu_hardware_setup(void)/* Are we using the domain P2M table as its IOMMU pagetable? */ #define iommu_use_hap_pt (d) \ - (hap_enabled(d) && need_iommu(d) && iommu_hap_pt_share) + (hap_enabled(d) && has_iommu_pt(d) && iommu_hap_pt_share)void iommu_update_ire_from_apic (unsigned int apic,unsigned int reg,unsigned int value);unsigned int iommu_read_apic_from_ire (unsigned int apic,unsigned int reg);
xen/include/xen/iommu.h +17 -0
@@ -111,6 +111,13 @@ enum iommu_featurebool_t iommu_has_feature (struct domain *d,enum iommu_feature feature); +enum iommu_status +{ + IOMMU_STATUS_disabled, + IOMMU_STATUS_initializing, + IOMMU_STATUS_initialized +}; +struct domain_iommu {struct arch_iommu arch;
@@ -130,6 +137,16 @@ struct domain_iommu {* must not be modified after initialization. */ struct rangeset *reserved_ranges; + +/* Status of guest IOMMU mappings */ +enum iommu_status status; + +/* +* Does the guest reqire mappings to be synchonized, to maintain +* the default bfn == pfn map. (See comment on bfn at the top of +* include/xen/mm.h). +*/ +bool need_sync; };#define dom_iommu (d) (&(d)->iommu)
xen/include/xen/sched.h +4 -5
@@ -371,9 +371,6 @@ struct domain#ifdef CONFIG_HAS_PASSTHROUGH struct domain_iommu iommu; - -/* Does this guest need iommu mappings (-1 meaning "being set up")? */ -s8 need_iommu;#endif /* is node-affinity automatically computed? */ bool auto_node_affinity;
@@ -889,9 +886,11 @@ void watchdog_domain_destroy(struct domain *d);#define is_pinned_vcpu (v) ((v)->domain->is_pinned || \ cpumask_weight((v)->cpu_hard_affinity) == 1)#ifdef CONFIG_HAS_PASSTHROUGH -#define need_iommu (d) ((d)->need_iommu) +#define has_iommu_pt (d) (dom_iommu(d)->status != IOMMU_STATUS_disabled) +#define need_iommu_pt_sync (d) (dom_iommu(d)->need_sync)#else -#define need_iommu (d) (0) +#define has_iommu_pt (d) (0) +#define need_iommu_pt_sync (d) (0)#endif static inline bool is_vcpu_online (const struct vcpu *v)
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 10/14] mm / iommu: split need_iommu() into has_iommu_pt() and need_iommu_pt_sync()
Quoted 27 lines · Paul Durrant
The name '
need_iommu()' is a little confusing as it suggests a domain needs to use the IOMMU but something might not be set up yet, when in fact it represents a tri-state value (not a boolean as might be expected) where -1 means 'IOMMU mappings being set up' and 1 means 'IOMMU mappings have been fully set up'.Two different meanings are also inferred from the macro it in various places in the code:
- Some callers want to test whether a domain has IOMMU mappings at all
- Some callers want to test whether they need to synchronize the domain's P2M and IOMMU mappingsThis patch replaces the '
need_iommu' tri-state value with a defined enumeration and adds a boolean flag 'need_sync' to separate these meanings, and places both of these in structdomain_iommu, rather than directly in struct domain.
This patch also creates two new boolean macros:- '
has_iommu_pt()' evaluates to true if a domain has IOMMU mappings, even if they are still under construction.
- 'need_iommu_pt_sync()' evaluates to true if a domain requires explicit synchronization of the P2M and IOMMU mappings.All callers of
need_iommu()are then modified to use the macro appropriate to what they are trying to test.
Quoted 6 lines
NOTE: The test of
need_iommu(d)in the AMD IOMMU initialization code has
been replaced with a test ofiommu_dom0_strictsince this more accurately reflects the meaning of the test and brings it into line with a similar test in the Intel VT-d code.Signed-off-by Paul Durrant <[email protected]>
For the vm_event part:
Thanks,
Razvan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 10/14] mm / iommu: split need_iommu() into has_iommu_pt() and need_iommu_pt_sync()
In reply to Paul Durrantxen/arch/x86/hvm/mtrr.c HVM_REGISTER_SAVE_RESTORE(MTRR, hvm_save_mtrr_msr, hvm_load_mtrr_msr, · line 800
Quoted 13 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/arch/x86/hvm/mtrr.c +2
-1@@ -793,7 +793,8 @@ HVM_REGISTER_SAVE_RESTORE(MTRR, hvm_save_mtrr_msr, hvm_load_mtrr_msr,void memory_type_changed (struct domain *d) { -if (need_iommu (d) && d->vcpu && d->vcpu[0 ] ) +if ( (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d)) && + d->vcpu && d->vcpu[0 ] ) {p2m_memory_type_changed (d);flush_all (FLUSH_CACHE );
How does need_iommu_pt_sync() come into play here? To me, has_iommu_pt() would seem to be a better match. The question here solely is (iirc) whether memory types take any effect in the first place (or else we can skip adjusting them), which in turn solely depends on whether there are any pass-through devices in the domain. This is along the lines of ...
xen/arch/x86/hvm/mtrr.c int epte_get_entry_emt(struct domain *d, unsigned long gfn, mfn_t mfn, · line 848
Quoted 9 lines
xen/arch/x86/hvm/mtrr.c +1
-1@@ -841,7 +842,7 @@ int epte_get_entry_emt(struct domain *d, unsigned long gfn, mfn_t mfn,return MTRR_TYPE_UNCACHABLE ; } -if ( !need_iommu (d) && !cache_flush_permitted (d) ) +if ( !has_iommu_pt (d) && !cache_flush_permitted (d) ) { *ipat =1 ;return MTRR_TYPE_WRBACK ;
... this.
xen/arch/x86/x86_64/mm.c int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm) · line 1433
Quoted 12 lines
xen/arch/x86/x86_64/mm.c +2
-1@@ -1426,7 +1426,8 @@ int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm)if ( ret ) goto destroy_m2p; -if ( iommu_enabled && !iommu_passthrough && !need_iommu (hardware_domain) ) +if ( iommu_enabled && !iommu_passthrough && + !need_iommu_pt_sync (hardware_domain) ) {for ( i = spfn; i < epfn; i++ ) if ( iommu_map_page(hardware_domain, _bfn(i), _mfn(i),
I'm confused - the condition you change looks to be inverted. Wouldn't we better fix this?
And then I again wonder whether you've chosen the right predicate:
Where would the equivalent mappings come from in the opposite case?
xen/common/memory.c int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, · line 810
Quoted 11 lines
xen/common/memory.c +2
-2@@ -805,8 +805,8 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, xatp->size -= start;#ifdef CONFIG_HAS_PASSTHROUGH -if (need_iommu (d) ) -this_cpu (iommu_dont_flush_iotlb) =1 ; +if (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d) ) +this_cpu (iommu_dont_flush_iotlb) =1 ;#endif
Rather than making the conditional more complicated, perhaps simply drop it (and move the reset-to-false code out of ...
xen/common/memory.c int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, · line 833
Quoted 8 lines
xen/common/memory.c +1
-1@@ -828,7 +828,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, }#ifdef CONFIG_HAS_PASSTHROUGH -if (need_iommu (d) ) +if (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d) ) {int ret;
... this if())?
Also it looks to me as if here you've got confused by the meaning you've assigned to need_iommu_pt_sync(): According to the description, it is about sync-ing of page tables. Here, however, we're checking whether to flush TLBs.
xen/common/memory.c void __hwdom_init iommu_hwdom_init(struct domain *d) · line 184
Quoted 9 lines
xen/common/memory.c +3
-2@@ -179,8 +179,10 @@ void __hwdom_init iommu_hwdom_init(struct domain *d)return ;register_keyhandler ('o' , &iommu_dump_p2m_table,"dump iommu p2m table" ,0 ); - d->need_iommu = !!iommu_dom0_strict; -if (need_iommu (d) && !iommu_use_hap_pt (d) ) + + hd->status = IOMMU_STATUS_initializing; + hd->need_sync = !!iommu_dom0_strict && !iommu_use_hap_pt (d);
Pointless !! afaict.
xen/common/memory.c void __hwdom_init iommu_hwdom_init(struct domain *d)
Quoted 10 lines
xen/common/memory.c +4
-2int iommu_construct (struct domain *d) { -if (need_iommu (d) >0 ) +struct domain_iommu *hd =dom_iommu (d); + +if ( hd->status == IOMMU_STATUS_initialized )return 0 ; -if ( !iommu_use_hap_pt (d) ) +if ( !iommu_use_hap_pt (d) && hd->status < IOMMU_STATUS_initialized )
Isn't the addition here redundant with the earlier if()?
xen/common/memory.c void watchdog_domain_destroy(struct domain *d); · line 893
Quoted 11 lines
xen/common/memory.c +4
-2@@ -889,9 +886,11 @@ void watchdog_domain_destroy(struct domain *d);#define is_pinned_vcpu (v) ((v)->domain->is_pinned || \ cpumask_weight((v)->cpu_hard_affinity) == 1)#ifdef CONFIG_HAS_PASSTHROUGH -#define need_iommu (d) ((d)->need_iommu) +#define has_iommu_pt (d) (dom_iommu(d)->status != IOMMU_STATUS_disabled) +#define need_iommu_pt_sync (d) (dom_iommu(d)->need_sync)#else -#define need_iommu (d) (0) +#define has_iommu_pt (d) (0) +#define need_iommu_pt_sync (d) (0)
Please use false here (and no need for the parentheses).
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 10/14] mm / iommu: split need_iommu() into has_iommu_pt() and need_iommu_pt_sync()
Quoted 33 lines
-----Original Message-----
From: Xen-devel [mailto:[email protected]] On Behalf Of Jan Beulich
Sent: 11 September 2018 15:31
To: Paul Durrant <[email protected]>
Cc: Kevin Tian <[email protected]>; Stefano Stabellini <[email protected]>; Wei Liu <[email protected]>; Jun Nakajima <[email protected]>; Razvan Cojocaru <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; George Dunlap <[email protected]>; Andrew Cooper <[email protected]>; Ian Jackson <[email protected]>; Tim (Xen.org) <[email protected]>; Julien Grall <[email protected]>; Tamas K Lengyel <[email protected]>; Suravee Suthikulpanit <[email protected]>; xen-devel <xen- [email protected]>; Brian Woods <[email protected]>
Subject: Re: [Xen-devel] [PATCH v6 10/14] mm / iommu: splitneed_iommu()intohas_iommu_pt()andneed_iommu_pt_sync()xen/arch/x86/hvm/mtrr.c HVM_REGISTER_SAVE_RESTORE(MTRR,
Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/arch/x86/hvm/mtrr.c +0
-0@@ -793,7 +793,8 @@ HVM_REGISTER_SAVE_RESTORE(MTRR,
hvm_save_mtrr_msr,hvm_load_mtrr_msr,xen/arch/x86/hvm/mtrr.c HVM_REGISTER_SAVE_RESTORE(MTRR, hvm_save_mtrr_msr, hvm_load_mtrr_msr, · line 800
Quoted 9 lines
xen/arch/x86/hvm/mtrr.c +2
-1void memory_type_changed (struct domain *d) { -if (need_iommu (d) && d->vcpu && d->vcpu[0 ] ) +if ( (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d)) && + d->vcpu && d->vcpu[0 ] ) {p2m_memory_type_changed (d);flush_all (FLUSH_CACHE );
Quoted 2 lines
How does
need_iommu_pt_sync()come into play here? To me,has_iommu_pt()would seem to be a better match.
Ok.
Quoted 28 lines
The question
here solely is (iirc) whether memory types take any effect in the first place (or else we can skip adjusting them), which in turn solely depends on whether there are any pass-through devices in the domain. This is along the lines of ...xen/arch/x86/hvm/mtrr.c int epte_get_entry_emt(struct domain *d,
Quoted 1 line
xen/arch/x86/hvm/mtrr.c +0
-0@@ -841,7 +842,7 @@ int epte_get_entry_emt(struct domain *d,unsigned long gfn,
mfn_tmfn,xen/arch/x86/hvm/mtrr.c int epte_get_entry_emt(struct domain *d, unsigned long gfn, mfn_t mfn, · line 848
Quoted 8 lines
xen/arch/x86/hvm/mtrr.c +1
-1return MTRR_TYPE_UNCACHABLE ; } -if ( !need_iommu (d) && !cache_flush_permitted (d) ) +if ( !has_iommu_pt (d) && !cache_flush_permitted (d) ) { *ipat =1 ;return MTRR_TYPE_WRBACK ;... this.
xen/arch/x86/x86_64/mm.c int memory_add(unsigned long spfn, unsigned
Quoted 3 lines
xen/arch/x86/x86_64/mm.c +0
-0@@ -1426,7 +1426,8 @@ int memory_add(unsigned long spfn, unsignedlong epfn, unsigned int pxm)
xen/arch/x86/x86_64/mm.c int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm)
Quoted 4 lines
xen/arch/x86/x86_64/mm.c +0
-1if ( ret ) goto destroy_m2p; - if ( iommu_enabled && !iommu_passthrough &&!need_iommu(hardware_domain) )
Quoted 8 lines
xen/arch/x86/x86_64/mm.c int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm) · line 1433
Quoted 5 lines
xen/arch/x86/x86_64/mm.c +2
-0+if ( iommu_enabled && !iommu_passthrough && + !need_iommu_pt_sync (hardware_domain) ) {for ( i = spfn; i < epfn; i++ ) if ( iommu_map_page(hardware_domain, _bfn(i), _mfn(i),I'm confused - the condition you change looks to be inverted. Wouldn't we better fix this?
I don't think it is inverted. I think this is to add new hotplugged memory to the 1:1 map in the case that dom0 is not in strict mode. I could be wrong. George?
Quoted 4 lines
And then I again wonder whether you've chosen the right predicate:
Where would the equivalent mappings come from in the opposite case?
If dom0 is in strict mode then I assume that the synchronization is handled when the calls are made to add memory into the p2m (which IIRC happens even for PV guests). My aim for this patch is to avoid any visible functional change.
Quoted 26 lines
xen/common/memory.c int xenmem_add_to_physmap(struct domain *d,
Quoted 3 lines
xen/common/memory.c +0
-0@@ -805,8 +805,8 @@ int xenmem_add_to_physmap(struct domain *d,struct
xen_add_to_physmap*xatp,xen/common/memory.c int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, · line 810
Quoted 8 lines
xen/common/memory.c +2
-2xatp->size -= start;#ifdef CONFIG_HAS_PASSTHROUGH -if (need_iommu (d) ) -this_cpu (iommu_dont_flush_iotlb) =1 ; +if (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d) ) +this_cpu (iommu_dont_flush_iotlb) =1 ;#endif Rather than making the conditional more complicated, perhaps simply drop it (and move the reset-to-false code out of ...
xen/common/memory.c int xenmem_add_to_physmap(struct domain *d,
Quoted 1 line
xen/common/memory.c +0
-0@@ -828,7 +828,7 @@ int xenmem_add_to_physmap(struct domain *d,struct
xen_add_to_physmap*xatp,xen/common/memory.c int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, · line 833
Quoted 7 lines
xen/common/memory.c +1
-1}#ifdef CONFIG_HAS_PASSTHROUGH -if (need_iommu (d) ) +if (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d) ) {int ret;
Quoted 7 lines
... this
if())?Also it looks to me as if here you've got confused by the meaning you've assigned to
need_iommu_pt_sync(): According to the description, it is about sync-ing of page tables. Here, however, we're checking whether to flush TLBs.
Yes, I may be confused here but it would seem to me that flushing the IOTLB would be necessary even in the case where the page tables are shared. I'll check the logic again.
Quoted 5 lines
xen/common/memory.c void __hwdom_init iommu_hwdom_init(struct
Quoted 1 line
xen/common/memory.c +0
-0@@ -179,8 +179,10 @@ void __hwdom_init iommu_hwdom_init(structdomain *d)
xen/common/memory.c void __hwdom_init iommu_hwdom_init(struct · line 181
Quoted 3 lines
xen/common/memory.c +0
-0return ; register_keyhandler('o', &iommu_dump_p2m_table, "dump iommu
Quoted 9 lines
p2m table", 0);
xen/common/memory.c void __hwdom_init iommu_hwdom_init(struct · line 184
Quoted 5 lines
xen/common/memory.c +3
-2- d->need_iommu = !!iommu_dom0_strict; -if (need_iommu (d) && !iommu_use_hap_pt (d) ) + + hd->status = IOMMU_STATUS_initializing; + hd->need_sync = !!iommu_dom0_strict && !iommu_use_hap_pt (d);Pointless !! afaict.
Ok, I'll drop it.
Quoted 8 lines
Quoted 8 lines
xen/common/memory.c +3
-1int iommu_construct (struct domain *d) { -if (need_iommu (d) >0 ) +struct domain_iommu *hd =dom_iommu (d); + +if ( hd->status == IOMMU_STATUS_initialized )return 0 ;
Quoted 6 lines
xen/common/memory.c
Quoted 2 lines
xen/common/memory.c +1
-1-if ( !iommu_use_hap_pt (d) ) + if ( !iommu_use_hap_pt(d) && hd->status <
IOMMU_STATUS_initialized)Isn't the addition here redundant with the earlier
if()?
Yes, it does appear to be.
Quoted 8 lines
xen/common/memory.c void watchdog_domain_destroy(struct domain
Quoted 1 line
xen/common/memory.c +0
-0@@ -889,9 +886,11 @@ void watchdog_domain_destroy(struct domain*d);
xen/common/memory.c void watchdog_domain_destroy(struct domain · line 889
Quoted 5 lines
xen/common/memory.c +1
-1#define is_pinned_vcpu (v) ((v)->domain->is_pinned || \ cpumask_weight((v)->cpu_hard_affinity) == 1)#ifdef CONFIG_HAS_PASSTHROUGH -#define need_iommu (d) ((d)->need_iommu) +#define has_iommu_pt (d) (dom_iommu(d)->status !=
IOMMU_STATUS_disabled)
Quoted 8 lines
xen/common/memory.c void watchdog_domain_destroy(struct domain · line 893
Quoted 5 lines
xen/common/memory.c +3
-1+#define need_iommu_pt_sync (d) (dom_iommu(d)->need_sync)#else -#define need_iommu (d) (0) +#define has_iommu_pt (d) (0) +#define need_iommu_pt_sync (d) (0)Please use false here (and no need for the parentheses).
Ok.
Paul
Quoted 7 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 10/14] mm / iommu: split need_iommu() into has_iommu_pt() and need_iommu_pt_sync()
Quoted 21 lines
Quoted 21 lines
Quoted 1 line · On 11.09.18 at 17:40, <[email protected]> wrote:
From: Xen-devel [mailto:[email protected]] On Behalf Of Jan Beulich
Sent: 11 September 2018 15:31xen/arch/x86/x86_64/mm.c int memory_add(unsigned long spfn, unsigned
Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/arch/x86/x86_64/mm.c +0
-0@@ -1426,7 +1426,8 @@ int memory_add(unsigned long spfn, unsignedlong epfn, unsigned int pxm)
xen/arch/x86/x86_64/mm.c int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm)
Quoted 4 lines
xen/arch/x86/x86_64/mm.c +0
-1if ( ret ) goto destroy_m2p; - if ( iommu_enabled && !iommu_passthrough &&!need_iommu(hardware_domain) )
Quoted 5 lines
xen/arch/x86/x86_64/mm.c +2
-0+if ( iommu_enabled && !iommu_passthrough && + !need_iommu_pt_sync (hardware_domain) ) {for ( i = spfn; i < epfn; i++ ) if ( iommu_map_page(hardware_domain, _bfn(i), _mfn(i),
Quoted 5 lines
Quoted 2 lines
I'm confused - the condition you change looks to be inverted. Wouldn't we better fix this?
I don't think it is inverted. I think this is to add new hotplugged memory to the 1:1 map in the case that dom0 is not in strict mode. I could be wrong.
Oh, I think you're right. It is just rather confusing to see an iommu_map_page() call qualified by !need_iommu(). But that's as confusing (to me) as the setup logic for Dom0's page tables.
Quoted 6 lines
Quoted 2 lines
And then I again wonder whether you've chosen the right predicate:
Where would the equivalent mappings come from in the opposite case?If dom0 is in strict mode then I assume that the synchronization is handled when the calls are made to add memory into the p2m (which IIRC happens even for PV guests).
Right you are.
Quoted 1 line
My aim for this patch is to avoid any visible functional change.
Sure - I didn't mean anything here (if at all) to be done in this patch (or perhaps even series), I've merely noticed this as an apparent oddity (which if I were right would perhaps better have been fixed before your transformations).
Quoted 28 lines
Quoted 28 lines
xen/common/memory.c int xenmem_add_to_physmap(struct domain *d,
Quoted 3 lines
xen/common/memory.c +0
-0@@ -805,8 +805,8 @@ int xenmem_add_to_physmap(struct domain *d,struct
xen_add_to_physmap*xatp,xen/common/memory.c int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, · line 810
Quoted 8 lines
xen/common/memory.c +2
-2xatp->size -= start;#ifdef CONFIG_HAS_PASSTHROUGH -if (need_iommu (d) ) -this_cpu (iommu_dont_flush_iotlb) =1 ; +if (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d) ) +this_cpu (iommu_dont_flush_iotlb) =1 ;#endif Rather than making the conditional more complicated, perhaps simply drop it (and move the reset-to-false code out of ...
xen/common/memory.c int xenmem_add_to_physmap(struct domain *d,
Quoted 1 line
xen/common/memory.c +0
-0@@ -828,7 +828,7 @@ int xenmem_add_to_physmap(struct domain *d,struct
xen_add_to_physmap*xatp,xen/common/memory.c int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, · line 833
Quoted 7 lines
xen/common/memory.c +1
-1}#ifdef CONFIG_HAS_PASSTHROUGH -if (need_iommu (d) ) +if (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d) ) {int ret;... this
if())?
Quoted 8 lines
Quoted 4 lines
Also it looks to me as if here you've got confused by the meaning you've assigned to
need_iommu_pt_sync(): According to the description, it is about sync-ing of page tables. Here, however, we're checking whether to flush TLBs.Yes, I may be confused here but it would seem to me that flushing the IOTLB would be necessary even in the case where the page tables are shared. I'll check the logic again.
Flushing is necessary always, and my comment didn't go in that direction. What I was trying to point out is that the value of iommu_dont_flush_iotlb doesn't matter when no flushing happens anyway. I.e. setting it to true unconditionally should not have any bad effect (but the non-strict-mode-Dom0 case may need double checking, albeit even in that case suppressing individual page flushing would be desirable, in which case - if needed - the second if() might need adjustment, independent of the change you're doing here).
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 10/14] mm / iommu: split need_iommu() into has_iommu_pt() and need_iommu_pt_sync()
Quoted 42 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 07:45
To: Paul Durrant <[email protected]>
Cc: Brian Woods <[email protected]>; Suravee Suthikulpanit <[email protected]>; Julien Grall <[email protected]>; Razvan Cojocaru <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Wei Liu <[email protected]>; Jun Nakajima <[email protected]>; Kevin Tian <[email protected]>; Stefano Stabellini <[email protected]>; xen- devel <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tamas K Lengyel <[email protected]>; Tim (Xen.org) <[email protected]>
Subject: RE: [Xen-devel] [PATCH v6 10/14] mm / iommu: splitneed_iommu()intohas_iommu_pt()andneed_iommu_pt_sync()Quoted 2 lines
Quoted 2 lines
Quoted 1 line · On 11.09.18 at 17:40, <[email protected]> wrote:
From: Xen-devel [mailto:[email protected]] On
Behalf
Quoted 22 lines
Quoted 21 lines
Of Jan Beulich
Sent: 11 September 2018 15:31xen/arch/x86/x86_64/mm.c int memory_add(unsigned long spfn, unsigned
Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/arch/x86/x86_64/mm.c +0
-0@@ -1426,7 +1426,8 @@ int memory_add(unsigned long spfn, unsignedlong epfn, unsigned int pxm)
xen/arch/x86/x86_64/mm.c int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm)
Quoted 4 lines
xen/arch/x86/x86_64/mm.c +0
-1if ( ret ) goto destroy_m2p; - if ( iommu_enabled && !iommu_passthrough &&!need_iommu(hardware_domain) )
xen/arch/x86/x86_64/mm.c int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm) · line 1433
Quoted 5 lines
xen/arch/x86/x86_64/mm.c +2
-0+if ( iommu_enabled && !iommu_passthrough && + !need_iommu_pt_sync (hardware_domain) ) {for ( i = spfn; i < epfn; i++ ) if ( iommu_map_page(hardware_domain, _bfn(i), _mfn(i),I'm confused - the condition you change looks to be inverted. Wouldn't we better fix this?
Quoted 7 lines
Quoted 2 lines
I don't think it is inverted. I think this is to add new hotplugged memory to the 1:1 map in the case that dom0 is not in strict mode. I could be wrong.
Oh, I think you're right. It is just rather confusing to see an
iommu_map_page()call qualified by !need_iommu(). But that's as confusing (to me) as the setup logic for Dom0's page tables.
I think it's generally confusing. I'll stick a comment in to explain.
Quoted 57 lines
Quoted 5 lines
Quoted 2 lines
And then I again wonder whether you've chosen the right predicate:
Where would the equivalent mappings come from in the opposite case?If dom0 is in strict mode then I assume that the synchronization is handled when the calls are made to add memory into the p2m (which IIRC happens
even
Quoted 1 line
for PV guests).
Right you are.
Quoted 1 line
My aim for this patch is to avoid any visible functional change.
Sure - I didn't mean anything here (if at all) to be done in this patch (or perhaps even series), I've merely noticed this as an apparent oddity (which if I were right would perhaps better have been fixed before your transformations).
Quoted 3 lines
Quoted 3 lines
xen/common/memory.c int xenmem_add_to_physmap(struct domain
Quoted 3 lines
xen/common/memory.c +0
-0@@ -805,8 +805,8 @@ int xenmem_add_to_physmap(struct domain*d,
Quoted 10 lines
Quoted 10 lines
struct
xen_add_to_physmap*xatp,Quoted 8 lines
xen/common/memory.c +2
-2xatp->size -= start;#ifdef CONFIG_HAS_PASSTHROUGH -if (need_iommu (d) ) -this_cpu (iommu_dont_flush_iotlb) =1 ; +if (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d) ) +this_cpu (iommu_dont_flush_iotlb) =1 ;#endif Quoted 4 lines
Quoted 4 lines
Rather than making the conditional more complicated, perhaps simply drop it (and move the reset-to-false code out of ...
xen/common/memory.c int xenmem_add_to_physmap(struct domain
Quoted 1 line
xen/common/memory.c +0
-0@@ -828,7 +828,7 @@ int xenmem_add_to_physmap(struct domain*d,
Quoted 9 lines
Quoted 9 lines
struct
xen_add_to_physmap*xatp,Quoted 7 lines
xen/common/memory.c +1
-1}#ifdef CONFIG_HAS_PASSTHROUGH -if (need_iommu (d) ) +if (need_iommu_pt_sync (d) ||iommu_use_hap_pt (d) ) {int ret;Quoted 8 lines
Quoted 6 lines
... this
if())?Also it looks to me as if here you've got confused by the meaning you've assigned to
need_iommu_pt_sync(): According to the description, it is about sync-ing of page tables. Here, however, we're checking whether to flush TLBs.Yes, I may be confused here but it would seem to me that flushing the
IOTLB
Quoted 2 lines
would be necessary even in the case where the page tables are shared. I'll check the logic again.
Quoted 10 lines
Flushing is necessary always, and my comment didn't go in that direction. What I was trying to point out is that the value of
iommu_dont_flush_iotlbdoesn't matter when no flushing happens anyway. I.e. setting it to true unconditionally should not have any bad effect (but the non-strict-mode-Dom0 case may need double checking, albeit even in that case suppressing individual page flushing would be desirable, in which case - if needed - the secondif()might need adjustment, independent of the change you're doing here).
Ok. I'll see if this needs correction and put a pre-requisite patch in if need be.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 11/14 x86: add iommu_op to enable modification of IOMMU mappings
In reply to Paul DurrantThis patch adds an iommu_op which checks whether it is possible or safe for a domain to modify its own IOMMU mappings and, if so, creates a rangeset to track modifications.
The op passes back a capabilities mask. There is only one bit currently defined for this mask: XEN_IOMMU_CAP_per_device_mappings. This bit is always left clear by the current implementation but may be set in future if the implementation is enhance to support per-device IOMMU mappings.
NOTE: The actual map and unmap operations are introduced by subsequent
patches.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
v6:
- Add struct xen_iommu_op_enable_modification to hold the capabilities mask.
v4:
- Set sync_iommu_pt to false instead of need_iommu.
v2:
- New in v2.
--- xen/common/iommu_op.c | 53 +++++++++++++++++++++++++++++++++++++++++ xen/drivers/passthrough/iommu.c | 2 +- xen/drivers/passthrough/pci.c | 4 ++-- xen/include/public/iommu_op.h | 20 ++++++++++++++++ xen/include/xen/iommu.h | 3 +++ xen/include/xlat.lst | 1 + 6 files changed, 80 insertions(+), 3 deletions(-)
xen/common/iommu_op.c +53 -0
@@ -78,6 +78,51 @@ static int iommu_op_query_reserved(struct xen_iommu_op_query_reserved *op)return 0 ; } +static int iommu_op_enable_modification ( +struct xen_iommu_op_enable_modification *op) +{ +struct domain *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); +const struct iommu_ops *ops = iommu->platform_ops; + +if ( op->cap || op->pad ) +return -EINVAL ; + +/* +* XEN_IOMMU_CAP_per_device_mappings is not supported yet so we can +* leave op->cap alone. +*/ + +/* Has modification already been enabled? */ +if ( iommu->iommu_op_ranges ) +return 0 ; + +/* +* The IOMMU mappings cannot be modified if: +* - the IOMMU is not enabled or, +* - the current domain is dom0 and tranlsation is disabled or, +* - HAP is enabled and the IOMMU shares the mappings. +*/ +if ( !iommu_enabled || + (is_hardware_domain (currd) && iommu_passthrough) || +iommu_use_hap_pt (currd) ) +return -EACCES ; + +/* +* The IOMMU implementation must provide the lookup method if +* modification of the mappings is to be supported. +*/ +if ( !ops->lookup_page ) +return -EOPNOTSUPP ; + + iommu->iommu_op_ranges =rangeset_new (currd,NULL ,0 ); +if ( !iommu->iommu_op_ranges ) +return -ENOMEM ; + + iommu->need_sync = false;/* Disable synchronization */ +return 0 ; +} +static void iommu_op (xen_iommu_op_t *op) { switch ( op->op )
@@ -86,6 +131,10 @@ static void iommu_op(xen_iommu_op_t *op) op->status =iommu_op_query_reserved (&op->u.query_reserved);break ; +case XEN_IOMMUOP_enable_modification: + op->status =iommu_op_enable_modification (&op->u.enable_modification); +break ; +default : op->status = -EOPNOTSUPP ;break ;
@@ -98,6 +147,7 @@ int do_one_iommu_op(xen_iommu_op_buf_t *buf)size_t offset;static const size_t op_size[] = { [XEN_IOMMUOP_query_reserved] =sizeof (struct xen_iommu_op_query_reserved ), + [XEN_IOMMUOP_enable_modification] =sizeof (struct xen_iommu_op_enable_modification ), };size_t size;int rc;
@@ -184,6 +234,7 @@ int compat_one_iommu_op(compat_iommu_op_buf_t *buf)size_t offset;static const size_t op_size[] = { [XEN_IOMMUOP_query_reserved] =sizeof (struct compat_iommu_op_query_reserved ), + [XEN_IOMMUOP_enable_modification] =sizeof (struct compat_iommu_op_enable_modification ), };size_t size;xen_iommu_op_t nat;
@@ -221,6 +272,7 @@ int compat_one_iommu_op(compat_iommu_op_buf_t *buf)* we need to fix things up here. */ #define XLAT_iommu_op_u_query_reserved XEN_IOMMUOP_query_reserved +#define XLAT_iommu_op_u_enable_modification XEN_IOMMUOP_enable_modification u = cmp.op;#define XLAT_iommu_op_query_reserved_HNDL_ranges (_d_, _s_) \
@@ -280,6 +332,7 @@ int compat_one_iommu_op(compat_iommu_op_buf_t *buf) &cmp, status) )return -EFAULT ; +#undef XLAT_iommu_op_u_enable_modification#undef XLAT_iommu_op_u_query_reservedreturn 0 ;
xen/drivers/passthrough/iommu.c +1 -1
@@ -26,7 +26,6 @@ static void iommu_dump_p2m_table(unsigned char key);unsigned int __read_mostly iommu_dev_iotlb_timeout =1000 ;integer_param ("iommu_dev_iotlb_timeout" , iommu_dev_iotlb_timeout); -/* * The 'iommu' parameter enables the IOMMU. Optional comma separated * value may contain:
@@ -280,6 +279,7 @@ void iommu_domain_destroy(struct domain *d)arch_iommu_domain_destroy (d);rangeset_destroy (hd->reserved_ranges); +rangeset_destroy (hd->iommu_op_ranges); } int iommu_map_page(struct domain *d, bfn_t bfn, mfn_t mfn,
xen/drivers/passthrough/pci.c +2 -2
@@ -1460,7 +1460,7 @@ static int assign_device(struct domain *d, u16 seg, u8 bus, u8 devfn, u32 flag) } done: -if ( !has_arch_pdevs (d) &&has_iommu_pt (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) && !hd->iommu_op_ranges )iommu_teardown (d);pcidevs_unlock ();
@@ -1510,7 +1510,7 @@ int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn) pdev->fault.count =0 ; -if ( !has_arch_pdevs (d) &&has_iommu_pt (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) && !hd->iommu_op_ranges )iommu_teardown (d);return ret;
xen/include/public/iommu_op.h +20 -0
@@ -61,6 +61,25 @@ struct xen_iommu_op_query_reserved {XEN_GUEST_HANDLE (xen_iommu_reserved_range_t) ranges; }; +/* +* XEN_IOMMUOP_enable_modification: Enable operations that modify IOMMU +* mappings. +*/ +#define XEN_IOMMUOP_enable_modification 2 + +struct xen_iommu_op_enable_modification { +/* +* OUT - On successful return this is set to the bitwise OR of capabilities +* defined below. On entry this must be set to zero. +*/ +unsigned int cap; +unsigned int pad; + +/* Does the implementation support per-device mappings? */ +#define _XEN_IOMMU_CAP_per_device_mappings 0 +#define XEN_IOMMU_CAP_per_device_mappings (1u << _XEN_IOMMU_CAP_per_device_mappings) +}; +struct xen_iommu_op {uint16_t op;/* op type */ uint16_t pad;
@@ -68,6 +87,7 @@ struct xen_iommu_op {/* 0 for success otherwise, negative errno */ union {struct xen_iommu_op_query_reserved query_reserved; +struct xen_iommu_op_enable_modification enable_modification; } u; };typedef struct xen_iommu_op xen_iommu_op_t ;
xen/include/xen/iommu.h +3 -0
@@ -147,6 +147,9 @@ struct domain_iommu {* include/xen/mm.h). */ bool need_sync; + +/* Ranges under the control of PV IOMMU interface */ +struct rangeset *iommu_op_ranges; };#define dom_iommu (d) (&(d)->iommu)
xen/include/xlat.lst +1 -0
@@ -79,6 +79,7 @@ ? vcpu_hvm_x86_64 hvm/hvm_vcpu.h ! iommu_op iommu_op.h ! iommu_op_buf iommu_op.h +! iommu_op_enable_modification iommu_op.h ! iommu_op_query_reserved iommu_op.h ? iommu_reserved_range iommu_op.h ? kexec_exec kexec.h
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 11/14] x86: add iommu_op to enable modification of IOMMU mappings
Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds an
iommu_opwhich checks whether it is possible or safe for a domain to modify its own IOMMU mappings and, if so, creates a rangeset to track modifications.
Now this can surely grow pretty big?
Quoted 14 lines
xen/common/iommu_op.c +8
-0@@ -78,6 +78,51 @@ static int iommu_op_query_reserved(struct xen_iommu_op_query_reserved *op)return 0 ; } +static int iommu_op_enable_modification ( +struct xen_iommu_op_enable_modification *op) +{ +struct domain *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); +const struct iommu_ops *ops = iommu->platform_ops; + +if ( op->cap || op->pad )
xen/common/iommu_op.c static int iommu_op_query_reserved(struct xen_iommu_op_query_reserved *op) · line 98
Quoted 10 lines
xen/common/iommu_op.c +10
-0+return -EINVAL ; + +/* +* XEN_IOMMU_CAP_per_device_mappings is not supported yet so we can +* leave op->cap alone. +*/ + +/* Has modification already been enabled? */ +if ( iommu->iommu_op_ranges ) +return 0 ;
I don't recall there being any synchronization around the check here until ...
Quoted 9 lines
xen/common/iommu_op.c +9
-0+/* +* The IOMMU mappings cannot be modified if: +* - the IOMMU is not enabled or, +* - the current domain is dom0 and tranlsation is disabled or, +* - HAP is enabled and the IOMMU shares the mappings. +*/ +if ( !iommu_enabled || + (is_hardware_domain (currd) && iommu_passthrough) || +iommu_use_hap_pt (currd) )
xen/common/iommu_op.c static int iommu_op_query_reserved(struct xen_iommu_op_query_reserved *op) · line 118
Quoted 10 lines
xen/common/iommu_op.c +10
-0+return -EACCES ; + +/* +* The IOMMU implementation must provide the lookup method if +* modification of the mappings is to be supported. +*/ +if ( !ops->lookup_page ) +return -EOPNOTSUPP ; + + iommu->iommu_op_ranges =rangeset_new (currd,NULL ,0 );
... the assignment here. In which case this is a (multi-vCPU) guest controllable leak.
xen/drivers/passthrough/iommu.c static void iommu_dump_p2m_table(unsigned char key); · line 31
Quoted 10 lines
xen/drivers/passthrough/iommu.c +0
-1@@ -26,7 +26,6 @@ static void iommu_dump_p2m_table(unsigned char key);unsigned int __read_mostly iommu_dev_iotlb_timeout =1000 ;integer_param ("iommu_dev_iotlb_timeout" , iommu_dev_iotlb_timeout); -/* * The 'iommu' parameter enables the IOMMU. Optional comma separated * value may contain:
Stray change?
Quoted 12 lines
xen/drivers/passthrough/pci.c +1
-1@@ -1460,7 +1460,7 @@ static int assign_device(struct domain *d, u16 seg, u8 bus, u8 devfn, u32 flag) } done: -if ( !has_arch_pdevs (d) &&has_iommu_pt (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) && !hd->iommu_op_ranges )iommu_teardown (d);pcidevs_unlock ();
xen/drivers/passthrough/pci.c int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn) · line 1514
Quoted 6 lines
xen/drivers/passthrough/pci.c +1
-1@@ -1510,7 +1510,7 @@ int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn) pdev->fault.count =0 ; -if ( !has_arch_pdevs (d) &&has_iommu_pt (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) && !hd->iommu_op_ranges )iommu_teardown (d);
These additions are pretty un-obvious, and hence at least need comments. But I'm also unclear about the intended behavior: For a guest not meaning to play with its mappings, why would you keep the tables around (and the memory uselessly allocated)?
Quoted 9 lines
xen/include/public/iommu_op.h +3
-0@@ -61,6 +61,25 @@ struct xen_iommu_op_query_reserved {XEN_GUEST_HANDLE (xen_iommu_reserved_range_t) ranges; }; +/* +* XEN_IOMMUOP_enable_modification: Enable operations that modify IOMMU +* mappings.
xen/include/public/iommu_op.h struct xen_iommu_op_query_reserved { · line 76
Quoted 10 lines
xen/include/public/iommu_op.h +10
-0+*/ +#define XEN_IOMMUOP_enable_modification 2 + +struct xen_iommu_op_enable_modification { +/* +* OUT - On successful return this is set to the bitwise OR of capabilities +* defined below. On entry this must be set to zero. +*/ +unsigned int cap; +unsigned int pad;
uint32_t
xen/include/xlat.lst line 82
Quoted 7 lines
xen/include/xlat.lst +1
-0@@ -79,6 +79,7 @@ ? vcpu_hvm_x86_64 hvm/hvm_vcpu.h ! iommu_op iommu_op.h ! iommu_op_buf iommu_op.h +! iommu_op_enable_modification iommu_op.h
The structure above looks to be 32-/64-bit agnostic. Why is this ! instead of ? ?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 11/14] x86: add iommu_op to enable modification of IOMMU mappings
Quoted 13 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 11 September 2018 15:48
To: Paul Durrant <[email protected]>
Cc: Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; Wei Liu <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>
Subject: Re: [PATCH v6 11/14] x86: addiommu_opto enable modification of IOMMU mappings
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds an
iommu_opwhich checks whether it is possible or safe for a domain to modify its own IOMMU mappings and, if so, creates a rangeset to track modifications.Now this can surely grow pretty big?
It could but I don't see any good way round it. The alternative is to shoot all RAM mappings out of the IOMMU when enabling PV IOMMU and determine the validity of an unmap op only on the basis of whether the mapping exists in the IOMMU. Is that option preferable?
Quoted 44 lines
xen/common/iommu_op.c static int iommu_op_query_reserved(struct
Quoted 3 lines
xen/common/iommu_op.c +0
-0@@ -78,6 +78,51 @@ static int iommu_op_query_reserved(struct
xen_iommu_op_query_reserved*op)Quoted 5 lines
xen/common/iommu_op.c +2
-0return 0 ; } +static int iommu_op_enable_modification ( +struct xen_iommu_op_enable_modification *op)xen/common/iommu_op.c static int iommu_op_query_reserved(struct xen_iommu_op_query_reserved *op)
Quoted 10 lines
xen/common/iommu_op.c +10
-0+{ +struct domain *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); +const struct iommu_ops *ops = iommu->platform_ops; + +if ( op->cap || op->pad ) +return -EINVAL ; + +/* +* XEN_IOMMU_CAP_per_device_mappings is not supported yet so we can
xen/common/iommu_op.c static int iommu_op_query_reserved(struct xen_iommu_op_query_reserved *op) · line 98
Quoted 6 lines
xen/common/iommu_op.c +6
-0+* leave op->cap alone. +*/ + +/* Has modification already been enabled? */ +if ( iommu->iommu_op_ranges ) +return 0 ;I don't recall there being any synchronization around the check here until ...
Quoted 14 lines
xen/common/iommu_op.c +14
-0+/* +* The IOMMU mappings cannot be modified if: +* - the IOMMU is not enabled or, +* - the current domain is dom0 and tranlsation is disabled or, +* - HAP is enabled and the IOMMU shares the mappings. +*/ +if ( !iommu_enabled || + (is_hardware_domain (currd) && iommu_passthrough) || +iommu_use_hap_pt (currd) ) +return -EACCES ; + +/* +* The IOMMU implementation must provide the lookup method if +* modification of the mappings is to be supported.
Quoted 8 lines
xen/common/iommu_op.c static int iommu_op_query_reserved(struct xen_iommu_op_query_reserved *op) · line 118
Quoted 5 lines
xen/common/iommu_op.c +5
-0+*/ +if ( !ops->lookup_page ) +return -EOPNOTSUPP ; + + iommu->iommu_op_ranges =rangeset_new (currd,NULL ,0 );... the assignment here. In which case this is a (multi-vCPU) guest controllable leak.
True, yes. This would need locking.
Quoted 11 lines
xen/drivers/passthrough/iommu.c static void iommu_dump_p2m_table(unsigned char
Quoted 4 lines
xen/drivers/passthrough/iommu.c +0
-0@@ -26,7 +26,6 @@ static void iommu_dump_p2m_table(unsigned charkey);
xen/drivers/passthrough/iommu.c static void iommu_dump_p2m_table(unsigned char key);
Quoted 6 lines
xen/drivers/passthrough/iommu.c +0
-1unsigned int __read_mostly iommu_dev_iotlb_timeout =1000 ;integer_param ("iommu_dev_iotlb_timeout" , iommu_dev_iotlb_timeout); -/* * The 'iommu' parameter enables the IOMMU. Optional comma
Quoted 4 lines
separated
xen/drivers/passthrough/iommu.c static void iommu_dump_p2m_table(unsigned char key); · line 31
Quoted 1 line
xen/drivers/passthrough/iommu.c +0
-0* value may contain: Stray change?
Yes.
Quoted 24 lines
xen/drivers/passthrough/pci.c static int assign_device(struct domain *d, u16 seg,
Quoted 4 lines
xen/drivers/passthrough/pci.c +0
-0@@ -1460,7 +1460,7 @@ static int assign_device(struct domain *d, u16 seg,u8 bus, u8 devfn, u32 flag)
xen/drivers/passthrough/pci.c int deassign_device(struct domain *d, u16 seg, u8
Quoted 10 lines
xen/drivers/passthrough/pci.c +1
-1} done: -if ( !has_arch_pdevs (d) &&has_iommu_pt (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) && !hd- iommu_op_ranges )iommu_teardown (d);pcidevs_unlock ();@@ -1510,7 +1510,7 @@ int deassign_device(struct domain *d, u16 seg, u8bus, u8 devfn)
xen/drivers/passthrough/pci.c int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn) · line 1514
Quoted 7 lines
xen/drivers/passthrough/pci.c +1
-1pdev->fault.count =0 ; -if ( !has_arch_pdevs (d) &&has_iommu_pt (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) && !hd- iommu_op_ranges )iommu_teardown (d);
Quoted 5 lines
These additions are pretty un-obvious, and hence at least need comments. But I'm also unclear about the intended behavior: For a guest not meaning to play with its mappings, why would you keep the tables around (and the memory uselessly allocated)?
No. If the guest has not enabled PV-IOMMu then hd->iommu_op_ranges would be NULL and so the tables will get torn down, otherwise the guest is in control and so Xen should leave the tables alone.
Quoted 9 lines
xen/include/public/iommu_op.h struct xen_iommu_op_query_reserved { · line 65
Quoted 8 lines
xen/include/public/iommu_op.h +2
-0@@ -61,6 +61,25 @@ struct xen_iommu_op_query_reserved {XEN_GUEST_HANDLE (xen_iommu_reserved_range_t) ranges; }; +/* +* XEN_IOMMUOP_enable_modification: Enable operations that modify IOMMU
Quoted 14 lines
xen/include/public/iommu_op.h struct xen_iommu_op_query_reserved { · line 76
Quoted 11 lines
xen/include/public/iommu_op.h +11
-0+* mappings. +*/ +#define XEN_IOMMUOP_enable_modification 2 + +struct xen_iommu_op_enable_modification { +/* +* OUT - On successful return this is set to the bitwise OR of capabilities +* defined below. On entry this must be set to zero. +*/ +unsigned int cap; +unsigned int pad;
uint32_t
Ok.
Quoted 10 lines
xen/include/xlat.lst line 82
Quoted 7 lines
xen/include/xlat.lst +1
-0@@ -79,6 +79,7 @@ ? vcpu_hvm_x86_64 hvm/hvm_vcpu.h ! iommu_op iommu_op.h ! iommu_op_buf iommu_op.h +! iommu_op_enable_modification iommu_op.hThe structure above looks to be 32-/64-bit agnostic. Why is this ! instead of ? ?
IIRC I investigated this. I think the reason is that the XLAT macro won't do the necessary copy-back of the struct in the compat path and so it messed up getting the status field out. Since you wanted by to adjust the compat code to avoid unnecessary copying (in an earlier patch) I can probably change these to ?.
Paul
Quoted 5 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 11/14] x86: add iommu_op to enable modification of IOMMU mappings
Quoted 9 lines
Quoted 9 lines
Quoted 1 line · On 11.09.18 at 17:52, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 11 September 2018 15:48Quoted 4 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
This patch adds an
iommu_opwhich checks whether it is possible or safe for a domain to modify its own IOMMU mappings and, if so, creates a rangeset to track modifications.
Quoted 6 lines
Quoted 1 line
Now this can surely grow pretty big?
It could but I don't see any good way round it. The alternative is to shoot all RAM mappings out of the IOMMU when enabling PV IOMMU and determine the validity of an unmap op only on the basis of whether the mapping exists in the IOMMU. Is that option preferable?
I'm not sure - it very much depends on the downsides. An obvious one is that dropping all RAM mappings is going to take quite some time for a big domain.
Quoted 29 lines
Quoted 23 lines
xen/drivers/passthrough/pci.c static int assign_device(struct domain *d, u16 seg,
Quoted 3 lines
xen/drivers/passthrough/pci.c +0
-0@@ -1460,7 +1460,7 @@ static int assign_device(struct domain *d, u16 seg,u8 bus, u8 devfn, u32 flag)
xen/drivers/passthrough/pci.c int deassign_device(struct domain *d, u16 seg, u8
Quoted 10 lines
xen/drivers/passthrough/pci.c +1
-1} done: -if ( !has_arch_pdevs (d) &&has_iommu_pt (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) && !hd- iommu_op_ranges )iommu_teardown (d);pcidevs_unlock ();@@ -1510,7 +1510,7 @@ int deassign_device(struct domain *d, u16 seg, u8bus, u8 devfn)
xen/drivers/passthrough/pci.c int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn) · line 1514
Quoted 7 lines
xen/drivers/passthrough/pci.c +1
-1pdev->fault.count =0 ; -if ( !has_arch_pdevs (d) &&has_iommu_pt (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) && !hd- iommu_op_ranges )iommu_teardown (d);Quoted 5 lines
These additions are pretty un-obvious, and hence at least need comments. But I'm also unclear about the intended behavior: For a guest not meaning to play with its mappings, why would you keep the tables around (and the memory uselessly allocated)?
Quoted 3 lines
No. If the guest has not enabled PV-IOMMu then
hd->iommu_op_rangeswould be NULL and so the tables will get torn down, otherwise the guest is in control and so Xen should leave the tables alone.
Oh, I've misread hd->iommu_op_ranges (taking it for a check for the required op that you check for in iommu_op_enable_modification()).
Quoted 11 lines
Quoted 10 lines
xen/include/xlat.lst line 82
Quoted 7 lines
xen/include/xlat.lst +1
-0@@ -79,6 +79,7 @@ ? vcpu_hvm_x86_64 hvm/hvm_vcpu.h ! iommu_op iommu_op.h ! iommu_op_buf iommu_op.h +! iommu_op_enable_modification iommu_op.hThe structure above looks to be 32-/64-bit agnostic. Why is this ! instead of ? ?
Quoted 3 lines
IIRC I investigated this. I think the reason is that the XLAT macro won't do the necessary copy-back of the struct in the compat path and so it messed up getting the status field out.
I'd be surprised, as I think we use such mix of ! and ? elsewhere too. I can't exclude that's in no-copy-back scenarios only, though.
Quoted 3 lines
Since you wanted by to adjust the compat code
to avoid unnecessary copying (in an earlier patch) I can probably change these to ?.
Perhaps I'm missing the connection, so I don't know for sure what to reply here - most likely the answer is "yes".
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 11/14] x86: add iommu_op to enable modification of IOMMU mappings
Quoted 92 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 07:54
To: Paul Durrant <[email protected]>
Cc: Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Wei Liu <[email protected]>; Stefano Stabellini <[email protected]>; xen- devel <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>
Subject: RE: [PATCH v6 11/14] x86: addiommu_opto enable modification of IOMMU mappingsQuoted 5 lines
Quoted 5 lines
Quoted 1 line · On 11.09.18 at 17:52, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 11 September 2018 15:48Quoted 1 line
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
This patch adds an
iommu_opwhich checks whether it is possible or safe for a domain to modify its own IOMMU mappings and, if so, creates a rangeset to track modifications.Now this can surely grow pretty big?
It could but I don't see any good way round it. The alternative is to shoot all RAM mappings out of the IOMMU when enabling PV IOMMU and
determine the
Quoted 2 lines
validity of an unmap op only on the basis of whether the mapping exists in the IOMMU. Is that option preferable?
I'm not sure - it very much depends on the downsides. An obvious one is that dropping all RAM mappings is going to take quite some time for a big domain.
Quoted 3 lines
Quoted 3 lines
xen/drivers/passthrough/pci.c static int assign_device(struct domain *d, u16
Quoted 3 lines
xen/drivers/passthrough/pci.c +0
-0@@ -1460,7 +1460,7 @@ static int assign_device(struct domain *d, u16seg,
Quoted 11 lines
Quoted 11 lines
u8 bus, u8 devfn, u32 flag)
xen/drivers/passthrough/pci.c int deassign_device(struct domain *d, u16 seg,
Quoted 10 lines
xen/drivers/passthrough/pci.c +1
-1} done: -if ( !has_arch_pdevs (d) &&has_iommu_pt (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) && !hd- iommu_op_ranges )iommu_teardown (d);pcidevs_unlock ();@@ -1510,7 +1510,7 @@ int deassign_device(struct domain *d, u16 seg,u8
Quoted 9 lines
Quoted 9 lines
bus, u8 devfn)
Quoted 7 lines
xen/drivers/passthrough/pci.c +1
-1pdev->fault.count =0 ; -if ( !has_arch_pdevs (d) &&has_iommu_pt (d) ) +if ( !has_arch_pdevs (d) &&has_iommu_pt (d) && !hd- iommu_op_ranges )iommu_teardown (d);Quoted 7 lines
Quoted 5 lines
These additions are pretty un-obvious, and hence at least need comments. But I'm also unclear about the intended behavior: For a guest not meaning to play with its mappings, why would you keep the tables around (and the memory uselessly allocated)?
No. If the guest has not enabled PV-IOMMu then
hd->iommu_op_rangeswould be
Quoted 2 lines
NULL and so the tables will get torn down, otherwise the guest is in control and so Xen should leave the tables alone.
Oh, I've misread
hd->iommu_op_ranges(taking it for a check for the required op that you check for iniommu_op_enable_modification()).Quoted 11 lines
Quoted 10 lines
xen/include/xlat.lst line 82
Quoted 7 lines
xen/include/xlat.lst +1
-0@@ -79,6 +79,7 @@ ? vcpu_hvm_x86_64 hvm/hvm_vcpu.h ! iommu_op iommu_op.h ! iommu_op_buf iommu_op.h +! iommu_op_enable_modification iommu_op.hThe structure above looks to be 32-/64-bit agnostic. Why is this ! instead of ? ?
Quoted 2 lines
IIRC I investigated this. I think the reason is that the XLAT macro won't do the necessary copy-back of the struct in the compat path and so it messed
up
Quoted 1 line
getting the status field out.
I'd be surprised, as I think we use such mix of ! and ? elsewhere too. I can't exclude that's in no-copy-back scenarios only, though.
Quoted 7 lines
Quoted 3 lines
Since you wanted by to adjust the compat code
to avoid unnecessary copying (in an earlier patch) I can probably change these to ?.Perhaps I'm missing the connection, so I don't know for sure what to reply here - most likely the answer is "yes".
It wasn't a question... perhaps I should have quoted that '?' :-) My point was that I'm going to have to re-investigate the copy-back anyway so I'll try to go for check rather than xlat where I can.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 12/14 memory: add get_paged_gfn() as a wrapper...
In reply to Paul Durrant...for some uses of get_page_from_gfn().
There are many occurences of the following pattern in the code:
q = <readonly look-up> ? P2M_ALLOC : P2M_UNSHARE;
page = get_page_from_gfn(d, gfn, &p2mt, q);
if ( p2m_is_paging(p2mt) )
{
if ( page )
put_page(page);
p2m_mem_paging_populate(d, gfn);
return <-EAGAIN or equivalent>;
}
if ( (q & P2M_UNSHARE) && p2m_is_shared(p2mt) )
{
if ( page )
put_page(page);
return <-EAGAIN or equivalent>;
}
if ( !page )
return <-EINVAL or equivalent>;
if ( !p2m_is_ram(p2mt) ||
(!<readonly look-up> && p2m_is_readonly(p2mt)) )
{
put_page(page);
return <-EINVAL or equivalent>;
}There are some small differences between the exact way the occurrences are coded but the desired semantic is the same.
This patch introduces a new common implementation of this code in get_paged_gfn() and then converts the various open-coded patterns into calls to this new function.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
v3:
- Addressed comments from George.
v2:
- New in v2.
--- xen/arch/x86/hvm/emulate.c | 32 ++++++-------------------- xen/arch/x86/hvm/hvm.c | 16 ++----------- xen/common/grant_table.c | 38 +++++++++---------------------- xen/common/memory.c | 56 +++++++++++++++++++++++++++++++++++++--------- xen/include/asm-arm/p2m.h | 3 +++ xen/include/asm-x86/p2m.h | 2 ++ 6 files changed, 71 insertions(+), 76 deletions(-)
xen/arch/x86/hvm/emulate.c +7 -25
@@ -350,34 +350,16 @@ static int hvmemul_do_io_buffer(static int hvmemul_acquire_page (unsigned long gmfn,struct page_info **page) { -struct domain *curr_d = current->domain; -p2m_type_t p2mt; - - *page =get_page_from_gfn (curr_d, gmfn, &p2mt,P2M_UNSHARE ); - -if ( *page ==NULL ) -return X86EMUL_UNHANDLEABLE ; - -if (p2m_is_paging (p2mt) ) - { -put_page (*page); -p2m_mem_paging_populate (curr_d, gmfn); -return X86EMUL_RETRY ; - } - -if (p2m_is_shared (p2mt) ) +switch (get_paged_gfn (current->domain,_gfn (gmfn), false,NULL , page) ) { -put_page (*page); +case -EAGAIN :return X86EMUL_RETRY ; - } - -/* This code should not be reached if the gmfn is not RAM */ -if (p2m_is_mmio (p2mt) ) - { -domain_crash (curr_d); - -put_page (*page); +case -EINVAL :return X86EMUL_UNHANDLEABLE ; +default : +ASSERT_UNREACHABLE (); +case 0 : +break ; }return X86EMUL_OKAY ;
xen/arch/x86/hvm/hvm.c +2 -14
@@ -2557,24 +2557,12 @@ static void *_hvm_map_guest_frame(unsigned long gfn, bool_t permanent, bool_t *writable) {void *map; -p2m_type_t p2mt;struct page_info *page;struct domain *d = current->domain; +p2m_type_t p2mt; - page =get_page_from_gfn (d, gfn, &p2mt, - writable ?P2M_UNSHARE :P2M_ALLOC ); -if ( (p2m_is_shared (p2mt) && writable) || !page ) - { -if ( page ) -put_page (page); -return NULL ; - } -if (p2m_is_paging (p2mt) ) - { -put_page (page); -p2m_mem_paging_populate (d, gfn); +if (get_paged_gfn (d,_gfn (gfn), !writable, &p2mt, &page) )return NULL ; - }if ( writable ) {
xen/common/grant_table.c +11 -27
@@ -374,39 +374,23 @@ static int get_paged_frame(unsigned long gfn, mfn_t *mfn, struct page_info **page, bool readonly, struct domain *rd) { - int rc = GNTST_okay; -p2m_type_t p2mt; - - *mfn =INVALID_MFN ; - *page =get_page_from_gfn (rd, gfn, &p2mt, - readonly ?P2M_ALLOC :P2M_UNSHARE ); -if ( !*page ) - { -#ifdef P2M_SHARED_TYPES -if (p2m_is_shared (p2mt) ) -return GNTST_eagain; -#endif -#ifdef P2M_PAGES_TYPES -if (p2m_is_paging (p2mt) ) - { -p2m_mem_paging_populate (rd, gfn); -return GNTST_eagain; - } -#endif -return GNTST_bad_page; - } + int rc; -if (p2m_is_foreign (p2mt) ) + rc =get_paged_gfn (rd,_gfn (gfn), readonly,NULL , page); +switch ( rc ) { -put_page (*page); - *page =NULL ; - +case -EAGAIN : +return GNTST_eagain; +case -EINVAL :return GNTST_bad_page; +default : +ASSERT_UNREACHABLE (); +case 0 : +break ; } *mfn =page_to_mfn (*page); - -return rc; +return GNTST_okay; } static inline void
xen/common/memory.c +46 -10
83 lines not on this page · raw message
xen/include/asm-arm/p2m.h +3 -0
@@ -303,6 +303,9 @@ static inline struct page_info *get_page_from_gfn(return page; } +int get_paged_gfn (struct domain *d,gfn_t gfn,bool readonly, +p2m_type_t *p2mt_p,struct page_info **page_p); +int get_page_type (struct page_info *page,unsigned long type);bool is_iomem_page (mfn_t mfn); static inline int get_page_and_type(struct page_info *page,
xen/include/asm-x86/p2m.h +2 -0
@@ -492,6 +492,8 @@ static inline struct page_info *get_page_from_gfn(return mfn_valid (_mfn (gfn)) &&get_page (page, d) ? page :NULL ; } +int get_paged_gfn (struct domain *d,gfn_t gfn,bool readonly, +p2m_type_t *p2mt_p,struct page_info **page_p);/* General conversion function from mfn to gfn */ static inline unsigned long mfn_to_gfn (struct domain *d,mfn_t mfn)
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 12/14] memory: add get_paged_gfn() as a wrapper...
Hi Paul,
Quoted 3 lines · Paul Durrant
...for some uses of
get_page_from_gfn().There are many occurences of the following pattern in the code:
NIT: s/occurences/occurences/
[...]
Quoted 30 lines
xen/common/memory.c +9
-6+int get_paged_gfn (struct domain *d,gfn_t gfn,bool readonly, +p2m_type_t *p2mt_p,struct page_info **page_p) { -struct page_info *page; +p2m_query_t q = readonly ?P2M_ALLOC :P2M_UNSHARE ;p2m_type_t p2mt; -void *va; +struct page_info *page; - page =get_page_from_gfn (d, gmfn, &p2mt,P2M_UNSHARE ); + page =get_page_from_gfn (d,gfn_x (gfn), &p2mt, q);#ifdef CONFIG_HAS_MEM_PAGING if (p2m_is_paging (p2mt) ) {if ( page )put_page (page); -p2m_mem_paging_populate (d, gmfn); -return -ENOENT ; + +p2m_mem_paging_populate (d,gfn_x (gfn)); +return -EAGAIN ; }#endif #ifdef CONFIG_HAS_MEM_SHARING -if (p2m_is_shared (p2mt) ) +if ( (q &P2M_UNSHARE ) &&p2m_is_shared (p2mt) ) {if ( page )put_page (page);
xen/common/memory.c void destroy_ring_for_helper( · line 1670
Quoted 10 lines
xen/common/memory.c +3
-1- return -ENOENT; + + return -EAGAIN; }#endif if ( !page )return -EINVAL ; +if ( !p2m_is_ram (p2mt) || (!readonly &&p2m_is_readonly (p2mt)) )
p2m_is_readonly does not exist on Arm. Can you please make sure this code build on Arm?
Cheers,
Signature · 6 lines
Julien Grall _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 12/14] memory: add get_paged_gfn() as a wrapper...
Quoted 59 lines
-----Original Message-----
From: Julien Grall [mailto:[email protected]]
Sent: 23 August 2018 11:25
To: Paul Durrant <[email protected]>; [email protected]
Cc: Jan Beulich <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Stefano Stabellini <[email protected]>; Tim (Xen.org) <[email protected]>; Wei Liu <[email protected]>
Subject: Re: [PATCH v6 12/14] memory: addget_paged_gfn()as a wrapper...Hi Paul,
Quoted 3 lines · Paul Durrant
...for some uses of
get_page_from_gfn().There are many occurences of the following pattern in the code:
NIT: s/occurences/occurences/
[...]
Quoted 36 lines
xen/common/memory.c +11
-7+int get_paged_gfn (struct domain *d,gfn_t gfn,bool readonly, +p2m_type_t *p2mt_p,struct page_info **page_p) { -struct page_info *page; +p2m_query_t q = readonly ?P2M_ALLOC :P2M_UNSHARE ;p2m_type_t p2mt; -void *va; +struct page_info *page; - page =get_page_from_gfn (d, gmfn, &p2mt,P2M_UNSHARE ); + page =get_page_from_gfn (d,gfn_x (gfn), &p2mt, q);#ifdef CONFIG_HAS_MEM_PAGING if (p2m_is_paging (p2mt) ) {if ( page )put_page (page); -p2m_mem_paging_populate (d, gmfn); -return -ENOENT ; + +p2m_mem_paging_populate (d,gfn_x (gfn)); +return -EAGAIN ; }#endif #ifdef CONFIG_HAS_MEM_SHARING -if (p2m_is_shared (p2mt) ) +if ( (q &P2M_UNSHARE ) &&p2m_is_shared (p2mt) ) {if ( page )put_page (page); -return -ENOENT ; + +return -EAGAIN ; }#endif
Quoted 8 lines
Quoted 4 lines
if ( !page )
return-EINVAL;+ if ( !p2m_is_ram(p2mt) || (!readonly &&
p2m_is_readonly(p2mt)) )
p2m_is_readonlydoes not exist on Arm. Can you please make sure this code build on Arm?
Ok. Will do.
Paul
Quoted 4 lines
Cheers,
--
Julien Grall
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 12/14] memory: add get_paged_gfn() as a wrapper...
In reply to Paul DurrantQuoted 6 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
...for some uses of
get_page_from_gfn().There are many occurences of the following pattern in the code:
q = <readonly look-up> ? P2M_ALLOC : P2M_UNSHARE;
Especially with this UNSHARE in mind - is "paged" in the helper function's name really suitable? Since we (I think) already have get_gfn(), how about try_get_gfn()?
Quoted 30 lines
xen/arch/x86/hvm/emulate.c +2
-20@@ -350,34 +350,16 @@ static int hvmemul_do_io_buffer(static int hvmemul_acquire_page (unsigned long gmfn,struct page_info **page) { -struct domain *curr_d = current->domain; -p2m_type_t p2mt; - - *page =get_page_from_gfn (curr_d, gmfn, &p2mt,P2M_UNSHARE ); - -if ( *page ==NULL ) -return X86EMUL_UNHANDLEABLE ; - -if (p2m_is_paging (p2mt) ) - { -put_page (*page); -p2m_mem_paging_populate (curr_d, gmfn); -return X86EMUL_RETRY ; - } - -if (p2m_is_shared (p2mt) ) +switch (get_paged_gfn (current->domain,_gfn (gmfn), false,NULL , page) ) { -put_page (*page); +case -EAGAIN :return X86EMUL_RETRY ; - } - -/* This code should not be reached if the gmfn is not RAM */
xen/arch/x86/hvm/emulate.c static int hvmemul_do_io_buffer( · line 361
Quoted 10 lines
xen/arch/x86/hvm/emulate.c +4
-5-if (p2m_is_mmio (p2mt) ) - { -domain_crash (curr_d); - -put_page (*page); +case -EINVAL :return X86EMUL_UNHANDLEABLE ; +default : +ASSERT_UNREACHABLE (); +case 0 :
I think you'd better have "default:" fall through to "case -EINVAL". Similarly elsewhere.
xen/arch/x86/hvm/hvm.c static void *_hvm_map_guest_frame(unsigned long gfn, bool_t permanent, · line 2562
Quoted 10 lines
xen/arch/x86/hvm/hvm.c +1
-1@@ -2557,24 +2557,12 @@ static void *_hvm_map_guest_frame(unsigned long gfn, bool_t permanent, bool_t *writable) {void *map; -p2m_type_t p2mt;struct page_info *page;struct domain *d = current->domain; +p2m_type_t p2mt;
???
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 12/14] memory: add get_paged_gfn() as a wrapper...
Quoted 17 lines
-----Original Message-----
From: Xen-devel [mailto:[email protected]] On Behalf Of Jan Beulich
Sent: 11 September 2018 15:56
To: Paul Durrant <[email protected]>
Cc: Stefano Stabellini <[email protected]>; Wei Liu <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; George Dunlap <[email protected]>; Andrew Cooper <[email protected]>; Ian Jackson <[email protected]>; Tim (Xen.org) <[email protected]>; Julien Grall <[email protected]>; xen-devel <[email protected]>
Subject: Re: [Xen-devel] [PATCH v6 12/14] memory: addget_paged_gfn()as a wrapper...Quoted 3 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
...for some uses of
get_page_from_gfn().
Quoted 7 lines
Quoted 3 lines
There are many occurences of the following pattern in the code:
q = <readonly look-up> ? P2M_ALLOC : P2M_UNSHARE;Especially with this UNSHARE in mind - is "paged" in the helper function's name really suitable? Since we (I think) already have
get_gfn(), how abouttry_get_gfn()?
That name may be a little misleading since it suggests a close functional relationship with get_gfn() whereas it does more than that. How about try_get_page_from_gfn()?
Quoted 38 lines
xen/arch/x86/hvm/emulate.c static int hvmemul_do_io_buffer( · line 351
Quoted 6 lines
xen/arch/x86/hvm/emulate.c +0
-0@@ -350,34 +350,16 @@ static int hvmemul_do_io_buffer( static int hvmemul_acquire_page(unsigned long gmfn, struct page_info**page)
Quoted 8 lines
xen/arch/x86/hvm/emulate.c +0
-7{ -struct domain *curr_d = current->domain; -p2m_type_t p2mt; - - *page =get_page_from_gfn (curr_d, gmfn, &p2mt,P2M_UNSHARE ); - -if ( *page ==NULL ) -return X86EMUL_UNHANDLEABLE ;xen/arch/x86/hvm/emulate.c static int hvmemul_do_io_buffer(
Quoted 10 lines
xen/arch/x86/hvm/emulate.c +1
-9- -if (p2m_is_paging (p2mt) ) - { -put_page (*page); -p2m_mem_paging_populate (curr_d, gmfn); -return X86EMUL_RETRY ; - } - -if (p2m_is_shared (p2mt) ) + switch ( get_paged_gfn(current->domain, _gfn(gmfn), false, NULL,page) )
Quoted 12 lines
xen/arch/x86/hvm/emulate.c +1
-9{ -put_page (*page); +case -EAGAIN :return X86EMUL_RETRY ; - } - -/* This code should not be reached if the gmfn is not RAM */ -if (p2m_is_mmio (p2mt) ) - { -domain_crash (curr_d); - -put_page (*page);
Quoted 8 lines
xen/arch/x86/hvm/emulate.c static int hvmemul_do_io_buffer( · line 361
Quoted 5 lines
xen/arch/x86/hvm/emulate.c +4
-0+case -EINVAL :return X86EMUL_UNHANDLEABLE ; +default : +ASSERT_UNREACHABLE (); +case 0 :I think you'd better have "default:" fall through to "case
-EINVAL". Similarly elsewhere.
Ok. I'll keep the ASSERT_UNREACHABLE() though.
Paul
Quoted 15 lines
xen/arch/x86/hvm/hvm.c static void *_hvm_map_guest_frame(unsigned
Quoted 4 lines
xen/arch/x86/hvm/hvm.c +0
-0@@ -2557,24 +2557,12 @@ static void *_hvm_map_guest_frame(unsignedlong gfn,
bool_tpermanent,xen/arch/x86/hvm/hvm.c static void *_hvm_map_guest_frame(unsigned long gfn, bool_t permanent, · line 2562
Quoted 7 lines
xen/arch/x86/hvm/hvm.c +1
-1bool_t *writable) {void *map; -p2m_type_t p2mt;struct page_info *page;struct domain *d = current->domain; +p2m_type_t p2mt;???
Quoted 8 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 12/14] memory: add get_paged_gfn() as a wrapper...
Quoted 12 lines
Quoted 12 lines
Quoted 1 line · On 12.09.18 at 11:10, <[email protected]> wrote:
From: Xen-devel [mailto:[email protected]] On Behalf Of Jan Beulich
Sent: 11 September 2018 15:56Quoted 6 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
...for some uses of
get_page_from_gfn().There are many occurences of the following pattern in the code:
q = <readonly look-up> ? P2M_ALLOC : P2M_UNSHARE;
Quoted 7 lines
Quoted 3 lines
Especially with this UNSHARE in mind - is "paged" in the helper function's name really suitable? Since we (I think) already have
get_gfn(), how abouttry_get_gfn()?That name may be a little misleading since it suggests a close functional relationship with
get_gfn()whereas it does more than that. How abouttry_get_page_from_gfn()?
Fine with me; George?
Quoted 46 lines
Quoted 37 lines
xen/arch/x86/hvm/emulate.c static int hvmemul_do_io_buffer( · line 351
Quoted 5 lines
xen/arch/x86/hvm/emulate.c +0
-0@@ -350,34 +350,16 @@ static int hvmemul_do_io_buffer( static int hvmemul_acquire_page(unsigned long gmfn, struct page_info**page)
Quoted 8 lines
xen/arch/x86/hvm/emulate.c +0
-7{ -struct domain *curr_d = current->domain; -p2m_type_t p2mt; - - *page =get_page_from_gfn (curr_d, gmfn, &p2mt,P2M_UNSHARE ); - -if ( *page ==NULL ) -return X86EMUL_UNHANDLEABLE ;xen/arch/x86/hvm/emulate.c static int hvmemul_do_io_buffer(
Quoted 10 lines
xen/arch/x86/hvm/emulate.c +1
-9- -if (p2m_is_paging (p2mt) ) - { -put_page (*page); -p2m_mem_paging_populate (curr_d, gmfn); -return X86EMUL_RETRY ; - } - -if (p2m_is_shared (p2mt) ) + switch ( get_paged_gfn(current->domain, _gfn(gmfn), false, NULL,page) )
Quoted 12 lines
xen/arch/x86/hvm/emulate.c +1
-9{ -put_page (*page); +case -EAGAIN :return X86EMUL_RETRY ; - } - -/* This code should not be reached if the gmfn is not RAM */ -if (p2m_is_mmio (p2mt) ) - { -domain_crash (curr_d); - -put_page (*page);Quoted 8 lines
xen/arch/x86/hvm/emulate.c static int hvmemul_do_io_buffer( · line 361
Quoted 5 lines
xen/arch/x86/hvm/emulate.c +4
-0+case -EINVAL :return X86EMUL_UNHANDLEABLE ; +default : +ASSERT_UNREACHABLE (); +case 0 :I think you'd better have "default:" fall through to "case
-EINVAL". Similarly elsewhere.
Quoted 1 line
Ok. I'll keep the
ASSERT_UNREACHABLE()though.
That's what I was implying by saying "fall through" - otherwise "case -EINVAL:" could as well have gone away.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 12/14] memory: add get_paged_gfn() as a wrapper...
Quoted 16 lines · Jan Beulich
Quoted 16 lines
Quoted 15 lines
Quoted 1 line · On 12.09.18 at 11:10, <[email protected]> wrote:
From: Xen-devel [mailto:[email protected]] On Behalf Of Jan Beulich
Sent: 11 September 2018 15:56Quoted 6 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
...for some uses of
get_page_from_gfn().There are many occurences of the following pattern in the code:
q = <readonly look-up> ? P2M_ALLOC : P2M_UNSHARE;Especially with this UNSHARE in mind - is "paged" in the helper function's name really suitable? Since we (I think) already have
get_gfn(), how abouttry_get_gfn()?
Quoted 5 lines
Quoted 3 lines
That name may be a little misleading since it suggests a close functional relationship with
get_gfn()whereas it does more than that. How abouttry_get_page_from_gfn()?Fine with me; George?
At the risk of bike shedding.. "try" to me means only pass/fail, with no side effects, and with no permissions checks. What about "check_and_get_page_from_gfn()"?
I'd prefer 'check' but if anyone objects I'd rather just go with 'try' and get things in -- the code is a definite improvement.
-George
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 12/14] memory: add get_paged_gfn() as a wrapper...
Quoted 35 lines
-----Original Message-----
From: George Dunlap [mailto:[email protected]]
Sent: 12 September 2018 11:02
To: Jan Beulich <[email protected]>; Paul Durrant <[email protected]>
Cc: Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; Ian Jackson <[email protected]>; Wei Liu <[email protected]>; Stefano Stabellini <[email protected]>; xen- devel <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>
Subject: Re: [Xen-devel] [PATCH v6 12/14] memory: addget_paged_gfn()as a wrapper...Quoted 2 lines · Jan Beulich
Quoted 2 lines
Quoted 2 lines
Quoted 1 line · On 12.09.18 at 11:10, <[email protected]> wrote:
From: Xen-devel [mailto:[email protected]] On
Behalf
Quoted 18 lines
Quoted 17 lines
Quoted 6 lines
Of Jan Beulich
Sent: 11 September 2018 15:56Quoted 3 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
...for some uses of
get_page_from_gfn().Quoted 7 lines
Quoted 3 lines
There are many occurences of the following pattern in the code:
q = <readonly look-up> ? P2M_ALLOC : P2M_UNSHARE;Especially with this UNSHARE in mind - is "paged" in the helper function's name really suitable? Since we (I think) already have
get_gfn(), how abouttry_get_gfn()?That name may be a little misleading since it suggests a close functional relationship with
get_gfn()whereas it does more than that. How abouttry_get_page_from_gfn()?
Quoted 9 lines
Quoted 1 line
Fine with me; George?
At the risk of bike shedding.. "try" to me means only pass/fail, with no side effects, and with no permissions checks. What about "
check_and_get_page_from_gfn()"?I'd prefer 'check' but if anyone objects I'd rather just go with 'try' and get things in -- the code is a definite improvement.
Jan?
Paul
Quoted 1 line
-George
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 12/14] memory: add get_paged_gfn() as a wrapper...
In reply to George DunlapQuoted 22 lines
Quoted 1 line
Quoted 1 line · On 12.09.18 at 12:01, <[email protected]> wrote:
Quoted 20 lines · Jan Beulich
Quoted 19 lines
Quoted 8 lines
Quoted 1 line · On 12.09.18 at 11:10, <[email protected]> wrote:
From: Xen-devel [mailto:[email protected]] On Behalf Of Jan Beulich
Sent: 11 September 2018 15:56Quoted 3 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
...for some uses of
get_page_from_gfn().Quoted 7 lines
Quoted 3 lines
There are many occurences of the following pattern in the code:
q = <readonly look-up> ? P2M_ALLOC : P2M_UNSHARE;Especially with this UNSHARE in mind - is "paged" in the helper function's name really suitable? Since we (I think) already have
get_gfn(), how abouttry_get_gfn()?That name may be a little misleading since it suggests a close functional relationship with
get_gfn()whereas it does more than that. How abouttry_get_page_from_gfn()?
Quoted 8 lines
Quoted 1 line
Fine with me; George?
At the risk of bike shedding.. "try" to me means only pass/fail, with no side effects, and with no permissions checks. What about "
check_and_get_page_from_gfn()"?I'd prefer 'check' but if anyone objects I'd rather just go with 'try' and get things in -- the code is a definite improvement.
I'm fine with "check", and indeed I wasn't really happy about the earlier proposed "try".
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 13/14 x86: add iommu_ops to modify and flush IOMMU mappings
In reply to Paul DurrantThis patch adds iommu_ops to add (map) or remove (unmap) frames in the domain's IOMMU mappings, and an iommu_op to synchronize (flush) those manipulations with the hardware.
Currently the flags value for each op must include the XEN_IOMMUOP_map/unmap/flush_all flag as the implementation does not yet support per-device mappings. The sbdf field of each hypercall is accordingly ignored.
Mappings added by the map operation are tracked and only those mappings may be removed by a subsequent unmap operation. Frames are specified by the owning domain and GFN. It is, of course, permissable for a domain to map its own frames using DOMID_SELF.
NOTE: The owning domain and GFN must also be specified in the unmap
operation, as well as the BFN, so that they can be cross-checked with the existent mapping.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
v6:
- Add placeholder sbdf field and flag to control scope of map, unmap and flush.
v4:
- Fixed logic inversion when checking return of iommu_unmap_page().
v3:
- Add type pinning.
v2:
- Heavily re-worked in v2, including explicit tracking of mappings. This avoids the need to clear non-reserved mappings from IOMMU at start of day, which would be prohibitively slow on a large host.
--- xen/common/iommu_op.c | 178 ++++++++++++++++++++++++++++++++++++++++++ xen/include/public/iommu_op.h | 110 ++++++++++++++++++++++++++ xen/include/xlat.lst | 3 + 3 files changed, 291 insertions(+)
xen/common/iommu_op.c +178 -0
@@ -123,6 +123,156 @@ static int iommu_op_enable_modification(return 0 ; } +static int iommuop_map (struct xen_iommu_op_map *op) +{ +struct domain *d, *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); +bool readonly = op->flags & XEN_IOMMUOP_map_readonly; +bfn_t bfn =_bfn (op->bfn); +struct page_info *page; +unsigned int prot; +int rc, ignore; + +if ( op->pad || + (op->flags & ~(XEN_IOMMUOP_map_all | + XEN_IOMMUOP_map_readonly)) ) +return -EINVAL ; + +if ( !iommu->iommu_op_ranges ) +return -EOPNOTSUPP ; + +/* Per-device mapping not yet supported */ +if ( !(op->flags & XEN_IOMMUOP_map_all) ) +return -EINVAL ; + +/* Check whether the specified BFN falls in a reserved region */ +if (rangeset_contains_singleton (iommu->reserved_ranges,bfn_x (bfn)) ) +return -EINVAL ; + + d =rcu_lock_domain_by_any_id (op->domid); +if ( !d ) +return -ESRCH ; + + rc =get_paged_gfn (d,_gfn (op->gfn), readonly,NULL , &page); +if ( rc ) + goto unlock; + + rc = -EINVAL ; +if ( !readonly && !get_page_type (page, PGT_writable_page) ) + { +put_page (page); + goto unlock; + } + + prot = IOMMUF_readable; +if ( !readonly ) + prot |= IOMMUF_writable; + + rc = -EIO ; +if (iommu_map_page (currd, bfn,page_to_mfn (page), prot) ) + goto release; + + rc =rangeset_add_singleton (iommu->iommu_op_ranges,bfn_x (bfn)); +if ( rc ) + goto unmap; + + rc =0 ; + goto unlock;/* retain mapping and references */ + + unmap: + ignore =iommu_unmap_page (currd, bfn); + + release: +if ( !readonly ) +put_page_type (page); +put_page (page); + + unlock: +rcu_unlock_domain (d); +return rc; +} + +static int iommuop_unmap (struct xen_iommu_op_unmap *op) +{ +struct domain *d, *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); +bfn_t bfn =_bfn (op->bfn); +mfn_t mfn; +bool readonly; +unsigned int prot; +struct page_info *page; +int rc; + +if ( op->pad || + (op->flags & ~XEN_IOMMUOP_unmap_all) ) +return -EINVAL ; + +if ( !iommu->iommu_op_ranges ) +return -EOPNOTSUPP ; + +/* Per-device unmapping not yet supported */ +if ( !(op->flags & XEN_IOMMUOP_unmap_all) ) +return -EINVAL ; + +if ( !rangeset_contains_singleton (iommu->iommu_op_ranges,bfn_x (bfn)) || +iommu_lookup_page (currd, bfn, &mfn, &prot) || + !mfn_valid (mfn) ) +return -ENOENT ; + + readonly = !(prot & IOMMUF_writable); + + d =rcu_lock_domain_by_any_id (op->domid); +if ( !d ) +return -ESRCH ; + + rc =get_paged_gfn (d,_gfn (op->gfn), !(prot & IOMMUF_writable),NULL , + &page); +if ( rc ) + goto unlock; + +put_page (page);/* release extra reference just taken */ + + rc = -EINVAL ; +if ( !mfn_eq (page_to_mfn (page), mfn) ) + goto unlock; + +/* release reference taken in map */ +if ( !readonly ) +put_page_type (page); +put_page (page); + + rc =rangeset_remove_singleton (iommu->iommu_op_ranges,bfn_x (bfn)); +if ( rc ) + goto unlock; + +if (iommu_unmap_page (currd, bfn) ) + rc = -EIO ; + + unlock: +rcu_unlock_domain (d); + +return rc; +} + +static int iommuop_flush (struct xen_iommu_op_flush *op) +{ +struct domain *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); + +if ( op->pad0 || op->pad1 || + (op->flags & ~XEN_IOMMUOP_flush_all) ) +return -EINVAL ; + +if ( !iommu->iommu_op_ranges ) +return -EOPNOTSUPP ; + +/* Per-device flushing not yet supported */ +if ( !(op->flags & XEN_IOMMUOP_flush_all) ) +return -EINVAL ; + +return !iommu_iotlb_flush_all (currd) ?0 : -EIO ; +} +static void iommu_op (xen_iommu_op_t *op) { switch ( op->op )
@@ -135,6 +285,22 @@ static void iommu_op(xen_iommu_op_t *op) op->status =iommu_op_enable_modification (&op->u.enable_modification);break ; +case XEN_IOMMUOP_map: +this_cpu (iommu_dont_flush_iotlb) =1 ; + op->status =iommuop_map (&op->u.map); +this_cpu (iommu_dont_flush_iotlb) =0 ; +break ; + +case XEN_IOMMUOP_unmap: +this_cpu (iommu_dont_flush_iotlb) =1 ; + op->status =iommuop_unmap (&op->u.unmap); +this_cpu (iommu_dont_flush_iotlb) =0 ; +break ; + +case XEN_IOMMUOP_flush: + op->status =iommuop_flush (&op->u.flush); +break ; +default : op->status = -EOPNOTSUPP ;break ;
@@ -148,6 +314,9 @@ int do_one_iommu_op(xen_iommu_op_buf_t *buf)static const size_t op_size[] = { [XEN_IOMMUOP_query_reserved] =sizeof (struct xen_iommu_op_query_reserved ), [XEN_IOMMUOP_enable_modification] =sizeof (struct xen_iommu_op_enable_modification ), + [XEN_IOMMUOP_map] =sizeof (struct xen_iommu_op_map ), + [XEN_IOMMUOP_unmap] =sizeof (struct xen_iommu_op_unmap ), + [XEN_IOMMUOP_flush] =sizeof (struct xen_iommu_op_flush ), };size_t size;int rc;
@@ -235,6 +404,9 @@ int compat_one_iommu_op(compat_iommu_op_buf_t *buf)static const size_t op_size[] = { [XEN_IOMMUOP_query_reserved] =sizeof (struct compat_iommu_op_query_reserved ), [XEN_IOMMUOP_enable_modification] =sizeof (struct compat_iommu_op_enable_modification ), + [XEN_IOMMUOP_map] =sizeof (struct compat_iommu_op_map ), + [XEN_IOMMUOP_unmap] =sizeof (struct compat_iommu_op_unmap ), + [XEN_IOMMUOP_flush] =sizeof (struct compat_iommu_op_flush ), };size_t size;xen_iommu_op_t nat;
@@ -273,6 +445,9 @@ int compat_one_iommu_op(compat_iommu_op_buf_t *buf)*/ #define XLAT_iommu_op_u_query_reserved XEN_IOMMUOP_query_reserved#define XLAT_iommu_op_u_enable_modification XEN_IOMMUOP_enable_modification +#define XLAT_iommu_op_u_map XEN_IOMMUOP_map +#define XLAT_iommu_op_u_unmap XEN_IOMMUOP_unmap +#define XLAT_iommu_op_u_flush XEN_IOMMUOP_flush u = cmp.op;#define XLAT_iommu_op_query_reserved_HNDL_ranges (_d_, _s_) \
@@ -332,6 +507,9 @@ int compat_one_iommu_op(compat_iommu_op_buf_t *buf) &cmp, status) )return -EFAULT ; +#undef XLAT_iommu_op_u_flush +#undef XLAT_iommu_op_u_unmap +#undef XLAT_iommu_op_u_map#undef XLAT_iommu_op_u_enable_modification#undef XLAT_iommu_op_u_query_reserved
xen/include/public/iommu_op.h +110 -0
122 lines not on this page · raw message
xen/include/xlat.lst +3 -0
@@ -80,7 +80,10 @@ ! iommu_op iommu_op.h ! iommu_op_buf iommu_op.h ! iommu_op_enable_modification iommu_op.h +! iommu_op_flush iommu_op.h +! iommu_op_map iommu_op.h ! iommu_op_query_reserved iommu_op.h +! iommu_op_unmap iommu_op.h ? iommu_reserved_range iommu_op.h ? kexec_exec kexec.h ! kexec_image kexec.h
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 13/14] x86: add iommu_ops to modify and flush IOMMU mappings
xen/common/iommu_op.c static int iommu_op_enable_modification( · line 129
Quoted 11 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/iommu_op.c +4
-0@@ -123,6 +123,156 @@ static int iommu_op_enable_modification(return 0 ; } +static int iommuop_map (struct xen_iommu_op_map *op) +{ +struct domain *d, *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd);
const (also in unmap)?
Quoted 13 lines
xen/common/iommu_op.c +13
-0+static int iommuop_unmap (struct xen_iommu_op_unmap *op) +{ +struct domain *d, *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); +bfn_t bfn =_bfn (op->bfn); +mfn_t mfn; +bool readonly; +unsigned int prot; +struct page_info *page; +int rc; + +if ( op->pad || + (op->flags & ~XEN_IOMMUOP_unmap_all) )
xen/common/iommu_op.c static int iommu_op_enable_modification( · line 217
Quoted 10 lines
xen/common/iommu_op.c +10
-0+return -EINVAL ; + +if ( !iommu->iommu_op_ranges ) +return -EOPNOTSUPP ; + +/* Per-device unmapping not yet supported */ +if ( !(op->flags & XEN_IOMMUOP_unmap_all) ) +return -EINVAL ; + + if ( !rangeset_contains_singleton(iommu->iommu_op_ranges, bfn_x(bfn)) ||
Again the question about a malicious multi-vCPU guest trying to utilize the gap between the check here and ...
Quoted 17 lines
xen/common/iommu_op.c +17
-0+iommu_lookup_page (currd, bfn, &mfn, &prot) || + !mfn_valid(mfn) ) + return -ENOENT; + + readonly = !(prot & IOMMUF_writable); + + d =rcu_lock_domain_by_any_id (op->domid); +if ( !d ) +return -ESRCH ; + + rc =get_paged_gfn (d,_gfn (op->gfn), !(prot & IOMMUF_writable),NULL , + &page); +if ( rc ) + goto unlock; + +put_page (page);/* release extra reference just taken */ +
xen/common/iommu_op.c static int iommu_op_enable_modification( · line 244
Quoted 10 lines
xen/common/iommu_op.c +10
-0+ rc = -EINVAL ; +if ( !mfn_eq (page_to_mfn (page), mfn) ) + goto unlock; + +/* release reference taken in map */ +if ( !readonly ) +put_page_type (page); +put_page (page); + + rc =rangeset_remove_singleton (iommu->iommu_op_ranges,bfn_x (bfn));
... the actual removal here.
xen/common/iommu_op.c static int iommu_op_enable_modification( · line 181
Quoted 2 lines
xen/common/iommu_op.c +2
-0+if ( rc ) + goto unlock;
You've already put the page ref - failure to remove needs to be fatal to the guest, or you'd need to re-obtain refs.
xen/common/iommu_op.c static int iommu_op_enable_modification( · line 249
Quoted 2 lines
xen/common/iommu_op.c +2
-0+if (iommu_unmap_page (currd, bfn) ) + rc = -EIO ;
Similarly here: All records of the page having a mapping are gone.
Quoted 5 lines
xen/common/iommu_op.c +1
-0@@ -135,6 +285,22 @@ static void iommu_op(xen_iommu_op_t *op) op->status =iommu_op_enable_modification (&op->u.enable_modification);break ; +case XEN_IOMMUOP_map:
xen/common/iommu_op.c static void iommu_op(xen_iommu_op_t *op) · line 298
Quoted 10 lines
xen/common/iommu_op.c +10
-0+this_cpu (iommu_dont_flush_iotlb) =1 ; + op->status =iommuop_map (&op->u.map); +this_cpu (iommu_dont_flush_iotlb) =0 ; +break ; + +case XEN_IOMMUOP_unmap: +this_cpu (iommu_dont_flush_iotlb) =1 ; + op->status =iommuop_unmap (&op->u.unmap); +this_cpu (iommu_dont_flush_iotlb) =0 ; +break ;
How is this flush suppression secure?
xen/include/public/iommu_op.h struct xen_iommu_op_enable_modification { · line 85
Quoted 9 lines
xen/include/public/iommu_op.h +3
-0@@ -80,6 +80,113 @@ struct xen_iommu_op_enable_modification {#define XEN_IOMMU_CAP_per_device_mappings (1u << _XEN_IOMMU_CAP_per_device_mappings) }; +/* +* XEN_IOMMUOP_map: Map a guest page in the IOMMU. +*/
Single line comment please (also below).
Quoted 9 lines
xen/include/public/iommu_op.h +9
-0+#define XEN_IOMMUOP_map 3 + +struct xen_iommu_op_map { +/* IN - The domid of the guest */ +domid_t domid; +/* +* IN - flags controlling the mapping. This should be a bitwise OR of the +* flags defined below. +*/
xen/include/public/iommu_op.h struct xen_iommu_op_enable_modification { · line 104
Quoted 10 lines
xen/include/public/iommu_op.h +10
-0+uint16_t flags; + +/* +* Should the mapping be created for all initiators? +* +* NOTE: This flag is currently required as the implementation does not yet +* support pre-device mappings. +*/ +#define _XEN_IOMMUOP_map_all 0 +#define XEN_IOMMUOP_map_all (1 << (_XEN_IOMMUOP_map_all))
Stray extra parentheses (also further down).
Quoted 20 lines
xen/include/public/iommu_op.h +20
-0+/* +* XEN_IOMMUOP_flush: Flush the IOMMU TLB. +*/ +#define XEN_IOMMUOP_flush 5 + +struct xen_iommu_op_flush { +/* +* IN - flags controlling flushing. This should be a bitwise OR of the +* flags defined below. +*/ +uint16_t flags; + +/* +* Should the mappings flushed for all initiators? +* +* NOTE: This flag is currently required as the implementation does not yet +* support pre-device mappings. +*/ +#define _XEN_IOMMUOP_flush_all 0 +#define XEN_IOMMUOP_flush_all (1 << (_XEN_IOMMUOP_flush_all))
xen/include/public/iommu_op.h struct xen_iommu_op_enable_modification { · line 188
Quoted 10 lines
xen/include/public/iommu_op.h +10
-0+ +uint16_t pad0; +uint32_t pad1; +/* +* IN - Segment/Bus/Device/Function of the initiator. +* +* NOTE: This is ignored if XEN_IOMMUOP_flush_all is set. +*/ +uint64_t sbdf; +};
No means to flush a single mapping, or a sub-range of the entire address space?
xen/include/xlat.lst line 83
Quoted 7 lines
xen/include/xlat.lst +1
-0@@ -80,7 +80,10 @@ ! iommu_op iommu_op.h ! iommu_op_buf iommu_op.h ! iommu_op_enable_modification iommu_op.h +! iommu_op_flush iommu_op.h
As long as there's no bfn_t in there, this could again be ? instead of ! ?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 13/14] x86: add iommu_ops to modify and flush IOMMU mappings
In reply to Paul DurrantQuoted 43 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/iommu_op.c +42
-0+static int iommuop_map (struct xen_iommu_op_map *op) +{ +struct domain *d, *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); +bool readonly = op->flags & XEN_IOMMUOP_map_readonly; +bfn_t bfn =_bfn (op->bfn); +struct page_info *page; +unsigned int prot; +int rc, ignore; + +if ( op->pad || + (op->flags & ~(XEN_IOMMUOP_map_all | + XEN_IOMMUOP_map_readonly)) ) +return -EINVAL ; + +if ( !iommu->iommu_op_ranges ) +return -EOPNOTSUPP ; + +/* Per-device mapping not yet supported */ +if ( !(op->flags & XEN_IOMMUOP_map_all) ) +return -EINVAL ; + +/* Check whether the specified BFN falls in a reserved region */ +if (rangeset_contains_singleton (iommu->reserved_ranges,bfn_x (bfn)) ) +return -EINVAL ; + + d =rcu_lock_domain_by_any_id (op->domid); +if ( !d ) +return -ESRCH ; + + rc =get_paged_gfn (d,_gfn (op->gfn), readonly,NULL , &page); +if ( rc ) + goto unlock; + + rc = -EINVAL ; +if ( !readonly && !get_page_type (page, PGT_writable_page) ) + { +put_page (page); + goto unlock; + } + + prot = IOMMUF_readable;
xen/common/iommu_op.c static int iommu_op_enable_modification( · line 177
Quoted 10 lines
xen/common/iommu_op.c +10
-0+if ( !readonly ) + prot |= IOMMUF_writable; + + rc = -EIO ; +if (iommu_map_page (currd, bfn,page_to_mfn (page), prot) ) + goto release; + + rc =rangeset_add_singleton (iommu->iommu_op_ranges,bfn_x (bfn)); +if ( rc ) + goto unmap;
When a mapping is requested for the same BFN that a prior mapping was already established for, the page refs of that prior mapping get leaked here. I don't think you want to require an intermediate unmap, so checking the rangeset first is not an option. Hence I think you need to look up the translation anyway, which may mean that the rangeset's usefulness is quite limited (relevant with the additional context of my question regarding it perhaps requiring a pretty much unbounded amount of memory).
In order to avoid shooting down all pre-existing RAM mappings - is there no way the page table entries could be marked to identify their origin?
I also have another more general concern: Allowing the guest to manipulate its IOMMU page tables means that it can deliberately shatter large pages, growing the overall memory footprint of the domain. I'm hesitant to say this, but I'm afraid that resource tracking of such "behind the scenes" allocations might be a necessary prereq for the PV IOMMU work.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 13/14] x86: add iommu_ops to modify and flush IOMMU mappings
Quoted 68 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 08:04
To: Paul Durrant <[email protected]>
Cc: Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; Wei Liu <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>
Subject: Re: [PATCH v6 13/14] x86: addiommu_opsto modify and flush IOMMU mappingsQuoted 15 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/iommu_op.c +14
-0+static int iommuop_map (struct xen_iommu_op_map *op) +{ +struct domain *d, *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); +bool readonly = op->flags & XEN_IOMMUOP_map_readonly; +bfn_t bfn =_bfn (op->bfn); +struct page_info *page; +unsigned int prot; +int rc, ignore; + +if ( op->pad || + (op->flags & ~(XEN_IOMMUOP_map_all | + XEN_IOMMUOP_map_readonly)) ) +return -EINVAL ;xen/common/iommu_op.c static int iommu_op_enable_modification(
Quoted 10 lines
xen/common/iommu_op.c +10
-0+ +if ( !iommu->iommu_op_ranges ) +return -EOPNOTSUPP ; + +/* Per-device mapping not yet supported */ +if ( !(op->flags & XEN_IOMMUOP_map_all) ) +return -EINVAL ; + +/* Check whether the specified BFN falls in a reserved region */ + if ( rangeset_contains_singleton(iommu->reserved_ranges,
bfn_x(bfn)) )Quoted 18 lines
xen/common/iommu_op.c +18
-0+return -EINVAL ; + + d =rcu_lock_domain_by_any_id (op->domid); +if ( !d ) +return -ESRCH ; + + rc =get_paged_gfn (d,_gfn (op->gfn), readonly,NULL , &page); +if ( rc ) + goto unlock; + + rc = -EINVAL ; +if ( !readonly && !get_page_type (page, PGT_writable_page) ) + { +put_page (page); + goto unlock; + } + + prot = IOMMUF_readable;xen/common/iommu_op.c static int iommu_op_enable_modification( · line 177
Quoted 10 lines
xen/common/iommu_op.c +10
-0+if ( !readonly ) + prot |= IOMMUF_writable; + + rc = -EIO ; +if (iommu_map_page (currd, bfn,page_to_mfn (page), prot) ) + goto release; + + rc =rangeset_add_singleton (iommu->iommu_op_ranges,bfn_x (bfn)); +if ( rc ) + goto unmap;
Quoted 9 lines
When a mapping is requested for the same BFN that a prior mapping was already established for, the page refs of that prior mapping get leaked here. I don't think you want to require an intermediate unmap, so checking the rangeset first is not an option. Hence I think you need to look up the translation anyway, which may mean that the rangeset's usefulness is quite limited (relevant with the additional context of my question regarding it perhaps requiring a pretty much unbounded amount of memory).
Yes, that's a good point. I could do a lookup to check whether the B[D]FN is already there though. Agreed that the memory is unounded unless the number of ranges is limited, which there is already a facility for. It is not ideal though.
Quoted 4 lines
In order to avoid shooting down all pre-existing RAM mappings - is there no way the page table entries could be marked to identify their origin?
I don't know whether that is possible; I'll have to find specs for Intel and AMD IOMMUs and see if they have PTE bits available for such a use.
Quoted 7 lines
I also have another more general concern: Allowing the guest to manipulate its IOMMU page tables means that it can deliberately shatter large pages, growing the overall memory footprint of the domain. I'm hesitant to say this, but I'm afraid that resource tracking of such "behind the scenes" allocations might be a necessary prereq for the PV IOMMU work.
Remember that PV-IOMMU is only available for dom0 as it stands (and that is the only use-case that XenServer currently has) so I think that, whilst the concern is valid, there is no need danger in putting the code without such tracking. Such work can be deferred to making PV-IOMMU for de-privileged guests... if that facility is needed.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 13/14] x86: add iommu_ops to modify and flush IOMMU mappings
Quoted 74 lines
Quoted 59 lines
Quoted 1 line · On 12.09.18 at 10:02, <[email protected]> wrote:
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 08:04Quoted 15 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/iommu_op.c +14
-0+static int iommuop_map (struct xen_iommu_op_map *op) +{ +struct domain *d, *currd = current->domain; +struct domain_iommu *iommu =dom_iommu (currd); +bool readonly = op->flags & XEN_IOMMUOP_map_readonly; +bfn_t bfn =_bfn (op->bfn); +struct page_info *page; +unsigned int prot; +int rc, ignore; + +if ( op->pad || + (op->flags & ~(XEN_IOMMUOP_map_all | + XEN_IOMMUOP_map_readonly)) ) +return -EINVAL ;xen/common/iommu_op.c static int iommu_op_enable_modification(
Quoted 10 lines
xen/common/iommu_op.c +10
-0+ +if ( !iommu->iommu_op_ranges ) +return -EOPNOTSUPP ; + +/* Per-device mapping not yet supported */ +if ( !(op->flags & XEN_IOMMUOP_map_all) ) +return -EINVAL ; + +/* Check whether the specified BFN falls in a reserved region */ + if ( rangeset_contains_singleton(iommu->reserved_ranges,
bfn_x(bfn)) )Quoted 18 lines
xen/common/iommu_op.c +18
-0+return -EINVAL ; + + d =rcu_lock_domain_by_any_id (op->domid); +if ( !d ) +return -ESRCH ; + + rc =get_paged_gfn (d,_gfn (op->gfn), readonly,NULL , &page); +if ( rc ) + goto unlock; + + rc = -EINVAL ; +if ( !readonly && !get_page_type (page, PGT_writable_page) ) + { +put_page (page); + goto unlock; + } + + prot = IOMMUF_readable;xen/common/iommu_op.c static int iommu_op_enable_modification( · line 177
Quoted 10 lines
xen/common/iommu_op.c +10
-0+if ( !readonly ) + prot |= IOMMUF_writable; + + rc = -EIO ; +if (iommu_map_page (currd, bfn,page_to_mfn (page), prot) ) + goto release; + + rc =rangeset_add_singleton (iommu->iommu_op_ranges,bfn_x (bfn)); +if ( rc ) + goto unmap;Quoted 9 lines
When a mapping is requested for the same BFN that a prior mapping was already established for, the page refs of that prior mapping get leaked here. I don't think you want to require an intermediate unmap, so checking the rangeset first is not an option. Hence I think you need to look up the translation anyway, which may mean that the rangeset's usefulness is quite limited (relevant with the additional context of my question regarding it perhaps requiring a pretty much unbounded amount of memory).
Yes, that's a good point. I could do a lookup to check whether the B[D]FN is already there though. Agreed that the memory is unounded unless the number of ranges is limited, which there is already a facility for. It is not ideal though.
Quoted 7 lines
Quoted 4 lines
In order to avoid shooting down all pre-existing RAM mappings - is there no way the page table entries could be marked to identify their origin?
I don't know whether that is possible; I'll have to find specs for Intel and AMD IOMMUs and see if they have PTE bits available for such a use.
I seem to vaguely recall the AMD side lacking any suitable bits.
Quoted 7 lines
Quoted 6 lines
I also have another more general concern: Allowing the guest to manipulate its IOMMU page tables means that it can deliberately shatter large pages, growing the overall memory footprint of the domain. I'm hesitant to say this, but I'm afraid that resource tracking of such "behind the scenes" allocations might be a necessary prereq for the PV IOMMU work.
Quoted 5 lines
Remember that PV-IOMMU is only available for dom0 as it stands (and that is the only use-case that XenServer currently has) so I think that, whilst the concern is valid, there is no need danger in putting the code without such tracking. Such work can be deferred to making PV-IOMMU for de-privileged guests... if that facility is needed.
Good point, but perhaps worth a prominent fixme note at a suitable place in code, like iirc Roger has done in a number of places for his vPCI work (which eventually we will want to extend to support DomU too).
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 13/14] x86: add iommu_ops to modify and flush IOMMU mappings
In reply to Paul DurrantQuoted 11 lines
From: Paul Durrant
Sent: Wednesday, September 12, 2018 4:02 PMQuoted 4 lines
In order to avoid shooting down all pre-existing RAM mappings - is there no way the page table entries could be marked to identify their origin?
I don't know whether that is possible; I'll have to find specs for Intel and AMD IOMMUs and see if they have PTE bits available for such a use.
there are ignored bits
Quoted 9 lines
Quoted 8 lines
I also have another more general concern: Allowing the guest to manipulate its IOMMU page tables means that it can deliberately shatter large pages, growing the overall memory footprint of the domain. I'm hesitant to say this, but I'm afraid that resource tracking of such "behind the scenes" allocations might be a necessary prereq for the PV IOMMU work.
Quoted 6 lines
Remember that PV-IOMMU is only available for dom0 as it stands (and that is the only use-case that XenServer currently has) so I think that, whilst the concern is valid, there is no need danger in putting the code without such tracking. Such work can be deferred to making PV-IOMMU for de-privileged guests... if that facility is needed.
I didn't get why this is PV-IOMMU specific. Guest can always manipulate guest CPU page table to shatter large pages too...
Thanks
Kevin
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 13/14] x86: add iommu_ops to modify and flush IOMMU mappings
Quoted 41 lines
-----Original Message-----
From: Tian, Kevin [mailto:[email protected]]
Sent: 13 September 2018 07:41
To: Paul Durrant <[email protected]>; 'Jan Beulich' <[email protected]>
Cc: Stefano Stabellini <[email protected]>; Wei Liu <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Andrew Cooper <[email protected]>; Tim (Xen.org) <[email protected]>; George Dunlap <[email protected]>; Julien Grall <[email protected]>; xen-devel <[email protected]>; Ian Jackson <[email protected]>
Subject: RE: [PATCH v6 13/14] x86: addiommu_opsto modify and flush IOMMU mappingsQuoted 11 lines
From: Paul Durrant
Sent: Wednesday, September 12, 2018 4:02 PMQuoted 4 lines
In order to avoid shooting down all pre-existing RAM mappings - is there no way the page table entries could be marked to identify their origin?
I don't know whether that is possible; I'll have to find specs for Intel and AMD IOMMUs and see if they have PTE bits available for such a use.
there are ignored bits
Quoted 9 lines
Quoted 8 lines
I also have another more general concern: Allowing the guest to manipulate its IOMMU page tables means that it can deliberately shatter large pages, growing the overall memory footprint of the domain. I'm hesitant to say this, but I'm afraid that resource tracking of such "behind the scenes" allocations might be a necessary prereq for the PV IOMMU work.
Quoted 4 lines
Remember that PV-IOMMU is only available for dom0 as it stands (and that is the only use-case that XenServer currently has) so I think that, whilst the concern is valid, there is no need danger in putting the code without such tracking. Such work can be deferred to making PV-IOMMU for de-
Quoted 7 lines
privileged
Quoted 2 lines
guests... if that facility is needed.
I didn't get why this is PV-IOMMU specific. Guest can always manipulate guest CPU page table to shatter large pages too...
At the moment that is true. I guess Jan doesn't want to introduce another way for a guest to cause Xen to consume large amounts of memory.
Paul
Quoted 2 lines
Thanks
Kevin
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 13/14] x86: add iommu_ops to modify and flush IOMMU mappings
In reply to Tian, KevinQuoted 12 lines
Quoted 12 lines
Quoted 1 line · On 13.09.18 at 08:41, <[email protected]> wrote:
From: Paul Durrant
Sent: Wednesday, September 12, 2018 4:02 PMQuoted 7 lines
I also have another more general concern: Allowing the guest to manipulate its IOMMU page tables means that it can deliberately shatter large pages, growing the overall memory footprint of the domain. I'm hesitant to say this, but I'm afraid that resource tracking of such "behind the scenes" allocations might be a necessary prereq for the PV IOMMU work.
Quoted 8 lines
Quoted 5 lines
Remember that PV-IOMMU is only available for dom0 as it stands (and that is the only use-case that XenServer currently has) so I think that, whilst the concern is valid, there is no need danger in putting the code without such tracking. Such work can be deferred to making PV-IOMMU for de-privileged guests... if that facility is needed.
I didn't get why this is PV-IOMMU specific. Guest can always manipulate guest CPU page table to shatter large pages too...
Hmm, good point. I keep forgetting that we allow guests to fiddle with their own p2m.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v6 14/14 x86: extend the map and unmap iommu_ops to support grant references
In reply to Paul DurrantThis patch allows a domain to add or remove foreign frames from its IOMMU mappings by grant reference as well as GFN. This is necessary, for example, to support a PV network backend that needs to construct a packet buffer that can be directly accessed by a NIC.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Wei Liu <[email protected]>
v6:
- Re-base.
v2:
- New in v2.
--- xen/common/grant_table.c | 143 ++++++++++++++++++++++++++++++++++++++++++ xen/common/iommu_op.c | 83 ++++++++++++++++-------- xen/include/public/iommu_op.h | 29 +++++++-- xen/include/xen/grant_table.h | 7 +++ 4 files changed, 232 insertions(+), 30 deletions(-)
xen/common/grant_table.c +143 -0
@@ -3961,6 +3961,149 @@ int gnttab_get_status_frame(struct domain *d, unsigned long idx,return rc; } +int +acquire_gref_for_iommu (struct domain *d,grant_ref_t gref, +bool readonly,mfn_t *mfn) +{ +struct domain *currd = current->domain; +struct grant_table *gt = d->grant_table; +grant_entry_header_t *shah; +struct active_grant_entry *act; +uint16_t *status; +int rc; + +grant_read_lock (gt); + + rc = -ENOENT ; +if ( gref >nr_grant_entries (gt) ) + goto unlock; + + act =active_entry_acquire (gt, gref); + shah =shared_entry_header (gt, gref); + status = ( gt->gt_version ==2 ) ? + &status_entry (gt, gref) : + &shah->flags; + + rc = -EACCES ; +if ( (shah->flags & GTF_type_mask) != GTF_permit_access || + (shah->flags & GTF_sub_page) ) + goto release; + + rc = -ERANGE ; +if ( act->pin && ((act->domid != currd->domain_id) || + (act->pin &0x80808080U ) !=0 ) ) + goto release; + + rc = -EINVAL ; +if ( !act->pin || + (!readonly && !(act->pin & GNTPIN_devw_mask)) ) { +if (_set_status (gt->gt_version, currd->domain_id, readonly, +0 , shah, act, status) != GNTST_okay ) + goto release; + } + +if ( !act->pin ) + { +gfn_t gfn = gt->gt_version ==1 ? +_gfn (shared_entry_v1 (gt, gref).frame) : +_gfn (shared_entry_v2 (gt, gref).full_page.frame); +struct page_info *page; + + rc =get_paged_gfn (d, gfn, readonly,NULL , &page); +if ( rc ) + goto clear; + +act_set_gfn (act, gfn); + act->mfn =page_to_mfn (page); + act->domid = currd->domain_id; + act->start =0 ; + act->length =PAGE_SIZE ; + act->is_sub_page = false; + act->trans_domain = d; + act->trans_gref = gref; + } +else + { +ASSERT (mfn_valid (act->mfn)); +if ( !get_page (mfn_to_page (act->mfn), d) ) + goto clear; + } + + rc =0 ; + act->pin += readonly ? GNTPIN_devr_inc : GNTPIN_devw_inc; + *mfn = act->mfn; + goto release; + + clear: +if ( !readonly && !(act->pin & GNTPIN_devw_mask) ) +gnttab_clear_flag (_GTF_writing, status); + +if ( !act->pin ) +gnttab_clear_flag (_GTF_reading, status); + + release: +active_entry_release (act); + + unlock: +grant_read_unlock (gt); + +return rc; +} + +int +release_gref_for_iommu (struct domain *d,grant_ref_t gref, +bool readonly,mfn_t mfn) +{ +struct domain *currd = current->domain; +struct grant_table *gt = d->grant_table; +grant_entry_header_t *shah; +struct active_grant_entry *act; +uint16_t *status; +int rc; + +grant_read_lock (gt); + + rc = -ENOENT ; +if ( gref >nr_grant_entries (gt) ) + goto unlock; + + act =active_entry_acquire (gt, gref); + shah =shared_entry_header (gt, gref); + status = ( gt->gt_version ==2 ) ? + &status_entry (gt, gref) : + &shah->flags; + + rc = -EINVAL ; +if ( !act->pin || (act->domid != currd->domain_id) || + !mfn_eq (act->mfn, mfn) ) + goto release; + + rc =0 ; +if ( readonly ) + act->pin -= GNTPIN_devr_inc; +else + { +gnttab_mark_dirty (d, mfn); + + act->pin -= GNTPIN_devw_inc; +if ( !(act->pin & GNTPIN_devw_mask) ) +gnttab_clear_flag (_GTF_writing, status); + } + +if ( !act->pin ) +gnttab_clear_flag (_GTF_reading, status); + +put_page (mfn_to_page (mfn)); + + release: +active_entry_release (act); + + unlock: +grant_read_unlock (gt); + +return rc; +} +static void gnttab_usage_print (struct domain *rd) {int first =1 ;
xen/common/iommu_op.c +57 -26
138 lines not on this page · raw message
xen/include/public/iommu_op.h +25 -4
@@ -24,6 +24,7 @@#define XEN_PUBLIC_IOMMU_OP_H #include "xen.h" +#include "grant_table.h" typedef uint64_t xen_bfn_t ;
@@ -107,6 +108,10 @@ struct xen_iommu_op_map {#define _XEN_IOMMUOP_map_readonly 1#define XEN_IOMMUOP_map_readonly (1 << (_XEN_IOMMUOP_map_readonly)) +/* Is the memory specified by gfn or grant reference? */ +#define _XEN_IOMMUOP_map_gref 2 +#define XEN_IOMMUOP_map_gref (1 << (_XEN_IOMMUOP_map_gref)) +uint32_t pad;/* * IN - Segment/Bus/Device/Function of the initiator.
@@ -116,8 +121,14 @@ struct xen_iommu_op_map {uint64_t sbdf;/* IN - The IOMMU frame number which will hold the new mapping */ xen_bfn_t bfn; -/* IN - The guest frame number of the page to be mapped */ -xen_pfn_t gfn; +/* +* IN - The guest frame number or grant reference of the page to +* be mapped. +*/ +union { +xen_pfn_t gfn; +grant_ref_t gref; + } u; };/*
@@ -143,6 +154,10 @@ struct xen_iommu_op_unmap {#define _XEN_IOMMUOP_unmap_all 0#define XEN_IOMMUOP_unmap_all (1 << (_XEN_IOMMUOP_unmap_all)) +/* Is the memory specified by gfn or grant reference? */ +#define _XEN_IOMMUOP_unmap_gref 1 +#define XEN_IOMMUOP_unmap_gref (1 << (_XEN_IOMMUOP_unmap_gref)) +uint32_t pad;/* * IN - Segment/Bus/Device/Function of the initiator.
@@ -152,8 +167,14 @@ struct xen_iommu_op_unmap {uint64_t sbdf;/* IN - The IOMMU frame number which holds the mapping to be removed */ xen_bfn_t bfn; -/* IN - The guest frame number of the page that is mapped */ -xen_pfn_t gfn; +/* +* IN - The guest frame number or grant reference of the page that +* is mapped. +*/ +union { +xen_pfn_t gfn; +grant_ref_t gref; + } u; };/*
xen/include/xen/grant_table.h +7 -0
@@ -63,6 +63,13 @@ int gnttab_get_shared_frame(struct domain *d, unsigned long idx,int gnttab_get_status_frame (struct domain *d,unsigned long idx,mfn_t *mfn); +int +acquire_gref_for_iommu (struct domain *d,grant_ref_t gref, +bool readonly,mfn_t *mfn); +int +release_gref_for_iommu (struct domain *d,grant_ref_t gref, +bool readonly,mfn_t mfn); +unsigned int gnttab_dom0_frames (void );#endif /* __XEN_GRANT_TABLE_H__ */
Signature · 7 lines
2.11.0 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 14/14] x86: extend the map and unmap iommu_ops to support grant references
Quoted 19 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/grant_table.c +15
-0+int +acquire_gref_for_iommu (struct domain *d,grant_ref_t gref, +bool readonly,mfn_t *mfn) +{ +struct domain *currd = current->domain; +struct grant_table *gt = d->grant_table; +grant_entry_header_t *shah; +struct active_grant_entry *act; +uint16_t *status; +int rc; + +grant_read_lock (gt); + + rc = -ENOENT ; +if ( gref >nr_grant_entries (gt) )= (also in the release counterpart)
xen/common/grant_table.c int gnttab_get_status_frame(struct domain *d, unsigned long idx, · line 3983
Quoted 5 lines
xen/common/grant_table.c +5
-0+ goto unlock; + + act =active_entry_acquire (gt, gref); + shah =shared_entry_header (gt, gref); + status = ( gt->gt_version == 2 ) ?
Stray blanks. Further down in a similar construct you even omit the parentheses, which you could as well do here too. Again also below.
Review
Quoted 2 lines
(quoted diff) +2
-0+ &status_entry(gt, gref) : + &shah->flags;
The whole thing does not fit on a line?
xen/common/grant_table.c int gnttab_get_status_frame(struct domain *d, unsigned long idx, · line 3990
Quoted 4 lines
xen/common/grant_table.c +4
-0+ rc = -EACCES ; +if ( (shah->flags & GTF_type_mask) != GTF_permit_access || + (shah->flags & GTF_sub_page) ) + goto release;
So transitive grants are okay despite there being no special handling anywhere in the function?
xen/common/grant_table.c int gnttab_get_status_frame(struct domain *d, unsigned long idx, · line 3994
Quoted 3 lines
xen/common/grant_table.c +3
-0+ rc = -ERANGE ; +if ( act->pin && ((act->domid != currd->domain_id) || + (act->pin &0x80808080U ) !=0 ) )
You want to check only two of the four top bits, as you only add in GNTPIN_dev{r,w}_inc below.
xen/common/grant_table.c int gnttab_get_status_frame(struct domain *d, unsigned long idx, · line 4003
Quoted 9 lines
xen/common/grant_table.c +9
-0+ goto release; + + rc = -EINVAL ; +if ( !act->pin || + (!readonly && !(act->pin & GNTPIN_devw_mask)) ) { +if (_set_status (gt->gt_version, currd->domain_id, readonly, +0 , shah, act, status) != GNTST_okay ) + goto release; + }
Please combine the two if()-s.
Quoted 16 lines
xen/common/grant_table.c +16
-0+if ( !act->pin ) + { +gfn_t gfn = gt->gt_version ==1 ? +_gfn (shared_entry_v1 (gt, gref).frame) : +_gfn (shared_entry_v2 (gt, gref).full_page.frame); +struct page_info *page; + + rc =get_paged_gfn (d, gfn, readonly,NULL , &page); +if ( rc ) + goto clear; + +act_set_gfn (act, gfn); + act->mfn =page_to_mfn (page); + act->domid = currd->domain_id; + act->start =0 ; + act->length =PAGE_SIZE ;
xen/common/grant_table.c int gnttab_get_status_frame(struct domain *d, unsigned long idx, · line 4030
Quoted 10 lines
xen/common/grant_table.c +10
-0+ act->is_sub_page = false; + act->trans_domain = d; + act->trans_gref = gref; + } +else + { +ASSERT (mfn_valid (act->mfn)); +if ( !get_page (mfn_to_page (act->mfn), d) ) + goto clear; + }
Don't you also need to also acquire a write ref here if !readonly?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [PATCH v6 14/14] x86: extend the map and unmap iommu_ops to support grant references
Quoted 4 lines
-----Original Message-----
From: Jan Beulich [mailto:[email protected]]
Sent: 12 September 2018 15:13
To: Paul Durrant <[email protected]>
Quoted 28 lines
Cc: Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; Wei Liu <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Stefano Stabellini <[email protected]>; xen-devel <xen- [email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>
Subject: Re: [PATCH v6 14/14] x86: extend the map and unmapiommu_opsto support grant referencesQuoted 18 lines
Quoted 1 line
Quoted 1 line · On 23.08.18 at 11:47, <[email protected]> wrote:
xen/common/grant_table.c +15
-0+int +acquire_gref_for_iommu (struct domain *d,grant_ref_t gref, +bool readonly,mfn_t *mfn) +{ +struct domain *currd = current->domain; +struct grant_table *gt = d->grant_table; +grant_entry_header_t *shah; +struct active_grant_entry *act; +uint16_t *status; +int rc; + +grant_read_lock (gt); + + rc = -ENOENT ; +if ( gref >nr_grant_entries (gt) )= (also in the release counterpart)
Yes, good point.
Quoted 8 lines
xen/common/grant_table.c int gnttab_get_status_frame(struct domain *d, unsigned long idx, · line 3983
Quoted 5 lines
xen/common/grant_table.c +5
-0+ goto unlock; + + act =active_entry_acquire (gt, gref); + shah =shared_entry_header (gt, gref); + status = ( gt->gt_version == 2 ) ?Stray blanks. Further down in a similar construct you even omit the parentheses, which you could as well do here too. Again also below.
Ok.
Quoted 5 lines
Review
Quoted 3 lines
(quoted diff) +2
-0+ &status_entry(gt, gref) : + &shah->flags;The whole thing does not fit on a line?
I'll check.
Quoted 9 lines
xen/common/grant_table.c int gnttab_get_status_frame(struct domain *d, unsigned long idx, · line 3990
Quoted 5 lines
xen/common/grant_table.c +4
-0+ rc = -EACCES ; +if ( (shah->flags & GTF_type_mask) != GTF_permit_access || + (shah->flags & GTF_sub_page) ) + goto release;So transitive grants are okay despite there being no special handling anywhere in the function?
No, they do need to be avoided.
Quoted 7 lines
xen/common/grant_table.c int gnttab_get_status_frame(struct domain *d, unsigned long idx, · line 3994
Quoted 3 lines
xen/common/grant_table.c +3
-0+ rc = -ERANGE ; +if ( act->pin && ((act->domid != currd->domain_id) || + (act->pin &0x80808080U ) !=0 ) )You want to check only two of the four top bits, as you only add in GNTPIN_dev{r,w}_inc below.
True.
Quoted 10 lines
xen/common/grant_table.c int gnttab_get_status_frame(struct domain *d, unsigned long idx, · line 4003
Quoted 9 lines
xen/common/grant_table.c +9
-0+ goto release; + + rc = -EINVAL ; +if ( !act->pin || + (!readonly && !(act->pin & GNTPIN_devw_mask)) ) { +if (_set_status (gt->gt_version, currd->domain_id, readonly, +0 , shah, act, status) != GNTST_okay ) + goto release; + }
Quoted 2 lines
Please combine the two if()-s.
Ok.
Quoted 20 lines
Quoted 20 lines
xen/common/grant_table.c +20
-0+if ( !act->pin ) + { +gfn_t gfn = gt->gt_version ==1 ? +_gfn (shared_entry_v1 (gt, gref).frame) : +_gfn (shared_entry_v2 (gt, gref).full_page.frame); +struct page_info *page; + + rc =get_paged_gfn (d, gfn, readonly,NULL , &page); +if ( rc ) + goto clear; + +act_set_gfn (act, gfn); + act->mfn =page_to_mfn (page); + act->domid = currd->domain_id; + act->start =0 ; + act->length =PAGE_SIZE ; + act->is_sub_page = false; + act->trans_domain = d; + act->trans_gref = gref; + }
Quoted 9 lines
xen/common/grant_table.c int gnttab_get_status_frame(struct domain *d, unsigned long idx, · line 4030
Quoted 6 lines
xen/common/grant_table.c +6
-0+else + { +ASSERT (mfn_valid (act->mfn)); +if ( !get_page (mfn_to_page (act->mfn), d) ) + goto clear; + }Don't you also need to also acquire a write ref here if !readonly?
Yes, perhaps I should do a get_page_type() in iommuop_map() regardless of whether the page comes from a gfn or a gref lookup.
Paul
Quoted 2 lines
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel