IOMMU cleanup
cover IOMMU cleanup
From: Paul Durrant <[email protected]>
Patches accumulated in my during 4.14 freeze...
Paul Durrant (6):
x86/iommu: re-arrange arch_iommu to separate common fields... x86/iommu: add common page-table allocator
iommu: remove iommu_lookup_page() and the lookup_page() method... remove remaining uses of iommu_legacy_map/unmap
iommu: remove the share_p2m operation
iommu: stop calling IOMMU page tables 'p2m tables'
xen/arch/x86/mm.c | 22 ++- xen/arch/x86/mm/p2m-ept.c | 22 ++- xen/arch/x86/mm/p2m-pt.c | 17 +- xen/arch/x86/mm/p2m.c | 31 ++- xen/arch/x86/tboot.c | 4 +- xen/arch/x86/x86_64/mm.c | 27 ++- xen/common/grant_table.c | 36 +++- xen/common/memory.c | 7 +- xen/drivers/passthrough/amd/iommu.h | 18 +- xen/drivers/passthrough/amd/iommu_guest.c | 8 +- xen/drivers/passthrough/amd/iommu_map.c | 22 +-- xen/drivers/passthrough/amd/pci_amd_iommu.c | 109 +++-------- xen/drivers/passthrough/iommu.c | 91 +-------- xen/drivers/passthrough/vtd/iommu.c | 206 ++++++-------------- xen/drivers/passthrough/x86/iommu.c | 42 +++- xen/include/asm-x86/iommu.h | 33 +++- xen/include/xen/iommu.h | 35 +--- 17 files changed, 303 insertions(+), 427 deletions(-)
Signature · 1 line
2.20.1
PATCH 1/6 x86/iommu: re-arrange arch_iommu to separate common fields...
From: Paul Durrant <[email protected]>
... from those specific to VT-d or AMD IOMMU, and put the latter in a union.
There is no functional change in this patch, although the initialization of the 'mapped_rmrrs' list occurs slightly later in iommu_domain_init() since it is now done (correctly) in VT-d specific code rather than in general x86 code.
---
Cc: Lukasz Hawrylko <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Kevin Tian <[email protected]>
--- xen/arch/x86/tboot.c | 4 +- xen/drivers/passthrough/amd/iommu_guest.c | 8 ++-- xen/drivers/passthrough/amd/iommu_map.c | 14 +++--- xen/drivers/passthrough/amd/pci_amd_iommu.c | 35 +++++++------- xen/drivers/passthrough/vtd/iommu.c | 53 +++++++++++---------- xen/drivers/passthrough/x86/iommu.c | 1 - xen/include/asm-x86/iommu.h | 27 +++++++---- 7 files changed, 78 insertions(+), 64 deletions(-)
xen/arch/x86/tboot.c +2 -2
@@ -230,8 +230,8 @@ static void tboot_gen_domain_integrity(const uint8_t key[TB_KEY_SIZE], {const struct domain_iommu *dio =dom_iommu (d); -update_iommu_mac (&ctx, dio->arch.pgd_maddr, -agaw_to_level (dio->arch.agaw)); +update_iommu_mac (&ctx, dio->arch.vtd.pgd_maddr, +agaw_to_level (dio->arch.vtd.agaw)); } }
xen/drivers/passthrough/amd/iommu_guest.c +4 -4
@@ -50,12 +50,12 @@ static uint16_t guest_bdf(struct domain *d, uint16_t machine_bdf)static inline struct guest_iommu *domain_iommu (struct domain *d) { -return dom_iommu (d)->arch.g_iommu; +return dom_iommu (d)->arch.amd_iommu.g_iommu; }static inline struct guest_iommu *vcpu_iommu (struct vcpu *v) { -return dom_iommu (v->domain)->arch.g_iommu; +return dom_iommu (v->domain)->arch.amd_iommu.g_iommu; }static void guest_iommu_enable (struct guest_iommu *iommu)
@@ -823,7 +823,7 @@ int guest_iommu_init(struct domain* d)guest_iommu_reg_init (iommu); iommu->mmio_base = ~0ULL ; iommu->domain = d; - hd->arch.g_iommu = iommu; + hd->arch.amd_iommu.g_iommu = iommu;tasklet_init (&iommu->cmd_buffer_tasklet, guest_iommu_process_command, d);
@@ -845,5 +845,5 @@ void guest_iommu_destroy(struct domain *d)tasklet_kill (&iommu->cmd_buffer_tasklet);xfree (iommu); -dom_iommu (d)->arch.g_iommu =NULL ; +dom_iommu (d)->arch.amd_iommu.g_iommu =NULL ; }
xen/drivers/passthrough/amd/iommu_map.c +7 -7
@@ -180,8 +180,8 @@ static int iommu_pde_from_dfn(struct domain *d, unsigned long dfn,struct page_info *table;const struct domain_iommu *hd =dom_iommu (d); - table = hd->arch.root_table; - level = hd->arch.paging_mode; + table = hd->arch.amd_iommu.root_table; + level = hd->arch.amd_iommu.paging_mode;BUG_ON ( table ==NULL || level <1 || level >6 );
@@ -325,7 +325,7 @@ int amd_iommu_unmap_page(struct domain *d, dfn_t dfn,spin_lock (&hd->arch.mapping_lock); -if ( !hd->arch.root_table ) +if ( !hd->arch.amd_iommu.root_table ) {spin_unlock (&hd->arch.mapping_lock);return 0 ;
@@ -450,7 +450,7 @@ int __init amd_iommu_quarantine_init(struct domain *d)unsigned int level =amd_iommu_get_paging_mode (end_gfn);struct amd_iommu_pte *table; -if ( hd->arch.root_table ) +if ( hd->arch.amd_iommu.root_table ) {ASSERT_UNREACHABLE ();return 0 ;
@@ -458,11 +458,11 @@ int __init amd_iommu_quarantine_init(struct domain *d)spin_lock (&hd->arch.mapping_lock); - hd->arch.root_table =alloc_amd_iommu_pgtable (); -if ( !hd->arch.root_table ) + hd->arch.amd_iommu.root_table =alloc_amd_iommu_pgtable (); +if ( !hd->arch.amd_iommu.root_table ) goto out; - table =__map_domain_page (hd->arch.root_table); + table =__map_domain_page (hd->arch.amd_iommu.root_table);while ( level ) {struct page_info *pg;
xen/drivers/passthrough/amd/pci_amd_iommu.c +19 -16
@@ -92,7 +92,8 @@ static void amd_iommu_setup_domain_device(u8 bus = pdev->bus;const struct domain_iommu *hd =dom_iommu (domain); -BUG_ON ( !hd->arch.root_table || !hd->arch.paging_mode || +BUG_ON ( !hd->arch.amd_iommu.root_table || + !hd->arch.amd_iommu.paging_mode || !iommu->dev_table.buffer );if ( iommu_hwdom_passthrough &&is_hardware_domain (domain) )
@@ -111,8 +112,8 @@ static void amd_iommu_setup_domain_device(/* bind DTE to domain page-tables */ amd_iommu_set_root_page_table ( - dte,page_to_maddr (hd->arch.root_table), domain->domain_id, - hd->arch.paging_mode, valid); + dte,page_to_maddr (hd->arch.amd_iommu.root_table), + domain->domain_id, hd->arch.amd_iommu.paging_mode, valid);/* Undo what amd_iommu_disable_domain_device() may have done. */ ivrs_dev = &get_ivrs_mappings (iommu->seg)[req_id];
@@ -132,8 +133,8 @@ static void amd_iommu_setup_domain_device("root table = %#" PRIx64", " "domain = %d, paging mode = %d\n" , req_id, pdev->type, -page_to_maddr (hd->arch.root_table), - domain->domain_id, hd->arch.paging_mode); +page_to_maddr (hd->arch.amd_iommu.root_table), + domain->domain_id, hd->arch.amd_iommu.paging_mode); }spin_unlock_irqrestore (&iommu->lock, flags);
@@ -207,10 +208,10 @@ static int iov_enable_xt(void)int amd_iommu_alloc_root (struct domain_iommu *hd) { -if (unlikely (!hd->arch.root_table) ) +if (unlikely (!hd->arch.amd_iommu.root_table) ) { - hd->arch.root_table =alloc_amd_iommu_pgtable (); -if ( !hd->arch.root_table ) + hd->arch.amd_iommu.root_table =alloc_amd_iommu_pgtable (); +if ( !hd->arch.amd_iommu.root_table )return -ENOMEM ; }
@@ -240,7 +241,7 @@ static int amd_iommu_domain_init(struct domain *d)* physical address space we give it, but this isn't known yet so use 4 * unilaterally. */ - hd->arch.paging_mode =amd_iommu_get_paging_mode ( + hd->arch.amd_iommu.paging_mode =amd_iommu_get_paging_mode (is_hvm_domain (d) ?1ul << (DEFAULT_DOMAIN_ADDRESS_WIDTH -PAGE_SHIFT ) :get_upper_mfn_bound () +1 );
@@ -306,7 +307,7 @@ static void amd_iommu_disable_domain_device(const struct domain *domain,AMD_IOMMU_DEBUG ("Disable: device id = %#x, " "domain = %d, paging mode = %d\n" , req_id, domain->domain_id, -dom_iommu (domain)->arch.paging_mode); +dom_iommu (domain)->arch.amd_iommu.paging_mode); }spin_unlock_irqrestore (&iommu->lock, flags);
@@ -422,10 +423,11 @@ static void deallocate_iommu_page_tables(struct domain *d)struct domain_iommu *hd =dom_iommu (d);spin_lock (&hd->arch.mapping_lock); -if ( hd->arch.root_table ) +if ( hd->arch.amd_iommu.root_table ) { -deallocate_next_page_table (hd->arch.root_table, hd->arch.paging_mode); - hd->arch.root_table =NULL ; +deallocate_next_page_table (hd->arch.amd_iommu.root_table, + hd->arch.amd_iommu.paging_mode); + hd->arch.amd_iommu.root_table =NULL ; }spin_unlock (&hd->arch.mapping_lock); }
@@ -605,11 +607,12 @@ static void amd_dump_p2m_table(struct domain *d) {const struct domain_iommu *hd =dom_iommu (d); -if ( !hd->arch.root_table ) +if ( !hd->arch.amd_iommu.root_table )return ; -printk ("p2m table has %d levels\n" , hd->arch.paging_mode); -amd_dump_p2m_table_level (hd->arch.root_table, hd->arch.paging_mode,0 ,0 ); +printk ("p2m table has %d levels\n" , hd->arch.amd_iommu.paging_mode); +amd_dump_p2m_table_level (hd->arch.amd_iommu.root_table, + hd->arch.amd_iommu.paging_mode,0 ,0 ); } static const struct iommu_ops __initconstrel _iommu_ops = {
xen/drivers/passthrough/vtd/iommu.c +29 -24
156 lines not on this page · raw message
xen/drivers/passthrough/x86/iommu.c +0 -1
@@ -139,7 +139,6 @@ int arch_iommu_domain_init(struct domain *d)struct domain_iommu *hd =dom_iommu (d);spin_lock_init (&hd->arch.mapping_lock); -INIT_LIST_HEAD (&hd->arch.mapped_rmrrs);return 0 ; }
xen/include/asm-x86/iommu.h +17 -10
@@ -45,16 +45,23 @@ typedef uint64_t daddr_t;struct arch_iommu { -u64 pgd_maddr;/* io page directory machine address */ -spinlock_t mapping_lock;/* io page table lock */ -int agaw;/* adjusted guest address width, 0 is level 2 30-bit */ -u64 iommu_bitmap;/* bitmap of iommu(s) that the domain uses */ -struct list_head mapped_rmrrs; - -/* amd iommu support */ -int paging_mode; -struct page_info *root_table; -struct guest_iommu *g_iommu; +spinlock_t mapping_lock;/* io page table lock */ + +union { +/* Intel VT-d */ +struct { +u64 pgd_maddr;/* io page directory machine address */ +int agaw;/* adjusted guest address width, 0 is level 2 30-bit */ +u64 iommu_bitmap;/* bitmap of iommu(s) that the domain uses */ +struct list_head mapped_rmrrs; + } vtd; +/* AMD IOMMU */ +struct { +int paging_mode; +struct page_info *root_table; +struct guest_iommu *g_iommu; + } amd_iommu; + }; };extern struct iommu_ops iommu_ops;
Signature · 1 line
2.20.1
Re: [PATCH 1/6] x86/iommu: re-arrange arch_iommu to separate common fields...
Quoted 25 lines · Paul Durrant
xen/include/asm-x86/iommu.h +7
-10@@ -45,16 +45,23 @@ typedef uint64_t daddr_t; struct arch_iommu { -u64 pgd_maddr;/* io page directory machine address */ -spinlock_t mapping_lock;/* io page table lock */ -int agaw;/* adjusted guest address width, 0 is level 2 30-bit */ -u64 iommu_bitmap;/* bitmap of iommu(s) that the domain uses */ -struct list_head mapped_rmrrs; - -/* amd iommu support */ -int paging_mode; -struct page_info *root_table; -struct guest_iommu *g_iommu; + spinlock_t mapping_lock; /* io page table lock */ + + union { + /* Intel VT-d */ + struct { + u64 pgd_maddr; /* io page directory machine address */ + int agaw; /* adjusted guest address width, 0 is level 2 30-bit */
xen/include/asm-x86/iommu.h typedef uint64_t daddr_t; · line 64
Quoted 10 lines
xen/include/asm-x86/iommu.h +10
-0+u64 iommu_bitmap;/* bitmap of iommu(s) that the domain uses */ +struct list_head mapped_rmrrs; + } vtd; +/* AMD IOMMU */ +struct { +int paging_mode; +struct page_info *root_table; +struct guest_iommu *g_iommu; + } amd_iommu; + };
The naming split here is weird.
Ideally we'd have struct {vtd,amd}_iommu in appropriate headers, and this would be simply
union {
struct vtd_iommu vtd;
struct amd_iommu amd;
};
If this isn't trivial to arrange, can we at least s/amd_iommu/amd/ here ?
~Andrew
RE: [PATCH 1/6] x86/iommu: re-arrange arch_iommu to separate common fields...
Quoted 51 lines
-----Original Message-----
From: Andrew Cooper <[email protected]>
Sent: 24 July 2020 18:29
To: Paul Durrant <[email protected]>; [email protected]
Cc: Paul Durrant <[email protected]>; Lukasz Hawrylko <[email protected]>; Jan Beulich <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monné <[email protected]>; Kevin Tian <[email protected]>
Subject: Re: [PATCH 1/6] x86/iommu: re-arrangearch_iommuto separate common fields...Quoted 25 lines · Paul Durrant
xen/include/asm-x86/iommu.h +7
-10@@ -45,16 +45,23 @@ typedef uint64_t daddr_t; struct arch_iommu { -u64 pgd_maddr;/* io page directory machine address */ -spinlock_t mapping_lock;/* io page table lock */ -int agaw;/* adjusted guest address width, 0 is level 2 30-bit */ -u64 iommu_bitmap;/* bitmap of iommu(s) that the domain uses */ -struct list_head mapped_rmrrs; - -/* amd iommu support */ -int paging_mode; -struct page_info *root_table; -struct guest_iommu *g_iommu; + spinlock_t mapping_lock; /* io page table lock */ + + union { + /* Intel VT-d */ + struct { + u64 pgd_maddr; /* io page directory machine address */ + int agaw; /* adjusted guest address width, 0 is level 2 30-bit */xen/include/asm-x86/iommu.h typedef uint64_t daddr_t; · line 64
Quoted 10 lines
xen/include/asm-x86/iommu.h +10
-0+u64 iommu_bitmap;/* bitmap of iommu(s) that the domain uses */ +struct list_head mapped_rmrrs; + } vtd; +/* AMD IOMMU */ +struct { +int paging_mode; +struct page_info *root_table; +struct guest_iommu *g_iommu; + } amd_iommu; + };The naming split here is weird.
Ideally we'd have struct {vtd,amd}_iommu in appropriate headers, and this would be simply
Quoted 6 lines
union { struct vtd_iommu vtd; struct amd_iommu amd; };If this isn't trivial to arrange, can we at least s/amd_iommu/amd/ here ?
I was in two minds. I tried to look for a TLA for the AMD IOMMU and 'amd' seemed a little too non-descript. I don't really mind though if there's a strong preference to shorted it. I can certainly try moving the struct definitions into the implementation headers.
Paul
Quoted 2 lines
~Andrew
Re: [PATCH 1/6] x86/iommu: re-arrange arch_iommu to separate common fields...
Quoted 45 lines · Paul Durrant
Quoted 45 lines
From: Andrew Cooper <[email protected]>
Sent: 24 July 2020 18:29Quoted 25 lines · Paul Durrant
xen/include/asm-x86/iommu.h +7
-10@@ -45,16 +45,23 @@ typedef uint64_t daddr_t; struct arch_iommu { -u64 pgd_maddr;/* io page directory machine address */ -spinlock_t mapping_lock;/* io page table lock */ -int agaw;/* adjusted guest address width, 0 is level 2 30-bit */ -u64 iommu_bitmap;/* bitmap of iommu(s) that the domain uses */ -struct list_head mapped_rmrrs; - -/* amd iommu support */ -int paging_mode; -struct page_info *root_table; -struct guest_iommu *g_iommu; + spinlock_t mapping_lock; /* io page table lock */ + + union { + /* Intel VT-d */ + struct { + u64 pgd_maddr; /* io page directory machine address */ + int agaw; /* adjusted guest address width, 0 is level 2 30-bit */xen/include/asm-x86/iommu.h typedef uint64_t daddr_t; · line 64
Quoted 10 lines
xen/include/asm-x86/iommu.h +10
-0+u64 iommu_bitmap;/* bitmap of iommu(s) that the domain uses */ +struct list_head mapped_rmrrs; + } vtd; +/* AMD IOMMU */ +struct { +int paging_mode; +struct page_info *root_table; +struct guest_iommu *g_iommu; + } amd_iommu; + };The naming split here is weird.
Ideally we'd have struct {vtd,amd}_iommu in appropriate headers, and this would be simply
Quoted 8 lines
Quoted 6 lines
union { struct vtd_iommu vtd; struct amd_iommu amd; };If this isn't trivial to arrange, can we at least s/amd_iommu/amd/ here ?
I was in two minds. I tried to look for a TLA for the AMD IOMMU and 'amd' seemed a little too non-descript. I don't really mind though if there's a strong preference to shorted it.
+1 for shortening in some way. Even amd_vi would already be better imo, albeit I'm with Andrew and would think just amd is fine here (and matches how things are in the file system structure).
While at it, may I ask that you also switch the plain "int" fields to "unsigned int" - I think that's doable for both of them.
Jan
RE: [PATCH 1/6] x86/iommu: re-arrange arch_iommu to separate common fields...
Quoted 65 lines
-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 26 July 2020 09:14
To: [email protected]
Cc: 'Andrew Cooper' <[email protected]>; [email protected]; Durrant, Paul <[email protected]>; 'Lukasz Hawrylko' <[email protected]>; 'Wei Liu' <[email protected]>; 'Roger Pau Monné' <[email protected]>; 'Kevin Tian' <[email protected]>
Subject: RE: [EXTERNAL] [PATCH 1/6] x86/iommu: re-arrangearch_iommuto separate common fields...CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.
Quoted 50 lines · Paul Durrant
Quoted 50 lines
From: Andrew Cooper <[email protected]>
Sent: 24 July 2020 18:29Quoted 25 lines · Paul Durrant
xen/include/asm-x86/iommu.h +7
-10@@ -45,16 +45,23 @@ typedef uint64_t daddr_t; struct arch_iommu { -u64 pgd_maddr;/* io page directory machine address */ -spinlock_t mapping_lock;/* io page table lock */ -int agaw;/* adjusted guest address width, 0 is level 2 30-bit */ -u64 iommu_bitmap;/* bitmap of iommu(s) that the domain uses */ -struct list_head mapped_rmrrs; - -/* amd iommu support */ -int paging_mode; -struct page_info *root_table; -struct guest_iommu *g_iommu; + spinlock_t mapping_lock; /* io page table lock */ + + union { + /* Intel VT-d */ + struct { + u64 pgd_maddr; /* io page directory machine address */ + int agaw; /* adjusted guest address width, 0 is level 2 30-bit */xen/include/asm-x86/iommu.h typedef uint64_t daddr_t; · line 64
Quoted 10 lines
xen/include/asm-x86/iommu.h +10
-0+u64 iommu_bitmap;/* bitmap of iommu(s) that the domain uses */ +struct list_head mapped_rmrrs; + } vtd; +/* AMD IOMMU */ +struct { +int paging_mode; +struct page_info *root_table; +struct guest_iommu *g_iommu; + } amd_iommu; + };The naming split here is weird.
Ideally we'd have struct {vtd,amd}_iommu in appropriate headers, and this would be simply
union { struct vtd_iommu vtd; struct amd_iommu amd; };
Quoted 9 lines
Quoted 3 lines
Quoted 1 line
If this isn't trivial to arrange, can we at least s/amd_iommu/amd/ here ?
I was in two minds. I tried to look for a TLA for the AMD IOMMU and 'amd' seemed a little too non-
descript. I don't really mind though if there's a strong preference to shorted it.
+1 for shortening in some way. Even
amd_viwould already be better imo, albeit I'm with Andrew and would think just amd is fine here (and matches how things are in the file system structure).
OK, I'll shorten to 'amd'.
Quoted 3 lines
While at it, may I ask that you also switch the plain "int" fields to "unsigned int" - I think that's doable for both of them.
Sure, I can do that.
Paul
Quoted 1 line
Jan
PATCH 2/6 x86/iommu: add common page-table allocator
In reply to Paul DurrantFrom: Paul Durrant <[email protected]>
Instead of having separate page table allocation functions in VT-d and AMD IOMMU code, use a common allocation function in the general x86 code. Also, rather than walking the page tables and using a tasklet to free them during iommu_domain_destroy(), add allocated page table pages to a list and then simply walk the list to free them. This saves ~90 lines of code overall.
NOTE: There is no need to clear and sync PTEs during teardown since the per-
device root entries will have already been cleared (when devices were de-assigned) so the page tables can no longer be accessed by the IOMMU.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Kevin Tian <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
--- xen/drivers/passthrough/amd/iommu.h | 18 +---- xen/drivers/passthrough/amd/iommu_map.c | 10 +-- xen/drivers/passthrough/amd/pci_amd_iommu.c | 74 +++------------------ xen/drivers/passthrough/iommu.c | 23 ------- xen/drivers/passthrough/vtd/iommu.c | 51 ++------------ xen/drivers/passthrough/x86/iommu.c | 41 ++++++++++++ xen/include/asm-x86/iommu.h | 6 ++ xen/include/xen/iommu.h | 5 -- 8 files changed, 68 insertions(+), 160 deletions(-)
xen/drivers/passthrough/amd/iommu.h +1 -17
@@ -226,7 +226,7 @@ int __must_check amd_iommu_map_page(struct domain *d, dfn_t dfn, unsigned int *flush_flags);int __must_check amd_iommu_unmap_page (struct domain *d,dfn_t dfn,unsigned int *flush_flags); -int __must_check amd_iommu_alloc_root (struct domain_iommu *hd); +int __must_check amd_iommu_alloc_root (struct domain *d);int amd_iommu_reserve_domain_unity_map (struct domain *domain,paddr_t phys_addr,unsigned long size,int iw,int ir);
@@ -356,22 +356,6 @@ static inline int amd_iommu_get_paging_mode(unsigned long max_frames)return level; } -static inline struct page_info *alloc_amd_iommu_pgtable (void ) -{ -struct page_info *pg =alloc_domheap_page (NULL ,0 ); - -if ( pg ) -clear_domain_page (page_to_mfn (pg)); - -return pg; -} - -static inline void free_amd_iommu_pgtable (struct page_info *pg) -{ -if ( pg ) -free_domheap_page (pg); -} -static inline void *__alloc_amd_iommu_tables (unsigned int order) {return alloc_xenheap_pages (order,0 );
xen/drivers/passthrough/amd/iommu_map.c +5 -5
@@ -217,7 +217,7 @@ static int iommu_pde_from_dfn(struct domain *d, unsigned long dfn, mfn = next_table_mfn;/* allocate lower level page table */ - table =alloc_amd_iommu_pgtable (); + table =iommu_alloc_pgtable (d);if ( table ==NULL ) {AMD_IOMMU_DEBUG ("Cannot allocate I/O page table\n" );
@@ -248,7 +248,7 @@ static int iommu_pde_from_dfn(struct domain *d, unsigned long dfn,if ( next_table_mfn ==0 ) { - table =alloc_amd_iommu_pgtable (); + table =iommu_alloc_pgtable (d);if ( table ==NULL ) {AMD_IOMMU_DEBUG ("Cannot allocate I/O page table\n" );
@@ -286,7 +286,7 @@ int amd_iommu_map_page(struct domain *d, dfn_t dfn, mfn_t mfn,spin_lock (&hd->arch.mapping_lock); - rc =amd_iommu_alloc_root (hd); + rc =amd_iommu_alloc_root (d);if ( rc ) {spin_unlock (&hd->arch.mapping_lock);
@@ -458,7 +458,7 @@ int __init amd_iommu_quarantine_init(struct domain *d)spin_lock (&hd->arch.mapping_lock); - hd->arch.amd_iommu.root_table =alloc_amd_iommu_pgtable (); + hd->arch.amd_iommu.root_table =iommu_alloc_pgtable (d);if ( !hd->arch.amd_iommu.root_table ) goto out;
@@ -473,7 +473,7 @@ int __init amd_iommu_quarantine_init(struct domain *d)* page table pages, and the resulting allocations are always * zeroed. */ - pg =alloc_amd_iommu_pgtable (); + pg =iommu_alloc_pgtable (d);if ( !pg )break ;
xen/drivers/passthrough/amd/pci_amd_iommu.c +10 -64
125 lines not on this page · raw message
xen/drivers/passthrough/iommu.c +0 -23
@@ -49,10 +49,6 @@ bool_t __read_mostly amd_iommu_perdev_intremap = 1;DEFINE_PER_CPU (bool_t, iommu_dont_flush_iotlb); -DEFINE_SPINLOCK (iommu_pt_cleanup_lock); -PAGE_LIST_HEAD (iommu_pt_cleanup_list); -static struct tasklet iommu_pt_cleanup_tasklet; -static int __init parse_iommu_param (const char *s) {const char *ss;
@@ -226,7 +222,6 @@ static void iommu_teardown(struct domain *d)struct domain_iommu *hd =dom_iommu (d); hd->platform_ops->teardown (d); -tasklet_schedule (&iommu_pt_cleanup_tasklet); }void iommu_domain_destroy (struct domain *d)
@@ -366,23 +361,6 @@ int iommu_lookup_page(struct domain *d, dfn_t dfn, mfn_t *mfn,return iommu_call (hd->platform_ops, lookup_page, d, dfn, mfn, flags); } -static void iommu_free_pagetables (void *unused) -{ -do { -struct page_info *pg; - -spin_lock (&iommu_pt_cleanup_lock); - pg =page_list_remove_head (&iommu_pt_cleanup_list); -spin_unlock (&iommu_pt_cleanup_lock); -if ( !pg ) -return ; -iommu_vcall (iommu_get_ops (), free_page_table, pg); - }while ( !softirq_pending (smp_processor_id ()) ); - -tasklet_schedule_on_cpu (&iommu_pt_cleanup_tasklet, -cpumask_cycle (smp_processor_id (), &cpu_online_map)); -} -int iommu_iotlb_flush (struct domain *d,dfn_t dfn,unsigned int page_count,unsigned int flush_flags) {
@@ -506,7 +484,6 @@ int __init iommu_setup(void)#ifndef iommu_intremapprintk ("Interrupt remapping %sabled\n" , iommu_intremap ?"en" :"dis" );#endif -tasklet_init (&iommu_pt_cleanup_tasklet, iommu_free_pagetables,NULL ); }return rc;
xen/drivers/passthrough/vtd/iommu.c +5 -46
81 lines not on this page · raw message
xen/drivers/passthrough/x86/iommu.c +41 -0
58 lines not on this page · raw message
xen/include/asm-x86/iommu.h +6 -0
@@ -46,6 +46,10 @@ typedef uint64_t daddr_t;struct arch_iommu {spinlock_t mapping_lock;/* io page table lock */ +struct { +struct page_list_head list; +spinlock_t lock; + } pgtables;union {/* Intel VT-d */
@@ -131,6 +135,8 @@ int pi_update_irte(const struct pi_desc *pi_desc, const struct pirq *pirq,iommu_vcall (ops, sync_cache, addr, size); \ }) +struct page_info *__must_check iommu_alloc_pgtable (struct domain *d); +#endif /* !__ARCH_X86_IOMMU_H__ */ /* * Local variables:
xen/include/xen/iommu.h +0 -5
@@ -263,8 +263,6 @@ struct iommu_ops {int __must_check (*lookup_page)(struct domain *d,dfn_t dfn,mfn_t *mfn,unsigned int *flags); -void (*free_page_table)(struct page_info *); -#ifdef CONFIG_X86 int (*enable_x2apic)(void );void (*disable_x2apic)(void );
@@ -381,9 +379,6 @@ void iommu_dev_iotlb_flush_timeout(struct domain *d, struct pci_dev *pdev);*/ DECLARE_PER_CPU (bool_t, iommu_dont_flush_iotlb); -extern struct spinlock iommu_pt_cleanup_lock; -extern struct page_list_head iommu_pt_cleanup_list; -#endif /* _IOMMU_H_ */ /*
Signature · 1 line
2.20.1
Re: [PATCH 2/6] x86/iommu: add common page-table allocator
Quoted 8 lines · Paul Durrant
From: Paul Durrant <[email protected]>
Instead of having separate page table allocation functions in VT-d and AMD IOMMU code, use a common allocation function in the general x86 code. Also, rather than walking the page tables and using a tasklet to free them during
iommu_domain_destroy(), add allocated page table pages to a list and then simply walk the list to free them. This saves ~90 lines of code overall.
Quoted 5 lines
NOTE: There is no need to clear and sync PTEs during teardown since the per-
device root entries will have already been cleared (when devices were de-assigned) so the page tables can no longer be accessed by the IOMMU.Signed-off-by Paul Durrant <[email protected]>
Oh wow - I don't have any polite words for how that code was organised before.
Instead of discussing the ~90 lines saved, what about the removal of a global bottleneck (the tasklet) or expand on the removal of redundant TLB/cache maintenance?
Quoted 47 lines
xen/drivers/passthrough/amd/pci_amd_iommu.c +0
-39@@ -378,64 +380,9 @@ static int amd_iommu_assign_device(struct domain *d, u8 devfn,return reassign_device (pdev->domain, d, devfn, pdev); } -static void deallocate_next_page_table (struct page_info *pg,int level) -{ -PFN_ORDER (pg) = level; -spin_lock (&iommu_pt_cleanup_lock); -page_list_add_tail (pg, &iommu_pt_cleanup_list); -spin_unlock (&iommu_pt_cleanup_lock); -} - -static void deallocate_page_table (struct page_info *pg) -{ -struct amd_iommu_pte *table_vaddr; -unsigned int index, level =PFN_ORDER (pg); - -PFN_ORDER (pg) =0 ; - -if ( level <=1 ) - { -free_amd_iommu_pgtable (pg); -return ; - } - - table_vaddr =__map_domain_page (pg); - -for ( index =0 ; index <PTE_PER_TABLE_SIZE ; index++ ) - { -struct amd_iommu_pte *pde = &table_vaddr[index]; - -if ( pde->mfn && pde->next_level && pde->pr ) - { -/* We do not support skip levels yet */ -ASSERT (pde->next_level == level -1 ); -deallocate_next_page_table (mfn_to_page (_mfn (pde->mfn)), - pde->next_level); - } - } - -unmap_domain_page (table_vaddr); -free_amd_iommu_pgtable (pg); -}
xen/drivers/passthrough/amd/pci_amd_iommu.c static int amd_iommu_assign_device(struct domain *d, u8 devfn, · line 429
Quoted 10 lines
xen/drivers/passthrough/amd/pci_amd_iommu.c +0
-10- -static void deallocate_iommu_page_tables (struct domain *d) -{ -struct domain_iommu *hd =dom_iommu (d); - -spin_lock (&hd->arch.mapping_lock); -if ( hd->arch.amd_iommu.root_table ) - { -deallocate_next_page_table (hd->arch.amd_iommu.root_table, - hd->arch.amd_iommu.paging_mode);
I really need to dust off my patch fixing up several bits of dubious logic, including the name "paging_mode" which is actually simply the number of levels.
At this point, it will probably be best to get this series in first, and for me to rebase.
xen/drivers/passthrough/amd/pci_amd_iommu.c static int amd_iommu_assign_device(struct domain *d, u8 devfn, · line 386
Quoted 11 lines
xen/drivers/passthrough/amd/pci_amd_iommu.c +1
-7- hd->arch.amd_iommu.root_table =NULL ; - } -spin_unlock (&hd->arch.mapping_lock); -} - -static void amd_iommu_domain_destroy (struct domain *d) { -deallocate_iommu_page_tables (d); +dom_iommu (d)->arch.amd_iommu.root_table =NULL ;amd_iommu_flush_all_pages (d);
Per your NOTE:, shouldn't this flush call be dropped?
Quoted 11 lines
xen/drivers/passthrough/x86/iommu.c +3
-0@@ -140,11 +140,19 @@ int arch_iommu_domain_init(struct domain *d)spin_lock_init (&hd->arch.mapping_lock); +INIT_PAGE_LIST_HEAD (&hd->arch.pgtables.list); +spin_lock_init (&hd->arch.pgtables.lock); +
xen/drivers/passthrough/x86/iommu.c int arch_iommu_domain_init(struct domain *d) · line 155
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +5
-0return 0 ; }void arch_iommu_domain_destroy (struct domain *d) { +struct domain_iommu *hd =dom_iommu (d); +struct page_info *pg; + +while ( (pg =page_list_remove_head (&hd->arch.pgtables.list)) ) +free_domheap_page (pg);
Some of those 90 lines saved were the logic to not suffer a watchdog timeout here.
To do it like this, it needs plumbing into the relinquish resources path.
Quoted 8 lines
xen/drivers/passthrough/x86/iommu.c +1
-0} static bool __hwdom_init hwdom_iommu_map(const struct domain *d,@@ -257,6 +265,39 @@ void __hwdom_init arch_iommu_hwdom_init(struct domain *d)return ; } +struct page_info *iommu_alloc_pgtable (struct domain *d)
xen/drivers/passthrough/x86/iommu.c void __hwdom_init arch_iommu_hwdom_init(struct domain *d) · line 278
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +10
-0+{ +struct domain_iommu *hd =dom_iommu (d); +#ifdef CONFIG_NUMA +unsigned int memflags = (hd->node ==NUMA_NO_NODE ) ? +0 :MEMF_node (hd->node); +#else +unsigned int memflags =0 ; +#endif +struct page_info *pg; +void *p;
The memflags code is very awkward. How about initialising it to 0, and having:
#ifdef CONFIG_NUMA
if ( hd->node != NUMA_NO_NODE )
memflags = MEMF_node(hd->node);
#endif
here?
xen/drivers/passthrough/x86/iommu.c void __hwdom_init arch_iommu_hwdom_init(struct domain *d) · line 280
Quoted 2 lines
xen/drivers/passthrough/x86/iommu.c +2
-0+ +BUG_ON (!iommu_enabled);
Is this really necessary? Can we plausibly end up in this function otherwise?
Overall, I wonder if this patch would better be split into several. One which introduces the common alloc/free implementation, two which switch the VT-d and AMD implementations over, and possibly one clean-up on the end?
~Andrew
Re: [PATCH 2/6] x86/iommu: add common page-table allocator
Quoted 24 lines · Andrew Cooper
Quoted 9 lines · Paul Durrant
xen/drivers/passthrough/x86/iommu.c +3
-0@@ -140,11 +140,19 @@ int arch_iommu_domain_init(struct domain *d)spin_lock_init (&hd->arch.mapping_lock); +INIT_PAGE_LIST_HEAD (&hd->arch.pgtables.list); +spin_lock_init (&hd->arch.pgtables.lock); +xen/drivers/passthrough/x86/iommu.c int arch_iommu_domain_init(struct domain *d) · line 155
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +5
-0return 0 ; }void arch_iommu_domain_destroy (struct domain *d) { +struct domain_iommu *hd =dom_iommu (d); +struct page_info *pg; + +while ( (pg =page_list_remove_head (&hd->arch.pgtables.list)) ) +free_domheap_page (pg);Some of those 90 lines saved were the logic to not suffer a watchdog timeout here.
Quoted 1 line
To do it like this, it needs plumbing into the relinquish resources path.
And indeed this is possible now only because we don't destroy page tables for still running domains anymore. Maybe the description should also make this connection.
Jan
RE: [PATCH 2/6] x86/iommu: add common page-table allocator
In reply to Andrew CooperQuoted 26 lines
-----Original Message-----
From: Andrew Cooper <[email protected]>
Sent: 24 July 2020 19:24
To: Paul Durrant <[email protected]>; [email protected]
Cc: Durrant, Paul <[email protected]>; Jan Beulich <[email protected]>; Kevin Tian <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monné <[email protected]>
Subject: RE: [EXTERNAL] [PATCH 2/6] x86/iommu: add common page-table allocatorCAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.
Quoted 12 lines · Paul Durrant
From: Paul Durrant <[email protected]>
Instead of having separate page table allocation functions in VT-d and AMD IOMMU code, use a common allocation function in the general x86 code. Also, rather than walking the page tables and using a tasklet to free them during
iommu_domain_destroy(), add allocated page table pages to a list and then simply walk the list to free them. This saves ~90 lines of code overall.NOTE: There is no need to clear and sync PTEs during teardown since the per-
device root entries will have already been cleared (when devices were de-assigned) so the page tables can no longer be accessed by the IOMMU.
Quoted 9 lines
Quoted 1 line
Signed-off-by Paul Durrant <[email protected]>Oh wow - I don't have any polite words for how that code was organised before.
Instead of discussing the ~90 lines saved, what about the removal of a global bottleneck (the tasklet) or expand on the removal of redundant TLB/cache maintenance?
Ok.
Quoted 63 lines
Review
Quoted 1 line
(quoted diff) +0
-0
b/xen/drivers/passthrough/amd/pci_amd_iommu.cQuoted 46 lines
index
c386dc4387..fd9b1e7bd5100644xen/drivers/passthrough/amd/pci_amd_iommu.c +0
-39@@ -378,64 +380,9 @@ static int amd_iommu_assign_device(struct domain *d, u8 devfn,return reassign_device (pdev->domain, d, devfn, pdev); } -static void deallocate_next_page_table (struct page_info *pg,int level) -{ -PFN_ORDER (pg) = level; -spin_lock (&iommu_pt_cleanup_lock); -page_list_add_tail (pg, &iommu_pt_cleanup_list); -spin_unlock (&iommu_pt_cleanup_lock); -} - -static void deallocate_page_table (struct page_info *pg) -{ -struct amd_iommu_pte *table_vaddr; -unsigned int index, level =PFN_ORDER (pg); - -PFN_ORDER (pg) =0 ; - -if ( level <=1 ) - { -free_amd_iommu_pgtable (pg); -return ; - } - - table_vaddr =__map_domain_page (pg); - -for ( index =0 ; index <PTE_PER_TABLE_SIZE ; index++ ) - { -struct amd_iommu_pte *pde = &table_vaddr[index]; - -if ( pde->mfn && pde->next_level && pde->pr ) - { -/* We do not support skip levels yet */ -ASSERT (pde->next_level == level -1 ); -deallocate_next_page_table (mfn_to_page (_mfn (pde->mfn)), - pde->next_level); - } - } - -unmap_domain_page (table_vaddr); -free_amd_iommu_pgtable (pg); -}xen/drivers/passthrough/amd/pci_amd_iommu.c static int amd_iommu_assign_device(struct domain *d, u8 devfn, · line 429
Quoted 10 lines
xen/drivers/passthrough/amd/pci_amd_iommu.c +0
-10- -static void deallocate_iommu_page_tables (struct domain *d) -{ -struct domain_iommu *hd =dom_iommu (d); - -spin_lock (&hd->arch.mapping_lock); -if ( hd->arch.amd_iommu.root_table ) - { -deallocate_next_page_table (hd->arch.amd_iommu.root_table, - hd->arch.amd_iommu.paging_mode);I really need to dust off my patch fixing up several bits of dubious logic, including the name "
paging_mode" which is actually simply the number of levels.
Quoted 3 lines
At this point, it will probably be best to get this series in first, and for me to rebase.
Ok.
Quoted 5 lines
Quoted 5 lines
xen/drivers/passthrough/amd/pci_amd_iommu.c +0
-5- hd->arch.amd_iommu.root_table =NULL ; - } -spin_unlock (&hd->arch.mapping_lock); -} -
Quoted 9 lines
xen/drivers/passthrough/amd/pci_amd_iommu.c static int amd_iommu_assign_device(struct domain *d, u8 devfn, · line 386
Quoted 6 lines
xen/drivers/passthrough/amd/pci_amd_iommu.c +1
-2-static void amd_iommu_domain_destroy (struct domain *d) { -deallocate_iommu_page_tables (d); +dom_iommu (d)->arch.amd_iommu.root_table =NULL ;amd_iommu_flush_all_pages (d);Per your NOTE:, shouldn't this flush call be dropped?
Indeed it should.
Quoted 25 lines
Quoted 11 lines
xen/drivers/passthrough/x86/iommu.c +3
-0@@ -140,11 +140,19 @@ int arch_iommu_domain_init(struct domain *d)spin_lock_init (&hd->arch.mapping_lock); +INIT_PAGE_LIST_HEAD (&hd->arch.pgtables.list); +spin_lock_init (&hd->arch.pgtables.lock); +xen/drivers/passthrough/x86/iommu.c int arch_iommu_domain_init(struct domain *d) · line 155
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +5
-0return 0 ; }void arch_iommu_domain_destroy (struct domain *d) { +struct domain_iommu *hd =dom_iommu (d); +struct page_info *pg; + +while ( (pg =page_list_remove_head (&hd->arch.pgtables.list)) ) +free_domheap_page (pg);Some of those 90 lines saved were the logic to not suffer a watchdog timeout here.
Quoted 2 lines
To do it like this, it needs plumbing into the relinquish resources path.
Ok. I does look like there could be other potentially lengthy destruction done off the back of the RCU call. Ought we have the ability to have a restartable domain_destroy()?
Quoted 22 lines
Quoted 8 lines
xen/drivers/passthrough/x86/iommu.c +1
-0} static bool __hwdom_init hwdom_iommu_map(const struct domain *d,@@ -257,6 +265,39 @@ void __hwdom_init arch_iommu_hwdom_init(struct domain *d)return ; } +struct page_info *iommu_alloc_pgtable (struct domain *d)xen/drivers/passthrough/x86/iommu.c void __hwdom_init arch_iommu_hwdom_init(struct domain *d) · line 278
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +10
-0+{ +struct domain_iommu *hd =dom_iommu (d); +#ifdef CONFIG_NUMA +unsigned int memflags = (hd->node ==NUMA_NO_NODE ) ? +0 :MEMF_node (hd->node); +#else +unsigned int memflags =0 ; +#endif +struct page_info *pg; +void *p;The memflags code is very awkward. How about initialising it to 0, and having:
Quoted 7 lines
#ifdef CONFIG_NUMA if ( hd->node != NUMA_NO_NODE ) memflags = MEMF_node(hd->node);#endif
here?
Sure.
Quoted 6 lines
xen/drivers/passthrough/x86/iommu.c void __hwdom_init arch_iommu_hwdom_init(struct domain *d) · line 280
Quoted 2 lines
xen/drivers/passthrough/x86/iommu.c +2
-0+ +BUG_ON (!iommu_enabled);Is this really necessary? Can we plausibly end up in this function otherwise?
Not really; I'll drop it.
Quoted 5 lines
Overall, I wonder if this patch would better be split into several. One which introduces the common alloc/free implementation, two which switch the VT-d and AMD implementations over, and possibly one clean-up on the end?
Ok, if you feel the patch is too large as-is then I'll split as you suggest.
Paul
Quoted 1 line
~Andrew
Re: [PATCH 2/6] x86/iommu: add common page-table allocator
Quoted 27 lines · Paul
Quoted 27 lines
From: Andrew Cooper <[email protected]>
Sent: 24 July 2020 19:24Quoted 9 lines · Paul Durrant
xen/drivers/passthrough/x86/iommu.c +3
-0@@ -140,11 +140,19 @@ int arch_iommu_domain_init(struct domain *d)spin_lock_init (&hd->arch.mapping_lock); +INIT_PAGE_LIST_HEAD (&hd->arch.pgtables.list); +spin_lock_init (&hd->arch.pgtables.lock); +xen/drivers/passthrough/x86/iommu.c int arch_iommu_domain_init(struct domain *d) · line 155
Quoted 10 lines
xen/drivers/passthrough/x86/iommu.c +5
-0return 0 ; }void arch_iommu_domain_destroy (struct domain *d) { +struct domain_iommu *hd =dom_iommu (d); +struct page_info *pg; + +while ( (pg =page_list_remove_head (&hd->arch.pgtables.list)) ) +free_domheap_page (pg);Some of those 90 lines saved were the logic to not suffer a watchdog timeout here.
Quoted 4 lines
Quoted 2 lines
To do it like this, it needs plumbing into the relinquish resources path.
Ok. I does look like there could be other potentially lengthy destruction done off the back of the RCU call. Ought we have the ability to have a restartable
domain_destroy()?
I don't see how this would be (easily) feasible. Instead - why do page tables not get cleaned up already at relinquish_resources time?
Jan
PATCH 3/6 iommu: remove iommu_lookup_page() and the lookup_page() method...
In reply to Paul DurrantFrom: Paul Durrant <[email protected]>
... from iommu_ops.
This patch is essentially a reversion of dd93d54f "vtd: add lookup_page method to iommu_ops". The code was intended to be used by a patch that has long- since been abandoned. Therefore it is dead code and can be removed.
---
Cc: Jan Beulich <[email protected]>
Cc: Kevin Tian <[email protected]>
--- xen/drivers/passthrough/iommu.c | 11 -------- xen/drivers/passthrough/vtd/iommu.c | 41 ----------------------------- xen/include/xen/iommu.h | 5 ---- 3 files changed, 57 deletions(-)
xen/drivers/passthrough/iommu.c +0 -11
@@ -350,17 +350,6 @@ int iommu_legacy_unmap(struct domain *d, dfn_t dfn, unsigned int page_order)return rc; } -int iommu_lookup_page (struct domain *d,dfn_t dfn,mfn_t *mfn, -unsigned int *flags) -{ -const struct domain_iommu *hd =dom_iommu (d); - -if ( !is_iommu_enabled (d) || !hd->platform_ops->lookup_page ) -return -EOPNOTSUPP ; - -return iommu_call (hd->platform_ops, lookup_page, d, dfn, mfn, flags); -} -int iommu_iotlb_flush (struct domain *d,dfn_t dfn,unsigned int page_count,unsigned int flush_flags) {
xen/drivers/passthrough/vtd/iommu.c +0 -41
53 lines not on this page · raw message
xen/include/xen/iommu.h +0 -5
@@ -158,9 +158,6 @@ int __must_check iommu_legacy_map(struct domain *d, dfn_t dfn, mfn_t mfn,int __must_check iommu_legacy_unmap (struct domain *d,dfn_t dfn,unsigned int page_order); -int __must_check iommu_lookup_page (struct domain *d,dfn_t dfn,mfn_t *mfn, -unsigned int *flags); -int __must_check iommu_iotlb_flush (struct domain *d,dfn_t dfn,unsigned int page_count,unsigned int flush_flags);
@@ -260,8 +257,6 @@ struct iommu_ops { unsigned int *flush_flags);int __must_check (*unmap_page)(struct domain *d,dfn_t dfn,unsigned int *flush_flags); -int __must_check (*lookup_page)(struct domain *d,dfn_t dfn,mfn_t *mfn, -unsigned int *flags);#ifdef CONFIG_X86 int (*enable_x2apic)(void );
Signature · 1 line
2.20.1
Re: [PATCH 3/6] iommu: remove iommu_lookup_page() and the lookup_page() method...
Quoted 7 lines · Paul Durrant
From: Paul Durrant <[email protected]>
... from
iommu_ops.This patch is essentially a reversion of
dd93d54f"vtd: addlookup_pagemethod toiommu_ops". The code was intended to be used by a patch that has long- since been abandoned. Therefore it is dead code and can be removed.
And by this, you mean the work that you only partial unstreamed, with the remainder of the feature still very much in use by XenServer?
Please don't go breaking in-use things, simply because we're fixing Xen's IOMMU mess once large XSA at a time...
As far as I can tell, this patch doesn't interact with any others in the series.
~Andrew
RE: [PATCH 3/6] iommu: remove iommu_lookup_page() and the lookup_page() method...
Quoted 11 lines
-----Original Message-----
From: Andrew Cooper <[email protected]>
Sent: 24 July 2020 19:39
To: Paul Durrant <[email protected]>; [email protected]
Cc: Paul Durrant <[email protected]>; Kevin Tian <[email protected]>; Jan Beulich <[email protected]>
Subject: Re: [PATCH 3/6] iommu: removeiommu_lookup_page()and thelookup_page()method...Quoted 2 lines · Paul Durrant
From: Paul Durrant <[email protected]>
Quoted 9 lines
Quoted 5 lines
... from
iommu_ops.This patch is essentially a reversion of
dd93d54f"vtd: addlookup_pagemethod toiommu_ops". The code was intended to be used by a patch that has long- since been abandoned. Therefore it is dead code and can be removed.And by this, you mean the work that you only partial unstreamed, with the remainder of the feature still very much in use by XenServer?
I thought we basically decided to bin the original PV IOMMU idea though?
Quoted 6 lines
Please don't go breaking in-use things, simply because we're fixing Xen's IOMMU mess once large XSA at a time...
As far as I can tell, this patch doesn't interact with any others in the series.
I can leave it, but I still don't think anything other than current XenServer will ever use it... so it really ought to just be in the downstream patch queue IMO.
Paul
Quoted 1 line
~Andrew
Re: [PATCH 3/6] iommu: remove iommu_lookup_page() and the lookup_page() method...
Quoted 12 lines · Paul Durrant
Quoted 12 lines
From: Andrew Cooper <[email protected]>
Sent: 24 July 2020 19:39Quoted 7 lines · Paul Durrant
From: Paul Durrant <[email protected]>
... from
iommu_ops.This patch is essentially a reversion of
dd93d54f"vtd: addlookup_pagemethod toiommu_ops". The code was intended to be used by a patch that has long- since been abandoned. Therefore it is dead code and can be removed.
Quoted 5 lines
Quoted 3 lines
And by this, you mean the work that you only partial unstreamed, with the remainder of the feature still very much in use by XenServer?
I thought we basically decided to bin the original PV IOMMU idea though?
Did we? It's the first time I hear of it, I think.
Jan
RE: [PATCH 3/6] iommu: remove iommu_lookup_page() and the lookup_page() method...
Quoted 26 lines
-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 26 July 2020 09:28
To: [email protected]
Cc: 'Andrew Cooper' <[email protected]>; [email protected]; Durrant, Paul <[email protected]>; 'Kevin Tian' <[email protected]>
Subject: RE: [EXTERNAL] [PATCH 3/6] iommu: removeiommu_lookup_page()and thelookup_page()method...CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.
Quoted 12 lines · Paul Durrant
Quoted 12 lines
From: Andrew Cooper <[email protected]>
Sent: 24 July 2020 19:39Quoted 7 lines · Paul Durrant
From: Paul Durrant <[email protected]>
... from
iommu_ops.This patch is essentially a reversion of
dd93d54f"vtd: addlookup_pagemethod toiommu_ops". The code was intended to be used by a patch that has long- since been abandoned. Therefore it is dead code and can be removed.
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines
And by this, you mean the work that you only partial unstreamed, with the remainder of the feature still very much in use by XenServer?
I thought we basically decided to bin the original PV IOMMU idea though?
Did we? It's the first time I hear of it, I think.
I circulated a doc. ages ago, while I was still at Citrix: https://docs.google.com/document/d/12-z6JD41J_oNrCg_c0yAxGWg5ADBQ8_bSiP_NH6Hqwo/edit?usp=sharing
In there I propose that we don't follow the original idea of keeping a single set of per-domain tables but instead have a set of tables (or IOMMU contexts) for groups of devices. 'Context 0' is the current set of static 1:1 tables but other contexts are manipulated by hypercall so, in this plan, I don't envisage the need to look up mappings in the tables in this way... but I guess I can't rule it out.
Paul
PATCH 4/6 remove remaining uses of iommu_legacy_map/unmap
In reply to Paul DurrantFrom: Paul Durrant <[email protected]>
The 'legacy' functions do implicit flushing so amend the callers to do the appropriate flushing.
Unfortunately, because of the structure of the P2M code, we cannot remove the per-CPU 'iommu_dont_flush_iotlb' global and the optimization it facilitates. It is now checked directly iommu_iotlb_flush(). Also, it is now declared as bool (rather than bool_t) and setting/clearing it are no longer pointlessly gated on is_iommu_enabled() returning true. (Arguably it is also pointless to gate the call to iommu_iotlb_flush() on that condition - since it is a no-op in that case - but the if clause allows the scope of a stack variable to be restricted).
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>
--- xen/arch/x86/mm.c | 22 +++++++++++++++----- xen/arch/x86/mm/p2m-ept.c | 22 +++++++++++++------- xen/arch/x86/mm/p2m-pt.c | 17 +++++++++++---- xen/arch/x86/mm/p2m.c | 28 ++++++++++++++++++------- xen/arch/x86/x86_64/mm.c | 27 ++++++++++++++++++------ xen/common/grant_table.c | 36 +++++++++++++++++++++++++------- xen/common/memory.c | 7 +++---- xen/drivers/passthrough/iommu.c | 37 +-------------------------------- xen/include/xen/iommu.h | 20 +++++------------- 9 files changed, 123 insertions(+), 93 deletions(-)
xen/arch/x86/mm.c +17 -5
@@ -2446,10 +2446,16 @@ static int cleanup_page_mappings(struct page_info *page)if ( d &&unlikely (need_iommu_pt_sync (d)) &&is_pv_domain (d) ) { -int rc2 =iommu_legacy_unmap (d,_dfn (mfn),PAGE_ORDER_4K ); +unsigned int flush_flags =0 ; +int err; + err =iommu_unmap (d,_dfn (mfn),PAGE_ORDER_4K , &flush_flags);if ( !rc ) - rc = rc2; + rc = err; + + err =iommu_iotlb_flush (d,_dfn (mfn),1 , flush_flags); +if ( !rc ) + rc = err; }if (likely (!is_special_page (page)) )
@@ -2971,13 +2977,19 @@ static int _get_page_type(struct page_info *page, unsigned long type,if ( d &&unlikely (need_iommu_pt_sync (d)) &&is_pv_domain (d) ) {mfn_t mfn =page_to_mfn (page); +dfn_t dfn =_dfn (mfn_x (mfn)); +unsigned int flush_flags =0 ; +int err;if ( (x & PGT_type_mask) == PGT_writable_page ) - rc =iommu_legacy_unmap (d,_dfn (mfn_x (mfn)),PAGE_ORDER_4K ); + rc =iommu_unmap (d, dfn,PAGE_ORDER_4K , &flush_flags);else - rc =iommu_legacy_map (d,_dfn (mfn_x (mfn)), mfn,PAGE_ORDER_4K , - IOMMUF_readable | IOMMUF_writable); + rc =iommu_map (d, dfn, mfn,PAGE_ORDER_4K , + IOMMUF_readable | IOMMUF_writable, &flush_flags); + err =iommu_iotlb_flush (d, dfn,1 , flush_flags); +if ( !rc ) + rc = err;if (unlikely (rc) ) {_put_page_type (page,0 ,NULL );
xen/arch/x86/mm/p2m-ept.c +14 -8
@@ -842,15 +842,21 @@ out:if ( rc ==0 &&p2m_is_hostp2m (p2m) && need_modify_vtd_table ) { -if (iommu_use_hap_pt (d) ) - rc =iommu_iotlb_flush (d,_dfn (gfn), (1u << order), - (iommu_flags ? IOMMU_FLUSHF_added :0 ) | - (vtd_pte_present ? IOMMU_FLUSHF_modified - :0 )); -else if (need_iommu_pt_sync (d) ) +unsigned int flush_flags =0 ; +int err; + +if (need_iommu_pt_sync (d) ) rc = iommu_flags ? -iommu_legacy_map (d,_dfn (gfn), mfn, order, iommu_flags) : -iommu_legacy_unmap (d,_dfn (gfn), order); +iommu_map (d,_dfn (gfn), mfn, order, iommu_flags, &flush_flags) : +iommu_unmap (d,_dfn (gfn), order, &flush_flags); +else if (iommu_use_hap_pt (d) ) + flush_flags = + (iommu_flags ? IOMMU_FLUSHF_added :0 ) | + (vtd_pte_present ? IOMMU_FLUSHF_modified :0 ); + + err =iommu_iotlb_flush (d,_dfn (gfn),1u << order, flush_flags); +if ( !rc ) + rc = err; }unmap_domain_page (table);
xen/arch/x86/mm/p2m-pt.c +13 -4
@@ -678,10 +678,19 @@ p2m_pt_set_entry(struct p2m_domain *p2m, gfn_t gfn_, mfn_t mfn,if (need_iommu_pt_sync (p2m->domain) && (iommu_old_flags != iommu_pte_flags || old_mfn !=mfn_x (mfn)) ) - rc = iommu_pte_flags - ?iommu_legacy_map (d,_dfn (gfn), mfn, page_order, - iommu_pte_flags) - :iommu_legacy_unmap (d,_dfn (gfn), page_order); + { +unsigned int flush_flags =0 ; +int err; + + rc = iommu_pte_flags ? +iommu_map (d,_dfn (gfn), mfn, page_order, iommu_pte_flags, + &flush_flags) : +iommu_unmap (d,_dfn (gfn), page_order, &flush_flags); + + err =iommu_iotlb_flush (d,_dfn (gfn),1u << page_order, flush_flags); +if ( !rc ) + rc = err; + }/* * Free old intermediate tables if necessary. This has to be the
xen/arch/x86/mm/p2m.c +21 -7
@@ -1350,10 +1350,17 @@ int set_identity_p2m_entry(struct domain *d, unsigned long gfn_l,if ( !paging_mode_translate (p2m->domain) ) { -if ( !is_iommu_enabled (d) ) -return 0 ; -return iommu_legacy_map (d,_dfn (gfn_l),_mfn (gfn_l),PAGE_ORDER_4K , - IOMMUF_readable | IOMMUF_writable); +unsigned int flush_flags =0 ; +int err; + + ret =iommu_map (d,_dfn (gfn_l),_mfn (gfn_l),PAGE_ORDER_4K , + IOMMUF_readable | IOMMUF_writable, &flush_flags); + + err =iommu_iotlb_flush (d,_dfn (gfn_l),1 , flush_flags); +if ( !ret ) + ret = err; + +return ret; }gfn_lock (p2m, gfn,0 );
@@ -1441,9 +1448,16 @@ int clear_identity_p2m_entry(struct domain *d, unsigned long gfn_l)if ( !paging_mode_translate (d) ) { -if ( !is_iommu_enabled (d) ) -return 0 ; -return iommu_legacy_unmap (d,_dfn (gfn_l),PAGE_ORDER_4K ); +unsigned int flush_flags =0 ; +int err; + + ret =iommu_unmap (d,_dfn (gfn_l),PAGE_ORDER_4K , &flush_flags); + + err =iommu_iotlb_flush (d,_dfn (gfn_l),1 , flush_flags); +if ( !ret ) + ret = err; + +return ret; }gfn_lock (p2m, gfn,0 );
xen/arch/x86/x86_64/mm.c +21 -6
@@ -1413,21 +1413,36 @@ int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm) !iommu_use_hap_pt(hardware_domain) && !need_iommu_pt_sync(hardware_domain) ) { +unsigned int flush_flags =0 ; +bool failed = false; +unsigned int n; +for ( i = spfn; i < epfn; i++ ) -if (iommu_legacy_map (hardware_domain,_dfn (i),_mfn (i), -PAGE_ORDER_4K , - IOMMUF_readable | IOMMUF_writable) ) +if (iommu_map (hardware_domain,_dfn (i),_mfn (i), +PAGE_ORDER_4K , IOMMUF_readable | IOMMUF_writable, + &flush_flags) )break ;if ( i != epfn ) { + failed = true; +while (i-- > old_max)/* If statement to satisfy __must_check. */ -if (iommu_legacy_unmap (hardware_domain,_dfn (i), -PAGE_ORDER_4K ) ) +if (iommu_unmap (hardware_domain,_dfn (i),PAGE_ORDER_4K , + &flush_flags) )continue ; + } - goto destroy_m2p; +for ( i = spfn; i < epfn; i += n ) + { + n = epfn - i;/* may truncate */ + +/* If statement to satisfy __must_check. */ +if (iommu_iotlb_flush (hardware_domain,_dfn (i), n, flush_flags) ) +continue ; } +if ( failed ) + goto destroy_m2p; }/* We can't revert any more */
xen/common/grant_table.c +28 -8
@@ -1225,11 +1225,25 @@ map_grant_ref( kind = IOMMUF_readable; else kind =0 ; -if ( kind &&iommu_legacy_map (ld,_dfn (mfn_x (mfn)), mfn,0 , kind) ) +if ( kind ) { -double_gt_unlock (lgt, rgt); - rc = GNTST_general_error; - goto undo_out; +dfn_t dfn =_dfn (mfn_x (mfn)); +unsigned int flush_flags =0 ; +int err; + + err =iommu_map (ld, dfn, mfn,0 , kind, &flush_flags); +if ( err ) + rc = GNTST_general_error; + + err =iommu_iotlb_flush (ld, dfn,1 , flush_flags); +if ( err ) + rc = GNTST_general_error; + +if ( rc != GNTST_okay ) + { +double_gt_unlock (lgt, rgt); + goto undo_out; + } } }
@@ -1473,21 +1487,27 @@ unmap_common(if ( rc == GNTST_okay &&gnttab_need_iommu_mapping (ld) ) {unsigned int kind; +dfn_t dfn =_dfn (mfn_x (op->mfn)); +unsigned int flush_flags =0 ;int err =0 ;double_gt_lock (lgt, rgt); kind =mapkind (lgt, rd, op->mfn);if ( !kind ) - err =iommu_legacy_unmap (ld,_dfn (mfn_x (op->mfn)),0 ); + err =iommu_unmap (ld, dfn,0 , &flush_flags);else if ( !(kind &MAPKIND_WRITE ) ) - err =iommu_legacy_map (ld,_dfn (mfn_x (op->mfn)), op->mfn,0 , - IOMMUF_readable); + err =iommu_map (ld, dfn, op->mfn,0 , IOMMUF_readable, + &flush_flags); -double_gt_unlock (lgt, rgt); +if ( err ) + rc = GNTST_general_error; + err =iommu_iotlb_flush (ld, dfn,1 , flush_flags);if ( err ) rc = GNTST_general_error; + +double_gt_unlock (lgt, rgt); }/* If just unmapped a writable mapping, mark as dirtied */
xen/common/memory.c +3 -4
@@ -824,8 +824,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, xatp->gpfn += start; xatp->size -= start; -if (is_iommu_enabled (d) ) -this_cpu (iommu_dont_flush_iotlb) =1 ; +this_cpu (iommu_dont_flush_iotlb) = true;while ( xatp->size > done ) {
@@ -845,12 +844,12 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp, } } +this_cpu (iommu_dont_flush_iotlb) = false; +if (is_iommu_enabled (d) ) {int ret; -this_cpu (iommu_dont_flush_iotlb) =0 ; - ret =iommu_iotlb_flush (d,_dfn (xatp->idx - done), done, IOMMU_FLUSHF_added | IOMMU_FLUSHF_modified);if (unlikely (ret) && rc >=0 )
xen/drivers/passthrough/iommu.c +1 -36
@@ -277,24 +277,6 @@ int iommu_map(struct domain *d, dfn_t dfn, mfn_t mfn,return rc; } -int iommu_legacy_map (struct domain *d,dfn_t dfn,mfn_t mfn, -unsigned int page_order,unsigned int flags) -{ -unsigned int flush_flags =0 ; -int rc =iommu_map (d, dfn, mfn, page_order, flags, &flush_flags); - -if ( !this_cpu (iommu_dont_flush_iotlb) ) - { -int err =iommu_iotlb_flush (d, dfn, (1u << page_order), - flush_flags); - -if ( !rc ) - rc = err; - } - -return rc; -} -int iommu_unmap (struct domain *d,dfn_t dfn,unsigned int page_order,unsigned int *flush_flags) {
@@ -333,23 +315,6 @@ int iommu_unmap(struct domain *d, dfn_t dfn, unsigned int page_order,return rc; } -int iommu_legacy_unmap (struct domain *d,dfn_t dfn,unsigned int page_order) -{ -unsigned int flush_flags =0 ; -int rc =iommu_unmap (d, dfn, page_order, &flush_flags); - -if ( !this_cpu (iommu_dont_flush_iotlb) ) - { -int err =iommu_iotlb_flush (d, dfn, (1u << page_order), - flush_flags); - -if ( !rc ) - rc = err; - } - -return rc; -} -int iommu_iotlb_flush (struct domain *d,dfn_t dfn,unsigned int page_count,unsigned int flush_flags) {
@@ -357,7 +322,7 @@ int iommu_iotlb_flush(struct domain *d, dfn_t dfn, unsigned int page_count,int rc;if ( !is_iommu_enabled (d) || !hd->platform_ops->iotlb_flush || - !page_count || !flush_flags ) + !page_count || !flush_flags ||this_cpu (iommu_dont_flush_iotlb) )return 0 ;if (dfn_eq (dfn,INVALID_DFN ) )
xen/include/xen/iommu.h +5 -15
@@ -151,13 +151,6 @@ int __must_check iommu_map(struct domain *d, dfn_t dfn, mfn_t mfn,int __must_check iommu_unmap (struct domain *d,dfn_t dfn,unsigned int page_order,unsigned int *flush_flags); - -int __must_check iommu_legacy_map (struct domain *d,dfn_t dfn,mfn_t mfn, -unsigned int page_order, -unsigned int flags); -int __must_check iommu_legacy_unmap (struct domain *d,dfn_t dfn, -unsigned int page_order); -int __must_check iommu_iotlb_flush (struct domain *d,dfn_t dfn,unsigned int page_count,unsigned int flush_flags);
@@ -364,15 +357,12 @@ void iommu_dev_iotlb_flush_timeout(struct domain *d, struct pci_dev *pdev);/* * The purpose of the iommu_dont_flush_iotlb optional cpu flag is to -* avoid unecessary iotlb_flush in the low level IOMMU code. -* -* iommu_map_page/iommu_unmap_page must flush the iotlb but somethimes -* this operation can be really expensive. This flag will be set by the -* caller to notify the low level IOMMU code to avoid the iotlb flushes. -* iommu_iotlb_flush/iommu_iotlb_flush_all will be explicitly called by -* the caller. +* avoid unecessary IOMMU flushing while updating the P2M. +* Setting the value to true will cause iommu_iotlb_flush() to return without +* actually performing a flush. A batch flush must therefore be done by the +* calling code after setting the value back to false. */ -DECLARE_PER_CPU (bool_t, iommu_dont_flush_iotlb); +DECLARE_PER_CPU (bool, iommu_dont_flush_iotlb);#endif /* _IOMMU_H_ */
Signature · 1 line
2.20.1
Re: [PATCH 4/6] remove remaining uses of iommu_legacy_map/unmap
Quoted 11 lines · Paul Durrant
--- xen/arch/x86/mm.c | 22 +++++++++++++++----- xen/arch/x86/mm/p2m-ept.c | 22 +++++++++++++------- xen/arch/x86/mm/p2m-pt.c | 17 +++++++++++---- xen/arch/x86/mm/p2m.c | 28 ++++++++++++++++++------- xen/arch/x86/x86_64/mm.c | 27 ++++++++++++++++++------ xen/common/grant_table.c | 36 +++++++++++++++++++++++++------- xen/common/memory.c | 7 +++---- xen/drivers/passthrough/iommu.c | 37 +-------------------------------- xen/include/xen/iommu.h | 20 +++++------------- 9 files changed, 123 insertions(+), 93 deletions(-)
Overall more code. I wonder whether a map-and-flush function (named differently than the current ones) wouldn't still be worthwhile to have.
Quoted 20 lines
xen/common/grant_table.c +9
-4@@ -1225,11 +1225,25 @@ map_grant_ref( kind = IOMMUF_readable; else kind =0 ; -if ( kind &&iommu_legacy_map (ld,_dfn (mfn_x (mfn)), mfn,0 , kind) ) +if ( kind ) { -double_gt_unlock (lgt, rgt); - rc = GNTST_general_error; - goto undo_out; +dfn_t dfn =_dfn (mfn_x (mfn)); +unsigned int flush_flags =0 ; +int err; + + err =iommu_map (ld, dfn, mfn,0 , kind, &flush_flags); +if ( err ) + rc = GNTST_general_error; +
xen/common/grant_table.c map_grant_ref( · line 1247
Quoted 10 lines
xen/common/grant_table.c +9
-0+ err =iommu_iotlb_flush (ld, dfn,1 , flush_flags); +if ( err ) + rc = GNTST_general_error; + +if ( rc != GNTST_okay ) + { +double_gt_unlock (lgt, rgt); + goto undo_out; + } }
The mapping needs to happen with at least ld's lock held, yes. But is the same true also for the flushing? Can't (not necessarily right in this change) the flush be pulled out of the function and instead done once per batch that got processed?
Jan
RE: [PATCH 4/6] remove remaining uses of iommu_legacy_map/unmap
Quoted 29 lines
-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 26 July 2020 09:36
To: Paul Durrant <[email protected]>
Cc: [email protected]; Durrant, Paul <[email protected]>; Andrew Cooper <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monné <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Julien Grall <[email protected]>; Stefano Stabellini <[email protected]>; Jun Nakajima <[email protected]>; Kevin Tian <[email protected]>
Subject: RE: [EXTERNAL] [PATCH 4/6] remove remaining uses of iommu_legacy_map/unmapCAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.
Quoted 11 lines · Paul Durrant
--- xen/arch/x86/mm.c | 22 +++++++++++++++----- xen/arch/x86/mm/p2m-ept.c | 22 +++++++++++++------- xen/arch/x86/mm/p2m-pt.c | 17 +++++++++++---- xen/arch/x86/mm/p2m.c | 28 ++++++++++++++++++------- xen/arch/x86/x86_64/mm.c | 27 ++++++++++++++++++------ xen/common/grant_table.c | 36 +++++++++++++++++++++++++------- xen/common/memory.c | 7 +++---- xen/drivers/passthrough/iommu.c | 37 +-------------------------------- xen/include/xen/iommu.h | 20 +++++------------- 9 files changed, 123 insertions(+), 93 deletions(-)
Quoted 3 lines
Overall more code. I wonder whether a map-and-flush function (named differently than the current ones) wouldn't still be worthwhile to have.
Agreed but an extra 30 lines is not huge. I'd still like to keep map/unmap and flush separate but I'll see if I can reduce the added lines.
Quoted 28 lines
Quoted 28 lines
xen/common/grant_table.c +16
-4@@ -1225,11 +1225,25 @@ map_grant_ref( kind = IOMMUF_readable; else kind =0 ; -if ( kind &&iommu_legacy_map (ld,_dfn (mfn_x (mfn)), mfn,0 , kind) ) +if ( kind ) { -double_gt_unlock (lgt, rgt); - rc = GNTST_general_error; - goto undo_out; +dfn_t dfn =_dfn (mfn_x (mfn)); +unsigned int flush_flags =0 ; +int err; + + err =iommu_map (ld, dfn, mfn,0 , kind, &flush_flags); +if ( err ) + rc = GNTST_general_error; + + err =iommu_iotlb_flush (ld, dfn,1 , flush_flags); +if ( err ) + rc = GNTST_general_error; + +if ( rc != GNTST_okay ) + { +double_gt_unlock (lgt, rgt);
Quoted 9 lines
xen/common/grant_table.c map_grant_ref( · line 1247
Quoted 3 lines
xen/common/grant_table.c +2
-0+ goto undo_out; + } }The mapping needs to happen with at least ld's lock held, yes. But is the same true also for the flushing? Can't (not necessarily right in this change) the flush be pulled out of the function and instead done once per batch that got processed?
True, the locks need not be held across the flush. I'll have a look at batching too.
Paul
Quoted 1 line
Jan
PATCH 5/6 iommu: remove the share_p2m operation
In reply to Paul DurrantFrom: Paul Durrant <[email protected]>
Sharing of HAP tables is VT-d specific so the operation is never defined for AMD IOMMU. There's also no need to pro-actively set vtd.pgd_maddr when using shared EPT as it is straightforward to simply define a helper function to return the appropriate value in the shared and non-shared cases.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Kevin Tian <[email protected]>
--- xen/arch/x86/mm/p2m.c | 3 -- xen/drivers/passthrough/iommu.c | 8 ----- xen/drivers/passthrough/vtd/iommu.c | 55 ++++++++++++++--------------- xen/include/xen/iommu.h | 3 -- 4 files changed, 27 insertions(+), 42 deletions(-)
xen/arch/x86/mm/p2m.c +0 -3
@@ -726,9 +726,6 @@ int p2m_alloc_table(struct p2m_domain *p2m) p2m->phys_table =pagetable_from_mfn (top_mfn); -if (hap_enabled (d) ) -iommu_share_p2m_table (d); -p2m_unlock (p2m);return 0 ; }
xen/drivers/passthrough/iommu.c +0 -8
@@ -478,14 +478,6 @@ int iommu_do_domctl(return ret; } -void iommu_share_p2m_table (struct domain * d) -{ -ASSERT (hap_enabled (d)); - -if (iommu_use_hap_pt (d) ) -iommu_get_ops ()->share_p2m (d); -} -void iommu_crash_shutdown (void ) {if ( !iommu_crash_disable )
xen/drivers/passthrough/vtd/iommu.c +27 -28
89 lines not on this page · raw message
xen/include/xen/iommu.h +0 -3
@@ -266,7 +266,6 @@ struct iommu_ops {int __must_check (*suspend)(void );void (*resume)(void ); -void (*share_p2m)(struct domain *d);void (*crash_shutdown)(void ); int __must_check (*iotlb_flush)(struct domain *d, dfn_t dfn, unsigned int page_count,
@@ -343,8 +342,6 @@ void iommu_resume(void);void iommu_crash_shutdown (void );int iommu_get_reserved_device_memory (iommu_grdm_t *,void *); -void iommu_share_p2m_table (struct domain *d); -#ifdef CONFIG_HAS_PCI int iommu_do_pci_domctl (struct xen_domctl *,struct domain *d,XEN_GUEST_HANDLE_PARAM (xen_domctl_t ));
Signature · 1 line
2.20.1
Re: [PATCH 5/6] iommu: remove the share_p2m operation
Quoted 4 lines · Paul Durrant
From: Paul Durrant <[email protected]>
Sharing of HAP tables is VT-d specific so the operation is never defined for AMD IOMMU.
It's not VT-d specific, and Xen really did used to share on AMD.
In fact, a remnant of this logic is still present in the form of the comment for p2m_type_t explaining why p2m_ram_rw needs to be 0.
That said, I agree it shouldn't be a hook, because it is statically known at domain create time based on flags and hardware properties.
Quoted 3 lines
There's also no need to pro-actively set vtd.pgd_maddr when using
shared EPT as it is straightforward to simply define a helper function to return the appropriate value in the shared and non-shared cases.
It occurs to me that vtd.pgd_maddr and amd.root_table really are the same thing, and furthermore are common to all IOMMU implementations on any architecture. Is it worth trying to make them common, rather than continuing to let each implementation handle them differently?
~Andrew
RE: [PATCH 5/6] iommu: remove the share_p2m operation
Quoted 14 lines
-----Original Message-----
From: Andrew Cooper <[email protected]>
Sent: 24 July 2020 20:01
To: Paul Durrant <[email protected]>; [email protected]
Cc: Durrant, Paul <[email protected]>; Jan Beulich <[email protected]>; George Dunlap <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monné <[email protected]>; Kevin Tian <[email protected]>
Subject: RE: [EXTERNAL] [PATCH 5/6] iommu: remove theshare_p2moperationCAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.
Quoted 8 lines
Quoted 4 lines · Paul Durrant
From: Paul Durrant <[email protected]>
Sharing of HAP tables is VT-d specific so the operation is never defined for AMD IOMMU.
It's not VT-d specific, and Xen really did used to share on AMD.
Oh, I never thought that ever worked.
Quoted 6 lines
In fact, a remnant of this logic is still present in the form of the comment for
p2m_type_texplaining whyp2m_ram_rwneeds to be 0.That said, I agree it shouldn't be a hook, because it is statically known at domain create time based on flags and hardware properties.
Well, for VT-d that may well change in future ;-)
Quoted 8 lines
Quoted 3 lines
There's also no need to pro-actively set vtd.pgd_maddr when using
shared EPT as it is straightforward to simply define a helper function to return the appropriate value in the shared and non-shared cases.It occurs to me that vtd.pgd_maddr and amd.root_table really are the same thing, and furthermore are common to all IOMMU implementations on any architecture. Is it worth trying to make them common, rather than continuing to let each implementation handle them differently?
I looked at this. The problem really lies in terminology. The 'root table' in the VT-d code refers to the single root table which IIRC is called the device table in the AMD IOMMU code so, whilst I could convert both to use a single common field, I think it's not really that valuable to do so since it's likely to make the code slightly more confusing to read (for someone expecting the names to tally with the spec).
Paul
Quoted 2 lines
~Andrew
Re: [PATCH 5/6] iommu: remove the share_p2m operation
In reply to Paul Durrantxen/drivers/passthrough/vtd/iommu.c static u64 addr_to_dma_page_maddr(struct domain *domain, u64 addr, int alloc) · line 316
Quoted 7 lines · Paul Durrant
xen/drivers/passthrough/vtd/iommu.c +1
-0@@ -313,6 +313,26 @@ static u64 addr_to_dma_page_maddr(struct domain *domain, u64 addr, int alloc)return pte_maddr; } +static u64 domain_pgd_maddr (struct domain *d)
uint64_t please.
Quoted 36 lines
xen/drivers/passthrough/vtd/iommu.c +21
-7+{ +struct domain_iommu *hd =dom_iommu (d); + +ASSERT (spin_is_locked (&hd->arch.mapping_lock)); + +if (iommu_use_hap_pt (d) ) + { +mfn_t pgd_mfn = +pagetable_get_mfn (p2m_get_pagetable (p2m_get_hostp2m (d))); + +return pagetable_get_paddr (pagetable_from_mfn (pgd_mfn)); + } + +if ( !hd->arch.vtd.pgd_maddr ) +addr_to_dma_page_maddr (d,0 ,1 ); + +return hd->arch.vtd.pgd_maddr; +} +static void iommu_flush_write_buffer (struct vtd_iommu *iommu) {u32 val;@@ -1347,22 +1367,17 @@ int domain_context_mapping_one( {spin_lock (&hd->arch.mapping_lock); -/* Ensure we have pagetables allocated down to leaf PTE. */ -if ( hd->arch.vtd.pgd_maddr ==0 ) + pgd_maddr =domain_pgd_maddr (domain); +if ( !pgd_maddr ) { -addr_to_dma_page_maddr (domain,0 ,1 ); -if ( hd->arch.vtd.pgd_maddr ==0 ) - { - nomem: -spin_unlock (&hd->arch.mapping_lock);
xen/drivers/passthrough/vtd/iommu.c int domain_context_mapping_one( · line 1378
Quoted 10 lines
xen/drivers/passthrough/vtd/iommu.c +5
-4-spin_unlock (&iommu->lock); -unmap_vtd_domain_page (context_entries); -return -ENOMEM ; - } + nomem: +spin_unlock (&hd->arch.mapping_lock); +spin_unlock (&iommu->lock); +unmap_vtd_domain_page (context_entries); +return -ENOMEM ; }
This renders all calls bogus in shared mode - the function, if it ended up getting called nevertheless, would then still alloc the root table. Therefore I'd like to suggest that at least all its callers get an explicit check. That's really just dma_pte_clear_one() as it looks.
Jan
RE: [PATCH 5/6] iommu: remove the share_p2m operation
Quoted 21 lines
-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 26 July 2020 09:50
To: Paul Durrant <[email protected]>
Cc: [email protected]; Durrant, Paul <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Wei Liu <[email protected]>; Roger Pau Monné <[email protected]>; Kevin Tian <[email protected]>
Subject: RE: [EXTERNAL] [PATCH 5/6] iommu: remove theshare_p2moperationCAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.
Quoted 6 lines · Paul Durrant
xen/drivers/passthrough/vtd/iommu.c +0
-0@@ -313,6 +313,26 @@ static u64 addr_to_dma_page_maddr(struct domain *domain, u64 addr, int alloc)return pte_maddr; }
Quoted 4 lines
xen/drivers/passthrough/vtd/iommu.c static u64 addr_to_dma_page_maddr(struct domain *domain, u64 addr, int alloc) · line 316
Quoted 1 line
xen/drivers/passthrough/vtd/iommu.c +1
-0+static u64 domain_pgd_maddr (struct domain *d)
uint64_tplease.
Ok.
Quoted 44 lines
Quoted 44 lines
xen/drivers/passthrough/vtd/iommu.c +25
-11+{ +struct domain_iommu *hd =dom_iommu (d); + +ASSERT (spin_is_locked (&hd->arch.mapping_lock)); + +if (iommu_use_hap_pt (d) ) + { +mfn_t pgd_mfn = +pagetable_get_mfn (p2m_get_pagetable (p2m_get_hostp2m (d))); + +return pagetable_get_paddr (pagetable_from_mfn (pgd_mfn)); + } + +if ( !hd->arch.vtd.pgd_maddr ) +addr_to_dma_page_maddr (d,0 ,1 ); + +return hd->arch.vtd.pgd_maddr; +} +static void iommu_flush_write_buffer (struct vtd_iommu *iommu) {u32 val;@@ -1347,22 +1367,17 @@ int domain_context_mapping_one( {spin_lock (&hd->arch.mapping_lock); -/* Ensure we have pagetables allocated down to leaf PTE. */ -if ( hd->arch.vtd.pgd_maddr ==0 ) + pgd_maddr =domain_pgd_maddr (domain); +if ( !pgd_maddr ) { -addr_to_dma_page_maddr (domain,0 ,1 ); -if ( hd->arch.vtd.pgd_maddr ==0 ) - { - nomem: -spin_unlock (&hd->arch.mapping_lock); -spin_unlock (&iommu->lock); -unmap_vtd_domain_page (context_entries); -return -ENOMEM ; - } + nomem: +spin_unlock (&hd->arch.mapping_lock); +spin_unlock (&iommu->lock); +unmap_vtd_domain_page (context_entries);
Quoted 9 lines
xen/drivers/passthrough/vtd/iommu.c int domain_context_mapping_one( · line 1378
Quoted 2 lines
xen/drivers/passthrough/vtd/iommu.c +1
-0+return -ENOMEM ; }This renders all calls bogus in shared mode - the function, if it ended up getting called nevertheless, would then still alloc the root table. Therefore I'd like to suggest that at least all its callers get an explicit check. That's really just
dma_pte_clear_one()as it looks.
Ok, I think I may move this code out into a separate function too since the nomem label is slightly awkward, so I'll re-factor it.
Paul
Quoted 1 line
Jan
PATCH 6/6 iommu: stop calling IOMMU page tables 'p2m tables'
In reply to Paul DurrantFrom: Paul Durrant <[email protected]>
It's confusing and not consistent with the terminology introduced with 'dfn_t'. Just call them IOMMU page tables.
Also remove a pointless check of the 'acpi_drhd_units' list in vtd_dump_page_table_level(). If the list is empty then IOMMU mappings would not have been enabled for the domain in the first place.
NOTE: There's also a bit of cosmetic clean-up in iommu_dump_page_tables()
to make use of %pd.
---
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Paul Durrant <[email protected]>
Cc: Kevin Tian <[email protected]>
--- xen/drivers/passthrough/amd/pci_amd_iommu.c | 16 +++++++------- xen/drivers/passthrough/iommu.c | 12 +++++------ xen/drivers/passthrough/vtd/iommu.c | 24 +++++++++------------ xen/include/xen/iommu.h | 2 +- 4 files changed, 25 insertions(+), 29 deletions(-)
xen/drivers/passthrough/amd/pci_amd_iommu.c +8 -8
@@ -499,8 +499,8 @@ static int amd_iommu_group_id(u16 seg, u8 bus, u8 devfn)#include <asm/io_apic.h> -static void amd_dump_p2m_table_level (struct page_info * pg,int level, -paddr_t gpa,int indent) +static void amd_dump_page_table_level (struct page_info * pg,int level, +paddr_t gpa,int indent) {paddr_t address;struct amd_iommu_pte *table_vaddr;
@@ -537,7 +537,7 @@ static void amd_dump_p2m_table_level(struct page_info* pg, int level, address = gpa +amd_offset_level_address (index, level);if ( pde->next_level >=1 ) -amd_dump_p2m_table_level ( +amd_dump_page_table_level (mfn_to_page (_mfn (pde->mfn)), pde->next_level, address, indent +1 );else
@@ -550,16 +550,16 @@ static void amd_dump_p2m_table_level(struct page_info* pg, int level,unmap_domain_page (table_vaddr); } -static void amd_dump_p2m_table (struct domain *d) +static void amd_dump_page_tables (struct domain *d) {const struct domain_iommu *hd =dom_iommu (d);if ( !hd->arch.amd_iommu.root_table )return ; -printk ("p2m table has %d levels\n" , hd->arch.amd_iommu.paging_mode); -amd_dump_p2m_table_level (hd->arch.amd_iommu.root_table, - hd->arch.amd_iommu.paging_mode,0 ,0 ); +printk ("table has %d levels\n" , hd->arch.amd_iommu.paging_mode); +amd_dump_page_table_level (hd->arch.amd_iommu.root_table, + hd->arch.amd_iommu.paging_mode,0 ,0 ); } static const struct iommu_ops __initconstrel _iommu_ops = {
@@ -586,7 +586,7 @@ static const struct iommu_ops __initconstrel _iommu_ops = { .suspend = amd_iommu_suspend, .resume = amd_iommu_resume, .crash_shutdown = amd_iommu_crash_shutdown, - .dump_p2m_table = amd_dump_p2m_table, + .dump_page_tables = amd_dump_page_tables, }; static const struct iommu_init_ops __initconstrel _iommu_init_ops = {
xen/drivers/passthrough/iommu.c +6 -6
@@ -22,7 +22,7 @@#include <xen/keyhandler.h> #include <xsm/xsm.h> -static void iommu_dump_p2m_table (unsigned char key); +static void iommu_dump_page_tables (unsigned char key);unsigned int __read_mostly iommu_dev_iotlb_timeout =1000 ;integer_param ("iommu_dev_iotlb_timeout" , iommu_dev_iotlb_timeout);
@@ -212,7 +212,7 @@ void __hwdom_init iommu_hwdom_init(struct domain *d)if ( !is_iommu_enabled (d) )return ; -register_keyhandler ('o' , &iommu_dump_p2m_table,"dump iommu p2m table" ,0 ); +register_keyhandler ('o' , &iommu_dump_page_tables,"dump iommu page tables" ,0 ); hd->platform_ops->hwdom_init (d); }
@@ -513,7 +513,7 @@ bool_t iommu_has_feature(struct domain *d, enum iommu_feature feature)return is_iommu_enabled (d) &&test_bit (feature,dom_iommu (d)->features); } -static void iommu_dump_p2m_table (unsigned char key) +static void iommu_dump_page_tables (unsigned char key) {struct domain *d;const struct iommu_ops *ops;
@@ -535,12 +535,12 @@ static void iommu_dump_p2m_table(unsigned char key)if (iommu_use_hap_pt (d) ) { -printk ("\ndomain%d IOMMU p2m table shared with MMU: \n" , d->domain_id); +printk ("%pd: IOMMU page tables shared with MMU\n" , d);continue ; } -printk ("\ndomain%d IOMMU p2m table: \n" , d->domain_id); - ops->dump_p2m_table (d); +printk ("%pd: IOMMU page tables: \n" , d); + ops->dump_page_tables (d); }rcu_read_unlock (&domlist_read_lock);
xen/drivers/passthrough/vtd/iommu.c +10 -14
@@ -2545,8 +2545,8 @@ static void vtd_resume(void) } } -static void vtd_dump_p2m_table_level (paddr_t pt_maddr,int level,paddr_t gpa, -int indent) +static void vtd_dump_page_table_level (paddr_t pt_maddr,int level,paddr_t gpa, +int indent) {paddr_t address;int i;
@@ -2575,8 +2575,8 @@ static void vtd_dump_p2m_table_level(paddr_t pt_maddr, int level, paddr_t gpa, address = gpa +offset_level_address (i, level);if ( next_level >=1 ) -vtd_dump_p2m_table_level (dma_pte_addr (*pte), next_level, - address, indent +1 ); +vtd_dump_page_table_level (dma_pte_addr (*pte), next_level, + address, indent +1 );else printk("%*sdfn: %08lx mfn: %08lx\n", indent, "",
@@ -2587,17 +2587,13 @@ static void vtd_dump_p2m_table_level(paddr_t pt_maddr, int level, paddr_t gpa,unmap_vtd_domain_page (pt_vaddr); } -static void vtd_dump_p2m_table (struct domain *d) +static void vtd_dump_page_tables (struct domain *d) { -const struct domain_iommu *hd; +const struct domain_iommu *hd =dom_iommu (d); -if (list_empty (&acpi_drhd_units) ) -return ; - - hd =dom_iommu (d); -printk ("p2m table has %d levels\n" ,agaw_to_level (hd->arch.vtd.agaw)); -vtd_dump_p2m_table_level (hd->arch.vtd.pgd_maddr, -agaw_to_level (hd->arch.vtd.agaw),0 ,0 ); +printk ("table has %d levels\n" ,agaw_to_level (hd->arch.vtd.agaw)); +vtd_dump_page_table_level (hd->arch.vtd.pgd_maddr, +agaw_to_level (hd->arch.vtd.agaw),0 ,0 ); }static int __init intel_iommu_quarantine_init (struct domain *d)
@@ -2686,7 +2682,7 @@ static struct iommu_ops __initdata vtd_ops = { .iotlb_flush = iommu_flush_iotlb_pages, .iotlb_flush_all = iommu_flush_iotlb_all, .get_reserved_device_memory = intel_iommu_get_reserved_device_memory, - .dump_p2m_table = vtd_dump_p2m_table, + .dump_page_tables = vtd_dump_page_tables, }; const struct iommu_init_ops __initconstrel intel_iommu_init_ops = {
xen/include/xen/iommu.h +1 -1
@@ -272,7 +272,7 @@ struct iommu_ops { unsigned int flush_flags);int __must_check (*iotlb_flush_all)(struct domain *d);int (*get_reserved_device_memory)(iommu_grdm_t *,void *); -void (*dump_p2m_table)(struct domain *d); +void (*dump_page_tables)(struct domain *d);#ifdef CONFIG_HAS_DEVICE_TREE /*
Signature · 1 line
2.20.1
Re: [PATCH 6/6] iommu: stop calling IOMMU page tables 'p2m tables'
xen/drivers/passthrough/iommu.c static void iommu_dump_p2m_table(unsigned char key) · line 538
Quoted 10 lines · Paul Durrant
xen/drivers/passthrough/iommu.c +1
-1@@ -535,12 +535,12 @@ static void iommu_dump_p2m_table(unsigned char key)if (iommu_use_hap_pt (d) ) { -printk ("\ndomain%d IOMMU p2m table shared with MMU: \n" , d->domain_id); +printk ("%pd: IOMMU page tables shared with MMU\n" , d);
Change MMU to CPU? MMU is very ambiguous in this context.
xen/drivers/passthrough/iommu.c static void iommu_dump_p2m_table(unsigned char key) · line 542
Quoted 6 lines
xen/drivers/passthrough/iommu.c +1
-2continue ; } -printk ("\ndomain%d IOMMU p2m table: \n" , d->domain_id); - ops->dump_p2m_table (d); +printk ("%pd: IOMMU page tables: \n" , d);
Drop the trailing whitespace?
~Andrew
RE: [PATCH 6/6] iommu: stop calling IOMMU page tables 'p2m tables'
Quoted 20 lines
-----Original Message-----
From: Andrew Cooper <[email protected]>
Sent: 24 July 2020 20:09
To: Paul Durrant <[email protected]>; [email protected]
Cc: Durrant, Paul <[email protected]>; Jan Beulich <[email protected]>; Kevin Tian <[email protected]>
Subject: RE: [EXTERNAL] [PATCH 6/6] iommu: stop calling IOMMU page tables 'p2m tables'CAUTION: This email originated from outside of the organization. Do not click links or open attachments unless you can confirm the sender and know the content is safe.
Quoted 6 lines · Paul Durrant
xen/drivers/passthrough/iommu.c +0
-0@@ -535,12 +535,12 @@ static void iommu_dump_p2m_table(unsigned char key)
Quoted 7 lines
xen/drivers/passthrough/iommu.c static void iommu_dump_p2m_table(unsigned char key) · line 538
Quoted 4 lines
xen/drivers/passthrough/iommu.c +1
-1if (iommu_use_hap_pt (d) ) { -printk ("\ndomain%d IOMMU p2m table shared with MMU: \n" , d->domain_id); +printk ("%pd: IOMMU page tables shared with MMU\n" , d);Change MMU to CPU? MMU is very ambiguous in this context.
Actually I could push this into the VT-d code and just say something like 'shared EPT is enabled'. Would that be less ambiguous?
Quoted 9 lines
xen/drivers/passthrough/iommu.c static void iommu_dump_p2m_table(unsigned char key) · line 542
Quoted 6 lines
xen/drivers/passthrough/iommu.c +1
-2continue ; } -printk ("\ndomain%d IOMMU p2m table: \n" , d->domain_id); - ops->dump_p2m_table (d); +printk ("%pd: IOMMU page tables: \n" , d);Drop the trailing whitespace?
Sure.
Paul
Quoted 1 line
~Andrew