Skip to contentSkip to conversation

use stashed domain create flags...

PATCH Superseded Paul Durrant 7 participants 35 messages Aug 16, 2019 – Aug 29, 2019

v6 of 13 Superseded Latest is v13 →

lore
Share & tools
lore
#0 lore raw

v6 cover use stashed domain create flags...

...and add per-domain IOMMU control

This is a combination of my previously separate series [1] and [2].

[1] https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02253.html
[2] https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

Paul Durrant (10):
make passthrough/pci.c:deassign_device() static x86/hvm/domain: remove the 'hap_enabled' flag x86/domain: remove the 'oos_off' flag
domain: remove the 'is_xenstore' flag
x86/domain: remove the 's3_integrity' flag
domain: introduce XEN_DOMCTL_CDF_iommu flag
use is_iommu_enabled() where appropriate...
remove late (on-demand) construction of IOMMU page tables
iommu: tidy up iommu_use_hap_pt() and need_iommu_pt_sync() macros introduce a 'passthrough' configuration option to xl.cfg...

 docs/man/xl.cfg.5.pod.in                  |  52 +++++++
 tools/libxl/libxl.h                       |   5 +
 tools/libxl/libxl_create.c                |   9 ++
 tools/libxl/libxl_types.idl               |   7 +
 tools/xl/xl_parse.c                       |  38 +++++
 xen/arch/arm/domain.c                     |  10 +-
 xen/arch/arm/p2m.c                        |   4 +-
 xen/arch/arm/setup.c                      |   3 +
 xen/arch/x86/dom0_build.c                 |   4 +-
 xen/arch/x86/domain.c                     |  24 +--
 xen/arch/x86/domctl.c                     |   4 +-
 xen/arch/x86/hvm/hvm.c                    |   6 +-
 xen/arch/x86/hvm/mtrr.c                   |   5 +-
 xen/arch/x86/hvm/vioapic.c                |   2 +-
 xen/arch/x86/hvm/vmx/vmcs.c               |   2 +-
 xen/arch/x86/hvm/vmx/vmx.c                |   2 +-
 xen/arch/x86/mm/mem_sharing.c             |   2 +-
 xen/arch/x86/mm/p2m-ept.c                 |   4 +-
 xen/arch/x86/mm/paging.c                  |   6 +-
 xen/arch/x86/mm/shadow/common.c           |   7 +-
 xen/arch/x86/mm/shadow/none.c             |   2 +-
 xen/arch/x86/setup.c                      |   5 +-
 xen/arch/x86/tboot.c                      |   2 +-
 xen/arch/x86/x86_64/mm.c                  |   2 +-
 xen/common/domain.c                       |  40 ++++-
 xen/common/domctl.c                       |  10 +-
 xen/common/memory.c                       |   4 +-
 xen/common/vm_event.c                     |   2 +-
 xen/drivers/passthrough/amd/iommu_guest.c |   2 +-
 xen/drivers/passthrough/device_tree.c     |  18 +--
 xen/drivers/passthrough/io.c              |   8 +-
 xen/drivers/passthrough/iommu.c           | 182 +++++++---------------
 xen/drivers/passthrough/pci.c             | 119 +++++++-------
 xen/drivers/passthrough/vtd/iommu.c       |  12 +-
 xen/drivers/passthrough/vtd/x86/hvm.c     |   2 +-
 xen/drivers/passthrough/x86/iommu.c       |  97 +-----------
 xen/include/asm-arm/iommu.h               |   3 -
 xen/include/asm-x86/domain.h              |   3 -
 xen/include/asm-x86/hvm/domain.h          |   7 -
 xen/include/asm-x86/iommu.h               |   4 -
 xen/include/asm-x86/paging.h              |   2 +-
 xen/include/asm-x86/shadow.h              |   2 +-
 xen/include/public/domctl.h               |  10 +-
 xen/include/xen/iommu.h                   |  41 ++---
 xen/include/xen/sched.h                   |  26 ++--
 xen/include/xsm/dummy.h                   |   2 +-
 xen/xsm/flask/hooks.c                     |  18 +--
 47 files changed, 399 insertions(+), 422 deletions(-)

---
Cc: Andrew Cooper <[email protected]>
Cc: Anthony PERARD <[email protected]>
Cc: Brian Woods <[email protected]>
Cc: Daniel De Graaf <[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: Petre Pircalabu <[email protected]>
Cc: Razvan Cojocaru <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>
Cc: Tamas K Lengyel <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Volodymyr Babchuk <[email protected]>
Cc: Wei Liu <[email protected]>

Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#1 lore raw

PATCH v6 1/10 make passthrough/pci.c:deassign_device() static

This function is only ever called from within the same source module and really has no business being declared xen/iommu.h. This patch relocates the function ahead of the first called and makes it static.

---

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v6:
- Re-base due to re-positioning in series

v5:
- minor style fixes
- use %pd, rather than d%d

---
 xen/drivers/passthrough/pci.c | 99 ++++++++++++++++++-----------------
 xen/include/xen/iommu.h       |  1 -
 2 files changed, 50 insertions(+), 50 deletions(-)
xen/drivers/passthrough/pci.c +50 -49
index 7c196ba58b..89536cc067 100644
@@ -889,6 +889,56 @@ static int pci_clean_dpci_irqs(struct domain *d)
     return 0;
 }
 
+/* caller should hold the pcidevs_lock */
+static int deassign_device(struct domain *d, uint16_t seg, uint8_t bus,
+                           uint8_t devfn)
+{
+    const struct domain_iommu *hd = dom_iommu(d);
+    struct pci_dev *pdev;
+    int ret = 0;
+
+    if ( !iommu_enabled || !hd->platform_ops )
+        return -EINVAL;
+
+    ASSERT(pcidevs_locked());
+    pdev = pci_get_pdev_by_domain(d, seg, bus, devfn);
+    if ( !pdev )
+        return -ENODEV;
+
+    while ( pdev->phantom_stride )
+    {
+        devfn += pdev->phantom_stride;
+        if ( PCI_SLOT(devfn) != PCI_SLOT(pdev->devfn) )
+            break;
+        ret = hd->platform_ops->reassign_device(d, hardware_domain, devfn,
+                                                pci_to_dev(pdev));
+        if ( !ret )
+            continue;
+
+        printk(XENLOG_G_ERR "%pd: deassign %04x:%02x:%02x.%u failed (%d)\n",
+               d, seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn), ret);
+        return ret;
+    }
+
+    devfn = pdev->devfn;
+    ret = hd->platform_ops->reassign_device(d, hardware_domain, devfn,
+                                            pci_to_dev(pdev));
+    if ( ret )
+    {
+        dprintk(XENLOG_G_ERR,
+                "%pd: deassign device (%04x:%02x:%02x.%u) failed\n",
+                d, seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
+        return ret;
+    }
+
+    pdev->fault.count = 0;
+
+    if ( !has_arch_pdevs(d) && has_iommu_pt(d) )
+        iommu_teardown(d);
+
+    return ret;
+}
+
 int pci_release_devices(struct domain *d)
 {
     struct pci_dev *pdev;
@@ -1476,55 +1526,6 @@ static int assign_device(struct domain *d, u16 seg, u8 bus, u8 devfn, u32 flag)
     return rc;
 }
 
-/* caller should hold the pcidevs_lock */
-int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn)
-{
-    const struct domain_iommu *hd = dom_iommu(d);
-    struct pci_dev *pdev = NULL;
-    int ret = 0;
-
-    if ( !iommu_enabled || !hd->platform_ops )
-        return -EINVAL;
-
-    ASSERT(pcidevs_locked());
-    pdev = pci_get_pdev_by_domain(d, seg, bus, devfn);
-    if ( !pdev )
-        return -ENODEV;
-
-    while ( pdev->phantom_stride )
-    {
-        devfn += pdev->phantom_stride;
-        if ( PCI_SLOT(devfn) != PCI_SLOT(pdev->devfn) )
-            break;
-        ret = hd->platform_ops->reassign_device(d, hardware_domain, devfn,
-                                                pci_to_dev(pdev));
-        if ( !ret )
-            continue;
-
-        printk(XENLOG_G_ERR "d%d: deassign %04x:%02x:%02x.%u failed (%d)\n",
-               d->domain_id, seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn), ret);
-        return ret;
-    }
-
-    devfn = pdev->devfn;
-    ret = hd->platform_ops->reassign_device(d, hardware_domain, devfn,
-                                            pci_to_dev(pdev));
-    if ( ret )
-    {
-        dprintk(XENLOG_G_ERR,
-                "d%d: deassign device (%04x:%02x:%02x.%u) failed\n",
-                d->domain_id, seg, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
-        return ret;
-    }
-
-    pdev->fault.count = 0;
-
-    if ( !has_arch_pdevs(d) && has_iommu_pt(d) )
-        iommu_teardown(d);
-
-    return ret;
-}
-
 static int iommu_get_device_group(
     struct domain *d, u16 seg, u8 bus, u8 devfn,
     XEN_GUEST_HANDLE_64(uint32) buf, int max_sdevs)
xen/include/xen/iommu.h +0 -1
index 48f87480a7..314f28f323 100644
@@ -70,7 +70,6 @@ int iommu_hardware_setup(void);
 int iommu_domain_init(struct domain *d);
 void iommu_hwdom_init(struct domain *d);
 void iommu_domain_destroy(struct domain *d);
-int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn);
 
 void arch_iommu_domain_destroy(struct domain *d);
 int arch_iommu_domain_init(struct domain *d);
Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Reviewed-by on v6 1/10
#2 lore raw

Re: [Xen-devel] [PATCH v6 01/10] make passthrough/pci.c:deassign_device() static

Quoted 3 lines · Paul Durrant

This function is only ever called from within the same source module and really has no business being declared xen/iommu.h. This patch relocates the function ahead of the first called and makes it static.

^ caller

Quoted 3 lines

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#3 lore raw

PATCH v6 2/10 x86/hvm/domain: remove the 'hap_enabled' flag

In reply to Paul Durrant

The hap_enabled() macro can determine whether the feature is available using the domain 'options'; there is no need for a separate flag.

NOTE: Furthermore, by extending sanitiziing of the domain 'options', the
macro can be transformed into an inline function and re-located to xen/sched.h. This also makes hap_enabled() common, thus allowing removal of an ugly ifdef CONFIG_X86 from the common iommu code.

---
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: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: George Dunlap <[email protected]>

v4:
- Add missing perentheses and move hap_enabled()
- Fix the shim build

v3:
- Re-worked as suggested by Jan
- Not adding Roger's R-b as the patch has changed substantially

v2:
- Defer changes to shadow_domain_init() to patch #4

---
 xen/arch/x86/domain.c            | 13 +++++++------
 xen/arch/x86/mm/paging.c         |  4 ++--
 xen/common/domain.c              |  7 +++++++
 xen/drivers/passthrough/iommu.c  |  2 --
 xen/include/asm-x86/hvm/domain.h |  7 -------
 xen/include/asm-x86/paging.h     |  2 +-
 xen/include/xen/sched.h          |  6 ++++++
 7 files changed, 23 insertions(+), 18 deletions(-)
xen/arch/x86/domain.c +7 -6
index 9a6eb89ddc..bc0db03387 100644
@@ -460,6 +460,12 @@ int arch_sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( (config->flags & XEN_DOMCTL_CDF_hap) && !hvm_hap_supported() )
+    {
+        dprintk(XENLOG_INFO, "HAP enabled but not supported\n");
+        return -EINVAL;
+    }
+
     return 0;
 }
 
@@ -564,12 +570,7 @@ int arch_domain_create(struct domain *d,
     HYPERVISOR_COMPAT_VIRT_START(d) =
         is_pv_domain(d) ? __HYPERVISOR_COMPAT_VIRT_START : ~0u;
 
-    /* Need to determine if HAP is enabled before initialising paging */
-    if ( is_hvm_domain(d) )
-        d->arch.hvm.hap_enabled =
-            hvm_hap_supported() && (config->flags & XEN_DOMCTL_CDF_hap);
-
-    if ( (rc = paging_domain_init(d, config->flags)) != 0 )
+    if ( (rc = paging_domain_init(d)) != 0 )
         goto fail;
     paging_initialised = true;
 
xen/arch/x86/mm/paging.c +2 -2
index 011089368a..097a27f608 100644
@@ -632,7 +632,7 @@ void paging_log_dirty_init(struct domain *d, const struct log_dirty_ops *ops)
 /*           CODE FOR PAGING SUPPORT            */
 /************************************************/
 /* Domain paging struct initialization. */
-int paging_domain_init(struct domain *d, unsigned int domcr_flags)
+int paging_domain_init(struct domain *d)
 {
     int rc;
 
@@ -653,7 +653,7 @@ int paging_domain_init(struct domain *d, unsigned int domcr_flags)
     if ( hap_enabled(d) )
         hap_domain_init(d);
     else
-        rc = shadow_domain_init(d, domcr_flags);
+        rc = shadow_domain_init(d, d->options);
 
     return rc;
 }
xen/common/domain.c +7 -0
index 744b572195..6109623730 100644
@@ -313,6 +313,13 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( !(config->flags & XEN_DOMCTL_CDF_hvm_guest) &&
+         (config->flags & XEN_DOMCTL_CDF_hap) )
+    {
+        dprintk(XENLOG_INFO, "HAP enabled for non-HVM guest\n");
+        return -EINVAL;
+    }
+
     return arch_sanitise_domain_config(config);
 }
 
xen/drivers/passthrough/iommu.c +0 -2
index f8c3bf53bd..37eb0f7d01 100644
@@ -573,9 +573,7 @@ int iommu_do_domctl(
 
 void iommu_share_p2m_table(struct domain* d)
 {
-#ifdef CONFIG_X86
     ASSERT(hap_enabled(d));
-#endif
     /*
      * iommu_use_hap_pt(d) cannot be used here because during domain
      * construction need_iommu(d) will always return false here.
xen/include/asm-x86/hvm/domain.h +0 -7
index 6c7c4f5aa6..bcc5621797 100644
@@ -156,7 +156,6 @@ struct hvm_domain {
 
     struct viridian_domain *viridian;
 
-    bool_t                 hap_enabled;
     bool_t                 mem_sharing_enabled;
     bool_t                 qemu_mapcache_invalidate;
     bool_t                 is_s3_suspended;
@@ -195,12 +194,6 @@ struct hvm_domain {
     };
 };
 
-#ifdef CONFIG_HVM
-#define hap_enabled(d)  (is_hvm_domain(d) && (d)->arch.hvm.hap_enabled)
-#else
-#define hap_enabled(d)  ({(void)(d); false;})
-#endif
-
 #endif /* __ASM_X86_HVM_DOMAIN_H__ */
 
 /*
xen/include/asm-x86/paging.h +1 -1
index cf57ca708d..ab7887f23c 100644
@@ -207,7 +207,7 @@ void paging_vcpu_init(struct vcpu *v);
 
 /* Set up the paging-assistance-specific parts of a domain struct at
  * start of day.  Called for every domain from arch_domain_create() */
-int paging_domain_init(struct domain *d, unsigned int domcr_flags);
+int paging_domain_init(struct domain *d);
 
 /* Handler for paging-control ops: operations from user-space to enable
  * and disable ephemeral shadow modes (test mode and log-dirty mode) and
xen/include/xen/sched.h +6 -0
index 2e6e0d3488..07a64947ed 100644
@@ -954,6 +954,12 @@ static inline bool is_hvm_vcpu(const struct vcpu *v)
     return is_hvm_domain(v->domain);
 }
 
+static inline bool hap_enabled(const struct domain *d)
+{
+    return IS_ENABLED(CONFIG_HVM) && /* necessary for pv shim build */
+        evaluate_nospec(d->options & XEN_DOMCTL_CDF_hap);
+}
+
 static inline bool is_hwdom_pinned_vcpu(const struct vcpu *v)
 {
     return (is_hardware_domain(v->domain) &&
Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Reviewed-by on v6 2/10
#4 lore raw

Re: [Xen-devel] [PATCH v6 02/10] x86/hvm/domain: remove the 'hap_enabled' flag

Quoted 9 lines · Paul Durrant

The hap_enabled() macro can determine whether the feature is available using the domain 'options'; there is no need for a separate flag.

NOTE: Furthermore, by extending sanitiziing of the domain 'options', the
macro can be transformed into an inline function and re-located to xen/sched.h. This also makes hap_enabled() common, thus allowing removal of an ugly ifdef CONFIG_X86 from the common iommu code.

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#5 lore raw

Re: [Xen-devel] [PATCH v6 02/10] x86/hvm/domain: remove the 'hap_enabled' flag

In reply to Paul Durrant
Quoted 4 lines · Paul Durrant

The hap_enabled() macro can determine whether the feature is available using the domain 'options'; there is no need for a separate flag.

NOTE: Furthermore, by extending sanitiziing of the domain 'options', the

s/ii/i/

xen/arch/x86/domain.c int arch_sanitise_domain_config(struct xen_domctl_createdomain *config) · line 465

Quoted 11 lines
xen/arch/x86/domain.c +3 -0
index 9a6eb89ddc..bc0db03387 100644
@@ -460,6 +460,12 @@ int arch_sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( (config->flags & XEN_DOMCTL_CDF_hap) && !hvm_hap_supported() )
+    {
+        dprintk(XENLOG_INFO, "HAP enabled but not supported\n");

s/enabled/requested/

xen/common/domain.c static int sanitise_domain_config(struct xen_domctl_createdomain *config) · line 319

Quoted 12 lines
xen/common/domain.c +4 -0
index 744b572195..6109623730 100644
@@ -313,6 +313,13 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( !(config->flags & XEN_DOMCTL_CDF_hvm_guest) &&
+         (config->flags & XEN_DOMCTL_CDF_hap) )
+    {
+        dprintk(XENLOG_INFO, "HAP enabled for non-HVM guest\n");

Again, I think 'requested' would be better here.

xen/include/xen/sched.h static inline bool is_hvm_vcpu(const struct vcpu *v) · line 960

Quoted 12 lines
xen/include/xen/sched.h +4 -0
index 2e6e0d3488..07a64947ed 100644
@@ -954,6 +954,12 @@ static inline bool is_hvm_vcpu(const struct vcpu *v)
     return is_hvm_domain(v->domain);
 }
 
+static inline bool hap_enabled(const struct domain *d)
+{
+    return IS_ENABLED(CONFIG_HVM) && /* necessary for pv shim build */
+        evaluate_nospec(d->options & XEN_DOMCTL_CDF_hap);

I'm not sure how helpful this comment is.  What should be here however is a note saying that this logic depends on domain_create() rejecting !HVM  and HAP.

All can be adjusted on commit if there are no other concerns.

~Andrew

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#6 lore raw

Re: [Xen-devel] [PATCH v6 02/10] x86/hvm/domain: remove the 'hap_enabled' flag

Quoted 50 lines · Andrew Cooper
Quoted 4 lines · Paul Durrant

The hap_enabled() macro can determine whether the feature is available using the domain 'options'; there is no need for a separate flag.

NOTE: Furthermore, by extending sanitiziing of the domain 'options', the

s/ii/i/

xen/arch/x86/domain.c int arch_sanitise_domain_config(struct xen_domctl_createdomain *config) · line 465

Quoted 11 lines
xen/arch/x86/domain.c +3 -0
index 9a6eb89ddc..bc0db03387 100644
@@ -460,6 +460,12 @@ int arch_sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( (config->flags & XEN_DOMCTL_CDF_hap) && !hvm_hap_supported() )
+    {
+        dprintk(XENLOG_INFO, "HAP enabled but not supported\n");

s/enabled/requested/

xen/common/domain.c static int sanitise_domain_config(struct xen_domctl_createdomain *config) · line 319

Quoted 12 lines
xen/common/domain.c +4 -0
index 744b572195..6109623730 100644
@@ -313,6 +313,13 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( !(config->flags & XEN_DOMCTL_CDF_hvm_guest) &&
+         (config->flags & XEN_DOMCTL_CDF_hap) )
+    {
+        dprintk(XENLOG_INFO, "HAP enabled for non-HVM guest\n");

Again, I think 'requested' would be better here.

xen/include/xen/sched.h static inline bool is_hvm_vcpu(const struct vcpu *v) · line 960

Quoted 12 lines
xen/include/xen/sched.h +4 -0
index 2e6e0d3488..07a64947ed 100644
@@ -954,6 +954,12 @@ static inline bool is_hvm_vcpu(const struct vcpu *v)
     return is_hvm_domain(v->domain);
 }
 
+static inline bool hap_enabled(const struct domain *d)
+{
+    return IS_ENABLED(CONFIG_HVM) && /* necessary for pv shim build */
+        evaluate_nospec(d->options & XEN_DOMCTL_CDF_hap);

I'm not sure how helpful this comment is.  What should be here however is a note saying that this logic depends on domain_create() rejecting !HVM  and HAP.

Quoted 1 line

All can be adjusted on commit if there are no other concerns.

One other thing.  Why is this eval_nospec()?

~Andrew

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#7 lore raw

Re: [Xen-devel] [PATCH v6 02/10] x86/hvm/domain: remove the 'hap_enabled' flag

Quoted 10 lines

-----Original Message-----
From: Andrew Cooper <[email protected]>
Sent: 23 August 2019 13:26
To: Paul Durrant <[email protected]>; [email protected]
Cc: Stefano Stabellini <[email protected]>; Wei Liu <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; George Dunlap <[email protected]>; Tim (Xen.org) <[email protected]>; Ian Jackson <[email protected]>; Julien Grall <[email protected]>; Jan Beulich <[email protected]>; Roger Pau Monne <[email protected]>
Subject: Re: [Xen-devel] [PATCH v6 02/10] x86/hvm/domain: remove the 'hap_enabled' flag

Quoted 7 lines
Quoted 6 lines · Andrew Cooper
Quoted 4 lines · Paul Durrant

The hap_enabled() macro can determine whether the feature is available using the domain 'options'; there is no need for a separate flag.

NOTE: Furthermore, by extending sanitiziing of the domain 'options', the

s/ii/i/

Oh yes.

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
xen/arch/x86/domain.c +0 -0
index 9a6eb89ddc..bc0db03387 100644
@@ -460,6 +460,12 @@ int arch_sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
Quoted 5 lines
Quoted 5 lines

xen/arch/x86/domain.c int arch_sanitise_domain_config(struct xen_domctl_createdomain *config) · line 465

Quoted 3 lines
xen/arch/x86/domain.c +3 -0
+    if ( (config->flags & XEN_DOMCTL_CDF_hap) && !hvm_hap_supported() )
+    {
+        dprintk(XENLOG_INFO, "HAP enabled but not supported\n");

s/enabled/requested/

I'm not fussed... I just went with the incumbent flag name.

Quoted 30 lines
Quoted 30 lines

xen/common/domain.c static int sanitise_domain_config(struct xen_domctl_createdomain *config) · line 319

Quoted 12 lines
xen/common/domain.c +4 -0
index 744b572195..6109623730 100644
@@ -313,6 +313,13 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( !(config->flags & XEN_DOMCTL_CDF_hvm_guest) &&
+         (config->flags & XEN_DOMCTL_CDF_hap) )
+    {
+        dprintk(XENLOG_INFO, "HAP enabled for non-HVM guest\n");

Again, I think 'requested' would be better here.

xen/include/xen/sched.h static inline bool is_hvm_vcpu(const struct vcpu *v) · line 960

Quoted 12 lines
xen/include/xen/sched.h +4 -0
index 2e6e0d3488..07a64947ed 100644
@@ -954,6 +954,12 @@ static inline bool is_hvm_vcpu(const struct vcpu *v)
     return is_hvm_domain(v->domain);
 }
 
+static inline bool hap_enabled(const struct domain *d)
+{
+    return IS_ENABLED(CONFIG_HVM) && /* necessary for pv shim build */
+        evaluate_nospec(d->options & XEN_DOMCTL_CDF_hap);

I'm not sure how helpful this comment is.  What should be here however is a note saying that this logic depends on domain_create() rejecting !HVM  and HAP.

Quoted 2 lines
Quoted 1 line

All can be adjusted on commit if there are no other concerns.

Ok.

Quoted 2 lines

One other thing.  Why is this eval_nospec()?

General paranoia about what might happen in speculation if the inline evaluates false and we wander into e.g. shadow code.

Paul

Quoted 1 line

~Andrew

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#8 lore raw

PATCH v6 3/10 x86/domain: remove the 'oos_off' flag

In reply to Paul Durrant

The flag is not needed since the domain 'options' can now be tested directly.

---
Cc: Tim Deegan <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

v3:
- Force 'oos_off' to be set for PV guests (to avoid call to is_hvm_domain() except in ASSERT)
- Dropped Tim's A-b because of the change

v2:
- Move some of the hunks from patch #3
- Also update the definition of shadow_domain_init() in none.c

---
 xen/arch/x86/mm/paging.c        |  2 +-
 xen/arch/x86/mm/shadow/common.c |  7 ++++---
 xen/arch/x86/mm/shadow/none.c   |  2 +-
 xen/common/domain.c             | 16 ++++++++++++----
 xen/include/asm-x86/domain.h    |  1 -
 xen/include/asm-x86/shadow.h    |  2 +-
 6 files changed, 19 insertions(+), 11 deletions(-)
xen/arch/x86/mm/paging.c +1 -1
index 097a27f608..69aa228e46 100644
@@ -653,7 +653,7 @@ int paging_domain_init(struct domain *d)
     if ( hap_enabled(d) )
         hap_domain_init(d);
     else
-        rc = shadow_domain_init(d, d->options);
+        rc = shadow_domain_init(d);
 
     return rc;
 }
xen/arch/x86/mm/shadow/common.c +4 -3
index fa18de0bb6..1187176993 100644
@@ -46,7 +46,7 @@ static void sh_clean_dirty_bitmap(struct domain *);
 
 /* Set up the shadow-specific parts of a domain struct at start of day.
  * Called for every domain from arch_domain_create() */
-int shadow_domain_init(struct domain *d, unsigned int domcr_flags)
+int shadow_domain_init(struct domain *d)
 {
     static const struct log_dirty_ops sh_ops = {
         .enable  = sh_enable_log_dirty,
@@ -62,7 +62,6 @@ int shadow_domain_init(struct domain *d, unsigned int domcr_flags)
 
 #if (SHADOW_OPTIMIZATIONS & SHOPT_OUT_OF_SYNC)
     d->arch.paging.shadow.oos_active = 0;
-    d->arch.paging.shadow.oos_off = domcr_flags & XEN_DOMCTL_CDF_oos_off;
 #endif
     d->arch.paging.shadow.pagetable_dying_op = 0;
 
@@ -2523,11 +2522,13 @@ static void sh_update_paging_modes(struct vcpu *v)
 #if (SHADOW_OPTIMIZATIONS & SHOPT_OUT_OF_SYNC)
     /* We need to check that all the vcpus have paging enabled to
      * unsync PTs. */
-    if ( is_hvm_domain(d) && !d->arch.paging.shadow.oos_off )
+    if ( !(d->options & XEN_DOMCTL_CDF_oos_off) )
     {
         int pe = 1;
         struct vcpu *vptr;
 
+        ASSERT(is_hvm_domain(d));
+
         for_each_vcpu(d, vptr)
         {
             if ( !hvm_paging_enabled(vptr) )
xen/arch/x86/mm/shadow/none.c +1 -1
index a70888bd98..2fddf4274c 100644
@@ -18,7 +18,7 @@ static void _clean_dirty_bitmap(struct domain *d)
     ASSERT(is_pv_domain(d));
 }
 
-int shadow_domain_init(struct domain *d, unsigned int domcr_flags)
+int shadow_domain_init(struct domain *d)
 {
     static const struct log_dirty_ops sh_none_ops = {
         .enable  = _enable_log_dirty,
xen/common/domain.c +12 -4
index 6109623730..95321482ef 100644
@@ -313,11 +313,19 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
-    if ( !(config->flags & XEN_DOMCTL_CDF_hvm_guest) &&
-         (config->flags & XEN_DOMCTL_CDF_hap) )
+    if ( !(config->flags & XEN_DOMCTL_CDF_hvm_guest) )
     {
-        dprintk(XENLOG_INFO, "HAP enabled for non-HVM guest\n");
-        return -EINVAL;
+        if ( config->flags & XEN_DOMCTL_CDF_hap )
+        {
+            dprintk(XENLOG_INFO, "HAP enabled for non-HVM guest\n");
+            return -EINVAL;
+        }
+
+        /*
+         * It is only meaningful for XEN_DOMCTL_CDF_oos_off to be clear
+         * for HVM guests.
+         */
+        config->flags |= XEN_DOMCTL_CDF_oos_off;
     }
 
     return arch_sanitise_domain_config(config);
xen/include/asm-x86/domain.h +0 -1
index 933b85901f..5f9899469c 100644
@@ -115,7 +115,6 @@ struct shadow_domain {
 
     /* OOS */
     bool_t oos_active;
-    bool_t oos_off;
 
     /* Has this domain ever used HVMOP_pagetable_dying? */
     bool_t pagetable_dying_op;
xen/include/asm-x86/shadow.h +1 -1
index f29f0f652b..8ebb89c027 100644
@@ -49,7 +49,7 @@
 
 /* Set up the shadow-specific parts of a domain struct at start of day.
  * Called from paging_domain_init(). */
-int shadow_domain_init(struct domain *d, unsigned int domcr_flags);
+int shadow_domain_init(struct domain *d);
 
 /* Setup the shadow-specific parts of a vcpu struct. It is called by
  * paging_vcpu_init() in paging.c */
Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#9 lore raw

PATCH v6 4/10 domain: remove the 'is_xenstore' flag

In reply to Paul Durrant

This patch introduces a convenience macro, is_xenstore_domain(), which tests the domain 'options' directly and then uses that in place of the 'is_xenstore' flag.

---
Cc: Andrew Cooper <[email protected]>
Cc: Ian Jackson <[email protected]>
Cc: Jan Beulich <[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]>
Cc: Daniel De Graaf <[email protected]>

v2:
- Set 'disable_migrate' to true rather 1

---
 xen/common/domain.c     | 9 +++------
 xen/common/domctl.c     | 2 +-
 xen/include/xen/sched.h | 7 +++++--
 xen/include/xsm/dummy.h | 2 +-
 4 files changed, 10 insertions(+), 10 deletions(-)
xen/common/domain.c +3 -6
index 95321482ef..76e6976617 100644
@@ -363,7 +363,7 @@ struct domain *domain_create(domid_t domid,
         if ( hardware_domid < 0 || hardware_domid >= DOMID_FIRST_RESERVED )
             panic("The value of hardware_dom must be a valid domain ID\n");
 
-        d->disable_migrate = 1;
+        d->disable_migrate = true;
         old_hwdom = hardware_domain;
         hardware_domain = d;
     }
@@ -442,11 +442,8 @@ struct domain *domain_create(domid_t domid,
         watchdog_domain_init(d);
         init_status |= INIT_watchdog;
 
-        if ( d->options & XEN_DOMCTL_CDF_xs_domain )
-        {
-            d->is_xenstore = 1;
-            d->disable_migrate = 1;
-        }
+        if ( is_xenstore_domain(d) )
+            d->disable_migrate = true;
 
         d->iomem_caps = rangeset_new(d, "I/O Memory", RANGESETF_prettyprint_hex);
         d->irq_caps   = rangeset_new(d, "Interrupts", 0);
xen/common/domctl.c +1 -1
index b48e408583..6e6e9b9866 100644
@@ -186,7 +186,7 @@ void getdomaininfo(struct domain *d, struct xen_domctl_getdomaininfo *info)
         (d->is_shut_down                ? XEN_DOMINF_shutdown  : 0) |
         (d->controller_pause_count > 0  ? XEN_DOMINF_paused    : 0) |
         (d->debugger_attached           ? XEN_DOMINF_debugged  : 0) |
-        (d->is_xenstore                 ? XEN_DOMINF_xs_domain : 0) |
+        (is_xenstore_domain(d)          ? XEN_DOMINF_xs_domain : 0) |
         (is_hvm_domain(d)               ? XEN_DOMINF_hvm_guest : 0) |
         d->shutdown_code << XEN_DOMINF_shutdownshift;
 
xen/include/xen/sched.h +5 -2
index 07a64947ed..df331a3bb0 100644
@@ -375,8 +375,6 @@ struct domain
     bool             is_privileged;
     /* Can this guest access the Xen console? */
     bool             is_console;
-    /* Is this a xenstore domain (not dom0)? */
-    bool             is_xenstore;
     /* Non-migratable and non-restoreable? */
     bool             disable_migrate;
     /* Is this guest being debugged by dom0? */
@@ -979,6 +977,11 @@ static inline bool is_vcpu_online(const struct vcpu *v)
     return !test_bit(_VPF_down, &v->pause_flags);
 }
 
+static inline bool is_xenstore_domain(const struct domain *d)
+{
+    return d->options & XEN_DOMCTL_CDF_xs_domain;
+}
+
 extern bool sched_smt_power_savings;
 
 extern enum cpufreq_controller {
xen/include/xsm/dummy.h +1 -1
index ef52bb1764..b8e185e6fa 100644
@@ -79,7 +79,7 @@ static always_inline int xsm_default_action(
         {
             return 0;
     case XSM_XS_PRIV:
-            if ( src->is_xenstore )
+            if ( is_xenstore_domain(src) )
                 return 0;
         }
         /* fall through */
Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v6 4/10
#10 lore raw

Re: [Xen-devel] [PATCH v6 04/10] domain: remove the 'is_xenstore' flag

Quoted 7 lines · Paul Durrant

This patch introduces a convenience macro, is_xenstore_domain(), which tests the domain 'options' directly and then uses that in place of the 'is_xenstore' flag.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#11 lore raw

PATCH v6 5/10 x86/domain: remove the 's3_integrity' flag

In reply to Paul Durrant

The flag is not needed since the domain 'options' can now be tested directly.

---
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>

v4:
- s/TBOOT/CONFIG_TBOOT/g

v3:
- Also sanitise the flag against CONFIG_TBOOT being set

---
 xen/arch/x86/domain.c        | 9 +++++++--
 xen/arch/x86/setup.c         | 2 +-
 xen/arch/x86/tboot.c         | 2 +-
 xen/include/asm-x86/domain.h | 2 --
 4 files changed, 9 insertions(+), 6 deletions(-)
xen/arch/x86/domain.c +7 -2
index bc0db03387..f144d8fe9a 100644
@@ -466,6 +466,13 @@ int arch_sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( (config->flags & XEN_DOMCTL_CDF_s3_integrity) &&
+         !IS_ENABLED(CONFIG_TBOOT) )
+    {
+        dprintk(XENLOG_INFO, "S3 integrity check not valid without CONFIG_TBOOT\n");
+        return -EINVAL;
+    }
+
     return 0;
 }
 
@@ -544,8 +551,6 @@ int arch_domain_create(struct domain *d,
                d->domain_id);
     }
 
-    d->arch.s3_integrity = config->flags & XEN_DOMCTL_CDF_s3_integrity;
-
     emflags = config->arch.emulation_flags;
 
     if ( is_hardware_domain(d) && is_pv_domain(d) )
xen/arch/x86/setup.c +1 -1
index 87fc7c90da..d0b35b0ce2 100644
@@ -696,7 +696,7 @@ void __init noreturn __start_xen(unsigned long mbi_p)
         .stop_bits = 1
     };
     struct xen_domctl_createdomain dom0_cfg = {
-        .flags = XEN_DOMCTL_CDF_s3_integrity,
+        .flags = IS_ENABLED(CONFIG_TBOOT) ? XEN_DOMCTL_CDF_s3_integrity : 0,
         .max_evtchn_port = -1,
         .max_grant_frames = opt_max_grant_frames,
         .max_maptrack_frames = opt_max_maptrack_frames,
xen/arch/x86/tboot.c +1 -1
index f3fdee4d39..3db8a8a8d8 100644
@@ -212,7 +212,7 @@ static void tboot_gen_domain_integrity(const uint8_t key[TB_KEY_SIZE],
     vmac_set_key((uint8_t *)key, &ctx);
     for_each_domain( d )
     {
-        if ( !d->arch.s3_integrity )
+        if ( !(d->options & XEN_DOMCTL_CDF_s3_integrity) )
             continue;
         printk("MACing Domain %u\n", d->domain_id);
 
xen/include/asm-x86/domain.h +0 -2
index 5f9899469c..5c038a1065 100644
@@ -295,8 +295,6 @@ struct arch_domain
     uint32_t pci_cf8;
     uint8_t cmos_idx;
 
-    bool_t s3_integrity;
-
     union {
         struct pv_domain pv;
         struct hvm_domain hvm;
Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#12 lore raw

PATCH v6 6/10 domain: introduce XEN_DOMCTL_CDF_iommu flag

In reply to Paul Durrant

This patch introduces a common domain creation flag to determine whether the domain is permitted to make use of the IOMMU. Currently the flag is always set (for both dom0 and domU) if the IOMMU is globally enabled (i.e. iommu_enabled == 1). sanitise_domain_config() is modified to reject the flag if !iommu_enabled.

A new helper function, is_iommu_enabled(), is added to test the flag and iommu_domain_init() will return immediately if !is_iommu_enabled(). This is slightly different to the previous behaviour based on !iommu_enabled where the call to arch_iommu_domain_init() was made regardless, however it appears that this call was only necessary to initialize the dt_devices list for ARM such that iommu_release_dt_devices() can be called unconditionally by domain_relinquish_resources(). Adding a simple check of is_iommu_enabled() into iommu_release_dt_devices() keeps this unconditional call working.

No functional change should be observed with this patch applied.

Subsequent patches will allow the toolstack to control whether use of the IOMMU is enabled for a domain.

NOTE: The introduction of the is_iommu_enabled() helper function might
seem excessive but its use is expected to increase with subsequent patches. Also, having iommu_domain_init() bail before calling arch_iommu_domain_init() is not strictly necessary, but I think the consequent addition of the call to is_iommu_enabled() in iommu_release_dt_devices() makes the code clearer.

---
Cc: Ian Jackson <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Anthony PERARD <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Volodymyr Babchuk <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v6:
- Remove the toolstack parts as there's no nice method of testing whether the IOMMU is enabled in an architecture-neutral way

v5:
- Move is_iommu_enabled() check into iommu_domain_init()
- Reject XEN_DOMCTL_CDF_iommu in sanitise_domain_config() if !iommu_enabled
- Use evaluate_nospec() in defintion of is_iommu_enabled()

---
 xen/arch/arm/setup.c                  | 3 +++
 xen/arch/x86/setup.c                  | 3 +++
 xen/common/domain.c                   | 9 ++++++++-
 xen/common/domctl.c                   | 8 ++++++++
 xen/drivers/passthrough/device_tree.c | 3 +++
 xen/drivers/passthrough/iommu.c       | 6 +++---
 xen/include/public/domctl.h           | 4 ++++
 xen/include/xen/sched.h               | 5 +++++
 8 files changed, 37 insertions(+), 4 deletions(-)
xen/arch/arm/setup.c +3 -0
index 2c5d1372c0..20021ee0ca 100644
@@ -914,6 +914,9 @@ void __init start_xen(unsigned long boot_phys_offset,
     dom0_cfg.arch.tee_type = tee_get_type();
     dom0_cfg.max_vcpus = dom0_max_vcpus();
 
+    if ( iommu_enabled )
+        dom0_cfg.flags |= XEN_DOMCTL_CDF_iommu;
+
     dom0 = domain_create(0, &dom0_cfg, true);
     if ( IS_ERR(dom0) || (alloc_dom0_vcpu0(dom0) == NULL) )
         panic("Error creating domain 0\n");
xen/arch/x86/setup.c +3 -0
index d0b35b0ce2..fa226a2bab 100644
@@ -1733,6 +1733,9 @@ void __init noreturn __start_xen(unsigned long mbi_p)
     }
     dom0_cfg.max_vcpus = dom0_max_vcpus();
 
+    if ( iommu_enabled )
+        dom0_cfg.flags |= XEN_DOMCTL_CDF_iommu;
+
     /* Create initial domain 0. */
     dom0 = domain_create(get_initial_domain_id(), &dom0_cfg, !pv_shim);
     if ( IS_ERR(dom0) || (alloc_dom0_vcpu0(dom0) == NULL) )
xen/common/domain.c +8 -1
index 76e6976617..e832a5c4aa 100644
@@ -301,7 +301,8 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
                            XEN_DOMCTL_CDF_hap |
                            XEN_DOMCTL_CDF_s3_integrity |
                            XEN_DOMCTL_CDF_oos_off |
-                           XEN_DOMCTL_CDF_xs_domain) )
+                           XEN_DOMCTL_CDF_xs_domain |
+                           XEN_DOMCTL_CDF_iommu) )
     {
         dprintk(XENLOG_INFO, "Unknown CDF flags %#x\n", config->flags);
         return -EINVAL;
@@ -328,6 +329,12 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         config->flags |= XEN_DOMCTL_CDF_oos_off;
     }
 
+    if ( (config->flags & XEN_DOMCTL_CDF_iommu) && !iommu_enabled )
+    {
+        dprintk(XENLOG_INFO, "IOMMU is not enabled\n");
+        return -EINVAL;
+    }
+
     return arch_sanitise_domain_config(config);
 }
 
xen/common/domctl.c +8 -0
index 6e6e9b9866..fddf20f1b5 100644
@@ -515,6 +515,14 @@ long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)
             rover = dom;
         }
 
+        /*
+         * For now, make sure the createdomain IOMMU flag is set if the
+         * IOMMU is enabled. When the flag comes under toolstack control
+         * this can go away.
+         */
+        if ( iommu_enabled )
+            op->u.createdomain.flags |= XEN_DOMCTL_CDF_iommu;
+
         d = domain_create(dom, &op->u.createdomain, false);
         if ( IS_ERR(d) )
         {
xen/drivers/passthrough/device_tree.c +3 -0
index b6eaae7283..d32b172664 100644
@@ -119,6 +119,9 @@ int iommu_release_dt_devices(struct domain *d)
     struct dt_device_node *dev, *_dev;
     int rc;
 
+    if ( !is_iommu_enabled(d) )
+        return 0;
+
     list_for_each_entry_safe(dev, _dev, &hd->dt_devices, domain_list)
     {
         rc = iommu_deassign_dt_device(d, dev);
xen/drivers/passthrough/iommu.c +3 -3
index 37eb0f7d01..e61d3d1368 100644
@@ -151,13 +151,13 @@ int iommu_domain_init(struct domain *d)
     struct domain_iommu *hd = dom_iommu(d);
     int ret = 0;
 
+    if ( !is_iommu_enabled(d) )
+        return 0;
+
     ret = arch_iommu_domain_init(d);
     if ( ret )
         return ret;
 
-    if ( !iommu_enabled )
-        return 0;
-
     hd->platform_ops = iommu_get_ops();
     return hd->platform_ops->init(d);
 }
xen/include/public/domctl.h +4 -0
index 19486d5e32..3f82c78870 100644
@@ -64,6 +64,10 @@ struct xen_domctl_createdomain {
  /* Is this a xenstore domain? */
 #define _XEN_DOMCTL_CDF_xs_domain     4
 #define XEN_DOMCTL_CDF_xs_domain      (1U<<_XEN_DOMCTL_CDF_xs_domain)
+ /* Should this domain be permitted to use the IOMMU? */
+#define _XEN_DOMCTL_CDF_iommu         5
+#define XEN_DOMCTL_CDF_iommu          (1U<<_XEN_DOMCTL_CDF_iommu)
+
     uint32_t flags;
 
     /*
xen/include/xen/sched.h +5 -0
index df331a3bb0..258924242f 100644
@@ -982,6 +982,11 @@ static inline bool is_xenstore_domain(const struct domain *d)
     return d->options & XEN_DOMCTL_CDF_xs_domain;
 }
 
+static inline bool is_iommu_enabled(const struct domain *d)
+{
+    return evaluate_nospec(d->options & XEN_DOMCTL_CDF_iommu);
+}
+
 extern bool sched_smt_power_savings;
 
 extern enum cpufreq_controller {
Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Reviewed-by on v6 6/10
#13 lore raw

Re: [Xen-devel] [PATCH v6 06/10] domain: introduce XEN_DOMCTL_CDF_iommu flag

Quoted 20 lines · Paul Durrant

This patch introduces a common domain creation flag to determine whether the domain is permitted to make use of the IOMMU. Currently the flag is always set (for both dom0 and domU) if the IOMMU is globally enabled (i.e. iommu_enabled == 1). sanitise_domain_config() is modified to reject the flag if !iommu_enabled.

A new helper function, is_iommu_enabled(), is added to test the flag and iommu_domain_init() will return immediately if !is_iommu_enabled(). This is slightly different to the previous behaviour based on !iommu_enabled where the call to arch_iommu_domain_init() was made regardless, however it appears that this call was only necessary to initialize the dt_devices list for ARM such that iommu_release_dt_devices() can be called unconditionally by domain_relinquish_resources(). Adding a simple check of is_iommu_enabled() into iommu_release_dt_devices() keeps this unconditional call working.

No functional change should be observed with this patch applied.

Subsequent patches will allow the toolstack to control whether use of the IOMMU is enabled for a domain.

Quoted 8 lines

NOTE: The introduction of the is_iommu_enabled() helper function might
seem excessive but its use is expected to increase with subsequent patches. Also, having iommu_domain_init() bail before calling arch_iommu_domain_init() is not strictly necessary, but I think the consequent addition of the call to is_iommu_enabled() in iommu_release_dt_devices() makes the code clearer.

I have one ARM-related question and one 'nice to have', but the code LGTM:

Quoted 95 lines

---
Cc: Ian Jackson <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Anthony PERARD <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Volodymyr Babchuk <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v6:
- Remove the toolstack parts as there's no nice method of testing whether the IOMMU is enabled in an architecture-neutral way

v5:
- Move is_iommu_enabled() check into iommu_domain_init()
- Reject XEN_DOMCTL_CDF_iommu in sanitise_domain_config() if !iommu_enabled
- Use evaluate_nospec() in defintion of is_iommu_enabled()

---
 xen/arch/arm/setup.c                  | 3 +++
 xen/arch/x86/setup.c                  | 3 +++
 xen/common/domain.c                   | 9 ++++++++-
 xen/common/domctl.c                   | 8 ++++++++
 xen/drivers/passthrough/device_tree.c | 3 +++
 xen/drivers/passthrough/iommu.c       | 6 +++---
 xen/include/public/domctl.h           | 4 ++++
 xen/include/xen/sched.h               | 5 +++++
 8 files changed, 37 insertions(+), 4 deletions(-)
xen/arch/arm/setup.c +3 -0
index 2c5d1372c0..20021ee0ca 100644
@@ -914,6 +914,9 @@ void __init start_xen(unsigned long boot_phys_offset,
     dom0_cfg.arch.tee_type = tee_get_type();
     dom0_cfg.max_vcpus = dom0_max_vcpus();
 
+    if ( iommu_enabled )
+        dom0_cfg.flags |= XEN_DOMCTL_CDF_iommu;
+
     dom0 = domain_create(0, &dom0_cfg, true);
     if ( IS_ERR(dom0) || (alloc_dom0_vcpu0(dom0) == NULL) )
         panic("Error creating domain 0\n");
xen/arch/x86/setup.c +3 -0
index d0b35b0ce2..fa226a2bab 100644
@@ -1733,6 +1733,9 @@ void __init noreturn __start_xen(unsigned long mbi_p)
     }
     dom0_cfg.max_vcpus = dom0_max_vcpus();
 
+    if ( iommu_enabled )
+        dom0_cfg.flags |= XEN_DOMCTL_CDF_iommu;
+
     /* Create initial domain 0. */
     dom0 = domain_create(get_initial_domain_id(), &dom0_cfg, !pv_shim);
     if ( IS_ERR(dom0) || (alloc_dom0_vcpu0(dom0) == NULL) )
xen/common/domain.c +8 -1
index 76e6976617..e832a5c4aa 100644
@@ -301,7 +301,8 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
                            XEN_DOMCTL_CDF_hap |
                            XEN_DOMCTL_CDF_s3_integrity |
                            XEN_DOMCTL_CDF_oos_off |
-                           XEN_DOMCTL_CDF_xs_domain) )
+                           XEN_DOMCTL_CDF_xs_domain |
+                           XEN_DOMCTL_CDF_iommu) )
     {
         dprintk(XENLOG_INFO, "Unknown CDF flags %#x\n", config->flags);
         return -EINVAL;
@@ -328,6 +329,12 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         config->flags |= XEN_DOMCTL_CDF_oos_off;
     }
 
+    if ( (config->flags & XEN_DOMCTL_CDF_iommu) && !iommu_enabled )
+    {
+        dprintk(XENLOG_INFO, "IOMMU is not enabled\n");
+        return -EINVAL;
+    }
+
     return arch_sanitise_domain_config(config);
 }
 
xen/common/domctl.c +0 -0
index 6e6e9b9866..fddf20f1b5 100644

xen/common/domctl.c long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl) · line 524

Quoted 10 lines
xen/common/domctl.c +7 -0
@@ -515,6 +515,14 @@ long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)
             rover = dom;
         }
 
+        /*
+         * For now, make sure the createdomain IOMMU flag is set if the
+         * IOMMU is enabled. When the flag comes under toolstack control
+         * this can go away.
+         */
+        if ( iommu_enabled )
+            op->u.createdomain.flags |= XEN_DOMCTL_CDF_iommu;

Can you add some kind of safety check here to make sure this bodge is removed when the toolstack takes control of the flag, ie:

BUG_ON(op->u.createdomain.flags & XEN_DOMCTL_CDF_iommu);

Or maybe an ASSERT_UNREACHABLE() followed by returning EINVAL?

Quoted 9 lines
xen/common/domctl.c +1 -0
+
         d = domain_create(dom, &op->u.createdomain, false);
         if ( IS_ERR(d) )
         {
xen/drivers/passthrough/device_tree.c +0 -0
index b6eaae7283..d32b172664 100644

xen/drivers/passthrough/device_tree.c int iommu_release_dt_devices(struct domain *d) · line 123

Quoted 5 lines
xen/drivers/passthrough/device_tree.c +2 -0
@@ -119,6 +119,9 @@ int iommu_release_dt_devices(struct domain *d)
     struct dt_device_node *dev, *_dev;
     int rc;
 
+    if ( !is_iommu_enabled(d) )
+        return 0;

How could you get here? If the domain doesn't have an iommu how did it got the devices assigned in the first place?

The hardware domain on ARM would be an exception here, since it uses an identity second stage translation, but I don't think iommu_release_dt_devices is ever used against the hardware domain.

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#14 lore raw

Re: [Xen-devel] [PATCH v6 06/10] domain: introduce XEN_DOMCTL_CDF_iommu flag

Quoted 155 lines

-----Original Message-----
From: Roger Pau Monne <[email protected]>
Sent: 23 August 2019 11:33
To: Paul Durrant <[email protected]>
Cc: [email protected]; Ian Jackson <[email protected]>; Wei Liu <[email protected]>; Anthony Perard <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Jan Beulich <[email protected]>; Julien Grall <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Stefano Stabellini <[email protected]>; Tim (Xen.org) <[email protected]>; Volodymyr Babchuk <[email protected]>
Subject: Re: [PATCH v6 06/10] domain: introduce XEN_DOMCTL_CDF_iommu flag

Quoted 20 lines · Paul Durrant

This patch introduces a common domain creation flag to determine whether the domain is permitted to make use of the IOMMU. Currently the flag is always set (for both dom0 and domU) if the IOMMU is globally enabled (i.e. iommu_enabled == 1). sanitise_domain_config() is modified to reject the flag if !iommu_enabled.

A new helper function, is_iommu_enabled(), is added to test the flag and iommu_domain_init() will return immediately if !is_iommu_enabled(). This is slightly different to the previous behaviour based on !iommu_enabled where the call to arch_iommu_domain_init() was made regardless, however it appears that this call was only necessary to initialize the dt_devices list for ARM such that iommu_release_dt_devices() can be called unconditionally by domain_relinquish_resources(). Adding a simple check of is_iommu_enabled() into iommu_release_dt_devices() keeps this unconditional call working.

No functional change should be observed with this patch applied.

Subsequent patches will allow the toolstack to control whether use of the IOMMU is enabled for a domain.

Quoted 8 lines

NOTE: The introduction of the is_iommu_enabled() helper function might
seem excessive but its use is expected to increase with subsequent patches. Also, having iommu_domain_init() bail before calling arch_iommu_domain_init() is not strictly necessary, but I think the consequent addition of the call to is_iommu_enabled() in iommu_release_dt_devices() makes the code clearer.

I have one ARM-related question and one 'nice to have', but the code LGTM:

Quoted 95 lines

---
Cc: Ian Jackson <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Anthony PERARD <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Volodymyr Babchuk <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v6:
- Remove the toolstack parts as there's no nice method of testing whether the IOMMU is enabled in an architecture-neutral way

v5:
- Move is_iommu_enabled() check into iommu_domain_init()
- Reject XEN_DOMCTL_CDF_iommu in sanitise_domain_config() if !iommu_enabled
- Use evaluate_nospec() in defintion of is_iommu_enabled()

---
 xen/arch/arm/setup.c                  | 3 +++
 xen/arch/x86/setup.c                  | 3 +++
 xen/common/domain.c                   | 9 ++++++++-
 xen/common/domctl.c                   | 8 ++++++++
 xen/drivers/passthrough/device_tree.c | 3 +++
 xen/drivers/passthrough/iommu.c       | 6 +++---
 xen/include/public/domctl.h           | 4 ++++
 xen/include/xen/sched.h               | 5 +++++
 8 files changed, 37 insertions(+), 4 deletions(-)
xen/arch/arm/setup.c +3 -0
index 2c5d1372c0..20021ee0ca 100644
@@ -914,6 +914,9 @@ void __init start_xen(unsigned long boot_phys_offset,
     dom0_cfg.arch.tee_type = tee_get_type();
     dom0_cfg.max_vcpus = dom0_max_vcpus();
 
+    if ( iommu_enabled )
+        dom0_cfg.flags |= XEN_DOMCTL_CDF_iommu;
+
     dom0 = domain_create(0, &dom0_cfg, true);
     if ( IS_ERR(dom0) || (alloc_dom0_vcpu0(dom0) == NULL) )
         panic("Error creating domain 0\n");
xen/arch/x86/setup.c +3 -0
index d0b35b0ce2..fa226a2bab 100644
@@ -1733,6 +1733,9 @@ void __init noreturn __start_xen(unsigned long mbi_p)
     }
     dom0_cfg.max_vcpus = dom0_max_vcpus();
 
+    if ( iommu_enabled )
+        dom0_cfg.flags |= XEN_DOMCTL_CDF_iommu;
+
     /* Create initial domain 0. */
     dom0 = domain_create(get_initial_domain_id(), &dom0_cfg, !pv_shim);
     if ( IS_ERR(dom0) || (alloc_dom0_vcpu0(dom0) == NULL) )
xen/common/domain.c +8 -1
index 76e6976617..e832a5c4aa 100644
@@ -301,7 +301,8 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
                            XEN_DOMCTL_CDF_hap |
                            XEN_DOMCTL_CDF_s3_integrity |
                            XEN_DOMCTL_CDF_oos_off |
-                           XEN_DOMCTL_CDF_xs_domain) )
+                           XEN_DOMCTL_CDF_xs_domain |
+                           XEN_DOMCTL_CDF_iommu) )
     {
         dprintk(XENLOG_INFO, "Unknown CDF flags %#x\n", config->flags);
         return -EINVAL;
@@ -328,6 +329,12 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         config->flags |= XEN_DOMCTL_CDF_oos_off;
     }
 
+    if ( (config->flags & XEN_DOMCTL_CDF_iommu) && !iommu_enabled )
+    {
+        dprintk(XENLOG_INFO, "IOMMU is not enabled\n");
+        return -EINVAL;
+    }
+
     return arch_sanitise_domain_config(config);
 }
 
xen/common/domctl.c +0 -0
index 6e6e9b9866..fddf20f1b5 100644

xen/common/domctl.c long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl) · line 524

Quoted 10 lines
xen/common/domctl.c +7 -0
@@ -515,6 +515,14 @@ long do_domctl(XEN_GUEST_HANDLE_PARAM(xen_domctl_t) u_domctl)
             rover = dom;
         }
 
+        /*
+         * For now, make sure the createdomain IOMMU flag is set if the
+         * IOMMU is enabled. When the flag comes under toolstack control
+         * this can go away.
+         */
+        if ( iommu_enabled )
+            op->u.createdomain.flags |= XEN_DOMCTL_CDF_iommu;

Can you add some kind of safety check here to make sure this bodge is removed when the toolstack takes control of the flag, ie:

Quoted 3 lines

BUG_ON(op->u.createdomain.flags & XEN_DOMCTL_CDF_iommu);

Or maybe an ASSERT_UNREACHABLE() followed by returning EINVAL?

Ok, the former is a bit severe so I think I'll go for the latter.

Quoted 13 lines
Quoted 13 lines
xen/common/domctl.c +1 -0
+
         d = domain_create(dom, &op->u.createdomain, false);
         if ( IS_ERR(d) )
         {
xen/drivers/passthrough/device_tree.c +0 -0
index b6eaae7283..d32b172664 100644
@@ -119,6 +119,9 @@ int iommu_release_dt_devices(struct domain *d)
     struct dt_device_node *dev, *_dev;
     int rc;
 
Quoted 6 lines

xen/drivers/passthrough/device_tree.c int iommu_release_dt_devices(struct domain *d) · line 123

Quoted 2 lines
xen/drivers/passthrough/device_tree.c +2 -0
+    if ( !is_iommu_enabled(d) )
+        return 0;

How could you get here? If the domain doesn't have an iommu how did it got the devices assigned in the first place?

As I say in the commit comment, iommu_release_dt_device() is called unconditionally from domain_relinquish_resources().

Quoted 4 lines

The hardware domain on ARM would be an exception here, since it uses an identity second stage translation, but I don't think iommu_release_dt_devices is ever used against the hardware domain.

No, it is only called from domain_relinquish_resources() so should not apply to the h/w domain.

Paul

Quoted 1 line

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#15 lore raw

PATCH v6 7/10 use is_iommu_enabled() where appropriate...

In reply to Paul Durrant

...rather than testing the global iommu_enabled flag and ops pointer.

Now that there is a per-domain flag indicating whether the domain is permitted to use the IOMMU (which determines whether the ops pointer will be set), many tests of the global iommu_enabled flag and ops pointer can be translated into tests of the per-domain flag. Some of the other tests of purely the global iommu_enabled flag can also be translated into tests of the per-domain flag.

NOTE: The comment in iommu_share_p2m_table() is also fixed; need_iommu()
disappeared some time ago. Also, whilst the style of the 'if' in flask_iommu_resource_use_perm() is fixed, I have not translated any instances of u32 into uint32_t to keep consistency. IMO such a translation would be better done globally for the source module in a separate patch.
The change in the initialization of the 'hd' variable in iommu_map() and iommu_unmap() is done to keep the PV shim build happy. Without this change it will fail to compile with errors of the form:

iommu.c:361:32: error: unused variable ‘hd’ [-Werror=unused-variable]

     const struct domain_iommu *hd = dom_iommu(d);
                                     ^~

---
Cc: Stefano Stabellini <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Volodymyr Babchuk <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>
Cc: Jun Nakajima <[email protected]>
Cc: Kevin Tian <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Suravee Suthikulpanit <[email protected]>
Cc: Brian Woods <[email protected]>
Cc: Daniel De Graaf <[email protected]>

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v5:
- Fix logic in ARM p2m_init()
- Make iommu_do_domctl() return -EOPNOTSUPP rather than -ENOSYS if the IOMMU is not enabled
- Fix test in pci_enable_acs()
- Fix test in flask_iommu_resource_use_perm()

---
 xen/arch/arm/p2m.c                        |  2 +-
 xen/arch/x86/dom0_build.c                 |  2 +-
 xen/arch/x86/domctl.c                     |  4 +--
 xen/arch/x86/hvm/hvm.c                    |  6 ++--
 xen/arch/x86/hvm/vioapic.c                |  2 +-
 xen/arch/x86/hvm/vmx/vmcs.c               |  2 +-
 xen/arch/x86/hvm/vmx/vmx.c                |  2 +-
 xen/arch/x86/mm/p2m-ept.c                 |  4 +--
 xen/drivers/passthrough/amd/iommu_guest.c |  2 +-
 xen/drivers/passthrough/device_tree.c     |  4 +--
 xen/drivers/passthrough/io.c              |  8 ++---
 xen/drivers/passthrough/iommu.c           | 39 ++++++++++++-----------
 xen/drivers/passthrough/pci.c             | 16 +++++-----
 xen/drivers/passthrough/vtd/iommu.c       |  2 +-
 xen/drivers/passthrough/vtd/x86/hvm.c     |  2 +-
 xen/drivers/passthrough/x86/iommu.c       |  2 +-
 xen/xsm/flask/hooks.c                     | 18 +++++------
 17 files changed, 59 insertions(+), 58 deletions(-)
xen/arch/arm/p2m.c +1 -1
index e28ea1c85a..7f1442932a 100644
@@ -1531,7 +1531,7 @@ int p2m_init(struct domain *d)
      * shared with the CPU, Xen has to make sure that the PT changes have
      * reached the memory
      */
-    p2m->clean_pte = iommu_enabled &&
+    p2m->clean_pte = is_iommu_enabled(d) &&
         !iommu_has_feature(d, IOMMU_FEAT_COHERENT_WALK);
 
     rc = p2m_alloc_table(d);
xen/arch/x86/dom0_build.c +1 -1
index c69570920c..d381784edd 100644
@@ -356,7 +356,7 @@ unsigned long __init dom0_compute_nr_pages(
         avail -= d->max_vcpus - 1;
 
     /* Reserve memory for iommu_dom0_init() (rough estimate). */
-    if ( iommu_enabled )
+    if ( is_iommu_enabled(d) )
     {
         unsigned int s;
 
xen/arch/x86/domctl.c +2 -2
index 2d45e5b8a8..be4b206068 100644
@@ -715,7 +715,7 @@ long arch_do_domctl(
             break;
 
         ret = -ESRCH;
-        if ( iommu_enabled )
+        if ( is_iommu_enabled(d) )
         {
             pcidevs_lock();
             ret = pt_irq_create_bind(d, bind);
@@ -744,7 +744,7 @@ long arch_do_domctl(
         if ( ret )
             break;
 
-        if ( iommu_enabled )
+        if ( is_iommu_enabled(d) )
         {
             pcidevs_lock();
             ret = pt_irq_destroy_bind(d, bind);
xen/arch/x86/hvm/hvm.c +3 -3
index 029eea3b85..172c860acc 100644
@@ -465,7 +465,7 @@ void hvm_migrate_timers(struct vcpu *v)
 
 void hvm_migrate_pirq(struct hvm_pirq_dpci *pirq_dpci, const struct vcpu *v)
 {
-    ASSERT(iommu_enabled &&
+    ASSERT(is_iommu_enabled(v->domain) &&
            (is_hardware_domain(v->domain) || hvm_domain_irq(v->domain)->dpci));
 
     if ( (pirq_dpci->flags & HVM_IRQ_DPCI_MACH_MSI) &&
@@ -496,7 +496,7 @@ void hvm_migrate_pirqs(struct vcpu *v)
 {
     struct domain *d = v->domain;
 
-    if ( !iommu_enabled || !hvm_domain_irq(d)->dpci )
+    if ( !is_iommu_enabled(d) || !hvm_domain_irq(d)->dpci )
        return;
 
     spin_lock(&d->event_lock);
@@ -2264,7 +2264,7 @@ int hvm_set_cr0(unsigned long value, bool may_defer)
     }
 
     if ( ((value ^ old_value) & X86_CR0_CD) &&
-         iommu_enabled && hvm_funcs.handle_cd &&
+         is_iommu_enabled(d) && hvm_funcs.handle_cd &&
          (!rangeset_is_empty(d->iomem_caps) ||
           !rangeset_is_empty(d->arch.ioport_caps) ||
           has_arch_pdevs(d)) )
xen/arch/x86/hvm/vioapic.c +1 -1
index 9c25f72b4d..9aeef32a14 100644
@@ -536,7 +536,7 @@ void vioapic_update_EOI(struct domain *d, u8 vector)
 
             ent->fields.remote_irr = 0;
 
-            if ( iommu_enabled )
+            if ( is_iommu_enabled(d) )
             {
                 spin_unlock(&d->arch.hvm.irq_lock);
                 hvm_dpci_eoi(d, vioapic->base_gsi + pin, ent);
xen/arch/x86/hvm/vmx/vmcs.c +1 -1
index 16f14abe8f..ed27e8def7 100644
@@ -1087,7 +1087,7 @@ static int construct_vmcs(struct vcpu *v)
         vmx_clear_msr_intercept(v, MSR_IA32_SYSENTER_CS, VMX_MSR_RW);
         vmx_clear_msr_intercept(v, MSR_IA32_SYSENTER_ESP, VMX_MSR_RW);
         vmx_clear_msr_intercept(v, MSR_IA32_SYSENTER_EIP, VMX_MSR_RW);
-        if ( paging_mode_hap(d) && (!iommu_enabled || iommu_snoop) )
+        if ( paging_mode_hap(d) && (!is_iommu_enabled(d) || iommu_snoop) )
             vmx_clear_msr_intercept(v, MSR_IA32_CR_PAT, VMX_MSR_RW);
         if ( (vmexit_ctl & VM_EXIT_CLEAR_BNDCFGS) &&
              (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS) )
xen/arch/x86/hvm/vmx/vmx.c +1 -1
index 0060310d74..3b3d5b6250 100644
@@ -1210,7 +1210,7 @@ static void vmx_handle_cd(struct vcpu *v, unsigned long value)
         {
             v->arch.hvm.cache_mode = NORMAL_CACHE_MODE;
             vmx_set_guest_pat(v, *pat);
-            if ( !iommu_enabled || iommu_snoop )
+            if ( !is_iommu_enabled(v->domain) || iommu_snoop )
                 vmx_clear_msr_intercept(v, MSR_IA32_CR_PAT, VMX_MSR_RW);
             hvm_asid_flush_vcpu(v); /* no need to flush cache */
         }
xen/arch/x86/mm/p2m-ept.c +2 -2
index 6b8468c793..93d031cc6c 100644
@@ -260,7 +260,7 @@ static bool_t ept_split_super_page(struct p2m_domain *p2m,
         *epte = *ept_entry;
         epte->sp = (level > 1);
         epte->mfn += i * trunk;
-        epte->snp = (iommu_enabled && iommu_snoop);
+        epte->snp = is_iommu_enabled(p2m->domain) && iommu_snoop;
         epte->suppress_ve = 1;
 
         ept_p2m_type_to_flags(p2m, epte, epte->sa_p2mt, epte->access);
@@ -766,7 +766,7 @@ ept_set_entry(struct p2m_domain *p2m, gfn_t gfn_, mfn_t mfn,
         new_entry.sp = !!i;
         new_entry.sa_p2mt = p2mt;
         new_entry.access = p2ma;
-        new_entry.snp = (iommu_enabled && iommu_snoop);
+        new_entry.snp = is_iommu_enabled(d) && iommu_snoop;
 
         /* the caller should take care of the previous page */
         new_entry.mfn = mfn_x(mfn);
xen/drivers/passthrough/amd/iommu_guest.c +1 -1
index 7f2dd662af..1f2bcfbe15 100644
@@ -821,7 +821,7 @@ int guest_iommu_init(struct domain* d)
     struct guest_iommu *iommu;
     struct domain_iommu *hd = dom_iommu(d);
 
-    if ( !is_hvm_domain(d) || !iommu_enabled || !iommuv2_enabled ||
+    if ( !is_hvm_domain(d) || !is_iommu_enabled(d) || !iommuv2_enabled ||
          !has_viommu(d) )
         return 0;
 
xen/drivers/passthrough/device_tree.c +2 -2
index d32b172664..12f2c4c3f2 100644
@@ -29,7 +29,7 @@ int iommu_assign_dt_device(struct domain *d, struct dt_device_node *dev)
     int rc = -EBUSY;
     struct domain_iommu *hd = dom_iommu(d);
 
-    if ( !iommu_enabled || !hd->platform_ops )
+    if ( !is_iommu_enabled(d) )
         return -EINVAL;
 
     if ( !dt_device_is_protected(dev) )
@@ -71,7 +71,7 @@ int iommu_deassign_dt_device(struct domain *d, struct dt_device_node *dev)
     const struct domain_iommu *hd = dom_iommu(d);
     int rc;
 
-    if ( !iommu_enabled || !hd->platform_ops )
+    if ( !is_iommu_enabled(d) )
         return -EINVAL;
 
     if ( !dt_device_is_protected(dev) )
xen/drivers/passthrough/io.c +4 -4
index 4290c7c710..b292e79382 100644
@@ -434,7 +434,7 @@ int pt_irq_create_bind(
             if ( vcpu )
                 pirq_dpci->gmsi.posted = true;
         }
-        if ( vcpu && iommu_enabled )
+        if ( vcpu && is_iommu_enabled(d) )
             hvm_migrate_pirq(pirq_dpci, vcpu);
 
         /* Use interrupt posting if it is supported. */
@@ -817,7 +817,7 @@ int hvm_do_IRQ_dpci(struct domain *d, struct pirq *pirq)
 
     ASSERT(is_hvm_domain(d));
 
-    if ( !iommu_enabled || (!is_hardware_domain(d) && !dpci) ||
+    if ( !is_iommu_enabled(d) || (!is_hardware_domain(d) && !dpci) ||
          !pirq_dpci || !(pirq_dpci->flags & HVM_IRQ_DPCI_MAPPED) )
         return 0;
 
@@ -869,7 +869,7 @@ static int _hvm_dpci_msi_eoi(struct domain *d,
 
 void hvm_dpci_msi_eoi(struct domain *d, int vector)
 {
-    if ( !iommu_enabled ||
+    if ( !is_iommu_enabled(d) ||
          (!hvm_domain_irq(d)->dpci && !is_hardware_domain(d)) )
        return;
 
@@ -1001,7 +1001,7 @@ void hvm_dpci_eoi(struct domain *d, unsigned int guest_gsi,
     const struct hvm_irq_dpci *hvm_irq_dpci;
     const struct hvm_girq_dpci_mapping *girq;
 
-    if ( !iommu_enabled )
+    if ( !is_iommu_enabled(d) )
         return;
 
     if ( is_hardware_domain(d) )
xen/drivers/passthrough/iommu.c +20 -19
index e61d3d1368..27c75e0786 100644
@@ -179,7 +179,7 @@ void __hwdom_init iommu_hwdom_init(struct domain *d)
 
     check_hwdom_reqs(d);
 
-    if ( !iommu_enabled )
+    if ( !is_iommu_enabled(d) )
         return;
 
     register_keyhandler('o', &iommu_dump_p2m_table, "dump iommu p2m table", 0);
@@ -284,7 +284,7 @@ int iommu_construct(struct domain *d)
 
 void iommu_domain_destroy(struct domain *d)
 {
-    if ( !iommu_enabled || !dom_iommu(d)->platform_ops )
+    if ( !is_iommu_enabled(d) )
         return;
 
     iommu_teardown(d);
@@ -296,16 +296,18 @@ int iommu_map(struct domain *d, dfn_t dfn, mfn_t mfn,
               unsigned int page_order, unsigned int flags,
               unsigned int *flush_flags)
 {
-    const struct domain_iommu *hd = dom_iommu(d);
+    const struct domain_iommu *hd;
     unsigned long i;
     int rc = 0;
 
-    if ( !iommu_enabled || !hd->platform_ops )
+    if ( !is_iommu_enabled(d) )
         return 0;
 
     ASSERT(IS_ALIGNED(dfn_x(dfn), (1ul << page_order)));
     ASSERT(IS_ALIGNED(mfn_x(mfn), (1ul << page_order)));
 
+    hd = dom_iommu(d);
+
     for ( i = 0; i < (1ul << page_order); i++ )
     {
         rc = iommu_call(hd->platform_ops, map_page, d, dfn_add(dfn, i),
@@ -356,15 +358,17 @@ int iommu_legacy_map(struct domain *d, dfn_t dfn, mfn_t mfn,
 int iommu_unmap(struct domain *d, dfn_t dfn, unsigned int page_order,
                 unsigned int *flush_flags)
 {
-    const struct domain_iommu *hd = dom_iommu(d);
+    const struct domain_iommu *hd;
     unsigned long i;
     int rc = 0;
 
-    if ( !iommu_enabled || !hd->platform_ops )
+    if ( !is_iommu_enabled(d) )
         return 0;
 
     ASSERT(IS_ALIGNED(dfn_x(dfn), (1ul << page_order)));
 
+    hd = dom_iommu(d);
+
     for ( i = 0; i < (1ul << page_order); i++ )
     {
         int err = iommu_call(hd->platform_ops, unmap_page, d, dfn_add(dfn, i),
@@ -413,7 +417,7 @@ int iommu_lookup_page(struct domain *d, dfn_t dfn, mfn_t *mfn,
 {
     const struct domain_iommu *hd = dom_iommu(d);
 
-    if ( !iommu_enabled || !hd->platform_ops || !hd->platform_ops->lookup_page )
+    if ( !is_iommu_enabled(d) || !hd->platform_ops->lookup_page )
         return -EOPNOTSUPP;
 
     return iommu_call(hd->platform_ops, lookup_page, d, dfn, mfn, flags);
@@ -442,8 +446,8 @@ int iommu_iotlb_flush(struct domain *d, dfn_t dfn, unsigned int page_count,
     const struct domain_iommu *hd = dom_iommu(d);
     int rc;
 
-    if ( !iommu_enabled || !hd->platform_ops ||
-         !hd->platform_ops->iotlb_flush || !page_count || !flush_flags )
+    if ( !is_iommu_enabled(d) || !hd->platform_ops->iotlb_flush ||
+         !page_count || !flush_flags )
         return 0;
 
     if ( dfn_eq(dfn, INVALID_DFN) )
@@ -470,8 +474,8 @@ int iommu_iotlb_flush_all(struct domain *d, unsigned int flush_flags)
     const struct domain_iommu *hd = dom_iommu(d);
     int rc;
 
-    if ( !iommu_enabled || !hd->platform_ops ||
-         !hd->platform_ops->iotlb_flush_all || !flush_flags )
+    if ( !is_iommu_enabled(d) || !hd->platform_ops->iotlb_flush_all ||
+         !flush_flags )
         return 0;
 
     /*
@@ -556,8 +560,8 @@ int iommu_do_domctl(
 {
     int ret = -ENODEV;
 
-    if ( !iommu_enabled )
-        return -ENOSYS;
+    if ( !is_iommu_enabled(d) )
+        return -EOPNOTSUPP;
 
 #ifdef CONFIG_HAS_PCI
     ret = iommu_do_pci_domctl(domctl, d, u_domctl);
@@ -576,9 +580,9 @@ void iommu_share_p2m_table(struct domain* d)
     ASSERT(hap_enabled(d));
     /*
      * iommu_use_hap_pt(d) cannot be used here because during domain
-     * construction need_iommu(d) will always return false here.
+     * construction has_iommu_pt(d) will always return false here.
      */
-    if ( iommu_enabled && iommu_hap_pt_share )
+    if ( is_iommu_enabled(d) && iommu_hap_pt_share )
         iommu_get_ops()->share_p2m(d);
 }
 
@@ -608,10 +612,7 @@ int iommu_get_reserved_device_memory(iommu_grdm_t *func, void *ctxt)
 
 bool_t iommu_has_feature(struct domain *d, enum iommu_feature feature)
 {
-    if ( !iommu_enabled )
-        return 0;
-
-    return test_bit(feature, dom_iommu(d)->features);
+    return is_iommu_enabled(d) && test_bit(feature, dom_iommu(d)->features);
 }
 
 static void iommu_dump_p2m_table(unsigned char key)
xen/drivers/passthrough/pci.c +8 -8
index 89536cc067..dee47fe08b 100644
@@ -595,7 +595,7 @@ static void pci_enable_acs(struct pci_dev *pdev)
     u16 cap, ctrl, seg = pdev->seg;
     u8 bus = pdev->bus;
 
-    if ( !iommu_enabled )
+    if ( !is_iommu_enabled(pdev->domain) )
         return;
 
     pos = pci_find_ext_capability(seg, bus, pdev->devfn, PCI_EXT_CAP_ID_ACS);
@@ -864,7 +864,7 @@ static int pci_clean_dpci_irqs(struct domain *d)
 {
     struct hvm_irq_dpci *hvm_irq_dpci = NULL;
 
-    if ( !iommu_enabled )
+    if ( !is_iommu_enabled(d) )
         return 0;
 
     if ( !is_hvm_domain(d) )
@@ -897,7 +897,7 @@ static int deassign_device(struct domain *d, uint16_t seg, uint8_t bus,
     struct pci_dev *pdev;
     int ret = 0;
 
-    if ( !iommu_enabled || !hd->platform_ops )
+    if ( !is_iommu_enabled(d) )
         return -EINVAL;
 
     ASSERT(pcidevs_locked());
@@ -1383,7 +1383,7 @@ static int iommu_add_device(struct pci_dev *pdev)
     ASSERT(pcidevs_locked());
 
     hd = dom_iommu(pdev->domain);
-    if ( !iommu_enabled || !hd->platform_ops )
+    if ( !is_iommu_enabled(pdev->domain) )
         return 0;
 
     rc = hd->platform_ops->add_device(pdev->devfn, pci_to_dev(pdev));
@@ -1412,7 +1412,7 @@ static int iommu_enable_device(struct pci_dev *pdev)
     ASSERT(pcidevs_locked());
 
     hd = dom_iommu(pdev->domain);
-    if ( !iommu_enabled || !hd->platform_ops ||
+    if ( !is_iommu_enabled(pdev->domain) ||
          !hd->platform_ops->enable_device )
         return 0;
 
@@ -1428,7 +1428,7 @@ static int iommu_remove_device(struct pci_dev *pdev)
         return -EINVAL;
 
     hd = dom_iommu(pdev->domain);
-    if ( !iommu_enabled || !hd->platform_ops )
+    if ( !is_iommu_enabled(pdev->domain) )
         return 0;
 
     for ( devfn = pdev->devfn ; pdev->phantom_stride; )
@@ -1471,7 +1471,7 @@ static int assign_device(struct domain *d, u16 seg, u8 bus, u8 devfn, u32 flag)
     struct pci_dev *pdev;
     int rc = 0;
 
-    if ( !iommu_enabled || !hd->platform_ops )
+    if ( !is_iommu_enabled(d) )
         return 0;
 
     /* Prevent device assign if mem paging or mem sharing have been 
@@ -1537,7 +1537,7 @@ static int iommu_get_device_group(
     int i = 0;
     const struct iommu_ops *ops = hd->platform_ops;
 
-    if ( !iommu_enabled || !ops || !ops->get_device_group_id )
+    if ( !is_iommu_enabled(d) || !ops->get_device_group_id )
         return 0;
 
     group_id = ops->get_device_group_id(seg, bus, devfn);
xen/drivers/passthrough/vtd/iommu.c +1 -1
index 5d72270c5b..01f0bc4689 100644
@@ -1757,7 +1757,7 @@ static void iommu_domain_teardown(struct domain *d)
         xfree(mrmrr);
     }
 
-    ASSERT(iommu_enabled);
+    ASSERT(is_iommu_enabled(d));
 
     /*
      * We can't use iommu_use_hap_pt here because either IOMMU state
xen/drivers/passthrough/vtd/x86/hvm.c +1 -1
index 6675dca027..f77b35815c 100644
@@ -51,7 +51,7 @@ void hvm_dpci_isairq_eoi(struct domain *d, unsigned int isairq)
     struct hvm_irq_dpci *dpci = NULL;
 
     ASSERT(isairq < NR_ISAIRQS);
-    if ( !iommu_enabled )
+    if ( !is_iommu_enabled(d) )
         return;
 
     spin_lock(&d->event_lock);
xen/drivers/passthrough/x86/iommu.c +1 -1
index fd05075bb5..9879558c17 100644
@@ -178,7 +178,7 @@ int arch_iommu_populate_page_table(struct domain *d)
 
 void __hwdom_init arch_iommu_check_autotranslated_hwdom(struct domain *d)
 {
-    if ( !iommu_enabled )
+    if ( !is_iommu_enabled(d) )
         panic("Presently, iommu must be enabled for PVH hardware domain\n");
 }
 
xen/xsm/flask/hooks.c +9 -9
index 791c1f66af..5a0f2e723e 100644
@@ -886,7 +886,7 @@ static int flask_map_domain_msi (struct domain *d, int irq, const void *data,
 #endif
 }
 
-static u32 flask_iommu_resource_use_perm(void)
+static u32 flask_iommu_resource_use_perm(struct domain *d)
 {
     /* Obtain the permission level required for allowing a domain
      * to use an assigned device.
@@ -899,7 +899,7 @@ static u32 flask_iommu_resource_use_perm(void)
      */
     u32 perm = RESOURCE__USE_NOIOMMU;
 
-    if (iommu_enabled)
+    if ( is_iommu_enabled(d) )
         perm = ( iommu_intremap ? RESOURCE__USE_IOMMU :
                                   RESOURCE__USE_IOMMU_NOINTREMAP );
     return perm;
@@ -910,7 +910,7 @@ static int flask_map_domain_irq (struct domain *d, int irq, const void *data)
     u32 sid, dsid;
     int rc = -EPERM;
     struct avc_audit_data ad;
-    u32 dperm = flask_iommu_resource_use_perm();
+    u32 dperm = flask_iommu_resource_use_perm(d);
 
     if ( irq >= nr_static_irqs && data ) {
         rc = flask_map_domain_msi(d, irq, data, &sid, &ad);
@@ -976,7 +976,7 @@ static int flask_bind_pt_irq (struct domain *d, struct xen_domctl_bind_pt_irq *b
     int rc = -EPERM;
     int irq;
     struct avc_audit_data ad;
-    u32 dperm = flask_iommu_resource_use_perm();
+    u32 dperm = flask_iommu_resource_use_perm(d);
 
     rc = current_has_perm(d, SECCLASS_RESOURCE, RESOURCE__ADD);
     if ( rc )
@@ -1049,7 +1049,7 @@ static int flask_iomem_permission(struct domain *d, uint64_t start, uint64_t end
 
     data.ssid = domain_sid(current->domain);
     data.dsid = domain_sid(d);
-    data.use_perm = flask_iommu_resource_use_perm();
+    data.use_perm = flask_iommu_resource_use_perm(d);
 
     return security_iterate_iomem_sids(start, end, _iomem_has_perm, &data);
 }
@@ -1074,7 +1074,7 @@ static int flask_pci_config_permission(struct domain *d, uint32_t machine_bdf, u
     if ( access && (end >= 0x10 && start < 0x28) )
         perm = RESOURCE__SETUP;
     else
-        perm = flask_iommu_resource_use_perm();
+        perm = flask_iommu_resource_use_perm(d);
 
     AVC_AUDIT_DATA_INIT(&ad, DEV);
     ad.device = (unsigned long) machine_bdf;
@@ -1299,7 +1299,7 @@ static int flask_assign_device(struct domain *d, uint32_t machine_bdf)
     u32 dsid, rsid;
     int rc = -EPERM;
     struct avc_audit_data ad;
-    u32 dperm = flask_iommu_resource_use_perm();
+    u32 dperm = flask_iommu_resource_use_perm(d);
 
     if ( !d )
         return flask_test_assign_device(machine_bdf);
@@ -1358,7 +1358,7 @@ static int flask_assign_dtdevice(struct domain *d, const char *dtpath)
     u32 dsid, rsid;
     int rc = -EPERM;
     struct avc_audit_data ad;
-    u32 dperm = flask_iommu_resource_use_perm();
+    u32 dperm = flask_iommu_resource_use_perm(d);
 
     if ( !d )
         return flask_test_assign_dtdevice(dtpath);
@@ -1543,7 +1543,7 @@ static int flask_ioport_permission(struct domain *d, uint32_t start, uint32_t en
 
     data.ssid = domain_sid(current->domain);
     data.dsid = domain_sid(d);
-    data.use_perm = flask_iommu_resource_use_perm();
+    data.use_perm = flask_iommu_resource_use_perm(d);
 
     return security_iterate_ioport_sids(start, end, _ioport_has_perm, &data);
 }
Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v6 7/10
#16 lore raw

Re: [Xen-devel] [PATCH v6 07/10] use is_iommu_enabled() where appropriate...

Quoted 19 lines · Paul Durrant

...rather than testing the global iommu_enabled flag and ops pointer.

Now that there is a per-domain flag indicating whether the domain is permitted to use the IOMMU (which determines whether the ops pointer will be set), many tests of the global iommu_enabled flag and ops pointer can be translated into tests of the per-domain flag. Some of the other tests of purely the global iommu_enabled flag can also be translated into tests of the per-domain flag.

NOTE: The comment in iommu_share_p2m_table() is also fixed; need_iommu()
disappeared some time ago. Also, whilst the style of the 'if' in flask_iommu_resource_use_perm() is fixed, I have not translated any instances of u32 into uint32_t to keep consistency. IMO such a translation would be better done globally for the source module in a separate patch.
The change in the initialization of the 'hd' variable in iommu_map() and iommu_unmap() is done to keep the PV shim build happy. Without this change it will fail to compile with errors of the form:

Quoted 5 lines

iommu.c:361:32: error: unused variable ‘hd’ [-Werror=unused-variable]

      const struct domain_iommu *hd = dom_iommu(d);
                                      ^~

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

Reviewed-by on v6 7/10
#17 lore raw

Re: [Xen-devel] [PATCH v6 07/10] use is_iommu_enabled() where appropriate...

In reply to Paul Durrant
Quoted 23 lines

From: Paul Durrant [mailto:[email protected]]
Sent: Saturday, August 17, 2019 1:20 AM

...rather than testing the global iommu_enabled flag and ops pointer.

Now that there is a per-domain flag indicating whether the domain is permitted to use the IOMMU (which determines whether the ops pointer will be set), many tests of the global iommu_enabled flag and ops pointer can be translated into tests of the per-domain flag. Some of the other tests of purely the global iommu_enabled flag can also be translated into tests of the per-domain flag.

NOTE: The comment in iommu_share_p2m_table() is also fixed; need_iommu()
disappeared some time ago. Also, whilst the style of the 'if' in flask_iommu_resource_use_perm() is fixed, I have not translated any instances of u32 into uint32_t to keep consistency. IMO such a translation would be better done globally for the source module in a separate patch.
The change in the initialization of the 'hd' variable in iommu_map() and iommu_unmap() is done to keep the PV shim build happy. Without this change it will fail to compile with errors of the form:

Quoted 5 lines

iommu.c:361:32: error: unused variable ‘hd’ [-Werror=unused-variable]

     const struct domain_iommu *hd = dom_iommu(d);
                                     ^~

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

Reviewed-by on v6 7/10
#18 lore raw

Re: [Xen-devel] [PATCH v6 07/10] use is_iommu_enabled() where appropriate...

In reply to Paul Durrant
Quoted 9 lines · Paul Durrant

...rather than testing the global iommu_enabled flag and ops pointer.

Now that there is a per-domain flag indicating whether the domain is permitted to use the IOMMU (which determines whether the ops pointer will be set), many tests of the global iommu_enabled flag and ops pointer can be translated into tests of the per-domain flag. Some of the other tests of purely the global iommu_enabled flag can also be translated into tests of the per-domain flag.

Quoted 6 lines

NOTE: The comment in iommu_share_p2m_table() is also fixed; need_iommu()
disappeared some time ago. Also, whilst the style of the 'if' in flask_iommu_resource_use_perm() is fixed, I have not translated any instances of u32 into uint32_t to keep consistency. IMO such a translation would be better done globally for the source module in a separate patch.

I've usually changed those instances as the line gets modified anyway, but I'm not going to ask everyone to do this if they don't feel like it.

The problem with doing wholesale changes of u32 -> uint32_t is that you introduce a lot of noise when trying to use git blame afterwards for example. Doing it when touching the line for legitimate changes avoids the noise.

Quoted 9 lines

The change in the initialization of the 'hd' variable in iommu_map() and iommu_unmap() is done to keep the PV shim build happy. Without this change it will fail to compile with errors of the form:

iommu.c:361:32: error: unused variable ‘hd’ [-Werror=unused-variable]

     const struct domain_iommu *hd = dom_iommu(d);
                                     ^~

I haven't looked however if there are further cases where iommu_enabled should be changed into is_iommu_enabled.

xen/drivers/passthrough/iommu.c int iommu_do_domctl( · line 564

Quoted 8 lines
xen/drivers/passthrough/iommu.c +2 -2
@@ -556,8 +560,8 @@ int iommu_do_domctl(
 {
     int ret = -ENODEV;
 
-    if ( !iommu_enabled )
-        return -ENOSYS;
+    if ( !is_iommu_enabled(d) )
+        return -EOPNOTSUPP;

I would use ENOENT here, but I don't have a strong opinion. The hypercall is supported, it's just that there's no iommu for this domain.

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#19 lore raw

Re: [Xen-devel] [PATCH v6 07/10] use is_iommu_enabled() where appropriate...

Quoted 42 lines

-----Original Message-----
From: Roger Pau Monne <[email protected]>
Sent: 23 August 2019 11:56
To: Paul Durrant <[email protected]>
Cc: [email protected]; Stefano Stabellini <[email protected]>; Julien Grall <[email protected]>; Volodymyr Babchuk <[email protected]>; Jan Beulich <[email protected]>; Andrew Cooper <[email protected]>; Wei Liu <[email protected]>; Jun Nakajima <[email protected]>; Kevin Tian <[email protected]>; George Dunlap <[email protected]>; Suravee Suthikulpanit <[email protected]>; Brian Woods <[email protected]>; Daniel De Graaf <[email protected]>
Subject: Re: [PATCH v6 07/10] use is_iommu_enabled() where appropriate...

Quoted 9 lines · Paul Durrant

...rather than testing the global iommu_enabled flag and ops pointer.

Now that there is a per-domain flag indicating whether the domain is permitted to use the IOMMU (which determines whether the ops pointer will be set), many tests of the global iommu_enabled flag and ops pointer can be translated into tests of the per-domain flag. Some of the other tests of purely the global iommu_enabled flag can also be translated into tests of the per-domain flag.

Quoted 6 lines

NOTE: The comment in iommu_share_p2m_table() is also fixed; need_iommu()
disappeared some time ago. Also, whilst the style of the 'if' in flask_iommu_resource_use_perm() is fixed, I have not translated any instances of u32 into uint32_t to keep consistency. IMO such a translation would be better done globally for the source module in a separate patch.

I've usually changed those instances as the line gets modified anyway, but I'm not going to ask everyone to do this if they don't feel like it.

The problem with doing wholesale changes of u32 -> uint32_t is that you introduce a lot of noise when trying to use git blame afterwards for example. Doing it when touching the line for legitimate changes avoids the noise.

Quoted 4 lines

The change in the initialization of the 'hd' variable in iommu_map() and iommu_unmap() is done to keep the PV shim build happy. Without this change it will fail to compile with errors of the form:

Quoted 8 lines
Quoted 5 lines

iommu.c:361:32: error: unused variable ‘hd’ [-Werror=unused-variable]

     const struct domain_iommu *hd = dom_iommu(d);
                                     ^~

Thanks.

Quoted 3 lines

I haven't looked however if there are further cases where iommu_enabled should be changed into is_iommu_enabled.

Jan had clearly checked on a previous review iteration because he spotted a couple of places I had missed. I'm reasonably confident I've found them all now.

Paul

Quoted 13 lines

xen/drivers/passthrough/iommu.c int iommu_do_domctl( · line 564

Quoted 8 lines
xen/drivers/passthrough/iommu.c +2 -2
@@ -556,8 +560,8 @@ int iommu_do_domctl(
 {
     int ret = -ENODEV;
 
-    if ( !iommu_enabled )
-        return -ENOSYS;
+    if ( !is_iommu_enabled(d) )
+        return -EOPNOTSUPP;

I would use ENOENT here, but I don't have a strong opinion. The hypercall is supported, it's just that there's no iommu for this domain.

Quoted 1 line

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

Reviewed-by on v6 7/10
#20 lore raw

Re: [Xen-devel] [PATCH v6 07/10] use is_iommu_enabled() where appropriate...

In reply to Paul Durrant
Quoted 19 lines · Paul Durrant

...rather than testing the global iommu_enabled flag and ops pointer.

Now that there is a per-domain flag indicating whether the domain is permitted to use the IOMMU (which determines whether the ops pointer will be set), many tests of the global iommu_enabled flag and ops pointer can be translated into tests of the per-domain flag. Some of the other tests of purely the global iommu_enabled flag can also be translated into tests of the per-domain flag.

NOTE: The comment in iommu_share_p2m_table() is also fixed; need_iommu()
disappeared some time ago. Also, whilst the style of the 'if' in flask_iommu_resource_use_perm() is fixed, I have not translated any instances of u32 into uint32_t to keep consistency. IMO such a translation would be better done globally for the source module in a separate patch.
The change in the initialization of the 'hd' variable in iommu_map() and iommu_unmap() is done to keep the PV shim build happy. Without this change it will fail to compile with errors of the form:

Quoted 3 lines

iommu.c:361:32: error: unused variable ‘hd’ [-Werror=unused-variable]

     const struct domain_iommu *hd = dom_iommu(d);
                                     ^~

Why would that be? Afaict there's no short-circuiting of is_iommu_enabled() (albeit there could be, as we don't mean the shim to have/use an IOMMU). Oh - I guess I see it: Instead of this change what I think we want is for x86's iommu_call() to evaluate its 1st argument. Otherwise we're liable to run into the same issue elsewhere again.

xen/xsm/flask/hooks.c static int flask_map_domain_msi (struct domain *d, int irq, const void *data, · line 889

Quoted 8 lines
xen/xsm/flask/hooks.c +1 -1
@@ -886,7 +886,7 @@ static int flask_map_domain_msi (struct domain *d, int irq, const void *data,
 #endif
 }
 
-static u32 flask_iommu_resource_use_perm(void)
+static u32 flask_iommu_resource_use_perm(struct domain *d)

const?

With these adjustments

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#21 lore raw

PATCH v6 8/10 remove late (on-demand) construction of IOMMU page tables

In reply to Paul Durrant

Now that there is a per-domain IOMMU enable flag, which should be enabled if any device is going to be passed through, stop deferring page table construction until the assignment is done. Also don't tear down the tables again when the last device is de-assigned; defer that task until domain destruction.

This allows the has_iommu_pt() helper and iommu_status enumeration to be removed. Calls to has_iommu_pt() are simply replaced by calls to is_iommu_enabled(). Remaining open-code tests of iommu_hap_pt_share can also be replaced by calls to iommu_use_hap_pt().
The arch_iommu_populate_page_table() and iommu_construct() functions become redundant, as does the 'strict mode' dom0 page_list mapping code in iommu_hwdom_init(), and iommu_teardown() can be made static is its only remaining caller, iommu_domain_destroy(), is within the same source module.

All in all, about 220 lines of code are removed.

NOTE: This patch will cause a small amount of extra resource to be used
to accommodate IOMMU page tables that may never be used, since the per-domain IOMMU flag enable flag is currently set to the value of the global iommu_enable flag. A subsequent patch will add an option to the toolstack to allow it to be turned off if there is no intention to assign passthrough hardware to the domain.

---
Cc: Stefano Stabellini <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Volodymyr Babchuk <[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: "Roger Pau Monné" <[email protected]>
Cc: Tamas K Lengyel <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Razvan Cojocaru <[email protected]>
Cc: Petre Pircalabu <[email protected]>

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v5:
- Minor style fixes

---
 xen/arch/arm/p2m.c                    |   2 +-
 xen/arch/x86/dom0_build.c             |   2 +-
 xen/arch/x86/hvm/mtrr.c               |   5 +-
 xen/arch/x86/mm/mem_sharing.c         |   2 +-
 xen/arch/x86/mm/paging.c              |   2 +-
 xen/arch/x86/x86_64/mm.c              |   2 +-
 xen/common/memory.c                   |   4 +-
 xen/common/vm_event.c                 |   2 +-
 xen/drivers/passthrough/device_tree.c |  11 ---
 xen/drivers/passthrough/iommu.c       | 134 ++++++--------------------
 xen/drivers/passthrough/pci.c         |  12 ---
 xen/drivers/passthrough/vtd/iommu.c   |  10 +-
 xen/drivers/passthrough/x86/iommu.c   |  95 ------------------
 xen/include/asm-arm/iommu.h           |   2 +-
 xen/include/asm-x86/iommu.h           |   2 +-
 xen/include/xen/iommu.h               |  16 ---
 xen/include/xen/sched.h               |   2 -
 17 files changed, 42 insertions(+), 263 deletions(-)
xen/arch/arm/p2m.c +1 -1
index 7f1442932a..692565757e 100644
@@ -1056,7 +1056,7 @@ static int __p2m_set_entry(struct p2m_domain *p2m,
          !mfn_eq(lpae_get_mfn(*entry), lpae_get_mfn(orig_pte)) )
         p2m_free_entry(p2m, orig_pte, level);
 
-    if ( has_iommu_pt(p2m->domain) &&
+    if ( is_iommu_enabled(p2m->domain) &&
          (lpae_is_valid(orig_pte) || lpae_is_valid(*entry)) )
     {
         unsigned int flush_flags = 0;
xen/arch/x86/dom0_build.c +1 -1
index d381784edd..7cfab2dc25 100644
@@ -365,7 +365,7 @@ unsigned long __init dom0_compute_nr_pages(
     }
 
     need_paging = is_hvm_domain(d) &&
-        (!iommu_hap_pt_share || !paging_mode_hap(d));
+        (!iommu_use_hap_pt(d) || !paging_mode_hap(d));
     for ( ; ; need_paging = false )
     {
         nr_pages = get_memsize(&dom0_size, avail);
xen/arch/x86/hvm/mtrr.c +3 -2
index 7ccd85bcea..5ad15eafe0 100644
@@ -783,7 +783,8 @@ HVM_REGISTER_SAVE_RESTORE(MTRR, hvm_save_mtrr_msr, hvm_load_mtrr_msr, 1,
 
 void memory_type_changed(struct domain *d)
 {
-    if ( (has_iommu_pt(d) || cache_flush_permitted(d)) && d->vcpu && d->vcpu[0] )
+    if ( (is_iommu_enabled(d) || cache_flush_permitted(d)) &&
+         d->vcpu && d->vcpu[0] )
     {
         p2m_memory_type_changed(d);
         flush_all(FLUSH_CACHE);
@@ -831,7 +832,7 @@ int epte_get_entry_emt(struct domain *d, unsigned long gfn, mfn_t mfn,
         return MTRR_TYPE_UNCACHABLE;
     }
 
-    if ( !has_iommu_pt(d) && !cache_flush_permitted(d) )
+    if ( !is_iommu_enabled(d) && !cache_flush_permitted(d) )
     {
         *ipat = 1;
         return MTRR_TYPE_WRBACK;
xen/arch/x86/mm/mem_sharing.c +1 -1
index a5fe89e339..efb8821768 100644
@@ -1664,7 +1664,7 @@ int mem_sharing_domctl(struct domain *d, struct xen_domctl_mem_sharing_op *mec)
         case XEN_DOMCTL_MEM_SHARING_CONTROL:
         {
             rc = 0;
-            if ( unlikely(has_iommu_pt(d) && mec->u.enable) )
+            if ( unlikely(is_iommu_enabled(d) && mec->u.enable) )
                 rc = -EXDEV;
             else
                 d->arch.hvm.mem_sharing_enabled = mec->u.enable;
xen/arch/x86/mm/paging.c +1 -1
index 69aa228e46..d9a52c4db4 100644
@@ -213,7 +213,7 @@ int paging_log_dirty_enable(struct domain *d, bool_t log_global)
 {
     int ret;
 
-    if ( has_iommu_pt(d) && log_global )
+    if ( is_iommu_enabled(d) && log_global )
     {
         /*
          * Refuse to turn on global log-dirty mode
xen/arch/x86/x86_64/mm.c +1 -1
index 1919cae18b..827b3f5e27 100644
@@ -1433,7 +1433,7 @@ int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm)
      * shared or being kept in sync then newly added memory needs to be
      * mapped here.
      */
-    if ( has_iommu_pt(hardware_domain) &&
+    if ( is_iommu_enabled(hardware_domain) &&
          !iommu_use_hap_pt(hardware_domain) &&
          !need_iommu_pt_sync(hardware_domain) )
     {
xen/common/memory.c +2 -2
index d9b35a608c..71445c2f53 100644
@@ -823,7 +823,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
     xatp->gpfn += start;
     xatp->size -= start;
 
-    if ( has_iommu_pt(d) )
+    if ( is_iommu_enabled(d) )
        this_cpu(iommu_dont_flush_iotlb) = 1;
 
     while ( xatp->size > done )
@@ -844,7 +844,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
         }
     }
 
-    if ( has_iommu_pt(d) )
+    if ( is_iommu_enabled(d) )
     {
         int ret;
 
xen/common/vm_event.c +1 -1
index 2a1c87e44b..3b18195ebf 100644
@@ -630,7 +630,7 @@ int vm_event_domctl(struct domain *d, struct xen_domctl_vm_event_op *vec)
 
             /* No paging if iommu is used */
             rc = -EMLINK;
-            if ( unlikely(has_iommu_pt(d)) )
+            if ( unlikely(is_iommu_enabled(d)) )
                 break;
 
             rc = -EXDEV;
xen/drivers/passthrough/device_tree.c +0 -11
index 12f2c4c3f2..ea9fd54e3b 100644
@@ -40,17 +40,6 @@ int iommu_assign_dt_device(struct domain *d, struct dt_device_node *dev)
     if ( !list_empty(&dev->domain_list) )
         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 +28 -106

182 lines not on this page · raw message

xen/drivers/passthrough/pci.c +0 -12
index dee47fe08b..28b77b3e30 100644
@@ -933,9 +933,6 @@ static int deassign_device(struct domain *d, uint16_t seg, uint8_t bus,
 
     pdev->fault.count = 0;
 
-    if ( !has_arch_pdevs(d) && has_iommu_pt(d) )
-        iommu_teardown(d);
-
     return ret;
 }
 
@@ -1484,13 +1481,6 @@ static int assign_device(struct domain *d, u16 seg, u8 bus, u8 devfn, u32 flag)
     if ( !pcidevs_trylock() )
         return -ERESTART;
 
-    rc = iommu_construct(d);
-    if ( rc )
-    {
-        pcidevs_unlock();
-        return rc;
-    }
-
     pdev = pci_get_pdev_by_domain(hardware_domain, seg, bus, devfn);
     if ( !pdev )
     {
@@ -1519,8 +1509,6 @@ 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) )
-        iommu_teardown(d);
     pcidevs_unlock();
 
     return rc;
xen/drivers/passthrough/vtd/iommu.c +1 -9
index 01f0bc4689..4ac5da197a 100644
@@ -1759,15 +1759,7 @@ static void iommu_domain_teardown(struct domain *d)
 
     ASSERT(is_iommu_enabled(d));
 
-    /*
-     * We can't use iommu_use_hap_pt here because either IOMMU state
-     * is already changed to IOMMU_STATUS_disabled at this point or
-     * has always been IOMMU_STATUS_disabled.
-     *
-     * We also need to test if HAP is enabled because PV guests can
-     * enter this path too.
-     */
-    if ( hap_enabled(d) && iommu_hap_pt_share )
+    if ( iommu_use_hap_pt(d) )
         return;
 
     spin_lock(&hd->arch.mapping_lock);
xen/drivers/passthrough/x86/iommu.c +0 -95

101 lines not on this page · raw message

xen/include/asm-arm/iommu.h +1 -1
index 904c9aec11..1577e83d2b 100644
@@ -21,7 +21,7 @@ struct arch_iommu
 };
 
 /* Always share P2M Table between the CPU and the IOMMU */
-#define iommu_use_hap_pt(d) (has_iommu_pt(d))
+#define iommu_use_hap_pt(d) is_iommu_enabled(d)
 
 const struct iommu_ops *iommu_get_ops(void);
 void iommu_set_ops(const struct iommu_ops *ops);
xen/include/asm-x86/iommu.h +1 -1
index facf835ada..6d024d5c0e 100644
@@ -81,7 +81,7 @@ extern const struct iommu_init_ops *iommu_init_ops;
 
 /* Are we using the domain P2M table as its IOMMU pagetable? */
 #define iommu_use_hap_pt(d) \
-    (hap_enabled(d) && has_iommu_pt(d) && iommu_hap_pt_share)
+    (hap_enabled(d) && is_iommu_enabled(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 +0 -16
index 314f28f323..4b6871936c 100644
@@ -73,15 +73,9 @@ void iommu_domain_destroy(struct domain *d);
 
 void arch_iommu_domain_destroy(struct domain *d);
 int arch_iommu_domain_init(struct domain *d);
-int arch_iommu_populate_page_table(struct domain *d);
 void arch_iommu_check_autotranslated_hwdom(struct domain *d);
 void arch_iommu_hwdom_init(struct domain *d);
 
-int iommu_construct(struct domain *d);
-
-/* Function used internally, use iommu_domain_destroy */
-void iommu_teardown(struct domain *d);
-
 /*
  * The following flags are passed to map operations and passed by lookup
  * operations.
@@ -248,13 +242,6 @@ struct iommu_ops {
 # define iommu_vcall iommu_call
 #endif
 
-enum iommu_status
-{
-    IOMMU_STATUS_disabled,
-    IOMMU_STATUS_initializing,
-    IOMMU_STATUS_initialized
-};
-
 struct domain_iommu {
     struct arch_iommu arch;
 
@@ -269,9 +256,6 @@ struct domain_iommu {
     /* Features supported by the IOMMU */
     DECLARE_BITMAP(features, IOMMU_FEAT_count);
 
-    /* Status of guest IOMMU mappings */
-    enum iommu_status status;
-
     /*
      * Does the guest reqire mappings to be synchonized, to maintain
      * the default dfn == pfn map. (See comment on dfn at the top of
xen/include/xen/sched.h +0 -2
index 258924242f..b376768424 100644
@@ -965,10 +965,8 @@ static inline bool is_hwdom_pinned_vcpu(const struct vcpu *v)
 }
 
 #ifdef CONFIG_HAS_PASSTHROUGH
-#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 has_iommu_pt(d) false
 #define need_iommu_pt_sync(d) false
 #endif
 
Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Acked-by on v6 8/10
#22 lore raw

Re: [Xen-devel] [PATCH v6 08/10] remove late (on-demand) construction of IOMMU page tables

Quoted 16 lines · Paul Durrant

Now that there is a per-domain IOMMU enable flag, which should be enabled if any device is going to be passed through, stop deferring page table construction until the assignment is done. Also don't tear down the tables again when the last device is de-assigned; defer that task until domain destruction.

This allows the has_iommu_pt() helper and iommu_status enumeration to be removed. Calls to has_iommu_pt() are simply replaced by calls to is_iommu_enabled(). Remaining open-code tests of iommu_hap_pt_share can also be replaced by calls to iommu_use_hap_pt().
The arch_iommu_populate_page_table() and iommu_construct() functions become redundant, as does the 'strict mode' dom0 page_list mapping code in iommu_hwdom_init(), and iommu_teardown() can be made static is its only remaining caller, iommu_domain_destroy(), is within the same source module.

Quoted 8 lines

All in all, about 220 lines of code are removed.

NOTE: This patch will cause a small amount of extra resource to be used
to accommodate IOMMU page tables that may never be used, since the per-domain IOMMU flag enable flag is currently set to the value of the global iommu_enable flag. A subsequent patch will add an option to the toolstack to allow it to be turned off if there is no intention to assign passthrough hardware to the domain.

This has slipped under my radar, sorry.

Thanks,
Razvan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#23 lore raw

Re: [Xen-devel] [PATCH v6 08/10] remove late (on-demand) construction of IOMMU page tables

In reply to Paul Durrant
Quoted 86 lines · Paul Durrant

Now that there is a per-domain IOMMU enable flag, which should be enabled if any device is going to be passed through, stop deferring page table construction until the assignment is done. Also don't tear down the tables again when the last device is de-assigned; defer that task until domain destruction.

This allows the has_iommu_pt() helper and iommu_status enumeration to be removed. Calls to has_iommu_pt() are simply replaced by calls to is_iommu_enabled(). Remaining open-code tests of iommu_hap_pt_share can also be replaced by calls to iommu_use_hap_pt().
The arch_iommu_populate_page_table() and iommu_construct() functions become redundant, as does the 'strict mode' dom0 page_list mapping code in iommu_hwdom_init(), and iommu_teardown() can be made static is its only remaining caller, iommu_domain_destroy(), is within the same source module.

All in all, about 220 lines of code are removed.

NOTE: This patch will cause a small amount of extra resource to be used
to accommodate IOMMU page tables that may never be used, since the per-domain IOMMU flag enable flag is currently set to the value of the global iommu_enable flag. A subsequent patch will add an option to the toolstack to allow it to be turned off if there is no intention to assign passthrough hardware to the domain.

---
Cc: Stefano Stabellini <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Volodymyr Babchuk <[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: "Roger Pau Monné" <[email protected]>
Cc: Tamas K Lengyel <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Razvan Cojocaru <[email protected]>
Cc: Petre Pircalabu <[email protected]>

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v5:
- Minor style fixes

---
 xen/arch/arm/p2m.c                    |   2 +-
 xen/arch/x86/dom0_build.c             |   2 +-
 xen/arch/x86/hvm/mtrr.c               |   5 +-
 xen/arch/x86/mm/mem_sharing.c         |   2 +-
 xen/arch/x86/mm/paging.c              |   2 +-
 xen/arch/x86/x86_64/mm.c              |   2 +-
 xen/common/memory.c                   |   4 +-
 xen/common/vm_event.c                 |   2 +-
 xen/drivers/passthrough/device_tree.c |  11 ---
 xen/drivers/passthrough/iommu.c       | 134 ++++++--------------------
 xen/drivers/passthrough/pci.c         |  12 ---
 xen/drivers/passthrough/vtd/iommu.c   |  10 +-
 xen/drivers/passthrough/x86/iommu.c   |  95 ------------------
 xen/include/asm-arm/iommu.h           |   2 +-
 xen/include/asm-x86/iommu.h           |   2 +-
 xen/include/xen/iommu.h               |  16 ---
 xen/include/xen/sched.h               |   2 -
 17 files changed, 42 insertions(+), 263 deletions(-)
xen/arch/arm/p2m.c +1 -1
index 7f1442932a..692565757e 100644
@@ -1056,7 +1056,7 @@ static int __p2m_set_entry(struct p2m_domain *p2m,
          !mfn_eq(lpae_get_mfn(*entry), lpae_get_mfn(orig_pte)) )
         p2m_free_entry(p2m, orig_pte, level);
 
-    if ( has_iommu_pt(p2m->domain) &&
+    if ( is_iommu_enabled(p2m->domain) &&
          (lpae_is_valid(orig_pte) || lpae_is_valid(*entry)) )
     {
         unsigned int flush_flags = 0;
xen/arch/x86/dom0_build.c +0 -0
index d381784edd..7cfab2dc25 100644

xen/arch/x86/dom0_build.c unsigned long __init dom0_compute_nr_pages( · line 368

Quoted 5 lines
xen/arch/x86/dom0_build.c +1 -1
@@ -365,7 +365,7 @@ unsigned long __init dom0_compute_nr_pages(
     }
 
     need_paging = is_hvm_domain(d) &&
-        (!iommu_hap_pt_share || !paging_mode_hap(d));
+        (!iommu_use_hap_pt(d) || !paging_mode_hap(d));

I'm not sure this is correct, at the point where dom0_compute_nr_pages gets called the iommu has not been initialized yet (the call to iommu_hwdom_init is done afterwards), so the iommu status field which is used by iommu_use_hap_pt is not yet initialized.

Quoted 209 lines
xen/arch/x86/dom0_build.c +0 -0
     for ( ; ; need_paging = false )
     {
         nr_pages = get_memsize(&dom0_size, avail);
xen/arch/x86/hvm/mtrr.c +3 -2
index 7ccd85bcea..5ad15eafe0 100644
@@ -783,7 +783,8 @@ HVM_REGISTER_SAVE_RESTORE(MTRR, hvm_save_mtrr_msr, hvm_load_mtrr_msr, 1,
 
 void memory_type_changed(struct domain *d)
 {
-    if ( (has_iommu_pt(d) || cache_flush_permitted(d)) && d->vcpu && d->vcpu[0] )
+    if ( (is_iommu_enabled(d) || cache_flush_permitted(d)) &&
+         d->vcpu && d->vcpu[0] )
     {
         p2m_memory_type_changed(d);
         flush_all(FLUSH_CACHE);
@@ -831,7 +832,7 @@ int epte_get_entry_emt(struct domain *d, unsigned long gfn, mfn_t mfn,
         return MTRR_TYPE_UNCACHABLE;
     }
 
-    if ( !has_iommu_pt(d) && !cache_flush_permitted(d) )
+    if ( !is_iommu_enabled(d) && !cache_flush_permitted(d) )
     {
         *ipat = 1;
         return MTRR_TYPE_WRBACK;
xen/arch/x86/mm/mem_sharing.c +1 -1
index a5fe89e339..efb8821768 100644
@@ -1664,7 +1664,7 @@ int mem_sharing_domctl(struct domain *d, struct xen_domctl_mem_sharing_op *mec)
         case XEN_DOMCTL_MEM_SHARING_CONTROL:
         {
             rc = 0;
-            if ( unlikely(has_iommu_pt(d) && mec->u.enable) )
+            if ( unlikely(is_iommu_enabled(d) && mec->u.enable) )
                 rc = -EXDEV;
             else
                 d->arch.hvm.mem_sharing_enabled = mec->u.enable;
xen/arch/x86/mm/paging.c +1 -1
index 69aa228e46..d9a52c4db4 100644
@@ -213,7 +213,7 @@ int paging_log_dirty_enable(struct domain *d, bool_t log_global)
 {
     int ret;
 
-    if ( has_iommu_pt(d) && log_global )
+    if ( is_iommu_enabled(d) && log_global )
     {
         /*
          * Refuse to turn on global log-dirty mode
xen/arch/x86/x86_64/mm.c +1 -1
index 1919cae18b..827b3f5e27 100644
@@ -1433,7 +1433,7 @@ int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm)
      * shared or being kept in sync then newly added memory needs to be
      * mapped here.
      */
-    if ( has_iommu_pt(hardware_domain) &&
+    if ( is_iommu_enabled(hardware_domain) &&
          !iommu_use_hap_pt(hardware_domain) &&
          !need_iommu_pt_sync(hardware_domain) )
     {
xen/common/memory.c +2 -2
index d9b35a608c..71445c2f53 100644
@@ -823,7 +823,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
     xatp->gpfn += start;
     xatp->size -= start;
 
-    if ( has_iommu_pt(d) )
+    if ( is_iommu_enabled(d) )
        this_cpu(iommu_dont_flush_iotlb) = 1;
 
     while ( xatp->size > done )
@@ -844,7 +844,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
         }
     }
 
-    if ( has_iommu_pt(d) )
+    if ( is_iommu_enabled(d) )
     {
         int ret;
 
xen/common/vm_event.c +1 -1
index 2a1c87e44b..3b18195ebf 100644
@@ -630,7 +630,7 @@ int vm_event_domctl(struct domain *d, struct xen_domctl_vm_event_op *vec)
 
             /* No paging if iommu is used */
             rc = -EMLINK;
-            if ( unlikely(has_iommu_pt(d)) )
+            if ( unlikely(is_iommu_enabled(d)) )
                 break;
 
             rc = -EXDEV;
xen/drivers/passthrough/device_tree.c +0 -11
index 12f2c4c3f2..ea9fd54e3b 100644
@@ -40,17 +40,6 @@ int iommu_assign_dt_device(struct domain *d, struct dt_device_node *dev)
     if ( !list_empty(&dev->domain_list) )
         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 +24 -34
index 27c75e0786..dc7b75fab6 100644
@@ -146,6 +146,17 @@ static int __init parse_dom0_iommu_param(const char *s)
 }
 custom_param("dom0-iommu", parse_dom0_iommu_param);
 
+static void __hwdom_init check_hwdom_reqs(struct domain *d)
+{
+    if ( iommu_hwdom_none || !paging_mode_translate(d) )
+        return;
+
+    arch_iommu_check_autotranslated_hwdom(d);
+
+    iommu_hwdom_passthrough = false;
+    iommu_hwdom_strict = true;
+}
+
 int iommu_domain_init(struct domain *d)
 {
     struct domain_iommu *hd = dom_iommu(d);
@@ -159,129 +170,44 @@ int iommu_domain_init(struct domain *d)
         return ret;
 
     hd->platform_ops = iommu_get_ops();
-    return hd->platform_ops->init(d);
-}
+    ret = hd->platform_ops->init(d);
+    if ( ret )
+        return ret;
 
-static void __hwdom_init check_hwdom_reqs(struct domain *d)
-{
-    if ( iommu_hwdom_none || !paging_mode_translate(d) )
-        return;
+    /*
+     * NB: 'relaxed' h/w domains don't need the IOMMU mappings to be kept
+     *     in-sync with their assigned pages because all host RAM will be
+     *     mapped during hwdom_init().
+     */
+    if ( is_hardware_domain(d) )
+        check_hwdom_reqs(d); /* may modify iommu_hwdom_strict */
 
-    arch_iommu_check_autotranslated_hwdom(d);
+    if ( !is_hardware_domain(d) || iommu_hwdom_strict )
+        hd->need_sync = !iommu_use_hap_pt(d);
 
-    iommu_hwdom_passthrough = false;
-    iommu_hwdom_strict = true;
+    return 0;
 }
 
 void __hwdom_init iommu_hwdom_init(struct domain *d)
 {
     struct domain_iommu *hd = dom_iommu(d);
 
-    check_hwdom_reqs(d);
-
     if ( !is_iommu_enabled(d) )
         return;
 
     register_keyhandler('o', &iommu_dump_p2m_table, "dump iommu p2m table", 0);
 
-    hd->status = IOMMU_STATUS_initializing;
-    /*
-     * NB: relaxed hw domains don't need sync because all ram is already
-     * mapped in the iommu page tables.
-     */
-    hd->need_sync = iommu_hwdom_strict && !iommu_use_hap_pt(d);
-    if ( need_iommu_pt_sync(d) )
-    {
-        struct page_info *page;
-        unsigned int i = 0, flush_flags = 0;
-        int rc = 0;
-
-        page_list_for_each ( page, &d->page_list )
-        {
-            unsigned long mfn = mfn_x(page_to_mfn(page));
-            unsigned long dfn = mfn_to_gmfn(d, mfn);
-            unsigned int mapping = IOMMUF_readable;
-            int ret;
-
-            if ( ((page->u.inuse.type_info & PGT_count_mask) == 0) ||
-                 ((page->u.inuse.type_info & PGT_type_mask)
-                  == PGT_writable_page) )
-                mapping |= IOMMUF_writable;

xen/drivers/passthrough/iommu.c int iommu_domain_init(struct domain *d) · line 219

Quoted 10 lines
xen/drivers/passthrough/iommu.c +0 -10
-
-            ret = iommu_map(d, _dfn(dfn), _mfn(mfn), 0, mapping,
-                            &flush_flags);
-
-            if ( !rc )
-                rc = ret;
-
-            if ( !(i++ & 0xfffff) )
-                process_pending_softirqs();
-        }

Don't you need to add the domain pages to the page-tables?

iommu_hwdom_init is called after the memory for the domain has been added. Maybe this is fine because the iommu would be enabled earlier and thus pages added to the domain would already be added to the iommu page-tables if required?

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#24 lore raw

Re: [Xen-devel] [PATCH v6 08/10] remove late (on-demand) construction of IOMMU page tables

Quoted 102 lines

-----Original Message-----
From: Roger Pau Monne <[email protected]>
Sent: 23 August 2019 12:34
To: Paul Durrant <[email protected]>
Cc: [email protected]; Alexandru Isaila <[email protected]>; Stefano Stabellini <[email protected]>; Julien Grall <[email protected]>; Volodymyr Babchuk <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Jan Beulich <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>; Wei Liu <[email protected]>; Tamas K Lengyel <[email protected]>; Razvan Cojocaru <[email protected]>; Petre Pircalabu <[email protected]>
Subject: Re: [PATCH v6 08/10] remove late (on-demand) construction of IOMMU page tables

Quoted 88 lines · Paul Durrant

Now that there is a per-domain IOMMU enable flag, which should be enabled if any device is going to be passed through, stop deferring page table construction until the assignment is done. Also don't tear down the tables again when the last device is de-assigned; defer that task until domain destruction.

This allows the has_iommu_pt() helper and iommu_status enumeration to be removed. Calls to has_iommu_pt() are simply replaced by calls to is_iommu_enabled(). Remaining open-code tests of iommu_hap_pt_share can also be replaced by calls to iommu_use_hap_pt().
The arch_iommu_populate_page_table() and iommu_construct() functions become redundant, as does the 'strict mode' dom0 page_list mapping code in iommu_hwdom_init(), and iommu_teardown() can be made static is its only remaining caller, iommu_domain_destroy(), is within the same source module.

All in all, about 220 lines of code are removed.

NOTE: This patch will cause a small amount of extra resource to be used
to accommodate IOMMU page tables that may never be used, since the per-domain IOMMU flag enable flag is currently set to the value of the global iommu_enable flag. A subsequent patch will add an option to the toolstack to allow it to be turned off if there is no intention to assign passthrough hardware to the domain.

---
Cc: Stefano Stabellini <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Volodymyr Babchuk <[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: "Roger Pau Monné" <[email protected]>
Cc: Tamas K Lengyel <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Razvan Cojocaru <[email protected]>
Cc: Petre Pircalabu <[email protected]>

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v5:
- Minor style fixes

---
 xen/arch/arm/p2m.c                    |   2 +-
 xen/arch/x86/dom0_build.c             |   2 +-
 xen/arch/x86/hvm/mtrr.c               |   5 +-
 xen/arch/x86/mm/mem_sharing.c         |   2 +-
 xen/arch/x86/mm/paging.c              |   2 +-
 xen/arch/x86/x86_64/mm.c              |   2 +-
 xen/common/memory.c                   |   4 +-
 xen/common/vm_event.c                 |   2 +-
 xen/drivers/passthrough/device_tree.c |  11 ---
 xen/drivers/passthrough/iommu.c       | 134 ++++++--------------------
 xen/drivers/passthrough/pci.c         |  12 ---
 xen/drivers/passthrough/vtd/iommu.c   |  10 +-
 xen/drivers/passthrough/x86/iommu.c   |  95 ------------------
 xen/include/asm-arm/iommu.h           |   2 +-
 xen/include/asm-x86/iommu.h           |   2 +-
 xen/include/xen/iommu.h               |  16 ---
 xen/include/xen/sched.h               |   2 -
 17 files changed, 42 insertions(+), 263 deletions(-)
xen/arch/arm/p2m.c +1 -1
index 7f1442932a..692565757e 100644
@@ -1056,7 +1056,7 @@ static int __p2m_set_entry(struct p2m_domain *p2m,
          !mfn_eq(lpae_get_mfn(*entry), lpae_get_mfn(orig_pte)) )
         p2m_free_entry(p2m, orig_pte, level);
 
-    if ( has_iommu_pt(p2m->domain) &&
+    if ( is_iommu_enabled(p2m->domain) &&
          (lpae_is_valid(orig_pte) || lpae_is_valid(*entry)) )
     {
         unsigned int flush_flags = 0;
xen/arch/x86/dom0_build.c +0 -0
index d381784edd..7cfab2dc25 100644
@@ -365,7 +365,7 @@ unsigned long __init dom0_compute_nr_pages(
     }
 
Quoted 8 lines

xen/arch/x86/dom0_build.c unsigned long __init dom0_compute_nr_pages( · line 368

Quoted 3 lines
xen/arch/x86/dom0_build.c +1 -1
     need_paging = is_hvm_domain(d) &&
-        (!iommu_hap_pt_share || !paging_mode_hap(d));
+        (!iommu_use_hap_pt(d) || !paging_mode_hap(d));

I'm not sure this is correct, at the point where dom0_compute_nr_pages gets called the iommu has not been initialized yet (the call to iommu_hwdom_init is done afterwards), so the iommu status field which is used by iommu_use_hap_pt is not yet initialized.

Note that this patch removes the iommu status field.

Quoted 215 lines

xen/arch/x86/mm/mem_sharing.c int mem_sharing_domctl(struct domain *d, struct xen_domctl_mem_sharing_op

Quoted 32 lines
xen/arch/x86/dom0_build.c +0 -0
     for ( ; ; need_paging = false )
     {
         nr_pages = get_memsize(&dom0_size, avail);
xen/arch/x86/hvm/mtrr.c +3 -2
index 7ccd85bcea..5ad15eafe0 100644
@@ -783,7 +783,8 @@ HVM_REGISTER_SAVE_RESTORE(MTRR, hvm_save_mtrr_msr, hvm_load_mtrr_msr, 1,
 
 void memory_type_changed(struct domain *d)
 {
-    if ( (has_iommu_pt(d) || cache_flush_permitted(d)) && d->vcpu && d->vcpu[0] )
+    if ( (is_iommu_enabled(d) || cache_flush_permitted(d)) &&
+         d->vcpu && d->vcpu[0] )
     {
         p2m_memory_type_changed(d);
         flush_all(FLUSH_CACHE);
@@ -831,7 +832,7 @@ int epte_get_entry_emt(struct domain *d, unsigned long gfn, mfn_t mfn,
         return MTRR_TYPE_UNCACHABLE;
     }
 
-    if ( !has_iommu_pt(d) && !cache_flush_permitted(d) )
+    if ( !is_iommu_enabled(d) && !cache_flush_permitted(d) )
     {
         *ipat = 1;
         return MTRR_TYPE_WRBACK;
xen/arch/x86/mm/mem_sharing.c +0 -0
index a5fe89e339..efb8821768 100644
@@ -1664,7 +1664,7 @@ int mem_sharing_domctl(struct domain *d, struct xen_domctl_mem_sharing_op

*mec)

Quoted 182 lines
xen/arch/x86/mm/mem_sharing.c +1 -1
         case XEN_DOMCTL_MEM_SHARING_CONTROL:
         {
             rc = 0;
-            if ( unlikely(has_iommu_pt(d) && mec->u.enable) )
+            if ( unlikely(is_iommu_enabled(d) && mec->u.enable) )
                 rc = -EXDEV;
             else
                 d->arch.hvm.mem_sharing_enabled = mec->u.enable;
xen/arch/x86/mm/paging.c +1 -1
index 69aa228e46..d9a52c4db4 100644
@@ -213,7 +213,7 @@ int paging_log_dirty_enable(struct domain *d, bool_t log_global)
 {
     int ret;
 
-    if ( has_iommu_pt(d) && log_global )
+    if ( is_iommu_enabled(d) && log_global )
     {
         /*
          * Refuse to turn on global log-dirty mode
xen/arch/x86/x86_64/mm.c +1 -1
index 1919cae18b..827b3f5e27 100644
@@ -1433,7 +1433,7 @@ int memory_add(unsigned long spfn, unsigned long epfn, unsigned int pxm)
      * shared or being kept in sync then newly added memory needs to be
      * mapped here.
      */
-    if ( has_iommu_pt(hardware_domain) &&
+    if ( is_iommu_enabled(hardware_domain) &&
          !iommu_use_hap_pt(hardware_domain) &&
          !need_iommu_pt_sync(hardware_domain) )
     {
xen/common/memory.c +2 -2
index d9b35a608c..71445c2f53 100644
@@ -823,7 +823,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
     xatp->gpfn += start;
     xatp->size -= start;
 
-    if ( has_iommu_pt(d) )
+    if ( is_iommu_enabled(d) )
        this_cpu(iommu_dont_flush_iotlb) = 1;
 
     while ( xatp->size > done )
@@ -844,7 +844,7 @@ int xenmem_add_to_physmap(struct domain *d, struct xen_add_to_physmap *xatp,
         }
     }
 
-    if ( has_iommu_pt(d) )
+    if ( is_iommu_enabled(d) )
     {
         int ret;
 
xen/common/vm_event.c +1 -1
index 2a1c87e44b..3b18195ebf 100644
@@ -630,7 +630,7 @@ int vm_event_domctl(struct domain *d, struct xen_domctl_vm_event_op *vec)
 
             /* No paging if iommu is used */
             rc = -EMLINK;
-            if ( unlikely(has_iommu_pt(d)) )
+            if ( unlikely(is_iommu_enabled(d)) )
                 break;
 
             rc = -EXDEV;
xen/drivers/passthrough/device_tree.c +0 -11
index 12f2c4c3f2..ea9fd54e3b 100644
@@ -40,17 +40,6 @@ int iommu_assign_dt_device(struct domain *d, struct dt_device_node *dev)
     if ( !list_empty(&dev->domain_list) )
         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 +24 -38
index 27c75e0786..dc7b75fab6 100644
@@ -146,6 +146,17 @@ static int __init parse_dom0_iommu_param(const char *s)
 }
 custom_param("dom0-iommu", parse_dom0_iommu_param);
 
+static void __hwdom_init check_hwdom_reqs(struct domain *d)
+{
+    if ( iommu_hwdom_none || !paging_mode_translate(d) )
+        return;
+
+    arch_iommu_check_autotranslated_hwdom(d);
+
+    iommu_hwdom_passthrough = false;
+    iommu_hwdom_strict = true;
+}
+
 int iommu_domain_init(struct domain *d)
 {
     struct domain_iommu *hd = dom_iommu(d);
@@ -159,129 +170,44 @@ int iommu_domain_init(struct domain *d)
         return ret;
 
     hd->platform_ops = iommu_get_ops();
-    return hd->platform_ops->init(d);
-}
+    ret = hd->platform_ops->init(d);
+    if ( ret )
+        return ret;
 
-static void __hwdom_init check_hwdom_reqs(struct domain *d)
-{
-    if ( iommu_hwdom_none || !paging_mode_translate(d) )
-        return;
+    /*
+     * NB: 'relaxed' h/w domains don't need the IOMMU mappings to be kept
+     *     in-sync with their assigned pages because all host RAM will be
+     *     mapped during hwdom_init().
+     */
+    if ( is_hardware_domain(d) )
+        check_hwdom_reqs(d); /* may modify iommu_hwdom_strict */
 
-    arch_iommu_check_autotranslated_hwdom(d);
+    if ( !is_hardware_domain(d) || iommu_hwdom_strict )
+        hd->need_sync = !iommu_use_hap_pt(d);
 
-    iommu_hwdom_passthrough = false;
-    iommu_hwdom_strict = true;
+    return 0;
 }
 
 void __hwdom_init iommu_hwdom_init(struct domain *d)
 {
     struct domain_iommu *hd = dom_iommu(d);
 
-    check_hwdom_reqs(d);
-
     if ( !is_iommu_enabled(d) )
         return;
 
     register_keyhandler('o', &iommu_dump_p2m_table, "dump iommu p2m table", 0);
 
-    hd->status = IOMMU_STATUS_initializing;
-    /*
-     * NB: relaxed hw domains don't need sync because all ram is already
-     * mapped in the iommu page tables.
-     */
-    hd->need_sync = iommu_hwdom_strict && !iommu_use_hap_pt(d);
-    if ( need_iommu_pt_sync(d) )
-    {
-        struct page_info *page;
-        unsigned int i = 0, flush_flags = 0;
-        int rc = 0;
-
-        page_list_for_each ( page, &d->page_list )
-        {
-            unsigned long mfn = mfn_x(page_to_mfn(page));
-            unsigned long dfn = mfn_to_gmfn(d, mfn);
-            unsigned int mapping = IOMMUF_readable;
-            int ret;
-
-            if ( ((page->u.inuse.type_info & PGT_count_mask) == 0) ||
-                 ((page->u.inuse.type_info & PGT_type_mask)
-                  == PGT_writable_page) )
-                mapping |= IOMMUF_writable;
-
-            ret = iommu_map(d, _dfn(dfn), _mfn(mfn), 0, mapping,
-                            &flush_flags);
-
Quoted 9 lines

xen/drivers/passthrough/iommu.c int iommu_domain_init(struct domain *d) · line 219

Quoted 6 lines
xen/drivers/passthrough/iommu.c +0 -6
-            if ( !rc )
-                rc = ret;
-
-            if ( !(i++ & 0xfffff) )
-                process_pending_softirqs();
-        }

Don't you need to add the domain pages to the page-tables?

Not any more.

Quoted 4 lines

iommu_hwdom_init is called after the memory for the domain has been added. Maybe this is fine because the iommu would be enabled earlier and thus pages added to the domain would already be added to the iommu page-tables if required?

Yes, that's right. Because we don't have the late init any more, the pages will be added to the IOMMU mappings when they get added to the domain.

Paul

Quoted 2 lines

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#25 lore raw

Re: [Xen-devel] [PATCH v6 08/10] remove late (on-demand) construction of IOMMU page tables

In reply to Paul Durrant

xen/drivers/passthrough/iommu.c static int __init parse_dom0_iommu_param(const char *s) · line 149

Quoted 7 lines · Paul Durrant
xen/drivers/passthrough/iommu.c +1 -0
@@ -146,6 +146,17 @@ static int __init parse_dom0_iommu_param(const char *s)
 }
 custom_param("dom0-iommu", parse_dom0_iommu_param);
 
+static void __hwdom_init check_hwdom_reqs(struct domain *d)

This really should have const, but I realize ...

xen/drivers/passthrough/iommu.c static int __init parse_dom0_iommu_param(const char *s) · line 154

Quoted 5 lines
xen/drivers/passthrough/iommu.c +5 -0
+{
+    if ( iommu_hwdom_none || !paging_mode_translate(d) )
+        return;
+
+    arch_iommu_check_autotranslated_hwdom(d);

... this one wants non-const (for - afaict - no reason).

Quoted 9 lines
xen/drivers/passthrough/iommu.c +3 -2
@@ -159,129 +170,44 @@ int iommu_domain_init(struct domain *d)
         return ret;
 
     hd->platform_ops = iommu_get_ops();
-    return hd->platform_ops->init(d);
-}
+    ret = hd->platform_ops->init(d);
+    if ( ret )
+        return ret;

xen/drivers/passthrough/iommu.c int iommu_domain_init(struct domain *d) · line 181

Quoted 10 lines
xen/drivers/passthrough/iommu.c +5 -4
 
-static void __hwdom_init check_hwdom_reqs(struct domain *d)
-{
-    if ( iommu_hwdom_none || !paging_mode_translate(d) )
-        return;
+    /*
+     * NB: 'relaxed' h/w domains don't need the IOMMU mappings to be kept
+     *     in-sync with their assigned pages because all host RAM will be
+     *     mapped during hwdom_init().
+     */

Doesn't this comment belong to ...

xen/drivers/passthrough/iommu.c int iommu_domain_init(struct domain *d) · line 186

Quoted 6 lines
xen/drivers/passthrough/iommu.c +4 -1
+    if ( is_hardware_domain(d) )
+        check_hwdom_reqs(d); /* may modify iommu_hwdom_strict */
 
-    arch_iommu_check_autotranslated_hwdom(d);
+    if ( !is_hardware_domain(d) || iommu_hwdom_strict )
+        hd->need_sync = !iommu_use_hap_pt(d);

... this if()?

xen/drivers/passthrough/iommu.c static void iommu_dump_p2m_table(unsigned char key) · line 556

Quoted 8 lines
xen/drivers/passthrough/iommu.c +1 -2
@@ -629,8 +552,7 @@ static void iommu_dump_p2m_table(unsigned char key)
     ops = iommu_get_ops();
     for_each_domain(d)
     {
-        if ( is_hardware_domain(d) ||
-             dom_iommu(d)->status < IOMMU_STATUS_initialized )
+        if ( !is_iommu_enabled(d) )
             continue;

Didn't you agree to retain the hwdom part of the condition here?

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#26 lore raw

Re: [Xen-devel] [PATCH v6 08/10] remove late (on-demand) construction of IOMMU page tables

Quoted 54 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 29 August 2019 14:39
To: Paul Durrant <[email protected]>
Cc: [email protected]; Julien Grall <[email protected]>; Alexandru Isaila <[email protected]>; Petre Pircalabu <[email protected]>; Razvan Cojocaru <[email protected]>; Andrew Cooper <[email protected]>; Roger Pau Monne <[email protected]>; Volodymyr Babchuk <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Stefano Stabellini <[email protected]>; KonradRzeszutek Wilk <[email protected]>; Tamas K Lengyel <[email protected]>; Tim (Xen.org) <[email protected]>; Wei Liu <[email protected]>
Subject: Re: [PATCH v6 08/10] remove late (on-demand) construction of IOMMU page tables

xen/drivers/passthrough/iommu.c static int __init parse_dom0_iommu_param(const char *s) · line 149

Quoted 7 lines · Paul Durrant
xen/drivers/passthrough/iommu.c +1 -0
@@ -146,6 +146,17 @@ static int __init parse_dom0_iommu_param(const char *s)
 }
 custom_param("dom0-iommu", parse_dom0_iommu_param);
 
+static void __hwdom_init check_hwdom_reqs(struct domain *d)

This really should have const, but I realize ...

xen/drivers/passthrough/iommu.c static int __init parse_dom0_iommu_param(const char *s) · line 154

Quoted 5 lines
xen/drivers/passthrough/iommu.c +5 -0
+{
+    if ( iommu_hwdom_none || !paging_mode_translate(d) )
+        return;
+
+    arch_iommu_check_autotranslated_hwdom(d);

... this one wants non-const (for - afaict - no reason).

Quoted 9 lines
xen/drivers/passthrough/iommu.c +3 -2
@@ -159,129 +170,44 @@ int iommu_domain_init(struct domain *d)
         return ret;
 
     hd->platform_ops = iommu_get_ops();
-    return hd->platform_ops->init(d);
-}
+    ret = hd->platform_ops->init(d);
+    if ( ret )
+        return ret;

xen/drivers/passthrough/iommu.c int iommu_domain_init(struct domain *d) · line 181

Quoted 10 lines
xen/drivers/passthrough/iommu.c +5 -4
 
-static void __hwdom_init check_hwdom_reqs(struct domain *d)
-{
-    if ( iommu_hwdom_none || !paging_mode_translate(d) )
-        return;
+    /*
+     * NB: 'relaxed' h/w domains don't need the IOMMU mappings to be kept
+     *     in-sync with their assigned pages because all host RAM will be
+     *     mapped during hwdom_init().
+     */

Doesn't this comment belong to ...

Quoted 8 lines

xen/drivers/passthrough/iommu.c int iommu_domain_init(struct domain *d) · line 186

Quoted 6 lines
xen/drivers/passthrough/iommu.c +4 -1
+    if ( is_hardware_domain(d) )
+        check_hwdom_reqs(d); /* may modify iommu_hwdom_strict */
 
-    arch_iommu_check_autotranslated_hwdom(d);
+    if ( !is_hardware_domain(d) || iommu_hwdom_strict )
+        hd->need_sync = !iommu_use_hap_pt(d);

... this if()?

Yes, it's probably adrift from where it should be now.

Quoted 12 lines

xen/drivers/passthrough/iommu.c static void iommu_dump_p2m_table(unsigned char key) · line 556

Quoted 9 lines
xen/drivers/passthrough/iommu.c +1 -2
@@ -629,8 +552,7 @@ static void iommu_dump_p2m_table(unsigned char key)
     ops = iommu_get_ops();
     for_each_domain(d)
     {
-        if ( is_hardware_domain(d) ||
-             dom_iommu(d)->status < IOMMU_STATUS_initialized )
+        if ( !is_iommu_enabled(d) )
             continue;

Didn't you agree to retain the hwdom part of the condition here?

Indeed I have as of today, but this was posted 2 weeks ago :-)

Paul

Quoted 1 line

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#27 lore raw

PATCH v6 9/10 iommu: tidy up iommu_use_hap_pt() and need_iommu_pt_sync() macros

In reply to Paul Durrant

Thes macros really ought to live in the common xen/iommu.h header rather then being distributed amongst architecture specific iommu headers and xen/sched.h. This patch moves them there.

NOTE: Disabling 'sharept' in the command line iommu options should really
be hard error on ARM (as opposed to just being ignored), so avoid parsing that option if CONFIG_ARM is set.

---
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]>
Cc: Volodymyr Babchuk <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

Previously part of https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

---
 xen/drivers/passthrough/iommu.c |  2 ++
 xen/include/asm-arm/iommu.h     |  3 ---
 xen/include/asm-x86/iommu.h     |  4 ----
 xen/include/xen/iommu.h         | 11 +++++++++++
 xen/include/xen/sched.h         |  6 ------
 5 files changed, 13 insertions(+), 13 deletions(-)
xen/drivers/passthrough/iommu.c +2 -0
index dc7b75fab6..b24be5ffa6 100644
@@ -102,8 +102,10 @@ static int __init parse_iommu_param(const char *s)
             iommu_hwdom_passthrough = val;
         else if ( (val = parse_boolean("dom0-strict", s, ss)) >= 0 )
             iommu_hwdom_strict = val;
+#ifndef CONFIG_ARM
         else if ( (val = parse_boolean("sharept", s, ss)) >= 0 )
             iommu_hap_pt_share = val;
+#endif
         else
             rc = -EINVAL;
 
xen/include/asm-arm/iommu.h +0 -3
index 1577e83d2b..77a94b29eb 100644
@@ -20,9 +20,6 @@ struct arch_iommu
     void *priv;
 };
 
-/* Always share P2M Table between the CPU and the IOMMU */
-#define iommu_use_hap_pt(d) is_iommu_enabled(d)
-
 const struct iommu_ops *iommu_get_ops(void);
 void iommu_set_ops(const struct iommu_ops *ops);
 
xen/include/asm-x86/iommu.h +0 -4
index 6d024d5c0e..25d2aee9a9 100644
@@ -79,10 +79,6 @@ struct iommu_init_ops {
 
 extern const struct iommu_init_ops *iommu_init_ops;
 
-/* Are we using the domain P2M table as its IOMMU pagetable? */
-#define iommu_use_hap_pt(d) \
-    (hap_enabled(d) && is_iommu_enabled(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);
 int iommu_setup_hpet_msi(struct msi_desc *);
xen/include/xen/iommu.h +11 -0
index 4b6871936c..5e7ca98170 100644
@@ -268,6 +268,17 @@ struct domain_iommu {
 #define iommu_set_feature(d, f)   set_bit(f, dom_iommu(d)->features)
 #define iommu_clear_feature(d, f) clear_bit(f, dom_iommu(d)->features)
 
+/* Are we using the domain P2M table as its IOMMU pagetable? */
+#define iommu_use_hap_pt(d) \
+    (hap_enabled(d) && is_iommu_enabled(d) && iommu_hap_pt_share)
+
+/* Does the IOMMU pagetable need to be kept synchronized with the P2M */
+#ifdef CONFIG_HAS_PASSTHROUGH
+#define need_iommu_pt_sync(d)     (dom_iommu(d)->need_sync)
+#else
+#define need_iommu_pt_sync(d)     false
+#endif
+
 int __must_check iommu_suspend(void);
 void iommu_resume(void);
 void iommu_crash_shutdown(void);
xen/include/xen/sched.h +0 -6
index b376768424..7e4c24f8d0 100644
@@ -964,12 +964,6 @@ static inline bool is_hwdom_pinned_vcpu(const struct vcpu *v)
             cpumask_weight(v->cpu_hard_affinity) == 1);
 }
 
-#ifdef CONFIG_HAS_PASSTHROUGH
-#define need_iommu_pt_sync(d) (dom_iommu(d)->need_sync)
-#else
-#define need_iommu_pt_sync(d) false
-#endif
-
 static inline bool is_vcpu_online(const struct vcpu *v)
 {
     return !test_bit(_VPF_down, &v->pause_flags);
Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#28 lore raw

Re: [Xen-devel] [PATCH v6 09/10] iommu: tidy up iommu_use_hap_pt() and need_iommu_pt_sync() macros

Quoted 36 lines · Paul Durrant

Thes macros really ought to live in the common xen/iommu.h header rather then being distributed amongst architecture specific iommu headers and xen/sched.h. This patch moves them there.

NOTE: Disabling 'sharept' in the command line iommu options should really
be hard error on ARM (as opposed to just being ignored), so avoid parsing that option if CONFIG_ARM is set.

---
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]>
Cc: Volodymyr Babchuk <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

Previously part of https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

---
 xen/drivers/passthrough/iommu.c |  2 ++
 xen/include/asm-arm/iommu.h     |  3 ---
 xen/include/asm-x86/iommu.h     |  4 ----
 xen/include/xen/iommu.h         | 11 +++++++++++
 xen/include/xen/sched.h         |  6 ------
 5 files changed, 13 insertions(+), 13 deletions(-)
xen/drivers/passthrough/iommu.c +0 -0
index dc7b75fab6..b24be5ffa6 100644

xen/drivers/passthrough/iommu.c static int __init parse_iommu_param(const char *s) · line 108

Quoted 7 lines
xen/drivers/passthrough/iommu.c +2 -0
@@ -102,8 +102,10 @@ static int __init parse_iommu_param(const char *s)
             iommu_hwdom_passthrough = val;
         else if ( (val = parse_boolean("dom0-strict", s, ss)) >= 0 )
             iommu_hwdom_strict = val;
+#ifndef CONFIG_ARM
         else if ( (val = parse_boolean("sharept", s, ss)) >= 0 )
             iommu_hap_pt_share = val;
+#endif

I would prefer if you would print a warning message in iommu_setup if iommu_hap_pt_share is set to false on ARM, and set it to true.

The rest LGTM.

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#29 lore raw

Re: [Xen-devel] [PATCH v6 09/10] iommu: tidy up iommu_use_hap_pt() and need_iommu_pt_sync() macros

In reply to Paul Durrant

xen/drivers/passthrough/iommu.c static int __init parse_iommu_param(const char *s) · line 110

Quoted 12 lines · Paul Durrant
xen/drivers/passthrough/iommu.c +2 -0
@@ -102,8 +102,10 @@ static int __init parse_iommu_param(const char *s)
             iommu_hwdom_passthrough = val;
         else if ( (val = parse_boolean("dom0-strict", s, ss)) >= 0 )
             iommu_hwdom_strict = val;
+#ifndef CONFIG_ARM
         else if ( (val = parse_boolean("sharept", s, ss)) >= 0 )
             iommu_hap_pt_share = val;
+#endif
         else
             rc = -EINVAL;

I think you/we should go further here: Arm should #define this to true, and here we should have "#ifndef iommu_hap_pt_share".

Quoted 5 lines
xen/include/xen/iommu.h +0 -0
@@ -268,6 +268,17 @@ struct domain_iommu {
 #define iommu_set_feature(d, f)   set_bit(f, dom_iommu(d)->features)
 #define iommu_clear_feature(d, f) clear_bit(f, dom_iommu(d)->features)

xen/include/xen/iommu.h struct domain_iommu { · line 279

Quoted 10 lines
xen/include/xen/iommu.h +9 -0
 
+/* Are we using the domain P2M table as its IOMMU pagetable? */
+#define iommu_use_hap_pt(d) \
+    (hap_enabled(d) && is_iommu_enabled(d) && iommu_hap_pt_share)
+
+/* Does the IOMMU pagetable need to be kept synchronized with the P2M */
+#ifdef CONFIG_HAS_PASSTHROUGH
+#define need_iommu_pt_sync(d)     (dom_iommu(d)->need_sync)
+#else
+#define need_iommu_pt_sync(d)     false

I think you'd better evaluate d here; one (somewhat in risk of opposition) variant would be

#define need_iommu_pt_sync(d) (!(d))

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#30 lore raw

PATCH v6 10/10 introduce a 'passthrough' configuration option to xl.cfg...

In reply to Paul Durrant

...and hence the ability to disable IOMMU mappings, and control EPT sharing.

This patch introduces a new 'libxl_passthrough' enumeration into libxl_domain_create_info. The value will be set by xl either when it parses a new 'passthrough' option in xl.cfg, or implicitly if there is passthrough hardware specified for the domain.

If the value of the passthrough configuration option is 'disabled' then the XEN_DOMCTL_CDF_iommu flag will be clear in the xen_domctl_createdomain flags, thus allowing the toolstack to control whether the domain gets IOMMU mappings or not (where previously they were globally set).

If the value of the passthrough configuration option is 'sync_pt' then a new 'iommu_opts' field in xen_domctl_createdomain will be set with the value XEN_DOMCTL_IOMMU_no_sharept. This will override the global default set in iommu_hap_pt_share, thus allowing the toolstack to control whether EPT sharing is used for the domain.

---
Cc: Ian Jackson <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Anthony PERARD <[email protected]>
Cc: Volodymyr Babchuk <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v6:
- Remove the libxl_physinfo() call since it's usefulness is limited to x86

v5:
- Expand xen_domctl_createdomain flags field and hence bump interface version
- Fix spelling mistakes in context line

---
 docs/man/xl.cfg.5.pod.in        | 52 +++++++++++++++++++++++++++++++++
 tools/libxl/libxl.h             |  5 ++++
 tools/libxl/libxl_create.c      |  9 ++++++
 tools/libxl/libxl_types.idl     |  7 +++++
 tools/xl/xl_parse.c             | 38 ++++++++++++++++++++++++
 xen/arch/arm/domain.c           | 10 ++++++-
 xen/arch/x86/domain.c           |  2 +-
 xen/common/domain.c             |  7 +++++
 xen/drivers/passthrough/iommu.c | 13 ++++++++-
 xen/include/public/domctl.h     |  6 +++-
 xen/include/xen/iommu.h         | 19 ++++++++----
 11 files changed, 158 insertions(+), 10 deletions(-)
docs/man/xl.cfg.5.pod.in +52 -0
index c99d40307e..fd35685e9e 100644
@@ -605,6 +605,58 @@ option should only be used with a trusted device tree.
 Note that the partial device tree should avoid using the phandle 65000
 which is reserved by the toolstack.
 
+=item B<passthrough="STRING">
+
+Specify whether IOMMU mappings are enabled for the domain and hence whether
+it will be enabled for passthrough hardware. Valid values for this option
+are:
+
+=over 4
+
+=item B<disabled>
+
+IOMMU mappings are disabled for the domain and so hardware may not be
+passed through.
+
+This option is the default if no passthrough hardware is specified
+in the domain's configuration.
+
+=item B<sync_pt>
+
+This option means that IOMMU mappings will be synchronized with the
+domain's P2M table as follows:
+
+For a PV domain, all writable pages assigned to the domain are identity
+mapped by MFN in the IOMMU page table. Thus a device driver running in the
+domain may program passthrough hardware for DMA using MFN values
+(i.e. host/machine frame numbers) looked up in its P2M.
+
+For an HVM domain, all non-foreign RAM pages present in its P2M will be
+mapped by GFN in the IOMMU page table. Thus a device driver running in the
+domain may program passthrough hardware using GFN values (i.e. guest
+physical frame numbers) without any further translation.
+
+This option is the default if the domain is PV and passthrough hardware
+is specified in the configuration.
+
+This option is not available on Arm.
+
+=item B<share_pt>
+
+This option is unavailable for a PV domain. For an HVM domain, this option
+means that the IOMMU will be programmed to directly reference the domain's
+P2M table as its page table. From the point of view of a device driver
+running in the domain this is functionally equivalent to B<sync_pt> but
+places less load on the hypervisor and so should generally be selected in
+preference. However, the availability of this option is hardware specific
+and thus, if it is specified for a domain running on hardware that does
+not allow it (e.g. AMD), B<sync_pt> will be used instead.
+
+This option is the default if the domain is HVM and passthrough hardware
+is specified in the configuration.
+
+=back
+
 =back
 
 =head2 Devices
tools/libxl/libxl.h +5 -0
index 9bacfb97f0..5de7c07a41 100644
@@ -394,6 +394,11 @@
  */
 #define LIBXL_HAVE_EXTENDED_VKB 1
 
+/*
+ * libxl_domain_create_info has libxl_passthrough enumeration.
+ */
+#define LIBXL_HAVE_CREATEINFO_PASSTHROUGH 1
+
 /*
  * libxl ABI compatibility
  *
tools/libxl/libxl_create.c +9 -0
index 03ce166f4f..f288e13dc1 100644
@@ -564,6 +564,15 @@ int libxl__domain_make(libxl__gc *gc, libxl_domain_config *d_config,
                 libxl_defbool_val(info->oos) ? 0 : XEN_DOMCTL_CDF_oos_off;
         }
 
+        LOG(DETAIL, "passthrough: %s",
+            libxl_passthrough_to_string(info->passthrough));
+
+        if (info->passthrough != LIBXL_PASSTHROUGH_DISABLED)
+            create.flags |= XEN_DOMCTL_CDF_iommu;
+
+        if (info->passthrough == LIBXL_PASSTHROUGH_SYNC_PT)
+            create.iommu_opts |= XEN_DOMCTL_IOMMU_no_sharept;
+
         /* Ultimately, handle is an array of 16 uint8_t, same as uuid */
         libxl_uuid_copy(ctx, (libxl_uuid *)&create.handle, &info->uuid);
 
tools/libxl/libxl_types.idl +7 -0
index b61399ce36..7e37de8646 100644
@@ -263,6 +263,12 @@ libxl_vkb_backend = Enumeration("vkb_backend", [
     (2, "LINUX")
     ])
 
+libxl_passthrough = Enumeration("passthrough", [
+    (0, "disabled"),
+    (1, "sync_pt"),
+    (2, "share_pt"),
+    ])
+
 #
 # Complex libxl types
 #
@@ -408,6 +414,7 @@ libxl_domain_create_info = Struct("domain_create_info",[
     ("pool_name",    string),
     ("run_hotplug_scripts",libxl_defbool),
     ("driver_domain",libxl_defbool),
+    ("passthrough",  libxl_passthrough),
     ], dir=DIR_IN)
 
 libxl_domain_restore_params = Struct("domain_restore_params", [
tools/xl/xl_parse.c +38 -0
index e105bda2bb..c904604008 100644
@@ -2326,6 +2326,44 @@ skip_vfb:
         }
     }
 
+    if (!xlu_cfg_get_string(config, "passthrough", &buf, 0)) {
+        libxl_passthrough o;
+
+        e = libxl_passthrough_from_string(buf, &o);
+        if (e) {
+            fprintf(stderr,
+                    "ERROR: unknown passthrough option '%s'\n",
+                    buf);
+            exit(-ERROR_FAIL);
+        }
+
+        switch (o) {
+        case LIBXL_PASSTHROUGH_DISABLED:
+            if (d_config->num_pcidevs || d_config->num_dtdevs) {
+                fprintf(stderr,
+                        "ERROR: passthrough disabled but devices are specified\n");
+                exit(-ERROR_FAIL);
+            }
+        case LIBXL_PASSTHROUGH_SHARE_PT:
+            if (c_info->type == LIBXL_DOMAIN_TYPE_PV) {
+                fprintf(stderr,
+                        "ERROR: passthrough=\"share_pt\" not valid for PV domain\n");
+                exit(-ERROR_FAIL);
+            }
+        default:
+            break;
+        }
+
+        c_info->passthrough = o;
+    } else if (d_config->num_pcidevs || d_config->num_dtdevs) {
+        /*
+         * Passthrough devices are specified so set an appropriate
+         * default value.
+         */
+        c_info->passthrough = (c_info->type == LIBXL_DOMAIN_TYPE_PV) ?
+            LIBXL_PASSTHROUGH_SYNC_PT : LIBXL_PASSTHROUGH_SHARE_PT;
+    }
+
     if (!xlu_cfg_get_list(config, "usbctrl", &usbctrls, 0, 0)) {
         d_config->num_usbctrls = 0;
         d_config->usbctrls = NULL;
xen/arch/arm/domain.c +9 -1
index 941bbff4fe..b12de6ff3d 100644
@@ -614,6 +614,14 @@ int arch_sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    /* Always share P2M Table between the CPU and the IOMMU */
+    if ( config->iommu_opts & XEN_DOMCTL_IOMMU_no_sharept )
+    {
+        dprintk(XENLOG_INFO,
+                "Unsupported iommu option: XEN_DOMCTL_IOMMU_no_sharept\n");
+        return -EINVAL;
+    }
+
     /* Fill in the native GIC version, passed back to the toolstack. */
     if ( config->arch.gic_version == XEN_DOMCTL_CONFIG_GIC_NATIVE )
     {
@@ -674,7 +682,7 @@ int arch_domain_create(struct domain *d,
     ASSERT(config != NULL);
 
     /* p2m_init relies on some value initialized by the IOMMU subsystem */
-    if ( (rc = iommu_domain_init(d)) != 0 )
+    if ( (rc = iommu_domain_init(d, config->iommu_opts)) != 0 )
         goto fail;
 
     if ( (rc = p2m_init(d)) != 0 )
xen/arch/x86/domain.c +1 -1
index f144d8fe9a..4f7dad49c5 100644
@@ -604,7 +604,7 @@ int arch_domain_create(struct domain *d,
     if ( (rc = init_domain_irq_mapping(d)) != 0 )
         goto fail;
 
-    if ( (rc = iommu_domain_init(d)) != 0 )
+    if ( (rc = iommu_domain_init(d, config->iommu_opts)) != 0 )
         goto fail;
 
     psr_domain_init(d);
xen/common/domain.c +7 -0
index e832a5c4aa..142b08174b 100644
@@ -308,6 +308,13 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( !(config->flags & XEN_DOMCTL_CDF_iommu) && config->iommu_opts )
+    {
+        dprintk(XENLOG_INFO,
+                "IOMMU options specified but IOMMU not enabled\n");
+        return -EINVAL;
+    }
+
     if ( config->max_vcpus < 1 )
     {
         dprintk(XENLOG_INFO, "No vCPUS\n");
xen/drivers/passthrough/iommu.c +12 -1
index b24be5ffa6..a526aa6c09 100644
@@ -159,7 +159,7 @@ static void __hwdom_init check_hwdom_reqs(struct domain *d)
     iommu_hwdom_strict = true;
 }
 
-int iommu_domain_init(struct domain *d)
+int iommu_domain_init(struct domain *d, unsigned int opts)
 {
     struct domain_iommu *hd = dom_iommu(d);
     int ret = 0;
@@ -176,6 +176,15 @@ int iommu_domain_init(struct domain *d)
     if ( ret )
         return ret;
 
+    /*
+     * Use shared page tables for HAP and IOMMU if the global option
+     * is enabled (from which we can infer the h/w is capable) and
+     * the domain options do not disallow it. HAP must, of course, also
+     * be enabled.
+     */
+    hd->hap_pt_share = hap_enabled(d) && iommu_hap_pt_share &&
+        !(opts & XEN_DOMCTL_IOMMU_no_sharept);
+
     /*
      * NB: 'relaxed' h/w domains don't need the IOMMU mappings to be kept
      *     in-sync with their assigned pages because all host RAM will be
@@ -187,6 +196,8 @@ int iommu_domain_init(struct domain *d)
     if ( !is_hardware_domain(d) || iommu_hwdom_strict )
         hd->need_sync = !iommu_use_hap_pt(d);
 
+    ASSERT(!(hd->need_sync && hd->hap_pt_share));
+
     return 0;
 }
 
xen/include/public/domctl.h +5 -1
index 3f82c78870..922ed50a11 100644
@@ -38,7 +38,7 @@
 #include "hvm/save.h"
 #include "memory.h"
 
-#define XEN_DOMCTL_INTERFACE_VERSION 0x00000011
+#define XEN_DOMCTL_INTERFACE_VERSION 0x00000012
 
 /*
  * NB. xen_domctl.domain is an IN/OUT parameter for this operation.
@@ -70,6 +70,10 @@ struct xen_domctl_createdomain {
 
     uint32_t flags;
 
+#define _XEN_DOMCTL_IOMMU_no_sharept  0
+#define XEN_DOMCTL_IOMMU_no_sharept   (1U<<_XEN_DOMCTL_IOMMU_no_sharept)
+    uint32_t iommu_opts;
+
     /*
      * Various domain limits, which impact the quantity of resources (global
      * mapping space, xenheap, etc) a guest may consume.
xen/include/xen/iommu.h +13 -6
index 5e7ca98170..01025e372b 100644
@@ -67,7 +67,7 @@ extern unsigned int iommu_dev_iotlb_timeout;
 int iommu_setup(void);
 int iommu_hardware_setup(void);
 
-int iommu_domain_init(struct domain *d);
+int iommu_domain_init(struct domain *d, unsigned int opts);
 void iommu_hwdom_init(struct domain *d);
 void iommu_domain_destroy(struct domain *d);
 
@@ -257,9 +257,17 @@ struct domain_iommu {
     DECLARE_BITMAP(features, IOMMU_FEAT_count);
 
     /*
-     * Does the guest reqire mappings to be synchonized, to maintain
-     * the default dfn == pfn map. (See comment on dfn at the top of
-     * include/xen/mm.h).
+     * Does the guest share HAP mapping with the IOMMU? This is always
+     * true for ARM systems and may be true for x86 systems where the
+     * the hardware is capable.
+     */
+    bool hap_pt_share;
+
+    /*
+     * Does the guest require mappings to be synchronized, to maintain
+     * the default dfn == pfn map? (See comment on dfn at the top of
+     * include/xen/mm.h). Note that hap_pt_share == false does not
+     * necessarily imply this is true.
      */
     bool need_sync;
 };
@@ -269,8 +277,7 @@ struct domain_iommu {
 #define iommu_clear_feature(d, f) clear_bit(f, dom_iommu(d)->features)
 
 /* Are we using the domain P2M table as its IOMMU pagetable? */
-#define iommu_use_hap_pt(d) \
-    (hap_enabled(d) && is_iommu_enabled(d) && iommu_hap_pt_share)
+#define iommu_use_hap_pt(d)       (dom_iommu(d)->hap_pt_share)
 
 /* Does the IOMMU pagetable need to be kept synchronized with the P2M */
 #ifdef CONFIG_HAS_PASSTHROUGH
Signature · 7 lines
2.20.1.2.gb21ebb671


_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
#31 lore raw

Re: [Xen-devel] [PATCH v6 10/10] introduce a 'passthrough' configuration option to xl.cfg...

Quoted 72 lines · Paul Durrant

...and hence the ability to disable IOMMU mappings, and control EPT sharing.

This patch introduces a new 'libxl_passthrough' enumeration into libxl_domain_create_info. The value will be set by xl either when it parses a new 'passthrough' option in xl.cfg, or implicitly if there is passthrough hardware specified for the domain.

If the value of the passthrough configuration option is 'disabled' then the XEN_DOMCTL_CDF_iommu flag will be clear in the xen_domctl_createdomain flags, thus allowing the toolstack to control whether the domain gets IOMMU mappings or not (where previously they were globally set).

If the value of the passthrough configuration option is 'sync_pt' then a new 'iommu_opts' field in xen_domctl_createdomain will be set with the value XEN_DOMCTL_IOMMU_no_sharept. This will override the global default set in iommu_hap_pt_share, thus allowing the toolstack to control whether EPT sharing is used for the domain.

---
Cc: Ian Jackson <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Anthony PERARD <[email protected]>
Cc: Volodymyr Babchuk <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v6:
- Remove the libxl_physinfo() call since it's usefulness is limited to x86

v5:
- Expand xen_domctl_createdomain flags field and hence bump interface version
- Fix spelling mistakes in context line

---
 docs/man/xl.cfg.5.pod.in        | 52 +++++++++++++++++++++++++++++++++
 tools/libxl/libxl.h             |  5 ++++
 tools/libxl/libxl_create.c      |  9 ++++++
 tools/libxl/libxl_types.idl     |  7 +++++
 tools/xl/xl_parse.c             | 38 ++++++++++++++++++++++++
 xen/arch/arm/domain.c           | 10 ++++++-
 xen/arch/x86/domain.c           |  2 +-
 xen/common/domain.c             |  7 +++++
 xen/drivers/passthrough/iommu.c | 13 ++++++++-
 xen/include/public/domctl.h     |  6 +++-
 xen/include/xen/iommu.h         | 19 ++++++++----
 11 files changed, 158 insertions(+), 10 deletions(-)
docs/man/xl.cfg.5.pod.in +7 -0
index c99d40307e..fd35685e9e 100644
@@ -605,6 +605,58 @@ option should only be used with a trusted device tree.
 Note that the partial device tree should avoid using the phandle 65000
 which is reserved by the toolstack.
 
+=item B<passthrough="STRING">
+
+Specify whether IOMMU mappings are enabled for the domain and hence whether
+it will be enabled for passthrough hardware. Valid values for this option
+are:
+
+=over 4

docs/man/xl.cfg.5.pod.in option should only be used with a trusted device tree. · line 624

Quoted 10 lines
docs/man/xl.cfg.5.pod.in +10 -0
+
+=item B<disabled>
+
+IOMMU mappings are disabled for the domain and so hardware may not be
+passed through.
+
+This option is the default if no passthrough hardware is specified
+in the domain's configuration.
+
+=item B<sync_pt>

I would maybe name this exclusive_pt, but historically it's been named sync_pt.

Quoted 115 lines
docs/man/xl.cfg.5.pod.in +35 -0
+
+This option means that IOMMU mappings will be synchronized with the
+domain's P2M table as follows:
+
+For a PV domain, all writable pages assigned to the domain are identity
+mapped by MFN in the IOMMU page table. Thus a device driver running in the
+domain may program passthrough hardware for DMA using MFN values
+(i.e. host/machine frame numbers) looked up in its P2M.
+
+For an HVM domain, all non-foreign RAM pages present in its P2M will be
+mapped by GFN in the IOMMU page table. Thus a device driver running in the
+domain may program passthrough hardware using GFN values (i.e. guest
+physical frame numbers) without any further translation.
+
+This option is the default if the domain is PV and passthrough hardware
+is specified in the configuration.
+
+This option is not available on Arm.
+
+=item B<share_pt>
+
+This option is unavailable for a PV domain. For an HVM domain, this option
+means that the IOMMU will be programmed to directly reference the domain's
+P2M table as its page table. From the point of view of a device driver
+running in the domain this is functionally equivalent to B<sync_pt> but
+places less load on the hypervisor and so should generally be selected in
+preference. However, the availability of this option is hardware specific
+and thus, if it is specified for a domain running on hardware that does
+not allow it (e.g. AMD), B<sync_pt> will be used instead.
+
+This option is the default if the domain is HVM and passthrough hardware
+is specified in the configuration.
+
+=back
+
 =back
 
 =head2 Devices
tools/libxl/libxl.h +5 -0
index 9bacfb97f0..5de7c07a41 100644
@@ -394,6 +394,11 @@
  */
 #define LIBXL_HAVE_EXTENDED_VKB 1
 
+/*
+ * libxl_domain_create_info has libxl_passthrough enumeration.
+ */
+#define LIBXL_HAVE_CREATEINFO_PASSTHROUGH 1
+
 /*
  * libxl ABI compatibility
  *
tools/libxl/libxl_create.c +9 -0
index 03ce166f4f..f288e13dc1 100644
@@ -564,6 +564,15 @@ int libxl__domain_make(libxl__gc *gc, libxl_domain_config *d_config,
                 libxl_defbool_val(info->oos) ? 0 : XEN_DOMCTL_CDF_oos_off;
         }
 
+        LOG(DETAIL, "passthrough: %s",
+            libxl_passthrough_to_string(info->passthrough));
+
+        if (info->passthrough != LIBXL_PASSTHROUGH_DISABLED)
+            create.flags |= XEN_DOMCTL_CDF_iommu;
+
+        if (info->passthrough == LIBXL_PASSTHROUGH_SYNC_PT)
+            create.iommu_opts |= XEN_DOMCTL_IOMMU_no_sharept;
+
         /* Ultimately, handle is an array of 16 uint8_t, same as uuid */
         libxl_uuid_copy(ctx, (libxl_uuid *)&create.handle, &info->uuid);
 
tools/libxl/libxl_types.idl +7 -0
index b61399ce36..7e37de8646 100644
@@ -263,6 +263,12 @@ libxl_vkb_backend = Enumeration("vkb_backend", [
     (2, "LINUX")
     ])
 
+libxl_passthrough = Enumeration("passthrough", [
+    (0, "disabled"),
+    (1, "sync_pt"),
+    (2, "share_pt"),
+    ])
+
 #
 # Complex libxl types
 #
@@ -408,6 +414,7 @@ libxl_domain_create_info = Struct("domain_create_info",[
     ("pool_name",    string),
     ("run_hotplug_scripts",libxl_defbool),
     ("driver_domain",libxl_defbool),
+    ("passthrough",  libxl_passthrough),
     ], dir=DIR_IN)
 
 libxl_domain_restore_params = Struct("domain_restore_params", [
tools/xl/xl_parse.c +8 -0
index e105bda2bb..c904604008 100644
@@ -2326,6 +2326,44 @@ skip_vfb:
         }
     }
 
+    if (!xlu_cfg_get_string(config, "passthrough", &buf, 0)) {
+        libxl_passthrough o;
+
+        e = libxl_passthrough_from_string(buf, &o);
+        if (e) {
+            fprintf(stderr,
+                    "ERROR: unknown passthrough option '%s'\n",
+                    buf);

tools/xl/xl_parse.c skip_vfb: · line 2346

Quoted 10 lines
tools/xl/xl_parse.c +10 -0
+            exit(-ERROR_FAIL);
+        }
+
+        switch (o) {
+        case LIBXL_PASSTHROUGH_DISABLED:
+            if (d_config->num_pcidevs || d_config->num_dtdevs) {
+                fprintf(stderr,
+                        "ERROR: passthrough disabled but devices are specified\n");
+                exit(-ERROR_FAIL);
+            }

Don't you need a break here?

tools/xl/xl_parse.c skip_vfb: · line 2352

Quoted 6 lines
tools/xl/xl_parse.c +6 -0
+        case LIBXL_PASSTHROUGH_SHARE_PT:
+            if (c_info->type == LIBXL_DOMAIN_TYPE_PV) {
+                fprintf(stderr,
+                        "ERROR: passthrough=\"share_pt\" not valid for PV domain\n");
+                exit(-ERROR_FAIL);
+            }

And here likely (or a /* fallthrough */ comment)

Quoted 161 lines
tools/xl/xl_parse.c +14 -0
+        default:
+            break;
+        }
+
+        c_info->passthrough = o;
+    } else if (d_config->num_pcidevs || d_config->num_dtdevs) {
+        /*
+         * Passthrough devices are specified so set an appropriate
+         * default value.
+         */
+        c_info->passthrough = (c_info->type == LIBXL_DOMAIN_TYPE_PV) ?
+            LIBXL_PASSTHROUGH_SYNC_PT : LIBXL_PASSTHROUGH_SHARE_PT;
+    }
+
     if (!xlu_cfg_get_list(config, "usbctrl", &usbctrls, 0, 0)) {
         d_config->num_usbctrls = 0;
         d_config->usbctrls = NULL;
xen/arch/arm/domain.c +9 -1
index 941bbff4fe..b12de6ff3d 100644
@@ -614,6 +614,14 @@ int arch_sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    /* Always share P2M Table between the CPU and the IOMMU */
+    if ( config->iommu_opts & XEN_DOMCTL_IOMMU_no_sharept )
+    {
+        dprintk(XENLOG_INFO,
+                "Unsupported iommu option: XEN_DOMCTL_IOMMU_no_sharept\n");
+        return -EINVAL;
+    }
+
     /* Fill in the native GIC version, passed back to the toolstack. */
     if ( config->arch.gic_version == XEN_DOMCTL_CONFIG_GIC_NATIVE )
     {
@@ -674,7 +682,7 @@ int arch_domain_create(struct domain *d,
     ASSERT(config != NULL);
 
     /* p2m_init relies on some value initialized by the IOMMU subsystem */
-    if ( (rc = iommu_domain_init(d)) != 0 )
+    if ( (rc = iommu_domain_init(d, config->iommu_opts)) != 0 )
         goto fail;
 
     if ( (rc = p2m_init(d)) != 0 )
xen/arch/x86/domain.c +1 -1
index f144d8fe9a..4f7dad49c5 100644
@@ -604,7 +604,7 @@ int arch_domain_create(struct domain *d,
     if ( (rc = init_domain_irq_mapping(d)) != 0 )
         goto fail;
 
-    if ( (rc = iommu_domain_init(d)) != 0 )
+    if ( (rc = iommu_domain_init(d, config->iommu_opts)) != 0 )
         goto fail;
 
     psr_domain_init(d);
xen/common/domain.c +7 -0
index e832a5c4aa..142b08174b 100644
@@ -308,6 +308,13 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( !(config->flags & XEN_DOMCTL_CDF_iommu) && config->iommu_opts )
+    {
+        dprintk(XENLOG_INFO,
+                "IOMMU options specified but IOMMU not enabled\n");
+        return -EINVAL;
+    }
+
     if ( config->max_vcpus < 1 )
     {
         dprintk(XENLOG_INFO, "No vCPUS\n");
xen/drivers/passthrough/iommu.c +12 -1
index b24be5ffa6..a526aa6c09 100644
@@ -159,7 +159,7 @@ static void __hwdom_init check_hwdom_reqs(struct domain *d)
     iommu_hwdom_strict = true;
 }
 
-int iommu_domain_init(struct domain *d)
+int iommu_domain_init(struct domain *d, unsigned int opts)
 {
     struct domain_iommu *hd = dom_iommu(d);
     int ret = 0;
@@ -176,6 +176,15 @@ int iommu_domain_init(struct domain *d)
     if ( ret )
         return ret;
 
+    /*
+     * Use shared page tables for HAP and IOMMU if the global option
+     * is enabled (from which we can infer the h/w is capable) and
+     * the domain options do not disallow it. HAP must, of course, also
+     * be enabled.
+     */
+    hd->hap_pt_share = hap_enabled(d) && iommu_hap_pt_share &&
+        !(opts & XEN_DOMCTL_IOMMU_no_sharept);
+
     /*
      * NB: 'relaxed' h/w domains don't need the IOMMU mappings to be kept
      *     in-sync with their assigned pages because all host RAM will be
@@ -187,6 +196,8 @@ int iommu_domain_init(struct domain *d)
     if ( !is_hardware_domain(d) || iommu_hwdom_strict )
         hd->need_sync = !iommu_use_hap_pt(d);
 
+    ASSERT(!(hd->need_sync && hd->hap_pt_share));
+
     return 0;
 }
 
xen/include/public/domctl.h +5 -1
index 3f82c78870..922ed50a11 100644
@@ -38,7 +38,7 @@
 #include "hvm/save.h"
 #include "memory.h"
 
-#define XEN_DOMCTL_INTERFACE_VERSION 0x00000011
+#define XEN_DOMCTL_INTERFACE_VERSION 0x00000012
 
 /*
  * NB. xen_domctl.domain is an IN/OUT parameter for this operation.
@@ -70,6 +70,10 @@ struct xen_domctl_createdomain {
 
     uint32_t flags;
 
+#define _XEN_DOMCTL_IOMMU_no_sharept  0
+#define XEN_DOMCTL_IOMMU_no_sharept   (1U<<_XEN_DOMCTL_IOMMU_no_sharept)
+    uint32_t iommu_opts;
+
     /*
      * Various domain limits, which impact the quantity of resources (global
      * mapping space, xenheap, etc) a guest may consume.
xen/include/xen/iommu.h +4 -4
index 5e7ca98170..01025e372b 100644
@@ -67,7 +67,7 @@ extern unsigned int iommu_dev_iotlb_timeout;
 int iommu_setup(void);
 int iommu_hardware_setup(void);
 
-int iommu_domain_init(struct domain *d);
+int iommu_domain_init(struct domain *d, unsigned int opts);
 void iommu_hwdom_init(struct domain *d);
 void iommu_domain_destroy(struct domain *d);
 
@@ -257,9 +257,17 @@ struct domain_iommu {
     DECLARE_BITMAP(features, IOMMU_FEAT_count);
 
     /*
-     * Does the guest reqire mappings to be synchonized, to maintain
-     * the default dfn == pfn map. (See comment on dfn at the top of
-     * include/xen/mm.h).
+     * Does the guest share HAP mapping with the IOMMU? This is always
+     * true for ARM systems and may be true for x86 systems where the
+     * the hardware is capable.

xen/include/xen/iommu.h struct domain_iommu { · line 272

Quoted 10 lines
xen/include/xen/iommu.h +8 -0
+     */
+    bool hap_pt_share;
+
+    /*
+     * Does the guest require mappings to be synchronized, to maintain
+     * the default dfn == pfn map? (See comment on dfn at the top of
+     * include/xen/mm.h). Note that hap_pt_share == false does not
+     * necessarily imply this is true.
      */
     bool need_sync;

I'm lost here, doesn't hap_pt_share = !need_sync?

ie: sync is required because the page-tables are not shared?

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#32 lore raw

Re: [Xen-devel] [PATCH v6 10/10] introduce a 'passthrough' configuration option to xl.cfg...

Quoted 89 lines

-----Original Message-----
From: Roger Pau Monne <[email protected]>
Sent: 23 August 2019 15:17
To: Paul Durrant <[email protected]>
Cc: [email protected]; Ian Jackson <[email protected]>; Wei Liu <[email protected]>; Andrew Cooper <[email protected]>; George Dunlap <[email protected]>; Jan Beulich <[email protected]>; Julien Grall <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Stefano Stabellini <[email protected]>; Tim (Xen.org) <[email protected]>; Anthony Perard <[email protected]>; Volodymyr Babchuk <[email protected]>
Subject: Re: [PATCH v6 10/10] introduce a 'passthrough' configuration option to xl.cfg...

Quoted 77 lines · Paul Durrant

...and hence the ability to disable IOMMU mappings, and control EPT sharing.

This patch introduces a new 'libxl_passthrough' enumeration into libxl_domain_create_info. The value will be set by xl either when it parses a new 'passthrough' option in xl.cfg, or implicitly if there is passthrough hardware specified for the domain.

If the value of the passthrough configuration option is 'disabled' then the XEN_DOMCTL_CDF_iommu flag will be clear in the xen_domctl_createdomain flags, thus allowing the toolstack to control whether the domain gets IOMMU mappings or not (where previously they were globally set).

If the value of the passthrough configuration option is 'sync_pt' then a new 'iommu_opts' field in xen_domctl_createdomain will be set with the value XEN_DOMCTL_IOMMU_no_sharept. This will override the global default set in iommu_hap_pt_share, thus allowing the toolstack to control whether EPT sharing is used for the domain.

---
Cc: Ian Jackson <[email protected]>
Cc: Wei Liu <[email protected]>
Cc: Andrew Cooper <[email protected]>
Cc: George Dunlap <[email protected]>
Cc: Jan Beulich <[email protected]>
Cc: Julien Grall <[email protected]>
Cc: Konrad Rzeszutek Wilk <[email protected]>
Cc: Stefano Stabellini <[email protected]>
Cc: Tim Deegan <[email protected]>
Cc: Anthony PERARD <[email protected]>
Cc: Volodymyr Babchuk <[email protected]>
Cc: "Roger Pau Monné" <[email protected]>

Previously part of series https://lists.xenproject.org/archives/html/xen-devel/2019-07/msg02267.html

v6:
- Remove the libxl_physinfo() call since it's usefulness is limited to x86

v5:
- Expand xen_domctl_createdomain flags field and hence bump interface version
- Fix spelling mistakes in context line

---
 docs/man/xl.cfg.5.pod.in        | 52 +++++++++++++++++++++++++++++++++
 tools/libxl/libxl.h             |  5 ++++
 tools/libxl/libxl_create.c      |  9 ++++++
 tools/libxl/libxl_types.idl     |  7 +++++
 tools/xl/xl_parse.c             | 38 ++++++++++++++++++++++++
 xen/arch/arm/domain.c           | 10 ++++++-
 xen/arch/x86/domain.c           |  2 +-
 xen/common/domain.c             |  7 +++++
 xen/drivers/passthrough/iommu.c | 13 ++++++++-
 xen/include/public/domctl.h     |  6 +++-
 xen/include/xen/iommu.h         | 19 ++++++++----
 11 files changed, 158 insertions(+), 10 deletions(-)
docs/man/xl.cfg.5.pod.in +12 -0
index c99d40307e..fd35685e9e 100644
@@ -605,6 +605,58 @@ option should only be used with a trusted device tree.
 Note that the partial device tree should avoid using the phandle 65000
 which is reserved by the toolstack.
 
+=item B<passthrough="STRING">
+
+Specify whether IOMMU mappings are enabled for the domain and hence whether
+it will be enabled for passthrough hardware. Valid values for this option
+are:
+
+=over 4
+
+=item B<disabled>
+
+IOMMU mappings are disabled for the domain and so hardware may not be
+passed through.
Quoted 9 lines

docs/man/xl.cfg.5.pod.in option should only be used with a trusted device tree. · line 624

Quoted 5 lines
docs/man/xl.cfg.5.pod.in +5 -0
+
+This option is the default if no passthrough hardware is specified
+in the domain's configuration.
+
+=item B<sync_pt>

I would maybe name this exclusive_pt, but historically it's been named sync_pt.

Yes, I prefer to stick with the incumbent name.

[snip]

Quoted 29 lines
Quoted 16 lines
tools/xl/xl_parse.c +8 -0
index e105bda2bb..c904604008 100644
@@ -2326,6 +2326,44 @@ skip_vfb:
         }
     }
 
+    if (!xlu_cfg_get_string(config, "passthrough", &buf, 0)) {
+        libxl_passthrough o;
+
+        e = libxl_passthrough_from_string(buf, &o);
+        if (e) {
+            fprintf(stderr,
+                    "ERROR: unknown passthrough option '%s'\n",
+                    buf);

tools/xl/xl_parse.c skip_vfb: · line 2346

Quoted 10 lines
tools/xl/xl_parse.c +10 -0
+            exit(-ERROR_FAIL);
+        }
+
+        switch (o) {
+        case LIBXL_PASSTHROUGH_DISABLED:
+            if (d_config->num_pcidevs || d_config->num_dtdevs) {
+                fprintf(stderr,
+                        "ERROR: passthrough disabled but devices are specified\n");
+                exit(-ERROR_FAIL);
+            }

Don't you need a break here?

Quoted 9 lines

tools/xl/xl_parse.c skip_vfb: · line 2352

Quoted 6 lines
tools/xl/xl_parse.c +6 -0
+        case LIBXL_PASSTHROUGH_SHARE_PT:
+            if (c_info->type == LIBXL_DOMAIN_TYPE_PV) {
+                fprintf(stderr,
+                        "ERROR: passthrough=\"share_pt\" not valid for PV domain\n");
+                exit(-ERROR_FAIL);
+            }

And here likely (or a /* fallthrough */ comment)

Nope. Missing breaks in both cases. Good spot.

Quoted 174 lines
Quoted 161 lines
tools/xl/xl_parse.c +14 -0
+        default:
+            break;
+        }
+
+        c_info->passthrough = o;
+    } else if (d_config->num_pcidevs || d_config->num_dtdevs) {
+        /*
+         * Passthrough devices are specified so set an appropriate
+         * default value.
+         */
+        c_info->passthrough = (c_info->type == LIBXL_DOMAIN_TYPE_PV) ?
+            LIBXL_PASSTHROUGH_SYNC_PT : LIBXL_PASSTHROUGH_SHARE_PT;
+    }
+
     if (!xlu_cfg_get_list(config, "usbctrl", &usbctrls, 0, 0)) {
         d_config->num_usbctrls = 0;
         d_config->usbctrls = NULL;
xen/arch/arm/domain.c +9 -1
index 941bbff4fe..b12de6ff3d 100644
@@ -614,6 +614,14 @@ int arch_sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    /* Always share P2M Table between the CPU and the IOMMU */
+    if ( config->iommu_opts & XEN_DOMCTL_IOMMU_no_sharept )
+    {
+        dprintk(XENLOG_INFO,
+                "Unsupported iommu option: XEN_DOMCTL_IOMMU_no_sharept\n");
+        return -EINVAL;
+    }
+
     /* Fill in the native GIC version, passed back to the toolstack. */
     if ( config->arch.gic_version == XEN_DOMCTL_CONFIG_GIC_NATIVE )
     {
@@ -674,7 +682,7 @@ int arch_domain_create(struct domain *d,
     ASSERT(config != NULL);
 
     /* p2m_init relies on some value initialized by the IOMMU subsystem */
-    if ( (rc = iommu_domain_init(d)) != 0 )
+    if ( (rc = iommu_domain_init(d, config->iommu_opts)) != 0 )
         goto fail;
 
     if ( (rc = p2m_init(d)) != 0 )
xen/arch/x86/domain.c +1 -1
index f144d8fe9a..4f7dad49c5 100644
@@ -604,7 +604,7 @@ int arch_domain_create(struct domain *d,
     if ( (rc = init_domain_irq_mapping(d)) != 0 )
         goto fail;
 
-    if ( (rc = iommu_domain_init(d)) != 0 )
+    if ( (rc = iommu_domain_init(d, config->iommu_opts)) != 0 )
         goto fail;
 
     psr_domain_init(d);
xen/common/domain.c +7 -0
index e832a5c4aa..142b08174b 100644
@@ -308,6 +308,13 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( !(config->flags & XEN_DOMCTL_CDF_iommu) && config->iommu_opts )
+    {
+        dprintk(XENLOG_INFO,
+                "IOMMU options specified but IOMMU not enabled\n");
+        return -EINVAL;
+    }
+
     if ( config->max_vcpus < 1 )
     {
         dprintk(XENLOG_INFO, "No vCPUS\n");
xen/drivers/passthrough/iommu.c +12 -1
index b24be5ffa6..a526aa6c09 100644
@@ -159,7 +159,7 @@ static void __hwdom_init check_hwdom_reqs(struct domain *d)
     iommu_hwdom_strict = true;
 }
 
-int iommu_domain_init(struct domain *d)
+int iommu_domain_init(struct domain *d, unsigned int opts)
 {
     struct domain_iommu *hd = dom_iommu(d);
     int ret = 0;
@@ -176,6 +176,15 @@ int iommu_domain_init(struct domain *d)
     if ( ret )
         return ret;
 
+    /*
+     * Use shared page tables for HAP and IOMMU if the global option
+     * is enabled (from which we can infer the h/w is capable) and
+     * the domain options do not disallow it. HAP must, of course, also
+     * be enabled.
+     */
+    hd->hap_pt_share = hap_enabled(d) && iommu_hap_pt_share &&
+        !(opts & XEN_DOMCTL_IOMMU_no_sharept);
+
     /*
      * NB: 'relaxed' h/w domains don't need the IOMMU mappings to be kept
      *     in-sync with their assigned pages because all host RAM will be
@@ -187,6 +196,8 @@ int iommu_domain_init(struct domain *d)
     if ( !is_hardware_domain(d) || iommu_hwdom_strict )
         hd->need_sync = !iommu_use_hap_pt(d);
 
+    ASSERT(!(hd->need_sync && hd->hap_pt_share));
+
     return 0;
 }
 
xen/include/public/domctl.h +5 -1
index 3f82c78870..922ed50a11 100644
@@ -38,7 +38,7 @@
 #include "hvm/save.h"
 #include "memory.h"
 
-#define XEN_DOMCTL_INTERFACE_VERSION 0x00000011
+#define XEN_DOMCTL_INTERFACE_VERSION 0x00000012
 
 /*
  * NB. xen_domctl.domain is an IN/OUT parameter for this operation.
@@ -70,6 +70,10 @@ struct xen_domctl_createdomain {
 
     uint32_t flags;
 
+#define _XEN_DOMCTL_IOMMU_no_sharept  0
+#define XEN_DOMCTL_IOMMU_no_sharept   (1U<<_XEN_DOMCTL_IOMMU_no_sharept)
+    uint32_t iommu_opts;
+
     /*
      * Various domain limits, which impact the quantity of resources (global
      * mapping space, xenheap, etc) a guest may consume.
xen/include/xen/iommu.h +4 -4
index 5e7ca98170..01025e372b 100644
@@ -67,7 +67,7 @@ extern unsigned int iommu_dev_iotlb_timeout;
 int iommu_setup(void);
 int iommu_hardware_setup(void);
 
-int iommu_domain_init(struct domain *d);
+int iommu_domain_init(struct domain *d, unsigned int opts);
 void iommu_hwdom_init(struct domain *d);
 void iommu_domain_destroy(struct domain *d);
 
@@ -257,9 +257,17 @@ struct domain_iommu {
     DECLARE_BITMAP(features, IOMMU_FEAT_count);
 
     /*
-     * Does the guest reqire mappings to be synchonized, to maintain
-     * the default dfn == pfn map. (See comment on dfn at the top of
-     * include/xen/mm.h).
+     * Does the guest share HAP mapping with the IOMMU? This is always
+     * true for ARM systems and may be true for x86 systems where the
+     * the hardware is capable.

xen/include/xen/iommu.h struct domain_iommu { · line 272

Quoted 10 lines
xen/include/xen/iommu.h +8 -0
+     */
+    bool hap_pt_share;
+
+    /*
+     * Does the guest require mappings to be synchronized, to maintain
+     * the default dfn == pfn map? (See comment on dfn at the top of
+     * include/xen/mm.h). Note that hap_pt_share == false does not
+     * necessarily imply this is true.
      */
     bool need_sync;

I'm lost here, doesn't hap_pt_share = !need_sync?

Quoted 2 lines

ie: sync is required because the page-tables are not shared?

I thought the comment was clear, but maybe not... Consider the 'inclusive' PV h/w domain mappings. No sharing (because it's a PV domain) and no sync, because all RAM is mapped anyway. Can I phrase it better?

Paul

Quoted 1 line

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

Reviewed-by on v6 10/10
#33 lore raw

Re: [Xen-devel] [PATCH v6 10/10] introduce a 'passthrough' configuration option to xl.cfg...

In reply to Paul Durrant
Quoted 13 lines · Paul Durrant

...and hence the ability to disable IOMMU mappings, and control EPT sharing.

This patch introduces a new 'libxl_passthrough' enumeration into libxl_domain_create_info. The value will be set by xl either when it parses a new 'passthrough' option in xl.cfg, or implicitly if there is passthrough hardware specified for the domain.

If the value of the passthrough configuration option is 'disabled' then the XEN_DOMCTL_CDF_iommu flag will be clear in the xen_domctl_createdomain flags, thus allowing the toolstack to control whether the domain gets IOMMU mappings or not (where previously they were globally set).

Quoted 7 lines

If the value of the passthrough configuration option is 'sync_pt' then a new 'iommu_opts' field in xen_domctl_createdomain will be set with the value XEN_DOMCTL_IOMMU_no_sharept. This will override the global default set in iommu_hap_pt_share, thus allowing the toolstack to control whether EPT sharing is used for the domain.

with a question/suggestion and a nit:

xen/common/domain.c static int sanitise_domain_config(struct xen_domctl_createdomain *config) · line 316

Quoted 12 lines
xen/common/domain.c +6 -0
@@ -308,6 +308,13 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
+    if ( !(config->flags & XEN_DOMCTL_CDF_iommu) && config->iommu_opts )
+    {
+        dprintk(XENLOG_INFO,
+                "IOMMU options specified but IOMMU not enabled\n");
+        return -EINVAL;
+    }

Seeing this I wonder whether XEN_DOMCTL_CDF_iommu wouldn't better be bit 0 of iommu_opts.

xen/common/domain.c struct xen_domctl_createdomain { · line 74

Quoted 6 lines
xen/common/domain.c +2 -0
@@ -70,6 +70,10 @@ struct xen_domctl_createdomain {
 
     uint32_t flags;
 
+#define _XEN_DOMCTL_IOMMU_no_sharept  0
+#define XEN_DOMCTL_IOMMU_no_sharept   (1U<<_XEN_DOMCTL_IOMMU_no_sharept)

Please can you add the missing blanks around << ?

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel

#34 lore raw

Re: [Xen-devel] [PATCH v6 10/10] introduce a 'passthrough' configuration option to xl.cfg...

Quoted 32 lines

-----Original Message-----
From: Jan Beulich <[email protected]>
Sent: 29 August 2019 15:07
To: Paul Durrant <[email protected]>
Cc: [email protected]; Julien Grall <[email protected]>; Andrew Cooper <[email protected]>; Anthony Perard <[email protected]>; Roger Pau Monne <[email protected]>; Volodymyr Babchuk <[email protected]>; George Dunlap <[email protected]>; Ian Jackson <[email protected]>; Stefano Stabellini <[email protected]>; Konrad Rzeszutek Wilk <[email protected]>; Tim (Xen.org) <[email protected]>; Wei Liu <[email protected]>
Subject: Re: [PATCH v6 10/10] introduce a 'passthrough' configuration option to xl.cfg...

Quoted 19 lines · Paul Durrant

...and hence the ability to disable IOMMU mappings, and control EPT sharing.

This patch introduces a new 'libxl_passthrough' enumeration into libxl_domain_create_info. The value will be set by xl either when it parses a new 'passthrough' option in xl.cfg, or implicitly if there is passthrough hardware specified for the domain.

If the value of the passthrough configuration option is 'disabled' then the XEN_DOMCTL_CDF_iommu flag will be clear in the xen_domctl_createdomain flags, thus allowing the toolstack to control whether the domain gets IOMMU mappings or not (where previously they were globally set).

If the value of the passthrough configuration option is 'sync_pt' then a new 'iommu_opts' field in xen_domctl_createdomain will be set with the value XEN_DOMCTL_IOMMU_no_sharept. This will override the global default set in iommu_hap_pt_share, thus allowing the toolstack to control whether EPT sharing is used for the domain.

Quoted 3 lines
Quoted 1 line

Thanks.

Quoted 8 lines

with a question/suggestion and a nit:

Quoted 6 lines
xen/common/domain.c +0 -0
@@ -308,6 +308,13 @@ static int sanitise_domain_config(struct xen_domctl_createdomain *config)
         return -EINVAL;
     }
 
Quoted 10 lines

xen/common/domain.c static int sanitise_domain_config(struct xen_domctl_createdomain *config) · line 316

Quoted 6 lines
xen/common/domain.c +6 -0
+    if ( !(config->flags & XEN_DOMCTL_CDF_iommu) && config->iommu_opts )
+    {
+        dprintk(XENLOG_INFO,
+                "IOMMU options specified but IOMMU not enabled\n");
+        return -EINVAL;
+    }

Seeing this I wonder whether XEN_DOMCTL_CDF_iommu wouldn't better be bit 0 of iommu_opts.

I debated this with myself and I prefer to stick with separate flag and options.

Quoted 9 lines

xen/common/domain.c struct xen_domctl_createdomain { · line 74

Quoted 6 lines
xen/common/domain.c +2 -0
@@ -70,6 +70,10 @@ struct xen_domctl_createdomain {
 
     uint32_t flags;
 
+#define _XEN_DOMCTL_IOMMU_no_sharept  0
+#define XEN_DOMCTL_IOMMU_no_sharept   (1U<<_XEN_DOMCTL_IOMMU_no_sharept)

Please can you add the missing blanks around << ?

Sure.

Paul

Quoted 1 line

Jan

_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel