Skip to contentSkip to conversation

ARM SMMUv3: Fix spurious notification errors and stall with vfio-pci

PATCH Superseded Eric Auger 3 participants 17 messages Jul 1, 2019 – Jul 4, 2019

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Eric Auger

author
#0 lore raw

v2 cover ARM SMMUv3: Fix spurious notification errors and stall with vfio-pci

This series fixes the guest stall observed when attempting to run a guest exposed with a SMMUv3 and a VFIO-PCI device. As a reminder SMMUv3 is not yet integrated with VFIO (the device will not work properly) but this shouldn't prevent the guest from booting.

It also silences some spurious translation configuration decoding errors (STE out of span or invalid STE) that may happen on guest IOVA invalidation notifications.

Best Regards

Eric

History:

v1 -> v2:
- Added "memory: Remove unused memory_region_iommu_replay_all()" & "hw/arm/smmuv3: Log a guest error when decoding an invalid STE"
- do not attempt to implement replay Cb but rather remove the call in case it is not needed
- explain why we do not remove other log messages on config decoding

Eric Auger (6):
memory: Remove unused memory_region_iommu_replay_all()
memory: Add IOMMU_ATTR_VFIO_NESTED IOMMU memory region attribute hw/vfio/common: Do not replay IOMMU mappings in nested case hw/arm/smmuv3: Advertise VFIO_NESTED
hw/arm/smmuv3: Log a guest error when decoding an invalid STE hw/arm/smmuv3: Remove spurious error messages on IOVA invalidations

 hw/arm/smmuv3-internal.h |  1 +
 hw/arm/smmuv3.c          | 26 ++++++++++++++++++++------
 hw/vfio/common.c         |  7 ++++++-
 include/exec/memory.h    | 13 ++-----------
 memory.c                 |  9 ---------
 5 files changed, 29 insertions(+), 27 deletions(-)
Signature · 1 line
2.20.1
author
#1 lore raw

PATCH v2 1/6 memory: Remove unused memory_region_iommu_replay_all()

memory_region_iommu_replay_all is not used. Remove it.

---
 include/exec/memory.h | 10 ----------
 memory.c              |  9 ---------
 2 files changed, 19 deletions(-)
include/exec/memory.h +0 -10
index e6140e8a04..bdd76653a8 100644
@@ -1076,16 +1076,6 @@ void memory_region_register_iommu_notifier(MemoryRegion *mr,
  */
 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n);
 
-/**
- * memory_region_iommu_replay_all: replay existing IOMMU translations
- * to all the notifiers registered.
- *
- * Note: this is not related to record-and-replay functionality.
- *
- * @iommu_mr: the memory region to observe
- */
-void memory_region_iommu_replay_all(IOMMUMemoryRegion *iommu_mr);
-
 /**
  * memory_region_unregister_iommu_notifier: unregister a notifier for
  * changes to IOMMU translation entries.
memory.c +0 -9
index 0a089a73ae..290a1493ef 100644
@@ -1910,15 +1910,6 @@ void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
     }
 }
 
-void memory_region_iommu_replay_all(IOMMUMemoryRegion *iommu_mr)
-{
-    IOMMUNotifier *notifier;
-
-    IOMMU_NOTIFIER_FOREACH(notifier, iommu_mr) {
-        memory_region_iommu_replay(iommu_mr, notifier);
-    }
-}
-
 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
                                              IOMMUNotifier *n)
 {
Signature · 1 line
2.20.1
Reviewed-by on v2 1/6
#2 lore raw

Re: [Qemu-arm] [Qemu-devel] [PATCH v2 1/6] memory: Remove unused memory_region_iommu_replay_all()

Quoted 41 lines · Eric Auger

memory_region_iommu_replay_all is not used. Remove it.

---
 include/exec/memory.h | 10 ----------
 memory.c              |  9 ---------
 2 files changed, 19 deletions(-)
include/exec/memory.h +0 -10
index e6140e8a04..bdd76653a8 100644
@@ -1076,16 +1076,6 @@ void memory_region_register_iommu_notifier(MemoryRegion *mr,
  */
 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n);
 
-/**
- * memory_region_iommu_replay_all: replay existing IOMMU translations
- * to all the notifiers registered.
- *
- * Note: this is not related to record-and-replay functionality.
- *
- * @iommu_mr: the memory region to observe
- */
-void memory_region_iommu_replay_all(IOMMUMemoryRegion *iommu_mr);
-
 /**
  * memory_region_unregister_iommu_notifier: unregister a notifier for
  * changes to IOMMU translation entries.
memory.c +0 -2
index 0a089a73ae..290a1493ef 100644
@@ -1910,15 +1910,6 @@ void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
     }
 }
 
-void memory_region_iommu_replay_all(IOMMUMemoryRegion *iommu_mr)
-{

memory.c void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n) · line 1915

Quoted 10 lines
memory.c +0 -7
-    IOMMUNotifier *notifier;
-
-    IOMMU_NOTIFIER_FOREACH(notifier, iommu_mr) {
-        memory_region_iommu_replay(iommu_mr, notifier);
-    }
-}
-
 void memory_region_unregister_iommu_notifier(MemoryRegion *mr,
                                              IOMMUNotifier *n)
 {
author
#4 lore raw

PATCH v2 2/6 memory: Add IOMMU_ATTR_VFIO_NESTED IOMMU memory region attribute

In reply to Eric Auger

We introduce a new IOMMU Memory Region attribute, IOMMU_ATTR_VFIO_NESTED that tells whether the virtual IOMMU requires physical nested stages for VFIO integration.

Current Intel virtual IOMMU device supports "Caching Mode" and does not require 2 stages at physical level to be integrated with VFIO. However SMMUv3 does not implement such "caching mode" and requires to use physical stage 1 for VFIO integration.

---
 include/exec/memory.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
include/exec/memory.h +2 -1
index bdd76653a8..dd7ef23f96 100644
@@ -204,7 +204,8 @@ struct MemoryRegionOps {
 };
 
 enum IOMMUMemoryRegionAttr {
-    IOMMU_ATTR_SPAPR_TCE_FD
+    IOMMU_ATTR_SPAPR_TCE_FD,
+    IOMMU_ATTR_VFIO_NESTED,
 };
 
 /**
Signature · 1 line
2.20.1
#5 lore raw

Re: [Qemu-arm] [PATCH v2 2/6] memory: Add IOMMU_ATTR_VFIO_NESTED IOMMU memory region attribute

Quoted 20 lines · Eric Auger

We introduce a new IOMMU Memory Region attribute, IOMMU_ATTR_VFIO_NESTED that tells whether the virtual IOMMU requires physical nested stages for VFIO integration.

Current Intel virtual IOMMU device supports "Caching Mode" and does not require 2 stages at physical level to be integrated with VFIO. However SMMUv3 does not implement such "caching mode" and requires to use physical stage 1 for VFIO integration.

---
 include/exec/memory.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
include/exec/memory.h +0 -0
index bdd76653a8..dd7ef23f96 100644

include/exec/memory.h struct MemoryRegionOps { · line 208

Quoted 6 lines
include/exec/memory.h +2 -1
@@ -204,7 +204,8 @@ struct MemoryRegionOps {
 };
 
 enum IOMMUMemoryRegionAttr {
-    IOMMU_ATTR_SPAPR_TCE_FD
+    IOMMU_ATTR_SPAPR_TCE_FD,
+    IOMMU_ATTR_VFIO_NESTED,

IMHO it'll be better if this patch can be squashed into the first user of the new flag to better clarify itself on why it will be needed (if finally we still would like to have this flag).

Regards,

Signature · 1 line
Peter Xu
author
#6 lore raw

Re: [Qemu-devel] [PATCH v2 2/6] memory: Add IOMMU_ATTR_VFIO_NESTED IOMMU memory region attribute

Hi Peter,

Quoted 23 lines · Peter Xu
Quoted 22 lines · Eric Auger

We introduce a new IOMMU Memory Region attribute, IOMMU_ATTR_VFIO_NESTED that tells whether the virtual IOMMU requires physical nested stages for VFIO integration.

Current Intel virtual IOMMU device supports "Caching Mode" and does not require 2 stages at physical level to be integrated with VFIO. However SMMUv3 does not implement such "caching mode" and requires to use physical stage 1 for VFIO integration.

---
 include/exec/memory.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
include/exec/memory.h +0 -0
index bdd76653a8..dd7ef23f96 100644
@@ -204,7 +204,8 @@ struct MemoryRegionOps {
 };
 
Quoted 8 lines

include/exec/memory.h struct MemoryRegionOps { · line 208

Quoted 4 lines
include/exec/memory.h +2 -1
 enum IOMMUMemoryRegionAttr {
-    IOMMU_ATTR_SPAPR_TCE_FD
+    IOMMU_ATTR_SPAPR_TCE_FD,
+    IOMMU_ATTR_VFIO_NESTED,

IMHO it'll be better if this patch can be squashed into the first user of the new flag to better clarify itself on why it will be needed (if finally we still would like to have this flag).

sure I will squash it.

Nested mode requires important adaptations in the current hw/vfio/common.c code to register specific notifiers: UNMAP, config change, MSI binding notifiers (this one actually uses a MAP notifier by the way). So there we need to recognize an IOMMU works in nested mode one way or another.

Thanks

Eric

Quoted 3 lines

Regards,

author
#7 lore raw

PATCH v2 3/6 hw/vfio/common: Do not replay IOMMU mappings in nested case

In reply to Eric Auger

In nested mode, the stage 1 translation tables are owned by the guest and there is no caching on host side. So there is no need to replay the mappings.

As of today, the SMMUv3 nested mode is not yet implemented and there is no functional VFIO integration without. But keeping the replay call would execute the default implementation of memory_region_iommu_replay and attempt to translate the whole address range, completely stalling qemu. Keeping the MAP/UNMAP notifier registration allows to hit a warning message in the SMMUv3 device that tells the user which VFIO device will not function properly:

"qemu-system-aarch64: -device vfio-pci,host=0000:89:00.0: warning:
SMMUv3 does not support notification on MAP: device vfio-pci will not function properly"

Besides, removing the replay call now allows the guest to boot.

---
 hw/vfio/common.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
hw/vfio/common.c +6 -1
index a859298fda..9ea58df67a 100644
@@ -604,6 +604,7 @@ static void vfio_listener_region_add(MemoryListener *listener,
     if (memory_region_is_iommu(section->mr)) {
         VFIOGuestIOMMU *giommu;
         IOMMUMemoryRegion *iommu_mr = IOMMU_MEMORY_REGION(section->mr);
+        bool nested = false;
         int iommu_idx;
 
         trace_vfio_listener_region_add_iommu(iova, end);
@@ -631,8 +632,12 @@ static void vfio_listener_region_add(MemoryListener *listener,
         QLIST_INSERT_HEAD(&container->giommu_list, giommu, giommu_next);
 
         memory_region_register_iommu_notifier(section->mr, &giommu->n);
-        memory_region_iommu_replay(giommu->iommu, &giommu->n);
 
+        memory_region_iommu_get_attr(iommu_mr, IOMMU_ATTR_VFIO_NESTED,
+                                     (void *)&nested);
+        if (!nested) {
+            memory_region_iommu_replay(iommu_mr, &giommu->n);
+        }
         return;
     }
 
Signature · 1 line
2.20.1
#8 lore raw

Re: [Qemu-arm] [PATCH v2 3/6] hw/vfio/common: Do not replay IOMMU mappings in nested case

Quoted 37 lines · Eric Auger

In nested mode, the stage 1 translation tables are owned by the guest and there is no caching on host side. So there is no need to replay the mappings.

As of today, the SMMUv3 nested mode is not yet implemented and there is no functional VFIO integration without. But keeping the replay call would execute the default implementation of memory_region_iommu_replay and attempt to translate the whole address range, completely stalling qemu. Keeping the MAP/UNMAP notifier registration allows to hit a warning message in the SMMUv3 device that tells the user which VFIO device will not function properly:

"qemu-system-aarch64: -device vfio-pci,host=0000:89:00.0: warning:
SMMUv3 does not support notification on MAP: device vfio-pci will not function properly"

Besides, removing the replay call now allows the guest to boot.

---
 hw/vfio/common.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
hw/vfio/common.c +1 -0
index a859298fda..9ea58df67a 100644
@@ -604,6 +604,7 @@ static void vfio_listener_region_add(MemoryListener *listener,
     if (memory_region_is_iommu(section->mr)) {
         VFIOGuestIOMMU *giommu;
         IOMMUMemoryRegion *iommu_mr = IOMMU_MEMORY_REGION(section->mr);
+        bool nested = false;
         int iommu_idx;
 
         trace_vfio_listener_region_add_iommu(iova, end);

hw/vfio/common.c static void vfio_listener_region_add(MemoryListener *listener, · line 640

Quoted 10 lines
hw/vfio/common.c +5 -1
@@ -631,8 +632,12 @@ static void vfio_listener_region_add(MemoryListener *listener,
         QLIST_INSERT_HEAD(&container->giommu_list, giommu, giommu_next);
 
         memory_region_register_iommu_notifier(section->mr, &giommu->n);
-        memory_region_iommu_replay(giommu->iommu, &giommu->n);
 
+        memory_region_iommu_get_attr(iommu_mr, IOMMU_ATTR_VFIO_NESTED,
+                                     (void *)&nested);
+        if (!nested) {
+            memory_region_iommu_replay(iommu_mr, &giommu->n);
+        }

For nested, do we need these IOMMU notifiers after all?

I'm asking because the no-IOMMU case of vfio_listener_region_add() seems to suite very well for nested page tables to me. For example, vfio does not need to listen to MAP events any more because we'll simply share the guest IOMMU page table to be the 1st level page table of the host SMMU IIUC. And if we have 2nd page table changes (like memory hotplug) then IMHO vfio_listener_region_add() will do this for us as well just like when there's no SMMU.

Another thing is that IOMMU_ATTR_VFIO_NESTED will be the same for all the memory regions, so it also seems a bit awkward to make it per memory region. If you see the other real user of this flag (which is IOMMU_ATTR_SPAPR_TCE_FD) it's per memory region.

Regards,

Signature · 1 line
Peter Xu
author
#9 lore raw

Re: [Qemu-arm] [PATCH v2 3/6] hw/vfio/common: Do not replay IOMMU mappings in nested case

Hi Peter,

Quoted 51 lines · Peter Xu
Quoted 37 lines · Eric Auger

In nested mode, the stage 1 translation tables are owned by the guest and there is no caching on host side. So there is no need to replay the mappings.

As of today, the SMMUv3 nested mode is not yet implemented and there is no functional VFIO integration without. But keeping the replay call would execute the default implementation of memory_region_iommu_replay and attempt to translate the whole address range, completely stalling qemu. Keeping the MAP/UNMAP notifier registration allows to hit a warning message in the SMMUv3 device that tells the user which VFIO device will not function properly:

"qemu-system-aarch64: -device vfio-pci,host=0000:89:00.0: warning:
SMMUv3 does not support notification on MAP: device vfio-pci will not function properly"

Besides, removing the replay call now allows the guest to boot.

---
 hw/vfio/common.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
hw/vfio/common.c +1 -0
index a859298fda..9ea58df67a 100644
@@ -604,6 +604,7 @@ static void vfio_listener_region_add(MemoryListener *listener,
     if (memory_region_is_iommu(section->mr)) {
         VFIOGuestIOMMU *giommu;
         IOMMUMemoryRegion *iommu_mr = IOMMU_MEMORY_REGION(section->mr);
+        bool nested = false;
         int iommu_idx;
 
         trace_vfio_listener_region_add_iommu(iova, end);

hw/vfio/common.c static void vfio_listener_region_add(MemoryListener *listener, · line 640

Quoted 10 lines
hw/vfio/common.c +5 -1
@@ -631,8 +632,12 @@ static void vfio_listener_region_add(MemoryListener *listener,
         QLIST_INSERT_HEAD(&container->giommu_list, giommu, giommu_next);
 
         memory_region_register_iommu_notifier(section->mr, &giommu->n);
-        memory_region_iommu_replay(giommu->iommu, &giommu->n);
 
+        memory_region_iommu_get_attr(iommu_mr, IOMMU_ATTR_VFIO_NESTED,
+                                     (void *)&nested);
+        if (!nested) {
+            memory_region_iommu_replay(iommu_mr, &giommu->n);
+        }

For nested, do we need these IOMMU notifiers after all?

Quoted 5 lines

I'm asking because the no-IOMMU case of vfio_listener_region_add() seems to suite very well for nested page tables to me. For example, vfio does not need to listen to MAP events any more because we'll simply share the guest IOMMU page table to be the 1st level page table of the host SMMU IIUC.

We don't need the MAP notifier but we need the UNMAP notifier: when the guest invalidates an ASID/IOVA we need to propagate this to the physical IOMMU.

As mentioned in the cover letter, at the moment, I still register both MAP/UNMAP notifiers as the MAP notifier registration produces an explicit warning message in the SMMUv3 device. If I remove the registration we will loose this message. I hope this code is just an intermediate state towards the actual nested stage support.

And if we have 2nd page table changes (like

Quoted 2 lines

memory hotplug) then IMHO vfio_listener_region_add() will do this for us as well just like when there's no SMMU.

In the current integration, see [RFC v4 20/27] hw/vfio/common: Setup nested stage mappings (https://patchwork.kernel.org/patch/10962721/) I use a prereg_listener for stage 2 mappings.

Quoted 5 lines

Another thing is that IOMMU_ATTR_VFIO_NESTED will be the same for all the memory regions, so it also seems a bit awkward to make it per memory region. If you see the other real user of this flag (which is IOMMU_ATTR_SPAPR_TCE_FD) it's per memory region.

That's correct all SMMUv3 regions will return this value. But what other API can be used to query IOMMU level attributes?

On the other hand,

Alexey's commit f1334de60b2 ("memory/iommu: Add get_attr()") says:
This adds get_attr() to IOMMUMemoryRegionClass, like iommu_ops::domain_get_attr in the Linux kernel.

and DOMAIN_ATTR_NESTING is part of enum iommu_attr at kernel level.

Thanks

Eric

Quoted 3 lines

Regards,

#10 lore raw

Re: [Qemu-arm] [PATCH v2 3/6] hw/vfio/common: Do not replay IOMMU mappings in nested case

Quoted 1 line · Auger Eric

Hi Peter,

Hi, Eric,

Quoted 53 lines · Peter Xu
Quoted 51 lines · Eric Auger
Quoted 37 lines

In nested mode, the stage 1 translation tables are owned by the guest and there is no caching on host side. So there is no need to replay the mappings.

As of today, the SMMUv3 nested mode is not yet implemented and there is no functional VFIO integration without. But keeping the replay call would execute the default implementation of memory_region_iommu_replay and attempt to translate the whole address range, completely stalling qemu. Keeping the MAP/UNMAP notifier registration allows to hit a warning message in the SMMUv3 device that tells the user which VFIO device will not function properly:

"qemu-system-aarch64: -device vfio-pci,host=0000:89:00.0: warning:
SMMUv3 does not support notification on MAP: device vfio-pci will not function properly"

Besides, removing the replay call now allows the guest to boot.

---
 hw/vfio/common.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
hw/vfio/common.c +1 -0
index a859298fda..9ea58df67a 100644
@@ -604,6 +604,7 @@ static void vfio_listener_region_add(MemoryListener *listener,
     if (memory_region_is_iommu(section->mr)) {
         VFIOGuestIOMMU *giommu;
         IOMMUMemoryRegion *iommu_mr = IOMMU_MEMORY_REGION(section->mr);
+        bool nested = false;
         int iommu_idx;
 
         trace_vfio_listener_region_add_iommu(iova, end);

hw/vfio/common.c static void vfio_listener_region_add(MemoryListener *listener, · line 640

Quoted 10 lines
hw/vfio/common.c +5 -1
@@ -631,8 +632,12 @@ static void vfio_listener_region_add(MemoryListener *listener,
         QLIST_INSERT_HEAD(&container->giommu_list, giommu, giommu_next);
 
         memory_region_register_iommu_notifier(section->mr, &giommu->n);
-        memory_region_iommu_replay(giommu->iommu, &giommu->n);
 
+        memory_region_iommu_get_attr(iommu_mr, IOMMU_ATTR_VFIO_NESTED,
+                                     (void *)&nested);
+        if (!nested) {
+            memory_region_iommu_replay(iommu_mr, &giommu->n);
+        }

For nested, do we need these IOMMU notifiers after all?

Quoted 8 lines
Quoted 5 lines

I'm asking because the no-IOMMU case of vfio_listener_region_add() seems to suite very well for nested page tables to me. For example, vfio does not need to listen to MAP events any more because we'll simply share the guest IOMMU page table to be the 1st level page table of the host SMMU IIUC.

We don't need the MAP notifier but we need the UNMAP notifier: when the guest invalidates an ASID/IOVA we need to propagate this to the physical IOMMU.

Indeed we need the unmaps. However I've got a major confusion here:
With nested mode, we should need unmap events for the 1st level rather than the 2nd level, am I right? I mean, the invalidate request should be a GVA range rather than GPA range? While here IIUC vfio_listener_region_add() should be working on GPA address space.

I don't know SMMU enough, but for Intel there should have two different kinds of invalidation messages. Currently because we still don't support nested on Intel so the 1st level invalidation is still not yet implemented (VTD_INV_DESC_PIOTLB). And IMHO if it is going to be implemented, I think it should be different comparing to current IOMMU_NOTIFIER_UNMAP in that it should not even need to bind to a memory region, and modules like vfio should simply deliver that exact message to the host IOMMU driver for the GVA range to be invalidated, just like what it will do with the root pointer of guest 1st level page table.

Quoted 6 lines

As mentioned in the cover letter, at the moment, I still register both MAP/UNMAP notifiers as the MAP notifier registration produces an explicit warning message in the SMMUv3 device. If I remove the registration we will loose this message. I hope this code is just an intermediate state towards the actual nested stage support.

I didn't see it in the cover letter. Would you please provide a link to the message?

Quoted 17 lines

And if we have 2nd page table changes (like

Quoted 2 lines

memory hotplug) then IMHO vfio_listener_region_add() will do this for us as well just like when there's no SMMU.

In the current integration, see [RFC v4 20/27] hw/vfio/common: Setup nested stage mappings (https://patchwork.kernel.org/patch/10962721/) I use a prereg_listener for stage 2 mappings.

Quoted 5 lines

Another thing is that IOMMU_ATTR_VFIO_NESTED will be the same for all the memory regions, so it also seems a bit awkward to make it per memory region. If you see the other real user of this flag (which is IOMMU_ATTR_SPAPR_TCE_FD) it's per memory region.

That's correct all SMMUv3 regions will return this value. But what other API can be used to query IOMMU level attributes?

Quoted 7 lines

On the other hand,

Alexey's commit f1334de60b2 ("memory/iommu: Add get_attr()") says:
This adds get_attr() to IOMMUMemoryRegionClass, like iommu_ops::domain_get_attr in the Linux kernel.

and DOMAIN_ATTR_NESTING is part of enum iommu_attr at kernel level.

Yeah it's fine to me.

Thanks,

Signature · 1 line
Peter Xu
author
#11 lore raw

Re: [Qemu-arm] [PATCH v2 3/6] hw/vfio/common: Do not replay IOMMU mappings in nested case

Hi Peter,

Quoted 63 lines · Peter Xu
Quoted 1 line · Auger Eric

Hi Peter,

Hi, Eric,

Quoted 58 lines · Peter Xu
Quoted 56 lines · Eric Auger
Quoted 37 lines

In nested mode, the stage 1 translation tables are owned by the guest and there is no caching on host side. So there is no need to replay the mappings.

As of today, the SMMUv3 nested mode is not yet implemented and there is no functional VFIO integration without. But keeping the replay call would execute the default implementation of memory_region_iommu_replay and attempt to translate the whole address range, completely stalling qemu. Keeping the MAP/UNMAP notifier registration allows to hit a warning message in the SMMUv3 device that tells the user which VFIO device will not function properly:

"qemu-system-aarch64: -device vfio-pci,host=0000:89:00.0: warning:
SMMUv3 does not support notification on MAP: device vfio-pci will not function properly"

Besides, removing the replay call now allows the guest to boot.

---
 hw/vfio/common.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
hw/vfio/common.c +1 -0
index a859298fda..9ea58df67a 100644
@@ -604,6 +604,7 @@ static void vfio_listener_region_add(MemoryListener *listener,
     if (memory_region_is_iommu(section->mr)) {
         VFIOGuestIOMMU *giommu;
         IOMMUMemoryRegion *iommu_mr = IOMMU_MEMORY_REGION(section->mr);
+        bool nested = false;
         int iommu_idx;
 
         trace_vfio_listener_region_add_iommu(iova, end);

hw/vfio/common.c static void vfio_listener_region_add(MemoryListener *listener, · line 640

Quoted 10 lines
hw/vfio/common.c +5 -1
@@ -631,8 +632,12 @@ static void vfio_listener_region_add(MemoryListener *listener,
         QLIST_INSERT_HEAD(&container->giommu_list, giommu, giommu_next);
 
         memory_region_register_iommu_notifier(section->mr, &giommu->n);
-        memory_region_iommu_replay(giommu->iommu, &giommu->n);
 
+        memory_region_iommu_get_attr(iommu_mr, IOMMU_ATTR_VFIO_NESTED,
+                                     (void *)&nested);
+        if (!nested) {
+            memory_region_iommu_replay(iommu_mr, &giommu->n);
+        }

For nested, do we need these IOMMU notifiers after all?

I'm asking because the no-IOMMU case of vfio_listener_region_add() seems to suite very well for nested page tables to me. For example, vfio does not need to listen to MAP events any more because we'll simply share the guest IOMMU page table to be the 1st level page table of the host SMMU IIUC.

Quoted 7 lines
Quoted 3 lines

We don't need the MAP notifier but we need the UNMAP notifier: when the guest invalidates an ASID/IOVA we need to propagate this to the physical IOMMU.

Indeed we need the unmaps. However I've got a major confusion here:
With nested mode, we should need unmap events for the 1st level rather than the 2nd level, am I right?

yes that's correct

I mean, the invalidate request should

Quoted 2 lines

be a GVA range rather than GPA range? While here IIUC vfio_listener_region_add() should be working on GPA address space.

Sorry I don't get your point. My understanding is in vfio_listener_region_add() we detect the addition of an IOMMU MR and init a notifier that covers the input AS it translates (GVA). When the guest sends an IOTLB invalidation on its first stage, this is trapped, we notify the UNMAP notifier and this eventually produces a stage1 invalidation at physical level (through VFIO/IOMMU kernel path). This piece is not yet implemented: see below.

Quoted 7 lines

I don't know SMMU enough, but for Intel there should have two different kinds of invalidation messages. Currently because we still don't support nested on Intel so the 1st level invalidation is still not yet implemented (VTD_INV_DESC_PIOTLB). And IMHO if it is going to be implemented, I think it should be different comparing to current IOMMU_NOTIFIER_UNMAP

Yes the UNMAP notifier implementation is definitively different. It calls a VFIO iotcl that eventually produces a physical IOMMU stage1 invalidation. See ttps://patchwork.kernel.org/patch/10962721/.

Maybe the confusion comes from the fact this patch is *not* an integration for nested SMMUv3 with VFIO. SMMUv3/VFIO still does not work. It just allows the guest to boot by bypassing the replay function. If things are clearer maybe I should simply assert() in case we detect a VFIO device protected by an SMMUv3.

in that it should not even need to bind to a

Quoted 6 lines

memory region, and modules like vfio should simply deliver that exact message to the host IOMMU driver for the GVA range to be invalidated, just like what it will do with the root pointer of guest 1st level page table.

Quoted 1 line
Quoted 8 lines
Quoted 5 lines

As mentioned in the cover letter, at the moment, I still register both MAP/UNMAP notifiers as the MAP notifier registration produces an explicit warning message in the SMMUv3 device. If I remove the registration we will loose this message. I hope this code is just an intermediate state towards the actual nested stage support.

I didn't see it in the cover letter. Would you please provide a link to the message?

Sorry it is in this commit message. Reference to

"qemu-system-aarch64: -device vfio-pci,host=0000:89:00.0: warning:
SMMUv3 does not support notification on MAP: device vfio-pci will not function properly"

Quoted 20 lines
Quoted 20 lines

And if we have 2nd page table changes (like

Quoted 2 lines

memory hotplug) then IMHO vfio_listener_region_add() will do this for us as well just like when there's no SMMU.

In the current integration, see [RFC v4 20/27] hw/vfio/common: Setup nested stage mappings (https://patchwork.kernel.org/patch/10962721/) I use a prereg_listener for stage 2 mappings.

Quoted 5 lines

Another thing is that IOMMU_ATTR_VFIO_NESTED will be the same for all the memory regions, so it also seems a bit awkward to make it per memory region. If you see the other real user of this flag (which is IOMMU_ATTR_SPAPR_TCE_FD) it's per memory region.

That's correct all SMMUv3 regions will return this value. But what other API can be used to query IOMMU level attributes?

On the other hand,

Quoted 7 lines
Quoted 5 lines

Alexey's commit f1334de60b2 ("memory/iommu: Add get_attr()") says:
This adds get_attr() to IOMMUMemoryRegionClass, like iommu_ops::domain_get_attr in the Linux kernel.

and DOMAIN_ATTR_NESTING is part of enum iommu_attr at kernel level.

Yeah it's fine to me.

Thanks

Eric

Quoted 3 lines

Thanks,

#12 lore raw

Re: [Qemu-arm] [PATCH v2 3/6] hw/vfio/common: Do not replay IOMMU mappings in nested case

Quoted 1 line · Auger Eric

Hi Peter,

Hi, Eric,

Quoted 94 lines · Peter Xu
Quoted 63 lines · Auger Eric
Quoted 1 line

Hi Peter,

Hi, Eric,

Quoted 58 lines · Peter Xu
Quoted 56 lines · Eric Auger
Quoted 37 lines

In nested mode, the stage 1 translation tables are owned by the guest and there is no caching on host side. So there is no need to replay the mappings.

As of today, the SMMUv3 nested mode is not yet implemented and there is no functional VFIO integration without. But keeping the replay call would execute the default implementation of memory_region_iommu_replay and attempt to translate the whole address range, completely stalling qemu. Keeping the MAP/UNMAP notifier registration allows to hit a warning message in the SMMUv3 device that tells the user which VFIO device will not function properly:

"qemu-system-aarch64: -device vfio-pci,host=0000:89:00.0: warning:
SMMUv3 does not support notification on MAP: device vfio-pci will not function properly"

Besides, removing the replay call now allows the guest to boot.

---
 hw/vfio/common.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
hw/vfio/common.c +1 -0
index a859298fda..9ea58df67a 100644
@@ -604,6 +604,7 @@ static void vfio_listener_region_add(MemoryListener *listener,
     if (memory_region_is_iommu(section->mr)) {
         VFIOGuestIOMMU *giommu;
         IOMMUMemoryRegion *iommu_mr = IOMMU_MEMORY_REGION(section->mr);
+        bool nested = false;
         int iommu_idx;
 
         trace_vfio_listener_region_add_iommu(iova, end);

hw/vfio/common.c static void vfio_listener_region_add(MemoryListener *listener, · line 640

Quoted 10 lines
hw/vfio/common.c +5 -1
@@ -631,8 +632,12 @@ static void vfio_listener_region_add(MemoryListener *listener,
         QLIST_INSERT_HEAD(&container->giommu_list, giommu, giommu_next);
 
         memory_region_register_iommu_notifier(section->mr, &giommu->n);
-        memory_region_iommu_replay(giommu->iommu, &giommu->n);
 
+        memory_region_iommu_get_attr(iommu_mr, IOMMU_ATTR_VFIO_NESTED,
+                                     (void *)&nested);
+        if (!nested) {
+            memory_region_iommu_replay(iommu_mr, &giommu->n);
+        }

For nested, do we need these IOMMU notifiers after all?

I'm asking because the no-IOMMU case of vfio_listener_region_add() seems to suite very well for nested page tables to me. For example, vfio does not need to listen to MAP events any more because we'll simply share the guest IOMMU page table to be the 1st level page table of the host SMMU IIUC.

Quoted 7 lines
Quoted 3 lines

We don't need the MAP notifier but we need the UNMAP notifier: when the guest invalidates an ASID/IOVA we need to propagate this to the physical IOMMU.

Indeed we need the unmaps. However I've got a major confusion here:
With nested mode, we should need unmap events for the 1st level rather than the 2nd level, am I right?

yes that's correct

I mean, the invalidate request should

Quoted 2 lines

be a GVA range rather than GPA range? While here IIUC vfio_listener_region_add() should be working on GPA address space.

Sorry I don't get your point. My understanding is in vfio_listener_region_add() we detect the addition of an IOMMU MR and init a notifier that covers the input AS it translates (GVA). When the guest sends an IOTLB invalidation on its first stage, this is trapped, we notify the UNMAP notifier and this eventually produces a stage1 invalidation at physical level (through VFIO/IOMMU kernel path). This piece is not yet implemented: see below.

Quoted 7 lines

I don't know SMMU enough, but for Intel there should have two different kinds of invalidation messages. Currently because we still don't support nested on Intel so the 1st level invalidation is still not yet implemented (VTD_INV_DESC_PIOTLB). And IMHO if it is going to be implemented, I think it should be different comparing to current IOMMU_NOTIFIER_UNMAP

Quoted 3 lines

Yes the UNMAP notifier implementation is definitively different. It calls a VFIO iotcl that eventually produces a physical IOMMU stage1 invalidation. See ttps://patchwork.kernel.org/patch/10962721/.

[1]

Quoted 6 lines

Maybe the confusion comes from the fact this patch is *not* an integration for nested SMMUv3 with VFIO. SMMUv3/VFIO still does not work. It just allows the guest to boot by bypassing the replay function. If things are clearer maybe I should simply assert() in case we detect a VFIO device protected by an SMMUv3.

Actually that's also my question to your other patch [1]:

+ if (container->iommu_type == VFIO_TYPE1_NESTING_IOMMU) {

+            /* Config notifier to propagate guest stage 1 config changes */
+            giommu = vfio_alloc_guest_iommu(container, iommu_mr, offset);
+            iommu_config_notifier_init(&giommu->n, vfio_iommu_nested_notify,

+ IOMMU_NOTIFIER_CONFIG_PASID, iommu_idx);
+ QLIST_INSERT_HEAD(&container->giommu_list, giommu, giommu_next);
+ memory_region_register_iommu_notifier(section->mr, &giommu->n);
+

+            /* IOTLB unmap notifier to propagate guest IOTLB invalidations */
+            giommu = vfio_alloc_guest_iommu(container, iommu_mr, offset);
+            iommu_iotlb_notifier_init(&giommu->n, vfio_iommu_unmap_notify,
+                                      IOMMU_NOTIFIER_IOTLB_UNMAP,
+                                      section->offset_within_region,
+                                      int128_get64(llend),
+                                      iommu_idx);

+ QLIST_INSERT_HEAD(&container->giommu_list, giommu, giommu_next);
+ memory_region_register_iommu_notifier(section->mr, &giommu->n);
+ } else {

It'll be fine if we want to do this way finally, but it just let me feel a bit confusing when we register these notifiers with current IOMMU notifiers, because IMHO all these two kinds of events:

- PASID root pointer
- PASID-based IOTLB invalidations

should not bind to any memory region at all, and should not have a concept of "memory range to register". It'll be easier for me to understand if vfio simply registers with IOMMU directly (or maybe registering with the PCI layer could be a bit better from code prospective?) in this case with these two notifiers and there seems to have nothing to do with current memory region framework.

My vague memory was that Liu Yi has had some similar work (e.g., introduce some PCI level notifers and let VFIO registers to that instead for the nested case, though that's for Intel but IMHO it suites too for ARM) but I've totally forgotten the details.

Thanks,

Signature · 1 line
Peter Xu
author
#13 lore raw

PATCH v2 4/6 hw/arm/smmuv3: Advertise VFIO_NESTED

In reply to Eric Auger

Virtual SMMUv3 requires physical nested stages for VFIO integration. Advertise this attribute.

---

---
 hw/arm/smmuv3.c | 12 ++++++++++++
 1 file changed, 12 insertions(+)
hw/arm/smmuv3.c +12 -0
index e96d5beb9a..384c02cb91 100644
@@ -1490,6 +1490,17 @@ static void smmuv3_notify_flag_changed(IOMMUMemoryRegion *iommu,
     }
 }
 
+static int smmuv3_get_attr(IOMMUMemoryRegion *iommu,
+                           enum IOMMUMemoryRegionAttr attr,
+                           void *data)
+{
+    if (attr == IOMMU_ATTR_VFIO_NESTED) {
+        *(bool *) data = true;
+        return 0;
+    }
+    return -EINVAL;
+}
+
 static void smmuv3_iommu_memory_region_class_init(ObjectClass *klass,
                                                   void *data)
 {
@@ -1497,6 +1508,7 @@ static void smmuv3_iommu_memory_region_class_init(ObjectClass *klass,
 
     imrc->translate = smmuv3_translate;
     imrc->notify_flag_changed = smmuv3_notify_flag_changed;
+    imrc->get_attr = smmuv3_get_attr;
 }
 
 static const TypeInfo smmuv3_type_info = {
Signature · 1 line
2.20.1
author
#14 lore raw

PATCH v2 5/6 hw/arm/smmuv3: Log a guest error when decoding an invalid STE

In reply to Eric Auger

Log a guest error when encountering an invalid STE.

---
 hw/arm/smmuv3.c | 1 +
 1 file changed, 1 insertion(+)
hw/arm/smmuv3.c +1 -0
index 384c02cb91..2e270a0f07 100644
@@ -320,6 +320,7 @@ static int decode_ste(SMMUv3State *s, SMMUTransCfg *cfg,
     uint32_t config;
 
     if (!STE_VALID(ste)) {
+        qemu_log_mask(LOG_GUEST_ERROR, "invalid STE\n");
         goto bad_ste;
     }
 
Signature · 1 line
2.20.1
Reviewed-by on v2 5/6
#15 lore raw

Re: [Qemu-devel] [PATCH v2 5/6] hw/arm/smmuv3: Log a guest error when decoding an invalid STE

Quoted 13 lines · Eric Auger

Log a guest error when encountering an invalid STE.

---
 hw/arm/smmuv3.c | 1 +
 1 file changed, 1 insertion(+)
hw/arm/smmuv3.c +0 -0
index 384c02cb91..2e270a0f07 100644

hw/arm/smmuv3.c static int decode_ste(SMMUv3State *s, SMMUTransCfg *cfg, · line 326

Quoted 7 lines
hw/arm/smmuv3.c +1 -0
@@ -320,6 +320,7 @@ static int decode_ste(SMMUv3State *s, SMMUTransCfg *cfg,
     uint32_t config;
 
     if (!STE_VALID(ste)) {
+        qemu_log_mask(LOG_GUEST_ERROR, "invalid STE\n");
         goto bad_ste;
     }
 
author
#16 lore raw

PATCH v2 6/6 hw/arm/smmuv3: Remove spurious error messages on IOVA invalidations

In reply to Eric Auger

An IOVA/ASID invalidation is notified to all IOMMU Memory Regions through smmuv3_inv_notifiers_iova/smmuv3_notify_iova.

When the notification occurs it is possible that some of the PCIe devices associated to the notified regions do not have a valid stream table entry. In that case we output a LOG_GUEST_ERROR message, for example:

invalid sid=<SID> (L1STD span=0)
"smmuv3_notify_iova error decoding the configuration for iommu mr=<MR>

This is unfortunate as the user gets the impression that there are some translation decoding errors whereas there are not.

This patch adds a new field in SMMUEventInfo that tells whether the detection of an invalid STE must lead to an error report. invalid_ste_allowed is set before doing the invalidations and kept unset on actual translation.

The other configuration decoding error messages are kept since if the STE is valid then the rest of the config must be correct.

---

v1 -> v2:
- explain why we keep the other config decoding errors
- handle the new guest error log on STE invalid

---
 hw/arm/smmuv3-internal.h |  1 +
 hw/arm/smmuv3.c          | 15 ++++++++-------
 2 files changed, 9 insertions(+), 7 deletions(-)
hw/arm/smmuv3-internal.h +1 -0
index b160289cd1..d190181ef1 100644
@@ -381,6 +381,7 @@ typedef struct SMMUEventInfo {
     uint32_t sid;
     bool recorded;
     bool record_trans_faults;
+    bool inval_ste_allowed;
     union {
         struct {
             uint32_t ssid;
hw/arm/smmuv3.c +8 -7
index 2e270a0f07..517755aed5 100644
@@ -320,7 +320,9 @@ static int decode_ste(SMMUv3State *s, SMMUTransCfg *cfg,
     uint32_t config;
 
     if (!STE_VALID(ste)) {
-        qemu_log_mask(LOG_GUEST_ERROR, "invalid STE\n");
+        if (!event->inval_ste_allowed) {
+            qemu_log_mask(LOG_GUEST_ERROR, "invalid STE\n");
+        }
         goto bad_ste;
     }
 
@@ -405,7 +407,7 @@ static int smmu_find_ste(SMMUv3State *s, uint32_t sid, STE *ste,
 
         span = L1STD_SPAN(&l1std);
 
-        if (!span) {
+        if (!span && !event->inval_ste_allowed) {
             /* l2ptr is not valid */
             qemu_log_mask(LOG_GUEST_ERROR,
                           "invalid sid=%d (L1STD span=0)\n", sid);
@@ -603,7 +605,9 @@ static IOMMUTLBEntry smmuv3_translate(IOMMUMemoryRegion *mr, hwaddr addr,
     SMMUDevice *sdev = container_of(mr, SMMUDevice, iommu);
     SMMUv3State *s = sdev->smmu;
     uint32_t sid = smmu_get_sid(sdev);
-    SMMUEventInfo event = {.type = SMMU_EVT_NONE, .sid = sid};
+    SMMUEventInfo event = {.type = SMMU_EVT_NONE,
+                           .sid = sid,
+                           .inval_ste_allowed = false};
     SMMUPTWEventInfo ptw_info = {};
     SMMUTranslationStatus status;
     SMMUState *bs = ARM_SMMU(s);
@@ -796,16 +800,13 @@ static void smmuv3_notify_iova(IOMMUMemoryRegion *mr,
                                dma_addr_t iova)
 {
     SMMUDevice *sdev = container_of(mr, SMMUDevice, iommu);
-    SMMUEventInfo event = {};
+    SMMUEventInfo event = {.inval_ste_allowed = true};
     SMMUTransTableInfo *tt;
     SMMUTransCfg *cfg;
     IOMMUTLBEntry entry;
 
     cfg = smmuv3_get_config(sdev, &event);
     if (!cfg) {
-        qemu_log_mask(LOG_GUEST_ERROR,
-                      "%s error decoding the configuration for iommu mr=%s\n",
-                      __func__, mr->parent_obj.name);
         return;
     }
 
Signature · 1 line
2.20.1