Skip to contentSkip to conversation

Support mapping virtio-gpu virgl hostmem blobs using MAP_FIXED API

RFC PATCH Superseded Dmitry Osipenko 3 participants 10 messages Nov 16, 2025 – Nov 17, 2025

v2 of 10 Superseded Latest is v10 →

lore
Share & tools
lore
#0 lore raw

v2 cover Support mapping virtio-gpu virgl hostmem blobs using MAP_FIXED API

Virglrender got a new unstable API that allows mapping host blobs at a given memory address using MAP_FIXED mmap flag [1]. Usage of this new API brings major performance and stability improvement for venus and drm native contexts, see commit message of the RFC patch for details.

Sending early to collect review feeback and have patch prepared by the time new version of libvirglrenderer will be released with the stabilized API.

[1] https://gitlab.freedesktop.org/virgl/virglrenderer/-/merge_requests/1374

Based-on: [email protected]

Changelog:

v2: - Addressed v1 review comments from Akihiko Odaki

- Added patch that removes unnecessary memory_region_set_enabled(), suggested by Akihiko Odaki

Dmitry Osipenko (2):
virtio-gpu: Remove superfluous memory_region_set_enabled()
virtio-gpu: Support mapping hostmem blobs with map_fixed

 hw/display/virtio-gpu-gl.c     | 45 +++++++++++++++++++-
 hw/display/virtio-gpu-virgl.c  | 76 ++++++++++++++++++++++++++++++++--
 include/hw/virtio/virtio-gpu.h |  3 ++
 3 files changed, 119 insertions(+), 5 deletions(-)
Signature · 1 line
2.51.1
#1 lore raw

RFC PATCH v2 1/2 virtio-gpu: Remove superfluous memory_region_set_enabled()

There is no need to explicitly enable/disable memory region when it's added or deleted respectively. Remove superfluous set_enabled() calls for consistency.

---
 hw/display/virtio-gpu-virgl.c | 2 --
 1 file changed, 2 deletions(-)
hw/display/virtio-gpu-virgl.c +0 -2
index 80f71c0b66aa..a6860f63b563 100644
@@ -139,7 +139,6 @@ virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g,
     mr = &vmr->mr;
     memory_region_init_ram_ptr(mr, OBJECT(mr), "blob", size, data);
     memory_region_add_subregion(&b->hostmem, offset, mr);
-    memory_region_set_enabled(mr, true);
 
     /*
      * MR could outlive the resource if MR's reference is held outside of
@@ -201,7 +200,6 @@ virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g,
         b->renderer_blocked++;
 
         /* memory region owns self res->mr object and frees it by itself */
-        memory_region_set_enabled(mr, false);
         memory_region_del_subregion(&b->hostmem, mr);
         object_unparent(OBJECT(mr));
     }
Signature · 1 line
2.51.1
#3 lore raw

RFC PATCH v2 2/2 virtio-gpu: Support mapping hostmem blobs with map_fixed

In reply to Dmitry Osipenko

Support mapping virgl blobs to a fixed location of a hostmem memory region using new virglrenderer MAP_FIXED API.

This new feature closes multiple problems for virtio-gpu on QEMU:

- Having dedicated memory region for each mapped blob works notoriously slow due to QEMU's memory region software design built around RCU that isn't optimized for frequent removal of the regions

- KVM isn't optimized for a frequent slot changes too

- QEMU/KVM has a limit for a total number of created memory regions, crashing QEMU when limit is reached

This patch makes virtio-gpu-gl to pre-create a single anonymous memory region covering whole hostmem area to which blobs will be mapped using the MAP_FIXED API.

Not all virgl resources will support mapping at a fixed memory address. For them, we will continue to create individual nested memory sub-regions. In particular, vrend resources may not have MAP_FIXED capability.

Venus and DRM native contexts will largely benefit from the MAP_FIXED feature in terms of performance and stability improvement.

---
 hw/display/virtio-gpu-gl.c     | 45 ++++++++++++++++++++-
 hw/display/virtio-gpu-virgl.c  | 74 +++++++++++++++++++++++++++++++++-
 include/hw/virtio/virtio-gpu.h |  3 ++
 3 files changed, 119 insertions(+), 3 deletions(-)
hw/display/virtio-gpu-gl.c +44 -1
index b640900fc6f1..23e0dd26c67b 100644
@@ -121,7 +121,12 @@ static void virtio_gpu_gl_reset(VirtIODevice *vdev)
 static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp)
 {
     ERRP_GUARD();
-    VirtIOGPU *g = VIRTIO_GPU(qdev);
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(qdev);
+    VirtIOGPU *g = VIRTIO_GPU(b);
+#if !defined(CONFIG_WIN32)
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
+    void *map;
+#endif
 
 #if HOST_BIG_ENDIAN
     error_setg(errp, "virgl is not supported on bigendian platforms");
@@ -152,6 +157,33 @@ static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp)
 #endif
 
     virtio_gpu_device_realize(qdev, errp);
+
+    /*
+     * Check whether virtio_gpu_device_realize() failed.
+     */
+    if (*errp) {
+        return;
+    }
+
+#if !defined(CONFIG_WIN32)
+    if (virtio_gpu_hostmem_enabled(b->conf)) {
+        map = mmap(NULL, b->conf.hostmem, PROT_READ | PROT_WRITE,
+                   MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+        if (map == MAP_FAILED) {
+            error_setg(errp,
+                       "virgl hostmem region could not be initialized: %s",
+                       strerror(errno));
+            return;
+        }
+
+        gl->hostmem_mmap = map;
+        gl->hostmem_background = g_new0(MemoryRegion, 1);
+        memory_region_init_ram_ptr(gl->hostmem_background, NULL,
+                                   "hostmem-background", b->conf.hostmem,
+                                   gl->hostmem_mmap);
+        memory_region_add_subregion(&b->hostmem, 0, gl->hostmem_background);
+    }
+#endif
 }
 
 static const Property virtio_gpu_gl_properties[] = {
@@ -165,6 +197,7 @@ static const Property virtio_gpu_gl_properties[] = {
 
 static void virtio_gpu_gl_device_unrealize(DeviceState *qdev)
 {
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(qdev);
     VirtIOGPU *g = VIRTIO_GPU(qdev);
     VirtIOGPUGL *gl = VIRTIO_GPU_GL(qdev);
 
@@ -187,6 +220,16 @@ static void virtio_gpu_gl_device_unrealize(DeviceState *qdev)
     gl->renderer_state = RS_START;
 
     g_array_unref(g->capset_ids);
+
+    /*
+     * It is not guaranteed that the memory region will be finalized
+     * immediately with memory_region_del_subregion(), there can be
+     * a remaining reference to gl->hostmem_mmap. VirtIO-GPU is not
+     * hotpluggable, hence no need to worry about the leaked mapping.
+     */
+    if (gl->hostmem_background) {
+        memory_region_del_subregion(&b->hostmem, gl->hostmem_background);
+    }
 }
 
 static void virtio_gpu_gl_class_init(ObjectClass *klass, const void *data)
hw/display/virtio-gpu-virgl.c +72 -2

122 lines not on this page · raw message

include/hw/virtio/virtio-gpu.h +3 -0
index 718332284305..bfe85a4e0449 100644
@@ -263,6 +263,9 @@ struct VirtIOGPUGL {
 
     QEMUBH *async_fence_bh;
     QSLIST_HEAD(, virtio_gpu_virgl_context_fence) async_fenceq;
+
+    MemoryRegion *hostmem_background;
+    void *hostmem_mmap;
 };
 
 struct VhostUserGPU {
Signature · 1 line
2.51.1
#4 lore raw

Re: [RFC PATCH v2 2/2] virtio-gpu: Support mapping hostmem blobs with map_fixed

Quoted 54 lines · Dmitry Osipenko

Support mapping virgl blobs to a fixed location of a hostmem memory region using new virglrenderer MAP_FIXED API.

This new feature closes multiple problems for virtio-gpu on QEMU:

- Having dedicated memory region for each mapped blob works notoriously slow due to QEMU's memory region software design built around RCU that isn't optimized for frequent removal of the regions

- KVM isn't optimized for a frequent slot changes too

- QEMU/KVM has a limit for a total number of created memory regions, crashing QEMU when limit is reached

This patch makes virtio-gpu-gl to pre-create a single anonymous memory region covering whole hostmem area to which blobs will be mapped using the MAP_FIXED API.

Not all virgl resources will support mapping at a fixed memory address. For them, we will continue to create individual nested memory sub-regions. In particular, vrend resources may not have MAP_FIXED capability.

Venus and DRM native contexts will largely benefit from the MAP_FIXED feature in terms of performance and stability improvement.

---
  hw/display/virtio-gpu-gl.c     | 45 ++++++++++++++++++++-
  hw/display/virtio-gpu-virgl.c  | 74 +++++++++++++++++++++++++++++++++-
  include/hw/virtio/virtio-gpu.h |  3 ++
  3 files changed, 119 insertions(+), 3 deletions(-)
hw/display/virtio-gpu-gl.c +6 -1
index b640900fc6f1..23e0dd26c67b 100644
@@ -121,7 +121,12 @@ static void virtio_gpu_gl_reset(VirtIODevice *vdev)
  static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp)
  {
      ERRP_GUARD();
-    VirtIOGPU *g = VIRTIO_GPU(qdev);
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(qdev);
+    VirtIOGPU *g = VIRTIO_GPU(b);
+#if !defined(CONFIG_WIN32)
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
+    void *map;
+#endif
 
  #if HOST_BIG_ENDIAN
      error_setg(errp, "virgl is not supported on bigendian platforms");
@@ -152,6 +157,33 @@ static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp)
  #endif
 
      virtio_gpu_device_realize(qdev, errp);

hw/display/virtio-gpu-gl.c static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp) · line 169

Quoted 10 lines
hw/display/virtio-gpu-gl.c +10 -0
+
+    /*
+     * Check whether virtio_gpu_device_realize() failed.
+     */
+    if (*errp) {
+        return;
+    }
+
+#if !defined(CONFIG_WIN32)
+    if (virtio_gpu_hostmem_enabled(b->conf)) {

hostmem should be kept enabled for Windows. I don't have any reason to believe hostmem does not work on Windows.

hw/display/virtio-gpu-gl.c static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp) · line 171

Quoted 2 lines
hw/display/virtio-gpu-gl.c +2 -0
+        map = mmap(NULL, b->conf.hostmem, PROT_READ | PROT_WRITE,
+                   MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

Use qemu_ram_mmap() that installs a guard page.

hw/display/virtio-gpu-gl.c static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp) · line 180

Quoted 9 lines
hw/display/virtio-gpu-gl.c +9 -0
+        if (map == MAP_FAILED) {
+            error_setg(errp,
+                       "virgl hostmem region could not be initialized: %s",
+                       strerror(errno));
+            return;
+        }
+
+        gl->hostmem_mmap = map;
+        gl->hostmem_background = g_new0(MemoryRegion, 1);

Embed hostmem_background to VirtIOGPUGL and avoid having g_new0(), just like hostmem of VirtIOGPUBase.

Quoted 21 lines
hw/display/virtio-gpu-gl.c +7 -0
+        memory_region_init_ram_ptr(gl->hostmem_background, NULL,
+                                   "hostmem-background", b->conf.hostmem,
+                                   gl->hostmem_mmap);
+        memory_region_add_subregion(&b->hostmem, 0, gl->hostmem_background);
+    }
+#endif
  }
 
  static const Property virtio_gpu_gl_properties[] = {
@@ -165,6 +197,7 @@ static const Property virtio_gpu_gl_properties[] = {
 
  static void virtio_gpu_gl_device_unrealize(DeviceState *qdev)
  {
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(qdev);
      VirtIOGPU *g = VIRTIO_GPU(qdev);
      VirtIOGPUGL *gl = VIRTIO_GPU_GL(qdev);
 
@@ -187,6 +220,16 @@ static void virtio_gpu_gl_device_unrealize(DeviceState *qdev)
      gl->renderer_state = RS_START;
 
      g_array_unref(g->capset_ids);

hw/display/virtio-gpu-gl.c static void virtio_gpu_gl_device_unrealize(DeviceState *qdev) · line 232

Quoted 10 lines
hw/display/virtio-gpu-gl.c +10 -0
+
+    /*
+     * It is not guaranteed that the memory region will be finalized
+     * immediately with memory_region_del_subregion(), there can be
+     * a remaining reference to gl->hostmem_mmap. VirtIO-GPU is not
+     * hotpluggable, hence no need to worry about the leaked mapping.
+     */
+    if (gl->hostmem_background) {
+        memory_region_del_subregion(&b->hostmem, gl->hostmem_background);
+    }

This memory_region_del_subregion() is unnecessary because b->hostmem and gl->hostmem_background belong to the same device and will be gone at the same time.

Quoted 8 lines
hw/display/virtio-gpu-gl.c +0 -0
  }
 
  static void virtio_gpu_gl_class_init(ObjectClass *klass, const void *data)
hw/display/virtio-gpu-virgl.c +0 -0
index a6860f63b563..5a663ad2acfa 100644

hw/display/virtio-gpu-virgl.c line 46

Quoted 6 lines
hw/display/virtio-gpu-virgl.c +3 -0
@@ -41,9 +41,14 @@
       VIRGL_VERSION_MICRO >= (micro))
  #endif
 
+#if VIRGL_CHECK_VERSION(1, 2, 1) && !defined(CONFIG_WIN32)
+    #define VIRGL_HAS_MAP_FIXED
+#endif

Simpler:

#define VIRGL_HAS_MAP_FIXED VIRGL_CHECK_VERSION(1, 2, 1) && !IS_ENABLED(CONFIG_WIN32)

And you can use it like:
#if VIRGL_HAS_MAP_FIXED

Quoted 25 lines
hw/display/virtio-gpu-virgl.c +8 -0
+
  struct virtio_gpu_virgl_resource {
      struct virtio_gpu_simple_resource base;
      MemoryRegion *mr;
+    void *map_fixed;
  };
 
  static struct virtio_gpu_virgl_resource *
@@ -116,6 +121,7 @@ virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g,
  {
      struct virtio_gpu_virgl_hostmem_region *vmr;
      VirtIOGPUBase *b = VIRTIO_GPU_BASE(g);
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
      MemoryRegion *mr;
      uint64_t size;
      void *data;
@@ -126,6 +132,41 @@ virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g,
          return -EOPNOTSUPP;
      }
 
+#ifdef VIRGL_HAS_MAP_FIXED
+    /*
+     * virgl_renderer_resource_map_fixed() allows to create multiple
+     * mappings of the same resource, while virgl_renderer_resource_map()
+     * not. Don't allow mapping same resource twice.

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g, · line 149

Quoted 10 lines
hw/display/virtio-gpu-virgl.c +10 -0
+     */
+    if (res->map_fixed || res->mr) {
+        qemu_log_mask(LOG_GUEST_ERROR,
+                      "%s: failed to map(fixed) virgl resource: already mapped\n",
+                      __func__);
+        return -EBUSY;
+    }
+
+    ret = virgl_renderer_resource_map_fixed(res->base.resource_id,
+                                            gl->hostmem_mmap + offset);

I think you need a boundary check here.

Quoted 24 lines
hw/display/virtio-gpu-virgl.c +20 -0
+    switch (ret) {
+    case 0:
+        res->map_fixed = gl->hostmem_mmap + offset;
+        return 0;
+
+    case -EOPNOTSUPP:
+        /*
+         * -MAP_FIXED is unsupported by this resource.
+         * Mapping falls back to a blob subregion method in that case.
+         */
+        break;
+
+    default:
+        qemu_log_mask(LOG_GUEST_ERROR,
+                      "%s: failed to map(fixed) virgl resource: %s\n",
+                      __func__, strerror(-ret));
+        return ret;
+    }
+#endif
+
      ret = virgl_renderer_resource_map(res->base.resource_id, &data, &size);
      if (ret) {
          qemu_log_mask(LOG_GUEST_ERROR, "%s: failed to map virgl resource: %s\n",

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g, · line 182

Quoted 5 lines
hw/display/virtio-gpu-virgl.c +1 -1
@@ -138,7 +179,7 @@ virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g,
 
      mr = &vmr->mr;
      memory_region_init_ram_ptr(mr, OBJECT(mr), "blob", size, data);
-    memory_region_add_subregion(&b->hostmem, offset, mr);
+    memory_region_add_subregion_overlap(gl->hostmem_background, offset, mr, 1);

This should be:
memory_region_add_subregion_overlap(&b->hostmem, offset, mr, 1);

Quoted 13 lines
hw/display/virtio-gpu-virgl.c +3 -0
      /*
       * MR could outlive the resource if MR's reference is held outside of
@@ -162,9 +203,28 @@ virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g,
  {
      struct virtio_gpu_virgl_hostmem_region *vmr;
      VirtIOGPUBase *b = VIRTIO_GPU_BASE(g);
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
      MemoryRegion *mr = res->mr;
      int ret;
 
+#ifdef VIRGL_HAS_MAP_FIXED
+    if (res->map_fixed) {

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g, · line 221

Quoted 10 lines
hw/display/virtio-gpu-virgl.c +10 -0
+        res->map_fixed = mmap(res->map_fixed, res->base.blob_size,
+                              PROT_READ | PROT_WRITE,
+                              MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED,
+                              -1, 0);
+        if (res->map_fixed == MAP_FAILED) {
+            ret = -errno;
+            qemu_log_mask(LOG_GUEST_ERROR,
+                          "%s: failed to unmap(fixed) virgl resource: %s\n",
+                          __func__, strerror(-ret));
+            return ret;

This code path leaves the blob mapped but overwrites res->map_fixed with MAP_FAILED.

Regards,
Akihiko Odaki

Quoted 49 lines
hw/display/virtio-gpu-virgl.c +17 -1
+        }
+
+        res->map_fixed = NULL;
+    }
+#endif
+
      if (!mr) {
          return 0;
      }
@@ -200,7 +260,7 @@ virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g,
          b->renderer_blocked++;
 
          /* memory region owns self res->mr object and frees it by itself */
-        memory_region_del_subregion(&b->hostmem, mr);
+        memory_region_del_subregion(gl->hostmem_background, mr);
          object_unparent(OBJECT(mr));
      }
 
@@ -1265,6 +1325,16 @@ void virtio_gpu_virgl_reset_scanout(VirtIOGPU *g)
 
  void virtio_gpu_virgl_reset(VirtIOGPU *g)
  {
+#ifdef VIRGL_HAS_MAP_FIXED
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(g);
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
+
+    if (gl->hostmem_mmap &&
+        mmap(gl->hostmem_mmap, b->conf.hostmem, PROT_READ | PROT_WRITE,
+             MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0) == MAP_FAILED) {
+        error_report("failed to reset virgl hostmem: %s", strerror(errno));
+    }
+#endif
      virgl_renderer_reset();
 
      virtio_gpu_virgl_reset_async_fences(g);
include/hw/virtio/virtio-gpu.h +3 -0
index 718332284305..bfe85a4e0449 100644
@@ -263,6 +263,9 @@ struct VirtIOGPUGL {
 
      QEMUBH *async_fence_bh;
      QSLIST_HEAD(, virtio_gpu_virgl_context_fence) async_fenceq;
+
+    MemoryRegion *hostmem_background;
+    void *hostmem_mmap;
  };
 
  struct VhostUserGPU {
#5 lore raw

Re: [RFC PATCH v2 2/2] virtio-gpu: Support mapping hostmem blobs with map_fixed

Quoted 261 lines · Akihiko Odaki
Quoted 53 lines · Dmitry Osipenko

Support mapping virgl blobs to a fixed location of a hostmem memory region using new virglrenderer MAP_FIXED API.

This new feature closes multiple problems for virtio-gpu on QEMU:

- Having dedicated memory region for each mapped blob works notoriously slow due to QEMU's memory region software design built around RCU that isn't optimized for frequent removal of the regions

- KVM isn't optimized for a frequent slot changes too

- QEMU/KVM has a limit for a total number of created memory regions, crashing QEMU when limit is reached

This patch makes virtio-gpu-gl to pre-create a single anonymous memory region covering whole hostmem area to which blobs will be mapped using the MAP_FIXED API.

Not all virgl resources will support mapping at a fixed memory address. For them, we will continue to create individual nested memory sub-regions. In particular, vrend resources may not have MAP_FIXED capability.

Venus and DRM native contexts will largely benefit from the MAP_FIXED feature in terms of performance and stability improvement.

---
  hw/display/virtio-gpu-gl.c     | 45 ++++++++++++++++++++- hw/display/virtio-gpu-virgl.c  | 74 +++++++++++++++++++++++++++++++++- include/hw/virtio/virtio-gpu.h |  3 ++

  3 files changed, 119 insertions(+), 3 deletions(-)
hw/display/virtio-gpu-gl.c +6 -1
index b640900fc6f1..23e0dd26c67b 100644
@@ -121,7 +121,12 @@ static void virtio_gpu_gl_reset(VirtIODevice *vdev)
   static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp)
   {
       ERRP_GUARD();
-    VirtIOGPU *g = VIRTIO_GPU(qdev);
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(qdev);
+    VirtIOGPU *g = VIRTIO_GPU(b);
+#if !defined(CONFIG_WIN32)
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
+    void *map;
+#endif
   #if HOST_BIG_ENDIAN
       error_setg(errp, "virgl is not supported on bigendian platforms");
@@ -152,6 +157,33 @@ static void 
 virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp)
   #endif
       virtio_gpu_device_realize(qdev, errp);

hw/display/virtio-gpu-gl.c static void · line 169

Quoted 10 lines
hw/display/virtio-gpu-gl.c +10 -0
+
+    /*
+     * Check whether virtio_gpu_device_realize() failed.
+     */
+    if (*errp) {
+        return;
+    }
+
+#if !defined(CONFIG_WIN32)
+    if (virtio_gpu_hostmem_enabled(b->conf)) {

hostmem should be kept enabled for Windows. I don't have any reason to believe hostmem does not work on Windows.

hw/display/virtio-gpu-gl.c static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp) · line 171

Quoted 2 lines
hw/display/virtio-gpu-gl.c +2 -0
+        map = mmap(NULL, b->conf.hostmem, PROT_READ | PROT_WRITE,
+                   MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

Use qemu_ram_mmap() that installs a guard page.

hw/display/virtio-gpu-gl.c static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp) · line 180

Quoted 10 lines
hw/display/virtio-gpu-gl.c +9 -0
+        if (map == MAP_FAILED) {
+            error_setg(errp,
+                       "virgl hostmem region could not be 
 initialized: %s",
+                       strerror(errno));
+            return;
+        }
+
+        gl->hostmem_mmap = map;
+        gl->hostmem_background = g_new0(MemoryRegion, 1);

Embed hostmem_background to VirtIOGPUGL and avoid having g_new0(), just like hostmem of VirtIOGPUBase.

Quoted 21 lines
hw/display/virtio-gpu-gl.c +6 -0
+        memory_region_init_ram_ptr(gl->hostmem_background, NULL,
+                                   "hostmem-background", b- 
 >conf.hostmem,
+                                   gl->hostmem_mmap);
+        memory_region_add_subregion(&b->hostmem, 0, gl- 
 >hostmem_background);
+    }
+#endif

  }
  static const Property virtio_gpu_gl_properties[] = {

hw/display/virtio-gpu-gl.c +2 -0
@@ -165,6 +197,7 @@ static const Property virtio_gpu_gl_properties[] = {
   static void virtio_gpu_gl_device_unrealize(DeviceState *qdev)
   {
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(qdev);
       VirtIOGPU *g = VIRTIO_GPU(qdev);
       VirtIOGPUGL *gl = VIRTIO_GPU_GL(qdev);
@@ -187,6 +220,16 @@ static void 
 virtio_gpu_gl_device_unrealize(DeviceState *qdev)
       gl->renderer_state = RS_START;
       g_array_unref(g->capset_ids);
+

hw/display/virtio-gpu-gl.c static void · line 233

Quoted 10 lines
hw/display/virtio-gpu-gl.c +9 -0
+    /*
+     * It is not guaranteed that the memory region will be finalized
+     * immediately with memory_region_del_subregion(), there can be
+     * a remaining reference to gl->hostmem_mmap. VirtIO-GPU is not
+     * hotpluggable, hence no need to worry about the leaked mapping.
+     */
+    if (gl->hostmem_background) {
+        memory_region_del_subregion(&b->hostmem, gl- 
 >hostmem_background);
+    }

This memory_region_del_subregion() is unnecessary because b->hostmem and gl->hostmem_background belong to the same device and will be gone at the same time.

hw/display/virtio-gpu-virgl.c line 45

Quoted 12 lines
hw/display/virtio-gpu-gl.c +0 -0
   }
   static void virtio_gpu_gl_class_init(ObjectClass *klass, const void *data)
hw/display/virtio-gpu-virgl.c +3 -0
index a6860f63b563..5a663ad2acfa 100644
@@ -41,9 +41,14 @@
        VIRGL_VERSION_MICRO >= (micro))
   #endif
+#if VIRGL_CHECK_VERSION(1, 2, 1) && !defined(CONFIG_WIN32)
+    #define VIRGL_HAS_MAP_FIXED
+#endif

Simpler:

#define VIRGL_HAS_MAP_FIXED VIRGL_CHECK_VERSION(1, 2, 1) && ! IS_ENABLED(CONFIG_WIN32)

And you can use it like:
#if VIRGL_HAS_MAP_FIXED

Quoted 25 lines
hw/display/virtio-gpu-virgl.c +2 -0
+
   struct virtio_gpu_virgl_resource {
       struct virtio_gpu_simple_resource base;
       MemoryRegion *mr;
+    void *map_fixed;
   };

  static struct virtio_gpu_virgl_resource *

hw/display/virtio-gpu-virgl.c +7 -0
@@ -116,6 +121,7 @@ virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g,
   {
       struct virtio_gpu_virgl_hostmem_region *vmr;
       VirtIOGPUBase *b = VIRTIO_GPU_BASE(g);
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
       MemoryRegion *mr;
       uint64_t size;
       void *data;
@@ -126,6 +132,41 @@ virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g,
           return -EOPNOTSUPP;
       }
+#ifdef VIRGL_HAS_MAP_FIXED
+    /*
+     * virgl_renderer_resource_map_fixed() allows to create multiple
+     * mappings of the same resource, while 
 virgl_renderer_resource_map()
+     * not. Don't allow mapping same resource twice.
+     */

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g, · line 150

Quoted 10 lines
hw/display/virtio-gpu-virgl.c +9 -0
+    if (res->map_fixed || res->mr) {
+        qemu_log_mask(LOG_GUEST_ERROR,
+                      "%s: failed to map(fixed) virgl resource: 
 already mapped\n",
+                      __func__);
+        return -EBUSY;
+    }
+
+    ret = virgl_renderer_resource_map_fixed(res->base.resource_id,
+                                            gl->hostmem_mmap + offset);

I think you need a boundary check here.

Quoted 24 lines
hw/display/virtio-gpu-virgl.c +20 -0
+    switch (ret) {
+    case 0:
+        res->map_fixed = gl->hostmem_mmap + offset;
+        return 0;
+
+    case -EOPNOTSUPP:
+        /*
+         * -MAP_FIXED is unsupported by this resource.
+         * Mapping falls back to a blob subregion method in that case.
+         */
+        break;
+
+    default:
+        qemu_log_mask(LOG_GUEST_ERROR,
+                      "%s: failed to map(fixed) virgl resource: %s\n",
+                      __func__, strerror(-ret));
+        return ret;
+    }
+#endif
+
       ret = virgl_renderer_resource_map(res->base.resource_id, &data, &size);
       if (ret) {

          qemu_log_mask(LOG_GUEST_ERROR, "%s: failed to map virgl resource: %s\n",

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g, · line 182

Quoted 5 lines
hw/display/virtio-gpu-virgl.c +1 -1
@@ -138,7 +179,7 @@ virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g,
       mr = &vmr->mr;
       memory_region_init_ram_ptr(mr, OBJECT(mr), "blob", size, data);
-    memory_region_add_subregion(&b->hostmem, offset, mr);
+    memory_region_add_subregion_overlap(gl->hostmem_background, 
 offset, mr, 1);

This should be:
memory_region_add_subregion_overlap(&b->hostmem, offset, mr, 1);

Quoted 12 lines
hw/display/virtio-gpu-virgl.c +4 -0
       /*
        * MR could outlive the resource if MR's reference is held outside of
@@ -162,9 +203,28 @@ virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g,
   {
       struct virtio_gpu_virgl_hostmem_region *vmr;
       VirtIOGPUBase *b = VIRTIO_GPU_BASE(g);
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
       MemoryRegion *mr = res->mr;
       int ret;
+#ifdef VIRGL_HAS_MAP_FIXED
+    if (res->map_fixed) {
+        res->map_fixed = mmap(res->map_fixed, res->base.blob_size,

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g, · line 221

Quoted 10 lines
hw/display/virtio-gpu-virgl.c +9 -0
+                              PROT_READ | PROT_WRITE,
+                              MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED,
+                              -1, 0);
+        if (res->map_fixed == MAP_FAILED) {
+            ret = -errno;
+            qemu_log_mask(LOG_GUEST_ERROR,
+                          "%s: failed to unmap(fixed) virgl resource: 
 %s\n",
+                          __func__, strerror(-ret));
+            return ret;

This code path leaves the blob mapped but overwrites res->map_fixed with MAP_FAILED.

Regards,
Akihiko Odaki

Quoted 23 lines
hw/display/virtio-gpu-virgl.c +11 -1
+        }
+
+        res->map_fixed = NULL;
+    }
+#endif
+
       if (!mr) {
           return 0;
       }
@@ -200,7 +260,7 @@ virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g,
           b->renderer_blocked++;
           /* memory region owns self res->mr object and frees it by itself */
-        memory_region_del_subregion(&b->hostmem, mr);
+        memory_region_del_subregion(gl->hostmem_background, mr);
           object_unparent(OBJECT(mr));
       }
@@ -1265,6 +1325,16 @@ void virtio_gpu_virgl_reset_scanout(VirtIOGPU *g)
   void virtio_gpu_virgl_reset(VirtIOGPU *g)
   {
+#ifdef VIRGL_HAS_MAP_FIXED
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(g);
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
+
Quoted 8 lines

hw/display/virtio-gpu-virgl.c void virtio_gpu_virgl_reset_scanout(VirtIOGPU *g) · line 1337

Quoted 8 lines
hw/display/virtio-gpu-virgl.c +6 -0
+    if (gl->hostmem_mmap &&
+        mmap(gl->hostmem_mmap, b->conf.hostmem, PROT_READ | PROT_WRITE,
+             MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0) == 
 MAP_FAILED) {
+        error_report("failed to reset virgl hostmem: %s", 
 strerror(errno));
+    }
+#endif

Please call virgl_cmd_resource_unref() instead to cover both fixed and conventional mappings.

vgc->resource_destroy also needs to be overriden or resources will be gone before reaching virtio_gpu_virgl_reset().

Quoted 15 lines
Quoted 14 lines
hw/display/virtio-gpu-virgl.c +0 -0
       virgl_renderer_reset();
       virtio_gpu_virgl_reset_async_fences(g);
include/hw/virtio/virtio-gpu.h +3 -0
index 718332284305..bfe85a4e0449 100644
@@ -263,6 +263,9 @@ struct VirtIOGPUGL {
       QEMUBH *async_fence_bh;
       QSLIST_HEAD(, virtio_gpu_virgl_context_fence) async_fenceq;
+
+    MemoryRegion *hostmem_background;
+    void *hostmem_mmap;
   };
   struct VhostUserGPU {
#6 lore raw

Re: [RFC PATCH v2 2/2] virtio-gpu: Support mapping hostmem blobs with map_fixed

Sorry for sending a reply again. I keep missing some code...

Quoted 178 lines · Akihiko Odaki
Quoted 177 lines · Akihiko Odaki
Quoted 53 lines · Dmitry Osipenko

Support mapping virgl blobs to a fixed location of a hostmem memory region using new virglrenderer MAP_FIXED API.

This new feature closes multiple problems for virtio-gpu on QEMU:

- Having dedicated memory region for each mapped blob works notoriously slow due to QEMU's memory region software design built around RCU that isn't optimized for frequent removal of the regions

- KVM isn't optimized for a frequent slot changes too

- QEMU/KVM has a limit for a total number of created memory regions, crashing QEMU when limit is reached

This patch makes virtio-gpu-gl to pre-create a single anonymous memory region covering whole hostmem area to which blobs will be mapped using the MAP_FIXED API.

Not all virgl resources will support mapping at a fixed memory address. For them, we will continue to create individual nested memory sub- regions. In particular, vrend resources may not have MAP_FIXED capability.

Venus and DRM native contexts will largely benefit from the MAP_FIXED feature in terms of performance and stability improvement.

---
  hw/display/virtio-gpu-gl.c     | 45 ++++++++++++++++++++- hw/display/virtio-gpu-virgl.c  | 74 +++++++++++++++++++++++++++++++++- include/hw/virtio/virtio-gpu.h |  3 ++

  3 files changed, 119 insertions(+), 3 deletions(-)
hw/display/virtio-gpu-gl.c +6 -1
index b640900fc6f1..23e0dd26c67b 100644
@@ -121,7 +121,12 @@ static void virtio_gpu_gl_reset(VirtIODevice *vdev)
   static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp)
   {
       ERRP_GUARD();
-    VirtIOGPU *g = VIRTIO_GPU(qdev);
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(qdev);
+    VirtIOGPU *g = VIRTIO_GPU(b);
+#if !defined(CONFIG_WIN32)
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
+    void *map;
+#endif
   #if HOST_BIG_ENDIAN
       error_setg(errp, "virgl is not supported on bigendian platforms");
@@ -152,6 +157,33 @@ static void 
 virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp)
   #endif
       virtio_gpu_device_realize(qdev, errp);

hw/display/virtio-gpu-gl.c static void · line 169

Quoted 10 lines
hw/display/virtio-gpu-gl.c +10 -0
+
+    /*
+     * Check whether virtio_gpu_device_realize() failed.
+     */
+    if (*errp) {
+        return;
+    }
+
+#if !defined(CONFIG_WIN32)
+    if (virtio_gpu_hostmem_enabled(b->conf)) {

hostmem should be kept enabled for Windows. I don't have any reason to believe hostmem does not work on Windows.

hw/display/virtio-gpu-gl.c static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp) · line 171

Quoted 2 lines
hw/display/virtio-gpu-gl.c +2 -0
+        map = mmap(NULL, b->conf.hostmem, PROT_READ | PROT_WRITE,
+                   MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);

Use qemu_ram_mmap() that installs a guard page.

hw/display/virtio-gpu-gl.c static void virtio_gpu_gl_device_realize(DeviceState *qdev, Error **errp) · line 180

Quoted 10 lines
hw/display/virtio-gpu-gl.c +9 -0
+        if (map == MAP_FAILED) {
+            error_setg(errp,
+                       "virgl hostmem region could not be 
 initialized: %s",
+                       strerror(errno));
+            return;
+        }
+
+        gl->hostmem_mmap = map;
+        gl->hostmem_background = g_new0(MemoryRegion, 1);

Embed hostmem_background to VirtIOGPUGL and avoid having g_new0(), just like hostmem of VirtIOGPUBase.

Quoted 21 lines
hw/display/virtio-gpu-gl.c +6 -0
+        memory_region_init_ram_ptr(gl->hostmem_background, NULL,
+                                   "hostmem-background", b- 
 >conf.hostmem,
+                                   gl->hostmem_mmap);
+        memory_region_add_subregion(&b->hostmem, 0, gl- 
 >hostmem_background);
+    }
+#endif

  }
  static const Property virtio_gpu_gl_properties[] = {

hw/display/virtio-gpu-gl.c +2 -0
@@ -165,6 +197,7 @@ static const Property virtio_gpu_gl_properties[] = {
   static void virtio_gpu_gl_device_unrealize(DeviceState *qdev)
   {
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(qdev);
       VirtIOGPU *g = VIRTIO_GPU(qdev);
       VirtIOGPUGL *gl = VIRTIO_GPU_GL(qdev);
@@ -187,6 +220,16 @@ static void 
 virtio_gpu_gl_device_unrealize(DeviceState *qdev)
       gl->renderer_state = RS_START;
       g_array_unref(g->capset_ids);
+

hw/display/virtio-gpu-gl.c static void · line 233

Quoted 10 lines
hw/display/virtio-gpu-gl.c +9 -0
+    /*
+     * It is not guaranteed that the memory region will be finalized
+     * immediately with memory_region_del_subregion(), there can be
+     * a remaining reference to gl->hostmem_mmap. VirtIO-GPU is not
+     * hotpluggable, hence no need to worry about the leaked mapping.
+     */
+    if (gl->hostmem_background) {
+        memory_region_del_subregion(&b->hostmem, gl- 
 >hostmem_background);
+    }

This memory_region_del_subregion() is unnecessary because b->hostmem and gl->hostmem_background belong to the same device and will be gone at the same time.

hw/display/virtio-gpu-virgl.c line 45

Quoted 12 lines
hw/display/virtio-gpu-gl.c +0 -0
   }
   static void virtio_gpu_gl_class_init(ObjectClass *klass, const void *data)
hw/display/virtio-gpu-virgl.c +3 -0
index a6860f63b563..5a663ad2acfa 100644
@@ -41,9 +41,14 @@
        VIRGL_VERSION_MICRO >= (micro))
   #endif
+#if VIRGL_CHECK_VERSION(1, 2, 1) && !defined(CONFIG_WIN32)
+    #define VIRGL_HAS_MAP_FIXED
+#endif

Simpler:

#define VIRGL_HAS_MAP_FIXED VIRGL_CHECK_VERSION(1, 2, 1) && ! IS_ENABLED(CONFIG_WIN32)

And you can use it like:
#if VIRGL_HAS_MAP_FIXED

Quoted 25 lines
hw/display/virtio-gpu-virgl.c +2 -0
+
   struct virtio_gpu_virgl_resource {
       struct virtio_gpu_simple_resource base;
       MemoryRegion *mr;
+    void *map_fixed;
   };

  static struct virtio_gpu_virgl_resource *

hw/display/virtio-gpu-virgl.c +7 -0
@@ -116,6 +121,7 @@ virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g,
   {
       struct virtio_gpu_virgl_hostmem_region *vmr;
       VirtIOGPUBase *b = VIRTIO_GPU_BASE(g);
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
       MemoryRegion *mr;
       uint64_t size;
       void *data;
@@ -126,6 +132,41 @@ virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g,
           return -EOPNOTSUPP;
       }
+#ifdef VIRGL_HAS_MAP_FIXED
+    /*
+     * virgl_renderer_resource_map_fixed() allows to create multiple
+     * mappings of the same resource, while 
 virgl_renderer_resource_map()
+     * not. Don't allow mapping same resource twice.
+     */

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g, · line 150

Quoted 10 lines
hw/display/virtio-gpu-virgl.c +9 -0
+    if (res->map_fixed || res->mr) {
+        qemu_log_mask(LOG_GUEST_ERROR,
+                      "%s: failed to map(fixed) virgl resource: 
 already mapped\n",
+                      __func__);
+        return -EBUSY;
+    }
+
+    ret = virgl_renderer_resource_map_fixed(res->base.resource_id,
+                                            gl->hostmem_mmap + offset);

I think you need a boundary check here.

Quoted 8 lines
Quoted 8 lines

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g, · line 157

Quoted 8 lines
hw/display/virtio-gpu-virgl.c +8 -0
+    switch (ret) {
+    case 0:
+        res->map_fixed = gl->hostmem_mmap + offset;
+        return 0;
+
+    case -EOPNOTSUPP:
+        /*
+         * -MAP_FIXED is unsupported by this resource.

s/-MAP_FIXED/MAP_FIXED/

Quoted 38 lines
Quoted 38 lines
Quoted 16 lines
hw/display/virtio-gpu-virgl.c +12 -0
+         * Mapping falls back to a blob subregion method in that case.
+         */
+        break;
+
+    default:
+        qemu_log_mask(LOG_GUEST_ERROR,
+                      "%s: failed to map(fixed) virgl resource: %s\n",
+                      __func__, strerror(-ret));
+        return ret;
+    }
+#endif
+
       ret = virgl_renderer_resource_map(res->base.resource_id, &data, &size);
       if (ret) {

          qemu_log_mask(LOG_GUEST_ERROR, "%s: failed to map virgl resource: %s\n",

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g, · line 182

Quoted 5 lines
hw/display/virtio-gpu-virgl.c +1 -1
@@ -138,7 +179,7 @@ virtio_gpu_virgl_map_resource_blob(VirtIOGPU *g,
       mr = &vmr->mr;
       memory_region_init_ram_ptr(mr, OBJECT(mr), "blob", size, data);
-    memory_region_add_subregion(&b->hostmem, offset, mr);
+    memory_region_add_subregion_overlap(gl->hostmem_background, 
 offset, mr, 1);

This should be:
memory_region_add_subregion_overlap(&b->hostmem, offset, mr, 1);

Quoted 13 lines
hw/display/virtio-gpu-virgl.c +5 -0
       /*
        * MR could outlive the resource if MR's reference is held outside of
@@ -162,9 +203,28 @@ virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g,
   {
       struct virtio_gpu_virgl_hostmem_region *vmr;
       VirtIOGPUBase *b = VIRTIO_GPU_BASE(g);
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
       MemoryRegion *mr = res->mr;
       int ret;
+#ifdef VIRGL_HAS_MAP_FIXED
+    if (res->map_fixed) {
+        res->map_fixed = mmap(res->map_fixed, res->base.blob_size,
+                              PROT_READ | PROT_WRITE,
Quoted 8 lines
Quoted 8 lines

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g, · line 220

Quoted 8 lines
hw/display/virtio-gpu-virgl.c +7 -0
+                              MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED,
+                              -1, 0);
+        if (res->map_fixed == MAP_FAILED) {
+            ret = -errno;
+            qemu_log_mask(LOG_GUEST_ERROR,
+                          "%s: failed to unmap(fixed) virgl 
 resource: %s\n",
+                          __func__, strerror(-ret));

This should be reported with error_report() since this error cannot be triggered by the guest.

Quoted 59 lines
Quoted 30 lines

hw/display/virtio-gpu-virgl.c virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g, · line 221

Quoted 1 line
hw/display/virtio-gpu-virgl.c +1 -0
+            return ret;

This code path leaves the blob mapped but overwrites res->map_fixed with MAP_FAILED.

Regards,
Akihiko Odaki

Quoted 23 lines
hw/display/virtio-gpu-virgl.c +11 -1
+        }
+
+        res->map_fixed = NULL;
+    }
+#endif
+
       if (!mr) {
           return 0;
       }
@@ -200,7 +260,7 @@ virtio_gpu_virgl_unmap_resource_blob(VirtIOGPU *g,
           b->renderer_blocked++;
           /* memory region owns self res->mr object and frees it by itself */
-        memory_region_del_subregion(&b->hostmem, mr);
+        memory_region_del_subregion(gl->hostmem_background, mr);
           object_unparent(OBJECT(mr));
       }
@@ -1265,6 +1325,16 @@ void virtio_gpu_virgl_reset_scanout(VirtIOGPU *g)
   void virtio_gpu_virgl_reset(VirtIOGPU *g)
   {
+#ifdef VIRGL_HAS_MAP_FIXED
+    VirtIOGPUBase *b = VIRTIO_GPU_BASE(g);
+    VirtIOGPUGL *gl = VIRTIO_GPU_GL(g);
+
Quoted 8 lines

hw/display/virtio-gpu-virgl.c void virtio_gpu_virgl_reset_scanout(VirtIOGPU *g) · line 1337

Quoted 8 lines
hw/display/virtio-gpu-virgl.c +6 -0
+    if (gl->hostmem_mmap &&
+        mmap(gl->hostmem_mmap, b->conf.hostmem, PROT_READ | PROT_WRITE,
+             MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED, -1, 0) == 
 MAP_FAILED) {
+        error_report("failed to reset virgl hostmem: %s", 
 strerror(errno));
+    }
+#endif

Please call virgl_cmd_resource_unref() instead to cover both fixed and conventional mappings.

vgc->resource_destroy also needs to be overriden or resources will be gone before reaching virtio_gpu_virgl_reset().

Quoted 15 lines
Quoted 14 lines
hw/display/virtio-gpu-virgl.c +0 -0
       virgl_renderer_reset();
       virtio_gpu_virgl_reset_async_fences(g);
include/hw/virtio/virtio-gpu.h +3 -0
index 718332284305..bfe85a4e0449 100644
@@ -263,6 +263,9 @@ struct VirtIOGPUGL {
       QEMUBH *async_fence_bh;
       QSLIST_HEAD(, virtio_gpu_virgl_context_fence) async_fenceq;
+
+    MemoryRegion *hostmem_background;
+    void *hostmem_mmap;
   };
   struct VhostUserGPU {
#7 lore raw

Re: [RFC PATCH v2 2/2] virtio-gpu: Support mapping hostmem blobs with map_fixed

Quoted 1 line · Akihiko Odaki

Sorry for sending a reply again. I keep missing some code...

Thanks a lot for the review. All comments look reasonable, will address in v3.

Signature · 2 lines
Best regards,
Dmitry
Tested-by on the v2 series
#8 lore raw
In reply to Dmitry Osipenko
Quoted 25 lines · Dmitry Osipenko

Virglrender got a new unstable API that allows mapping host blobs at a given memory address using MAP_FIXED mmap flag [1]. Usage of this new API brings major performance and stability improvement for venus and drm native contexts, see commit message of the RFC patch for details.

Sending early to collect review feeback and have patch prepared by the time new version of libvirglrenderer will be released with the stabilized API.

[1] https://gitlab.freedesktop.org/virgl/virglrenderer/-/merge_requests/1374

Based-on: [email protected]

Changelog:

v2: - Addressed v1 review comments from Akihiko Odaki

- Added patch that removes unnecessary memory_region_set_enabled(), suggested by Akihiko Odaki

Dmitry Osipenko (2):
virtio-gpu: Remove superfluous memory_region_set_enabled()
virtio-gpu: Support mapping hostmem blobs with map_fixed

Quoted 8 lines
 hw/display/virtio-gpu-gl.c     | 45 +++++++++++++++++++-
 hw/display/virtio-gpu-virgl.c  | 76 ++++++++++++++++++++++++++++++++--
 include/hw/virtio/virtio-gpu.h |  3 ++
 3 files changed, 119 insertions(+), 5 deletions(-)

--
2.51.1

Nice work! I'd say Venus loves it soooo much ; )

#9 lore raw
Quoted 32 lines · Yiwei Zhang
Quoted 30 lines · Dmitry Osipenko

Virglrender got a new unstable API that allows mapping host blobs at a given memory address using MAP_FIXED mmap flag [1]. Usage of this new API brings major performance and stability improvement for venus and drm native contexts, see commit message of the RFC patch for details.

Sending early to collect review feeback and have patch prepared by the time new version of libvirglrenderer will be released with the stabilized API.

[1] https://gitlab.freedesktop.org/virgl/virglrenderer/-/merge_requests/1374

Based-on: [email protected]

Changelog:

v2: - Addressed v1 review comments from Akihiko Odaki

- Added patch that removes unnecessary memory_region_set_enabled(), suggested by Akihiko Odaki

Dmitry Osipenko (2):
virtio-gpu: Remove superfluous memory_region_set_enabled()
virtio-gpu: Support mapping hostmem blobs with map_fixed

 hw/display/virtio-gpu-gl.c     | 45 +++++++++++++++++++-
 hw/display/virtio-gpu-virgl.c  | 76 ++++++++++++++++++++++++++++++++--
 include/hw/virtio/virtio-gpu.h |  3 ++
 3 files changed, 119 insertions(+), 5 deletions(-)
Quoted 7 lines
Quoted 3 lines

--
2.51.1

Nice work! I'd say Venus loves it soooo much ; )

Thanks!

Signature · 2 lines
Best regards,
Dmitry