[PATCH 0/6] virtio: Use bounce buffers when VIRTIO_F_IOMMU_PLATFORM set

[PATCH 0/6] virtio: Use bounce buffers when VIRTIO_F_IOMMU_PLATFORM set

[Ying-Chun Liu (PaulLiu)]

These patches will use bounce buffers when VIRTIO_F_IOMMU_PLATFORM feature
is in a virtio device.

This feature can be tested with qemu with -device virtio-iommu-pci.
So that when a -device virtio-blk-pci with iommu_platform=true, it will
uses the bounce buffer instead.

Will Deacon (6):
  virtio: Expose VIRTIO_F_IOMMU_PLATFORM in device features
  virtio: pci: Tear down VQs in virtio_pci_reset()
  virtio: Allocate virtqueue in page-size units
  virtio: Add helper functions to attach/detach vring descriptors
  virtio: Allocate bounce buffers for devices with
    VIRTIO_F_IOMMU_PLATFORM
  virtio: Use bounce buffers when VIRTIO_F_IOMMU_PLATFORM is set

 drivers/virtio/virtio-uclass.c     |   2 +-
 drivers/virtio/virtio_pci_modern.c |  38 +++++------
 drivers/virtio/virtio_ring.c       | 101 ++++++++++++++++++++++++++---
 include/virtio_ring.h              |  21 +++---
 4 files changed, 125 insertions(+), 37 deletions(-)

-- 
2.39.2

[PATCH 1/6] virtio: Expose VIRTIO_F_IOMMU_PLATFORM in device features

[Ying-Chun Liu (PaulLiu)]

From: Will Deacon <willdeacon@google.com>

If we detect the VIRTIO_F_IOMMU_PLATFORM transport feature for a device,
then expose it in the device features.

Signed-off-by: Will Deacon <willdeacon@google.com>
[ Paul: pick from the Android tree. Rebase to the upstream ]
Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
Cc: Bin Meng <bmeng.cn@gmail.com>
Link: https://android.googlesource.com/platform/external/u-boot/+/9693bd26bfcfe77d6a1295a561420e08c5daf019
---
 drivers/virtio/virtio-uclass.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/virtio/virtio-uclass.c b/drivers/virtio/virtio-uclass.c
index de9bc90359..b3fb3dbfad 100644
--- a/drivers/virtio/virtio-uclass.c
+++ b/drivers/virtio/virtio-uclass.c
@@ -336,7 +336,7 @@ static int virtio_uclass_child_pre_probe(struct udevice *vdev)
 	/* Transport features always preserved to pass to finalize_features */
 	for (i = VIRTIO_TRANSPORT_F_START; i < VIRTIO_TRANSPORT_F_END; i++)
 		if ((device_features & (1ULL << i)) &&
-		    (i == VIRTIO_F_VERSION_1))
+		    (i == VIRTIO_F_VERSION_1 || i == VIRTIO_F_IOMMU_PLATFORM))
 			__virtio_set_bit(vdev->parent, i);
 
 	debug("(%s) final negotiated features supported %016llx\n",
-- 
2.39.2

Re: [PATCH 1/6] virtio: Expose VIRTIO_F_IOMMU_PLATFORM in device features

[Simon Glass]

On Thu, 30 Mar 2023 at 03:25, Ying-Chun Liu (PaulLiu)
<paul.liu@linaro.org> wrote:
>
> From: Will Deacon <willdeacon@google.com>
>
> If we detect the VIRTIO_F_IOMMU_PLATFORM transport feature for a device,
> then expose it in the device features.
>
> Signed-off-by: Will Deacon <willdeacon@google.com>
> [ Paul: pick from the Android tree. Rebase to the upstream ]
> Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
> Cc: Bin Meng <bmeng.cn@gmail.com>
> Link: https://android.googlesource.com/platform/external/u-boot/+/9693bd26bfcfe77d6a1295a561420e08c5daf019
> ---
>  drivers/virtio/virtio-uclass.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>

Reviewed-by: Simon Glass <sjg@chromium.org>

[PATCH 2/6] virtio: pci: Tear down VQs in virtio_pci_reset()

[Ying-Chun Liu (PaulLiu)]

From: Will Deacon <willdeacon@google.com>

The pages backing the virtqueues for virtio PCI devices are not freed
on reset, despite the virtqueue structure being freed as part of the
driver '->priv_auto' destruction at ->remove() time.

Call virtio_pci_del_vqs() from virtio_pci_reset() to free the virtqueue
pages before freeing the virtqueue structure itself.

Signed-off-by: Will Deacon <willdeacon@google.com>
[ Paul: pick from the Android tree. Rebase to the upstream ]
Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
Cc: Bin Meng <bmeng.cn@gmail.com>
Link: https://android.googlesource.com/platform/external/u-boot/+/5ed54ccd83cbffd0d8719ce650604b4e44b5b0d8
---
 drivers/virtio/virtio_pci_modern.c | 38 +++++++++++++++---------------
 1 file changed, 19 insertions(+), 19 deletions(-)

diff --git a/drivers/virtio/virtio_pci_modern.c b/drivers/virtio/virtio_pci_modern.c
index cfde4007f5..3cdc2d2d6f 100644
--- a/drivers/virtio/virtio_pci_modern.c
+++ b/drivers/virtio/virtio_pci_modern.c
@@ -218,25 +218,6 @@ static int virtio_pci_set_status(struct udevice *udev, u8 status)
 	return 0;
 }
 
-static int virtio_pci_reset(struct udevice *udev)
-{
-	struct virtio_pci_priv *priv = dev_get_priv(udev);
-
-	/* 0 status means a reset */
-	iowrite8(0, &priv->common->device_status);
-
-	/*
-	 * After writing 0 to device_status, the driver MUST wait for a read
-	 * of device_status to return 0 before reinitializing the device.
-	 * This will flush out the status write, and flush in device writes,
-	 * including MSI-X interrupts, if any.
-	 */
-	while (ioread8(&priv->common->device_status))
-		udelay(1000);
-
-	return 0;
-}
-
 static int virtio_pci_get_features(struct udevice *udev, u64 *features)
 {
 	struct virtio_pci_priv *priv = dev_get_priv(udev);
@@ -363,6 +344,25 @@ static int virtio_pci_find_vqs(struct udevice *udev, unsigned int nvqs,
 	return 0;
 }
 
+static int virtio_pci_reset(struct udevice *udev)
+{
+	struct virtio_pci_priv *priv = dev_get_priv(udev);
+
+	/* 0 status means a reset */
+	iowrite8(0, &priv->common->device_status);
+
+	/*
+	 * After writing 0 to device_status, the driver MUST wait for a read
+	 * of device_status to return 0 before reinitializing the device.
+	 * This will flush out the status write, and flush in device writes,
+	 * including MSI-X interrupts, if any.
+	 */
+	while (ioread8(&priv->common->device_status))
+		udelay(1000);
+
+	return virtio_pci_del_vqs(udev);
+}
+
 static int virtio_pci_notify(struct udevice *udev, struct virtqueue *vq)
 {
 	struct virtio_pci_priv *priv = dev_get_priv(udev);
-- 
2.39.2

Re: [PATCH 2/6] virtio: pci: Tear down VQs in virtio_pci_reset()

[Simon Glass]

On Thu, 30 Mar 2023 at 03:26, Ying-Chun Liu (PaulLiu)
<paul.liu@linaro.org> wrote:
>
> From: Will Deacon <willdeacon@google.com>
>
> The pages backing the virtqueues for virtio PCI devices are not freed
> on reset, despite the virtqueue structure being freed as part of the
> driver '->priv_auto' destruction at ->remove() time.
>
> Call virtio_pci_del_vqs() from virtio_pci_reset() to free the virtqueue
> pages before freeing the virtqueue structure itself.
>
> Signed-off-by: Will Deacon <willdeacon@google.com>
> [ Paul: pick from the Android tree. Rebase to the upstream ]
> Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
> Cc: Bin Meng <bmeng.cn@gmail.com>
> Link: https://android.googlesource.com/platform/external/u-boot/+/5ed54ccd83cbffd0d8719ce650604b4e44b5b0d8
> ---
>  drivers/virtio/virtio_pci_modern.c | 38 +++++++++++++++---------------
>  1 file changed, 19 insertions(+), 19 deletions(-)
>

Reviewed-by: Simon Glass <sjg@chromium.org>

[PATCH 3/6] virtio: Allocate virtqueue in page-size units

[Ying-Chun Liu (PaulLiu)]

From: Will Deacon <willdeacon@google.com>

In preparation for explicit bouncing of virtqueue pages for devices
advertising the VIRTIO_F_IOMMU_PLATFORM feature, introduce a couple
of wrappers around virtqueue allocation and freeing operations,
ensuring that buffers are handled in terms of page-size units.

Signed-off-by: Will Deacon <willdeacon@google.com>
[ Paul: pick from the Android tree. Rebase to the upstream ]
Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
Cc: Bin Meng <bmeng.cn@gmail.com>
Link: https://android.googlesource.com/platform/external/u-boot/+/b4bb5227d4cf4fdfcd8b4e1ff2692d3a54d1482a
---
 drivers/virtio/virtio_ring.c | 24 ++++++++++++++++++++----
 include/virtio_ring.h        | 16 +++++++++-------
 2 files changed, 29 insertions(+), 11 deletions(-)

diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index f71bab7847..5aeb13fd59 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -15,6 +15,17 @@
 #include <virtio_ring.h>
 #include <linux/bug.h>
 #include <linux/compat.h>
+#include <linux/kernel.h>
+
+static void *virtio_alloc_pages(struct udevice *vdev, u32 npages)
+{
+	return memalign(PAGE_SIZE, npages * PAGE_SIZE);
+}
+
+static void virtio_free_pages(struct udevice *vdev, void *ptr, u32 npages)
+{
+	free(ptr);
+}
 
 static unsigned int virtqueue_attach_desc(struct virtqueue *vq, unsigned int i,
 					  struct virtio_sg *sg, u16 flags)
@@ -271,6 +282,8 @@ struct virtqueue *vring_create_virtqueue(unsigned int index, unsigned int num,
 					 unsigned int vring_align,
 					 struct udevice *udev)
 {
+	struct virtio_dev_priv *uc_priv = dev_get_uclass_priv(udev);
+	struct udevice *vdev = uc_priv->vdev;
 	struct virtqueue *vq;
 	void *queue = NULL;
 	struct vring vring;
@@ -283,7 +296,9 @@ struct virtqueue *vring_create_virtqueue(unsigned int index, unsigned int num,
 
 	/* TODO: allocate each queue chunk individually */
 	for (; num && vring_size(num, vring_align) > PAGE_SIZE; num /= 2) {
-		queue = memalign(PAGE_SIZE, vring_size(num, vring_align));
+		size_t sz = vring_size(num, vring_align);
+
+		queue = virtio_alloc_pages(vdev, DIV_ROUND_UP(sz, PAGE_SIZE));
 		if (queue)
 			break;
 	}
@@ -293,7 +308,7 @@ struct virtqueue *vring_create_virtqueue(unsigned int index, unsigned int num,
 
 	if (!queue) {
 		/* Try to get a single page. You are my only hope! */
-		queue = memalign(PAGE_SIZE, vring_size(num, vring_align));
+		queue = virtio_alloc_pages(vdev, 1);
 	}
 	if (!queue)
 		return NULL;
@@ -303,7 +318,7 @@ struct virtqueue *vring_create_virtqueue(unsigned int index, unsigned int num,
 
 	vq = __vring_new_virtqueue(index, vring, udev);
 	if (!vq) {
-		free(queue);
+		virtio_free_pages(vdev, queue, DIV_ROUND_UP(vring.size, PAGE_SIZE));
 		return NULL;
 	}
 	debug("(%s): created vring @ %p for vq @ %p with num %u\n", udev->name,
@@ -314,7 +329,8 @@ struct virtqueue *vring_create_virtqueue(unsigned int index, unsigned int num,
 
 void vring_del_virtqueue(struct virtqueue *vq)
 {
-	free(vq->vring.desc);
+	virtio_free_pages(vq->vdev, vq->vring.desc,
+			  DIV_ROUND_UP(vq->vring.size, PAGE_SIZE));
 	free(vq->vring_desc_shadow);
 	list_del(&vq->list);
 	free(vq);
diff --git a/include/virtio_ring.h b/include/virtio_ring.h
index c77c212cff..8f8a55c7bd 100644
--- a/include/virtio_ring.h
+++ b/include/virtio_ring.h
@@ -86,6 +86,7 @@ struct vring_used {
 
 struct vring {
 	unsigned int num;
+	size_t size;
 	struct vring_desc *desc;
 	struct vring_avail *avail;
 	struct vring_used *used;
@@ -137,23 +138,24 @@ struct virtqueue {
 #define vring_used_event(vr)	((vr)->avail->ring[(vr)->num])
 #define vring_avail_event(vr)	(*(__virtio16 *)&(vr)->used->ring[(vr)->num])
 
+static inline unsigned int vring_size(unsigned int num, unsigned long align)
+{
+	return ((sizeof(struct vring_desc) * num +
+		sizeof(__virtio16) * (3 + num)  + align - 1) & ~(align - 1)) +
+		sizeof(__virtio16) * 3 + sizeof(struct vring_used_elem) * num;
+}
+
 static inline void vring_init(struct vring *vr, unsigned int num, void *p,
 			      unsigned long align)
 {
 	vr->num = num;
+	vr->size = vring_size(num, align);
 	vr->desc = p;
 	vr->avail = p + num * sizeof(struct vring_desc);
 	vr->used = (void *)(((uintptr_t)&vr->avail->ring[num] +
 		   sizeof(__virtio16) + align - 1) & ~(align - 1));
 }
 
-static inline unsigned int vring_size(unsigned int num, unsigned long align)
-{
-	return ((sizeof(struct vring_desc) * num +
-		sizeof(__virtio16) * (3 + num)  + align - 1) & ~(align - 1)) +
-		sizeof(__virtio16) * 3 + sizeof(struct vring_used_elem) * num;
-}
-
 /*
  * The following is used with USED_EVENT_IDX and AVAIL_EVENT_IDX.
  * Assuming a given event_idx value from the other side, if we have just
-- 
2.39.2

Re: [PATCH 3/6] virtio: Allocate virtqueue in page-size units

[Simon Glass]

On Thu, 30 Mar 2023 at 03:26, Ying-Chun Liu (PaulLiu)
<paul.liu@linaro.org> wrote:
>
> From: Will Deacon <willdeacon@google.com>
>
> In preparation for explicit bouncing of virtqueue pages for devices
> advertising the VIRTIO_F_IOMMU_PLATFORM feature, introduce a couple
> of wrappers around virtqueue allocation and freeing operations,
> ensuring that buffers are handled in terms of page-size units.
>
> Signed-off-by: Will Deacon <willdeacon@google.com>
> [ Paul: pick from the Android tree. Rebase to the upstream ]
> Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
> Cc: Bin Meng <bmeng.cn@gmail.com>
> Link: https://android.googlesource.com/platform/external/u-boot/+/b4bb5227d4cf4fdfcd8b4e1ff2692d3a54d1482a
> ---
>  drivers/virtio/virtio_ring.c | 24 ++++++++++++++++++++----
>  include/virtio_ring.h        | 16 +++++++++-------
>  2 files changed, 29 insertions(+), 11 deletions(-)
>

Reviewed-by: Simon Glass <sjg@chromium.org>

[PATCH 4/6] virtio: Add helper functions to attach/detach vring descriptors

[Ying-Chun Liu (PaulLiu)]

From: Will Deacon <willdeacon@google.com>

Move the attach and detach logic for manipulating vring descriptors
out into their own functions so that we can later extend these to
bounce the data for devices with VIRTIO_F_IOMMU_PLATFORM set.

Signed-off-by: Will Deacon <willdeacon@google.com>
[ Paul: pick from the Android tree. Rebase to the upstream ]
Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
Cc: Bin Meng <bmeng.cn@gmail.com>
Link: https://android.googlesource.com/platform/external/u-boot/+/f73258a4bfe968c5f935db45f2ec5cc0104ee796
---
 drivers/virtio/virtio_ring.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 5aeb13fd59..1bd19add75 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -47,6 +47,10 @@ static unsigned int virtqueue_attach_desc(struct virtqueue *vq, unsigned int i,
 	return desc_shadow->next;
 }
 
+static void virtqueue_detach_desc(struct virtqueue *vq, unsigned int idx)
+{
+}
+
 int virtqueue_add(struct virtqueue *vq, struct virtio_sg *sgs[],
 		  unsigned int out_sgs, unsigned int in_sgs)
 {
@@ -165,10 +169,12 @@ static void detach_buf(struct virtqueue *vq, unsigned int head)
 	i = head;
 
 	while (vq->vring_desc_shadow[i].flags & VRING_DESC_F_NEXT) {
+		virtqueue_detach_desc(vq, i);
 		i = vq->vring_desc_shadow[i].next;
 		vq->num_free++;
 	}
 
+	virtqueue_detach_desc(vq, i);
 	vq->vring_desc_shadow[i].next = vq->free_head;
 	vq->free_head = head;
 
-- 
2.39.2

Re: [PATCH 4/6] virtio: Add helper functions to attach/detach vring descriptors

[Simon Glass]

On Thu, 30 Mar 2023 at 03:26, Ying-Chun Liu (PaulLiu)
<paul.liu@linaro.org> wrote:
>
> From: Will Deacon <willdeacon@google.com>
>
> Move the attach and detach logic for manipulating vring descriptors
> out into their own functions so that we can later extend these to
> bounce the data for devices with VIRTIO_F_IOMMU_PLATFORM set.
>
> Signed-off-by: Will Deacon <willdeacon@google.com>
> [ Paul: pick from the Android tree. Rebase to the upstream ]
> Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
> Cc: Bin Meng <bmeng.cn@gmail.com>
> Link: https://android.googlesource.com/platform/external/u-boot/+/f73258a4bfe968c5f935db45f2ec5cc0104ee796
> ---
>  drivers/virtio/virtio_ring.c | 6 ++++++
>  1 file changed, 6 insertions(+)
>

Reviewed-by: Simon Glass <sjg@chromium.org>

[PATCH 5/6] virtio: Allocate bounce buffers for devices with VIRTIO_F_IOMMU_PLATFORM

[Ying-Chun Liu (PaulLiu)]

From: Will Deacon <willdeacon@google.com>

In preparation for bouncing virtio data for devices advertising the
VIRTIO_F_IOMMU_PLATFORM feature, allocate an array of bounce buffer
structures in the vring, one per descriptor.

Signed-off-by: Will Deacon <willdeacon@google.com>
[ Paul: pick from the Android tree. Rebase to the upstream ]
Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
Cc: Bin Meng <bmeng.cn@gmail.com>
Link: https://android.googlesource.com/platform/external/u-boot/+/3e052749e7c50c4c1a6014e645ae3b9be3710c07
---
 drivers/virtio/virtio_ring.c | 25 ++++++++++++++++++++-----
 include/virtio_ring.h        |  5 ++++-
 2 files changed, 24 insertions(+), 6 deletions(-)

diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index 1bd19add75..de75786ca7 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -6,6 +6,7 @@
  * virtio ring implementation
  */
 
+#include <bouncebuf.h>
 #include <common.h>
 #include <dm.h>
 #include <log.h>
@@ -292,6 +293,7 @@ struct virtqueue *vring_create_virtqueue(unsigned int index, unsigned int num,
 	struct udevice *vdev = uc_priv->vdev;
 	struct virtqueue *vq;
 	void *queue = NULL;
+	struct bounce_buffer *bbs = NULL;
 	struct vring vring;
 
 	/* We assume num is a power of 2 */
@@ -320,17 +322,29 @@ struct virtqueue *vring_create_virtqueue(unsigned int index, unsigned int num,
 		return NULL;
 
 	memset(queue, 0, vring_size(num, vring_align));
-	vring_init(&vring, num, queue, vring_align);
 
-	vq = __vring_new_virtqueue(index, vring, udev);
-	if (!vq) {
-		virtio_free_pages(vdev, queue, DIV_ROUND_UP(vring.size, PAGE_SIZE));
-		return NULL;
+	if (virtio_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
+		bbs = calloc(num, sizeof(*bbs));
+		if (!bbs)
+			goto err_free_queue;
 	}
+
+	vring_init(&vring, num, queue, vring_align, bbs);
+
+	vq = __vring_new_virtqueue(index, vring, udev);
+	if (!vq)
+		goto err_free_bbs;
+
 	debug("(%s): created vring @ %p for vq @ %p with num %u\n", udev->name,
 	      queue, vq, num);
 
 	return vq;
+
+err_free_bbs:
+	free(bbs);
+err_free_queue:
+	virtio_free_pages(vdev, queue, DIV_ROUND_UP(vring.size, PAGE_SIZE));
+	return NULL;
 }
 
 void vring_del_virtqueue(struct virtqueue *vq)
@@ -339,6 +353,7 @@ void vring_del_virtqueue(struct virtqueue *vq)
 			  DIV_ROUND_UP(vq->vring.size, PAGE_SIZE));
 	free(vq->vring_desc_shadow);
 	list_del(&vq->list);
+	free(vq->vring.bouncebufs);
 	free(vq);
 }
 
diff --git a/include/virtio_ring.h b/include/virtio_ring.h
index 8f8a55c7bd..e8e91044a2 100644
--- a/include/virtio_ring.h
+++ b/include/virtio_ring.h
@@ -87,6 +87,7 @@ struct vring_used {
 struct vring {
 	unsigned int num;
 	size_t size;
+	struct bounce_buffer *bouncebufs;
 	struct vring_desc *desc;
 	struct vring_avail *avail;
 	struct vring_used *used;
@@ -146,10 +147,12 @@ static inline unsigned int vring_size(unsigned int num, unsigned long align)
 }
 
 static inline void vring_init(struct vring *vr, unsigned int num, void *p,
-			      unsigned long align)
+			      unsigned long align,
+			      struct bounce_buffer *bouncebufs)
 {
 	vr->num = num;
 	vr->size = vring_size(num, align);
+	vr->bouncebufs = bouncebufs;
 	vr->desc = p;
 	vr->avail = p + num * sizeof(struct vring_desc);
 	vr->used = (void *)(((uintptr_t)&vr->avail->ring[num] +
-- 
2.39.2

Re: [PATCH 5/6] virtio: Allocate bounce buffers for devices with VIRTIO_F_IOMMU_PLATFORM

[Simon Glass]

On Thu, 30 Mar 2023 at 03:27, Ying-Chun Liu (PaulLiu)
<paul.liu@linaro.org> wrote:
>
> From: Will Deacon <willdeacon@google.com>
>
> In preparation for bouncing virtio data for devices advertising the
> VIRTIO_F_IOMMU_PLATFORM feature, allocate an array of bounce buffer
> structures in the vring, one per descriptor.
>
> Signed-off-by: Will Deacon <willdeacon@google.com>
> [ Paul: pick from the Android tree. Rebase to the upstream ]
> Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
> Cc: Bin Meng <bmeng.cn@gmail.com>
> Link: https://android.googlesource.com/platform/external/u-boot/+/3e052749e7c50c4c1a6014e645ae3b9be3710c07
> ---
>  drivers/virtio/virtio_ring.c | 25 ++++++++++++++++++++-----
>  include/virtio_ring.h        |  5 ++++-
>  2 files changed, 24 insertions(+), 6 deletions(-)

Reviewed-by: Simon Glass <sjg@chromium.org>

[PATCH 6/6] virtio: Use bounce buffers when VIRTIO_F_IOMMU_PLATFORM is set

[Ying-Chun Liu (PaulLiu)]

From: Will Deacon <willdeacon@google.com>

Devices advertising the VIRTIO_F_IOMMU_PLATFORM feature require
platform-specific handling to configure their DMA transactions.

When handling virtio descriptors for such a device, use bounce
buffers to ensure that the underlying buffers are always aligned
to and padded to PAGE_SIZE in preparation for platform specific
handling at page granularity.

Signed-off-by: Will Deacon <willdeacon@google.com>
[ Paul: pick from the Android tree. Rebase to the upstream ]
Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
Cc: Bin Meng <bmeng.cn@gmail.com>
Link: https://android.googlesource.com/platform/external/u-boot/+/1eff171e613ee67dca71dbe97be7282e2db17011
---
 drivers/virtio/virtio_ring.c | 48 +++++++++++++++++++++++++++++++++++-
 1 file changed, 47 insertions(+), 1 deletion(-)

diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index de75786ca7..c9adcce5c0 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -28,14 +28,51 @@ static void virtio_free_pages(struct udevice *vdev, void *ptr, u32 npages)
 	free(ptr);
 }
 
+static int __bb_force_page_align(struct bounce_buffer *state)
+{
+	const ulong align_mask = PAGE_SIZE - 1;
+
+	if ((ulong)state->user_buffer & align_mask)
+		return 0;
+
+	if (state->len != state->len_aligned)
+		return 0;
+
+	return 1;
+}
+
 static unsigned int virtqueue_attach_desc(struct virtqueue *vq, unsigned int i,
 					  struct virtio_sg *sg, u16 flags)
 {
 	struct vring_desc_shadow *desc_shadow = &vq->vring_desc_shadow[i];
 	struct vring_desc *desc = &vq->vring.desc[i];
+	void *addr;
+
+	if (IS_ENABLED(CONFIG_BOUNCE_BUFFER) && vq->vring.bouncebufs) {
+		struct bounce_buffer *bb = &vq->vring.bouncebufs[i];
+		unsigned int bbflags;
+		int ret;
+
+		if (flags & VRING_DESC_F_WRITE)
+			bbflags = GEN_BB_WRITE;
+		else
+			bbflags = GEN_BB_READ;
+
+		ret = bounce_buffer_start_extalign(bb, sg->addr, sg->length,
+						   bbflags, PAGE_SIZE,
+						   __bb_force_page_align);
+		if (ret) {
+			debug("%s: failed to allocate bounce buffer (length 0x%zx)\n",
+			      vq->vdev->name, sg->length);
+		}
+
+		addr = bb->bounce_buffer;
+	} else {
+		addr = sg->addr;
+	}
 
 	/* Update the shadow descriptor. */
-	desc_shadow->addr = (u64)(uintptr_t)sg->addr;
+	desc_shadow->addr = (u64)(uintptr_t)addr;
 	desc_shadow->len = sg->length;
 	desc_shadow->flags = flags;
 
@@ -50,6 +87,15 @@ static unsigned int virtqueue_attach_desc(struct virtqueue *vq, unsigned int i,
 
 static void virtqueue_detach_desc(struct virtqueue *vq, unsigned int idx)
 {
+	struct vring_desc *desc = &vq->vring.desc[idx];
+	struct bounce_buffer *bb;
+
+	if (!IS_ENABLED(CONFIG_BOUNCE_BUFFER) || !vq->vring.bouncebufs)
+		return;
+
+	bb = &vq->vring.bouncebufs[idx];
+	bounce_buffer_stop(bb);
+	desc->addr = cpu_to_virtio64(vq->vdev, (u64)(uintptr_t)bb->user_buffer);
 }
 
 int virtqueue_add(struct virtqueue *vq, struct virtio_sg *sgs[],
-- 
2.39.2

Re: [PATCH 6/6] virtio: Use bounce buffers when VIRTIO_F_IOMMU_PLATFORM is set

[Simon Glass]

On Thu, 30 Mar 2023 at 03:27, Ying-Chun Liu (PaulLiu)
<paul.liu@linaro.org> wrote:
>
> From: Will Deacon <willdeacon@google.com>
>
> Devices advertising the VIRTIO_F_IOMMU_PLATFORM feature require
> platform-specific handling to configure their DMA transactions.
>
> When handling virtio descriptors for such a device, use bounce
> buffers to ensure that the underlying buffers are always aligned
> to and padded to PAGE_SIZE in preparation for platform specific
> handling at page granularity.
>
> Signed-off-by: Will Deacon <willdeacon@google.com>
> [ Paul: pick from the Android tree. Rebase to the upstream ]
> Signed-off-by: Ying-Chun Liu (PaulLiu) <paul.liu@linaro.org>
> Cc: Bin Meng <bmeng.cn@gmail.com>
> Link: https://android.googlesource.com/platform/external/u-boot/+/1eff171e613ee67dca71dbe97be7282e2db17011
> ---
>  drivers/virtio/virtio_ring.c | 48 +++++++++++++++++++++++++++++++++++-
>  1 file changed, 47 insertions(+), 1 deletion(-)

Reviewed-by: Simon Glass <sjg@chromium.org>

Re: [PATCH 0/6] virtio: Use bounce buffers when VIRTIO_F_IOMMU_PLATFORM set

[Tom Rini]

On Wed, 29 Mar 2023 22:24:54 +0800, Ying-Chun Liu (PaulLiu) wrote:

> These patches will use bounce buffers when VIRTIO_F_IOMMU_PLATFORM feature
> is in a virtio device.
> 
> This feature can be tested with qemu with -device virtio-iommu-pci.
> So that when a -device virtio-blk-pci with iommu_platform=true, it will
> uses the bounce buffer instead.
> 
> [...]

Applied to u-boot/master, thanks!

-- 
Tom