Merge branch '2023-04-25-use-bounce-buffers-for-VIRTIO_F_IOMMU_PLATFORM'
To quote the author:
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.
diff --git a/drivers/virtio/virtio-uclass.c b/drivers/virtio/virtio-uclass.c
index de9bc90..b3fb3db 100644
--- a/drivers/virtio/virtio-uclass.c
+++ b/drivers/virtio/virtio-uclass.c
@@ -336,7 +336,7 @@
/* 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",
diff --git a/drivers/virtio/virtio_pci_modern.c b/drivers/virtio/virtio_pci_modern.c
index cfde400..3cdc2d2 100644
--- a/drivers/virtio/virtio_pci_modern.c
+++ b/drivers/virtio/virtio_pci_modern.c
@@ -218,25 +218,6 @@
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 @@
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);
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index f71bab7..c9adcce 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>
@@ -15,15 +16,63 @@
#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 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;
@@ -36,6 +85,19 @@
return desc_shadow->next;
}
+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[],
unsigned int out_sgs, unsigned int in_sgs)
{
@@ -154,10 +216,12 @@
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;
@@ -271,8 +335,11 @@
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 bounce_buffer *bbs = NULL;
struct vring vring;
/* We assume num is a power of 2 */
@@ -283,7 +350,9 @@
/* 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;
}
@@ -293,30 +362,44 @@
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;
memset(queue, 0, vring_size(num, vring_align));
- vring_init(&vring, num, queue, vring_align);
- vq = __vring_new_virtqueue(index, vring, udev);
- if (!vq) {
- free(queue);
- 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)
{
- 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->vring.bouncebufs);
free(vq);
}
diff --git a/include/virtio_ring.h b/include/virtio_ring.h
index c77c212..e8e9104 100644
--- a/include/virtio_ring.h
+++ b/include/virtio_ring.h
@@ -86,6 +86,8 @@
struct vring {
unsigned int num;
+ size_t size;
+ struct bounce_buffer *bouncebufs;
struct vring_desc *desc;
struct vring_avail *avail;
struct vring_used *used;
@@ -137,23 +139,26 @@
#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)
+ 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] +
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