Skip to contentSkip to conversation

virtio indirect ring entries

PATCH In review Mark McLoughlin 2 participants 4 messages Jun 17, 2009

lore
Share & tools
lore
#0 lore raw

cover virtio indirect ring entries

Hi,
Indirect ring entries has been merged for 2.6.31, these patches
implement the qemu side.

Cheers,
Mark.

#1 lore raw

PATCH 1/2 virtio: make vring_desc_*() take phys addrs

Change the vring descriptor helpers to take the physical address of the descriptor table rather than a virtqueue.

This is needed in order to allow these helpers to be used with an indirect descriptor table.

---
 hw/virtio.c |   59 +++++++++++++++++++++++++++++++++--------------------------
 1 files changed, 33 insertions(+), 26 deletions(-)
hw/virtio.c +33 -26
index 45a49fa..1e8376d 100644
@@ -85,31 +85,31 @@ static void virtqueue_init(VirtQueue *vq)
                                  VIRTIO_PCI_VRING_ALIGN);
 }
 
-static inline uint64_t vring_desc_addr(VirtQueue *vq, int i)
+static inline uint64_t vring_desc_addr(target_phys_addr_t desc_pa, int i)
 {
     target_phys_addr_t pa;
-    pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, addr);
+    pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, addr);
     return ldq_phys(pa);
 }
 
-static inline uint32_t vring_desc_len(VirtQueue *vq, int i)
+static inline uint32_t vring_desc_len(target_phys_addr_t desc_pa, int i)
 {
     target_phys_addr_t pa;
-    pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, len);
+    pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, len);
     return ldl_phys(pa);
 }
 
-static inline uint16_t vring_desc_flags(VirtQueue *vq, int i)
+static inline uint16_t vring_desc_flags(target_phys_addr_t desc_pa, int i)
 {
     target_phys_addr_t pa;
-    pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, flags);
+    pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, flags);
     return lduw_phys(pa);
 }
 
-static inline uint16_t vring_desc_next(VirtQueue *vq, int i)
+static inline uint16_t vring_desc_next(target_phys_addr_t desc_pa, int i)
 {
     target_phys_addr_t pa;
-    pa = vq->vring.desc + sizeof(VRingDesc) * i + offsetof(VRingDesc, next);
+    pa = desc_pa + sizeof(VRingDesc) * i + offsetof(VRingDesc, next);
     return lduw_phys(pa);
 }
 
@@ -269,20 +269,21 @@ static unsigned int virtqueue_get_head(VirtQueue *vq, unsigned int idx)
     return head;
 }
 
-static unsigned virtqueue_next_desc(VirtQueue *vq, unsigned int i)
+static unsigned virtqueue_next_desc(target_phys_addr_t desc_pa,
+                                    unsigned int i, unsigned int max)
 {
     unsigned int next;
 
     /* If this descriptor says it doesn't chain, we're done. */
-    if (!(vring_desc_flags(vq, i) & VRING_DESC_F_NEXT))
-        return vq->vring.num;
+    if (!(vring_desc_flags(desc_pa, i) & VRING_DESC_F_NEXT))
+        return max;
 
     /* Check they're not leading us off end of descriptors. */
-    next = vring_desc_next(vq, i);
+    next = vring_desc_next(desc_pa, i);
     /* Make sure compiler knows to grab that: we don't want it changing! */
     wmb();
 
-    if (next >= vq->vring.num) {
+    if (next >= max) {
         fprintf(stderr, "Desc next is %u", next);
         exit(1);
     }
@@ -292,10 +293,12 @@ static unsigned virtqueue_next_desc(VirtQueue *vq, unsigned int i)
 
 int virtqueue_avail_bytes(VirtQueue *vq, int in_bytes, int out_bytes)
 {
-    unsigned int idx;
+    target_phys_addr_t desc_pa = vq->vring.desc;
+    unsigned int idx, max;
     int num_bufs, in_total, out_total;
 
     idx = vq->last_avail_idx;
+    max = vq->vring.num;
 
     num_bufs = in_total = out_total = 0;
     while (virtqueue_num_heads(vq, idx)) {
@@ -304,21 +307,21 @@ int virtqueue_avail_bytes(VirtQueue *vq, int in_bytes, int out_bytes)
         i = virtqueue_get_head(vq, idx++);
         do {
             /* If we've got too many, that implies a descriptor loop. */
-            if (++num_bufs > vq->vring.num) {
+            if (++num_bufs > max) {
                 fprintf(stderr, "Looped descriptor");
                 exit(1);
             }
 
-            if (vring_desc_flags(vq, i) & VRING_DESC_F_WRITE) {
+            if (vring_desc_flags(desc_pa, i) & VRING_DESC_F_WRITE) {
                 if (in_bytes > 0 &&
-                    (in_total += vring_desc_len(vq, i)) >= in_bytes)
+                    (in_total += vring_desc_len(desc_pa, i)) >= in_bytes)
                     return 1;
             } else {
                 if (out_bytes > 0 &&
-                    (out_total += vring_desc_len(vq, i)) >= out_bytes)
+                    (out_total += vring_desc_len(desc_pa, i)) >= out_bytes)
                     return 1;
             }
-        } while ((i = virtqueue_next_desc(vq, i)) != vq->vring.num);
+        } while ((i = virtqueue_next_desc(desc_pa, i, max)) != max);
     }
 
     return 0;
@@ -326,7 +329,8 @@ int virtqueue_avail_bytes(VirtQueue *vq, int in_bytes, int out_bytes)
 
 int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
 {
-    unsigned int i, head;
+    unsigned int i, head, max;
+    target_phys_addr_t desc_pa = vq->vring.desc;
     target_phys_addr_t len;
 
     if (!virtqueue_num_heads(vq, vq->last_avail_idx))
@@ -335,23 +339,26 @@ int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
     /* When we start there are none of either input nor output. */
     elem->out_num = elem->in_num = 0;
 
+    max = vq->vring.num;
+
     i = head = virtqueue_get_head(vq, vq->last_avail_idx++);
     do {
         struct iovec *sg;
         int is_write = 0;
 
-        if (vring_desc_flags(vq, i) & VRING_DESC_F_WRITE) {
-            elem->in_addr[elem->in_num] = vring_desc_addr(vq, i);
+        if (vring_desc_flags(desc_pa, i) & VRING_DESC_F_WRITE) {
+            elem->in_addr[elem->in_num] = vring_desc_addr(desc_pa, i);
             sg = &elem->in_sg[elem->in_num++];
             is_write = 1;
         } else
             sg = &elem->out_sg[elem->out_num++];
 
         /* Grab the first descriptor, and check it's OK. */
-        sg->iov_len = vring_desc_len(vq, i);
+        sg->iov_len = vring_desc_len(desc_pa, i);
         len = sg->iov_len;
 
-        sg->iov_base = cpu_physical_memory_map(vring_desc_addr(vq, i), &len, is_write);
+        sg->iov_base = cpu_physical_memory_map(vring_desc_addr(desc_pa, i),
+                                               &len, is_write);
 
         if (sg->iov_base == NULL || len != sg->iov_len) {
             fprintf(stderr, "virtio: trying to map MMIO memory\n");
@@ -359,11 +366,11 @@ int virtqueue_pop(VirtQueue *vq, VirtQueueElement *elem)
         }
 
         /* If we've got too many, that implies a descriptor loop. */
-        if ((elem->in_num + elem->out_num) > vq->vring.num) {
+        if ((elem->in_num + elem->out_num) > max) {
             fprintf(stderr, "Looped descriptor");
             exit(1);
         }
-    } while ((i = virtqueue_next_desc(vq, i)) != vq->vring.num);
+    } while ((i = virtqueue_next_desc(desc_pa, i, max)) != max);
 
     elem->index = head;
 
Signature · 1 line
1.6.0.6
#2 lore raw

PATCH 2/2 virtio: add support for indirect ring entries

Support a new feature flag for indirect ring entries. These are ring entries which point to a table of buffer descriptors.

The idea here is to increase the ring capacity by allowing a larger effective ring size whereby the ring size dictates the number of requests that may be outstanding, rather than the size of those requests.

This should be most effective in the case of block I/O where we can potentially benefit by concurrently dispatching a large number of large requests. Even in the simple case of single segment block requests, this results in a threefold increase in ring capacity.

---
 hw/virtio-pci.c |    4 +++-
 hw/virtio.c     |   51 ++++++++++++++++++++++++++++++++++++++++++++++-----
 hw/virtio.h     |    4 ++++
 3 files changed, 53 insertions(+), 6 deletions(-)
hw/virtio-pci.c +3 -1
index 24fe837..df45036 100644
@@ -136,7 +136,9 @@ static uint32_t virtio_ioport_read(void *opaque, uint32_t addr)
     switch (addr) {
     case VIRTIO_PCI_HOST_FEATURES:
         ret = vdev->get_features(vdev);
-        ret |= (1 << VIRTIO_F_NOTIFY_ON_EMPTY) | (1 << VIRTIO_F_BAD_FEATURE);
+        ret |= (1 << VIRTIO_F_NOTIFY_ON_EMPTY);
+        ret |= (1 << VIRTIO_RING_F_INDIRECT_DESC);
+        ret |= (1 << VIRTIO_F_BAD_FEATURE);
         break;
     case VIRTIO_PCI_GUEST_FEATURES:
         ret = vdev->features;
hw/virtio.c +46 -5

76 lines not on this page · raw message

hw/virtio.h +4 -0
index 425727e..46ed0b2 100644
@@ -32,6 +32,8 @@
 /* We notify when the ring is completely used, even if the guest is supressing
  * callbacks */
 #define VIRTIO_F_NOTIFY_ON_EMPTY        24
+/* We support indirect buffer descriptors */
+#define VIRTIO_RING_F_INDIRECT_DESC     28
 /* A guest should never accept this.  It implies negotiation is broken. */
 #define VIRTIO_F_BAD_FEATURE		30
 
@@ -41,6 +43,8 @@
 #define VRING_DESC_F_NEXT       1
 /* This marks a buffer as write-only (otherwise read-only). */
 #define VRING_DESC_F_WRITE      2
+/* This means the buffer contains a list of buffer descriptors. */
+#define VRING_DESC_F_INDIRECT  4
 
 /* This means don't notify other side when buffer added. */
 #define VRING_USED_F_NO_NOTIFY  1
Signature · 1 line
1.6.0.6
#3 lore raw
In reply to Mark McLoughlin
Quoted 2 lines · Mark McLoughlin

Indirect ring entries has been merged for 2.6.31, these patches
implement the qemu side.

I like the idea of increasing the ring capacity, but I see that this method still has limitations, and the effective capacity depends on the I/O submission pattern, which isn't very friendly.

Has arbitrarily expandable rings (using tail pointer ring entries) ever been considered?

-- Jamie