aboutsummaryrefslogtreecommitdiff
path: root/hw/net
diff options
context:
space:
mode:
authorThomas Huth <thuth@redhat.com>2016-03-21 17:25:23 +0100
committerDavid Gibson <david@gibson.dropbear.id.au>2016-03-24 11:17:34 +1100
commit831e8822530bb511a24329494f9121ad1f8b94ab (patch)
tree53431c7dafe56ed798fc197dc358bd884e70ccbf /hw/net
parentd6f39fdfcd30b3f65f9aad396e7caa49285a5a0a (diff)
hw/net/spapr_llan: Fix receive buffer handling for better performance
tl;dr: This patch introduces an alternate way of handling the receive buffers of the spapr-vlan device, resulting in much better receive performance for the guest. Full story: One of our testers recently discovered that the performance of the spapr-vlan device is very poor compared to other NICs, and that a simple "ping -i 0.2 -s 65507 someip" in the guest can result in more than 50% lost ping packets (especially with older guest kernels < 3.17). After doing some analysis, it was clear that there is a problem with the way we handle the receive buffers in spapr_llan.c: The ibmveth driver of the guest Linux kernel tries to add a lot of buffers into several buffer pools (with 512, 2048 and 65536 byte sizes by default, but it can be changed via the entries in the /sys/devices/vio/1000/pool* directories of the guest). However, the spapr-vlan device of QEMU only tries to squeeze all receive buffer descriptors into one single page which has been supplied by the guest during the H_REGISTER_LOGICAL_LAN call, without taking care of different buffer sizes. This has two bad effects: First, only a very limited number of buffer descriptors is accepted at all. Second, we also hand 64k buffers to the guest even if the 2k buffers would fit better - and this results in dropped packets in the IP layer of the guest since too much skbuf memory is used. Though it seems at a first glance like PAPR says that we should store the receive buffer descriptors in the page that is supplied during the H_REGISTER_LOGICAL_LAN call, chapter 16.4.1.2 in the LoPAPR spec declares that "the contents of these descriptors are architecturally opaque, none of these descriptors are manipulated by code above the architected interfaces". That means we don't have to store the RX buffer descriptors in this page, but can also manage the receive buffers at the hypervisor level only. This is now what we are doing here: Introducing proper RX buffer pools which are also sorted by size of the buffers, so we can hand out a buffer with the best fitting size when a packet has been received. To avoid problems with migration from/to older version of QEMU, the old behavior is also retained and enabled by default. The new buffer management has to be enabled via a new "use-rx-buffer-pools" property. Now with the new buffer pool management enabled, the problem with "ping -s 65507" is fixed for me, and the throughput of a simple test with wget increases from creeping 3MB/s up to 20MB/s! Signed-off-by: Thomas Huth <thuth@redhat.com> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Diffstat (limited to 'hw/net')
-rw-r--r--hw/net/spapr_llan.c218
1 files changed, 216 insertions, 2 deletions
diff --git a/hw/net/spapr_llan.c b/hw/net/spapr_llan.c
index 39a1dd1d13..c7ce87017b 100644
--- a/hw/net/spapr_llan.c
+++ b/hw/net/spapr_llan.c
@@ -45,6 +45,10 @@
#define DPRINTF(fmt...)
#endif
+/* Compatibility flags for migration */
+#define SPAPRVLAN_FLAG_RX_BUF_POOLS_BIT 0
+#define SPAPRVLAN_FLAG_RX_BUF_POOLS (1 << SPAPRVLAN_FLAG_RX_BUF_POOLS_BIT)
+
/*
* Virtual LAN device
*/
@@ -86,6 +90,15 @@ typedef uint64_t vlan_bd_t;
#define VIO_SPAPR_VLAN_DEVICE(obj) \
OBJECT_CHECK(VIOsPAPRVLANDevice, (obj), TYPE_VIO_SPAPR_VLAN_DEVICE)
+#define RX_POOL_MAX_BDS 4096
+#define RX_MAX_POOLS 5
+
+typedef struct {
+ int32_t bufsize;
+ int32_t count;
+ vlan_bd_t bds[RX_POOL_MAX_BDS];
+} RxBufPool;
+
typedef struct VIOsPAPRVLANDevice {
VIOsPAPRDevice sdev;
NICConf nicconf;
@@ -94,6 +107,8 @@ typedef struct VIOsPAPRVLANDevice {
target_ulong buf_list;
uint32_t add_buf_ptr, use_buf_ptr, rx_bufs;
target_ulong rxq_ptr;
+ uint32_t compat_flags; /* Compatability flags for migration */
+ RxBufPool *rx_pool[RX_MAX_POOLS]; /* Receive buffer descriptor pools */
} VIOsPAPRVLANDevice;
static int spapr_vlan_can_receive(NetClientState *nc)
@@ -104,6 +119,37 @@ static int spapr_vlan_can_receive(NetClientState *nc)
}
/**
+ * Get buffer descriptor from one of our receive buffer pools
+ */
+static vlan_bd_t spapr_vlan_get_rx_bd_from_pool(VIOsPAPRVLANDevice *dev,
+ size_t size)
+{
+ vlan_bd_t bd;
+ int pool;
+
+ for (pool = 0; pool < RX_MAX_POOLS; pool++) {
+ if (dev->rx_pool[pool]->count > 0 &&
+ dev->rx_pool[pool]->bufsize >= size + 8) {
+ break;
+ }
+ }
+ if (pool == RX_MAX_POOLS) {
+ /* Failed to find a suitable buffer */
+ return 0;
+ }
+
+ DPRINTF("Found buffer: pool=%d count=%d rxbufs=%d\n", pool,
+ dev->rx_pool[pool]->count, dev->rx_bufs);
+
+ /* Remove the buffer from the pool */
+ dev->rx_pool[pool]->count--;
+ bd = dev->rx_pool[pool]->bds[dev->rx_pool[pool]->count];
+ dev->rx_pool[pool]->bds[dev->rx_pool[pool]->count] = 0;
+
+ return bd;
+}
+
+/**
* Get buffer descriptor from the receive buffer list page that has been
* supplied by the guest with the H_REGISTER_LOGICAL_LAN call
*/
@@ -160,7 +206,11 @@ static ssize_t spapr_vlan_receive(NetClientState *nc, const uint8_t *buf,
return -1;
}
- bd = spapr_vlan_get_rx_bd_from_page(dev, size);
+ if (dev->compat_flags & SPAPRVLAN_FLAG_RX_BUF_POOLS) {
+ bd = spapr_vlan_get_rx_bd_from_pool(dev, size);
+ } else {
+ bd = spapr_vlan_get_rx_bd_from_page(dev, size);
+ }
if (!bd) {
return -1;
}
@@ -213,13 +263,31 @@ static NetClientInfo net_spapr_vlan_info = {
.receive = spapr_vlan_receive,
};
+static void spapr_vlan_reset_rx_pool(RxBufPool *rxp)
+{
+ /*
+ * Use INT_MAX as bufsize so that unused buffers are moved to the end
+ * of the list during the qsort in spapr_vlan_add_rxbuf_to_pool() later.
+ */
+ rxp->bufsize = INT_MAX;
+ rxp->count = 0;
+ memset(rxp->bds, 0, sizeof(rxp->bds));
+}
+
static void spapr_vlan_reset(VIOsPAPRDevice *sdev)
{
VIOsPAPRVLANDevice *dev = VIO_SPAPR_VLAN_DEVICE(sdev);
+ int i;
dev->buf_list = 0;
dev->rx_bufs = 0;
dev->isopen = 0;
+
+ if (dev->compat_flags & SPAPRVLAN_FLAG_RX_BUF_POOLS) {
+ for (i = 0; i < RX_MAX_POOLS; i++) {
+ spapr_vlan_reset_rx_pool(dev->rx_pool[i]);
+ }
+ }
}
static void spapr_vlan_realize(VIOsPAPRDevice *sdev, Error **errp)
@@ -236,10 +304,31 @@ static void spapr_vlan_realize(VIOsPAPRDevice *sdev, Error **errp)
static void spapr_vlan_instance_init(Object *obj)
{
VIOsPAPRVLANDevice *dev = VIO_SPAPR_VLAN_DEVICE(obj);
+ int i;
device_add_bootindex_property(obj, &dev->nicconf.bootindex,
"bootindex", "",
DEVICE(dev), NULL);
+
+ if (dev->compat_flags & SPAPRVLAN_FLAG_RX_BUF_POOLS) {
+ for (i = 0; i < RX_MAX_POOLS; i++) {
+ dev->rx_pool[i] = g_new(RxBufPool, 1);
+ spapr_vlan_reset_rx_pool(dev->rx_pool[i]);
+ }
+ }
+}
+
+static void spapr_vlan_instance_finalize(Object *obj)
+{
+ VIOsPAPRVLANDevice *dev = VIO_SPAPR_VLAN_DEVICE(obj);
+ int i;
+
+ if (dev->compat_flags & SPAPRVLAN_FLAG_RX_BUF_POOLS) {
+ for (i = 0; i < RX_MAX_POOLS; i++) {
+ g_free(dev->rx_pool[i]);
+ dev->rx_pool[i] = NULL;
+ }
+ }
}
void spapr_vlan_create(VIOsPAPRBus *bus, NICInfo *nd)
@@ -390,6 +479,87 @@ static target_ulong h_free_logical_lan(PowerPCCPU *cpu,
return H_SUCCESS;
}
+/**
+ * Used for qsort, this function compares two RxBufPools by size.
+ */
+static int rx_pool_size_compare(const void *p1, const void *p2)
+{
+ const RxBufPool *pool1 = *(RxBufPool **)p1;
+ const RxBufPool *pool2 = *(RxBufPool **)p2;
+
+ if (pool1->bufsize < pool2->bufsize) {
+ return -1;
+ }
+ return pool1->bufsize > pool2->bufsize;
+}
+
+/**
+ * Search for a matching buffer pool with exact matching size,
+ * or return -1 if no matching pool has been found.
+ */
+static int spapr_vlan_get_rx_pool_id(VIOsPAPRVLANDevice *dev, int size)
+{
+ int pool;
+
+ for (pool = 0; pool < RX_MAX_POOLS; pool++) {
+ if (dev->rx_pool[pool]->bufsize == size) {
+ return pool;
+ }
+ }
+
+ return -1;
+}
+
+/**
+ * Enqueuing receive buffer by adding it to one of our receive buffer pools
+ */
+static target_long spapr_vlan_add_rxbuf_to_pool(VIOsPAPRVLANDevice *dev,
+ target_ulong buf)
+{
+ int size = VLAN_BD_LEN(buf);
+ int pool;
+
+ pool = spapr_vlan_get_rx_pool_id(dev, size);
+ if (pool < 0) {
+ /*
+ * No matching pool found? Try to use a new one. If the guest used all
+ * pools before, but changed the size of one pool inbetween, we might
+ * need to recycle that pool here (if it's empty already). Thus scan
+ * all buffer pools now, starting with the last (likely empty) one.
+ */
+ for (pool = RX_MAX_POOLS - 1; pool >= 0 ; pool--) {
+ if (dev->rx_pool[pool]->count == 0) {
+ dev->rx_pool[pool]->bufsize = size;
+ /*
+ * Sort pools by size so that spapr_vlan_receive()
+ * can later find the smallest buffer pool easily.
+ */
+ qsort(dev->rx_pool, RX_MAX_POOLS, sizeof(dev->rx_pool[0]),
+ rx_pool_size_compare);
+ pool = spapr_vlan_get_rx_pool_id(dev, size);
+ DPRINTF("created RX pool %d for size %lld\n", pool,
+ VLAN_BD_LEN(buf));
+ break;
+ }
+ }
+ }
+ /* Still no usable pool? Give up */
+ if (pool < 0 || dev->rx_pool[pool]->count >= RX_POOL_MAX_BDS) {
+ return H_RESOURCE;
+ }
+
+ DPRINTF("h_add_llan_buf(): Add buf using pool %i (size %lli, count=%i)\n",
+ pool, VLAN_BD_LEN(buf), dev->rx_pool[pool]->count);
+
+ dev->rx_pool[pool]->bds[dev->rx_pool[pool]->count++] = buf;
+
+ return 0;
+}
+
+/**
+ * This is the old way of enqueuing receive buffers: Add it to the rx queue
+ * page that has been supplied by the guest (which is quite limited in size).
+ */
static target_long spapr_vlan_add_rxbuf_to_page(VIOsPAPRVLANDevice *dev,
target_ulong buf)
{
@@ -445,7 +615,11 @@ static target_ulong h_add_logical_lan_buffer(PowerPCCPU *cpu,
return H_RESOURCE;
}
- ret = spapr_vlan_add_rxbuf_to_page(dev, buf);
+ if (dev->compat_flags & SPAPRVLAN_FLAG_RX_BUF_POOLS) {
+ ret = spapr_vlan_add_rxbuf_to_pool(dev, buf);
+ } else {
+ ret = spapr_vlan_add_rxbuf_to_page(dev, buf);
+ }
if (ret) {
return ret;
}
@@ -543,9 +717,44 @@ static target_ulong h_multicast_ctrl(PowerPCCPU *cpu, sPAPRMachineState *spapr,
static Property spapr_vlan_properties[] = {
DEFINE_SPAPR_PROPERTIES(VIOsPAPRVLANDevice, sdev),
DEFINE_NIC_PROPERTIES(VIOsPAPRVLANDevice, nicconf),
+ DEFINE_PROP_BIT("use-rx-buffer-pools", VIOsPAPRVLANDevice,
+ compat_flags, SPAPRVLAN_FLAG_RX_BUF_POOLS_BIT, false),
DEFINE_PROP_END_OF_LIST(),
};
+static bool spapr_vlan_rx_buffer_pools_needed(void *opaque)
+{
+ VIOsPAPRVLANDevice *dev = opaque;
+
+ return (dev->compat_flags & SPAPRVLAN_FLAG_RX_BUF_POOLS) != 0;
+}
+
+static const VMStateDescription vmstate_rx_buffer_pool = {
+ .name = "spapr_llan/rx_buffer_pool",
+ .version_id = 1,
+ .minimum_version_id = 1,
+ .needed = spapr_vlan_rx_buffer_pools_needed,
+ .fields = (VMStateField[]) {
+ VMSTATE_INT32(bufsize, RxBufPool),
+ VMSTATE_INT32(count, RxBufPool),
+ VMSTATE_UINT64_ARRAY(bds, RxBufPool, RX_POOL_MAX_BDS),
+ VMSTATE_END_OF_LIST()
+ }
+};
+
+static const VMStateDescription vmstate_rx_pools = {
+ .name = "spapr_llan/rx_pools",
+ .version_id = 1,
+ .minimum_version_id = 1,
+ .needed = spapr_vlan_rx_buffer_pools_needed,
+ .fields = (VMStateField[]) {
+ VMSTATE_ARRAY_OF_POINTER_TO_STRUCT(rx_pool, VIOsPAPRVLANDevice,
+ RX_MAX_POOLS, 1,
+ vmstate_rx_buffer_pool, RxBufPool),
+ VMSTATE_END_OF_LIST()
+ }
+};
+
static const VMStateDescription vmstate_spapr_llan = {
.name = "spapr_llan",
.version_id = 1,
@@ -562,6 +771,10 @@ static const VMStateDescription vmstate_spapr_llan = {
VMSTATE_END_OF_LIST()
},
+ .subsections = (const VMStateDescription * []) {
+ &vmstate_rx_pools,
+ NULL
+ }
};
static void spapr_vlan_class_init(ObjectClass *klass, void *data)
@@ -588,6 +801,7 @@ static const TypeInfo spapr_vlan_info = {
.instance_size = sizeof(VIOsPAPRVLANDevice),
.class_init = spapr_vlan_class_init,
.instance_init = spapr_vlan_instance_init,
+ .instance_finalize = spapr_vlan_instance_finalize,
};
static void spapr_vlan_register_types(void)