MFC 295748,295792,295793,295794

295748
    hyperv/hn: Use buf_ring for txdesc list

    So one spinlock is avoided, which would be potentially dangerous for
    virtual machine, if the spinlock holder was scheduled out by the host,
    as noted by royger.

    Old spinlock based txdesc list is still kept around, so we could have
    a safe fallback.

    No performance regression nor improvement is observed.

    Reviewed by:        adrian
    Approved by:        adrian (mentor)
    MFC after:  1 week
    Sponsored by:       Microsoft OSTC
    Differential Revision:      https://reviews.freebsd.org/D5290

295792
    hyperv/hn: Add option to bind TX taskqueues to the specified CPU

    It will be used to help tracking host side transmission ring selection
    issue; and it will be turned on by default, once we have concrete result.

    Reviewed by:        adrian, Jun Su <junsu microsoft com>
    Approved by:        adrian (mento)
    MFC after:  1 week
    Sponsored by:       Microsoft OSTC
    Differential Revision:      https://reviews.freebsd.org/D5316

295793
    hyperv/hn: Enable IP header checksum offloading for WIN8 (WinServ2012)

    Tested on Windows Server 2012.

    Reviewed by:        adrian
    Approved by:        adrian (mentor)
    MFC after:  1 week
    Sponsored by:       Microsoft OSTC
    Differential Revision:      https://reviews.freebsd.org/D5317

295794
    hyperv/hn: Free the txdesc buf_ring when the TX ring is destroyed

    Reviewed by:        adrian
    Approved by:        adrian (mentor)
    MFC after:  1 week
    Sponsored by:       Microsoft OSTC
    Differential Revision:      https://reviews.freebsd.org/D5318

2 files changed, 115 insertions, 11 deletions

diff --git a/sys/dev/hyperv/netvsc/hv_net_vsc.h b/sys/dev/hyperv/netvsc/hv_net_vsc.h
index 2c6c4b7..a7fa7698 100644
--- a/sys/dev/hyperv/netvsc/hv_net_vsc.h
+++ b/sys/dev/hyperv/netvsc/hv_net_vsc.h
@@ -58,6 +58,8 @@
 
 #include <dev/hyperv/include/hyperv.h>
 
+#define HN_USE_TXDESC_BUFRING
+
 MALLOC_DECLARE(M_NETVSC);
 
 #define NVSP_INVALID_PROTOCOL_VERSION           (0xFFFFFFFF)
@@ -990,8 +992,12 @@ typedef struct {
 	hv_bool_uint8_t	link_state;
 } netvsc_device_info;
 
+#ifndef HN_USE_TXDESC_BUFRING
 struct hn_txdesc;
 SLIST_HEAD(hn_txdesc_list, hn_txdesc);
+#else
+struct buf_ring;
+#endif
 
 struct hn_rx_ring {
 	struct lro_ctrl	hn_lro;
@@ -1012,8 +1018,12 @@ struct hn_rx_ring {
 #define HN_TRUST_HCSUM_UDP	0x0004
 
 struct hn_tx_ring {
+#ifndef HN_USE_TXDESC_BUFRING
 	struct mtx	hn_txlist_spin;
 	struct hn_txdesc_list hn_txlist;
+#else
+	struct buf_ring	*hn_txdesc_br;
+#endif
 	int		hn_txdesc_cnt;
 	int		hn_txdesc_avail;
 	int		hn_txeof;
diff --git a/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c b/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c
index 7502641..1ce1136 100644
--- a/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c
+++ b/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c
@@ -66,10 +66,12 @@ __FBSDID("$FreeBSD$");
 #include <sys/module.h>
 #include <sys/kernel.h>
 #include <sys/socket.h>
+#include <sys/proc.h>
 #include <sys/queue.h>
 #include <sys/lock.h>
 #include <sys/sx.h>
 #include <sys/sysctl.h>
+#include <sys/buf_ring.h>
 
 #include <net/if.h>
 #include <net/if_arp.h>
@@ -151,7 +153,9 @@ __FBSDID("$FreeBSD$");
 #define HN_DIRECT_TX_SIZE_DEF		128
 
 struct hn_txdesc {
+#ifndef HN_USE_TXDESC_BUFRING
 	SLIST_ENTRY(hn_txdesc) link;
+#endif
 	struct mbuf	*m;
 	struct hn_tx_ring *txr;
 	int		refs;
@@ -173,7 +177,7 @@ struct hn_txdesc {
  * later.  UDP checksum offloading doesn't work on earlier
  * Windows releases.
  */
-#define HN_CSUM_ASSIST_WIN8	(CSUM_TCP)
+#define HN_CSUM_ASSIST_WIN8	(CSUM_IP | CSUM_TCP)
 #define HN_CSUM_ASSIST		(CSUM_IP | CSUM_UDP | CSUM_TCP)
 
 #define HN_LRO_LENLIM_DEF		(25 * ETHERMTU)
@@ -258,6 +262,18 @@ SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
 
 static struct taskqueue	*hn_tx_taskq;
 
+#ifndef HN_USE_TXDESC_BUFRING
+static int hn_use_txdesc_bufring = 0;
+#else
+static int hn_use_txdesc_bufring = 1;
+#endif
+SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
+    &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
+
+static int hn_bind_tx_taskq = -1;
+SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
+    &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
+
 /*
  * Forward declarations
  */
@@ -343,6 +359,19 @@ netvsc_probe(device_t dev)
 	return (ENXIO);
 }
 
+static void
+hn_cpuset_setthread_task(void *xmask, int pending __unused)
+{
+	cpuset_t *mask = xmask;
+	int error;
+
+	error = cpuset_setthread(curthread->td_tid, mask);
+	if (error) {
+		panic("curthread=%ju: can't pin; error=%d",
+		    (uintmax_t)curthread->td_tid, error);
+	}
+}
+
 /*
  * Standard attach entry point.
  *
@@ -376,6 +405,19 @@ netvsc_attach(device_t dev)
 		    taskqueue_thread_enqueue, &sc->hn_tx_taskq);
 		taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx",
 		    device_get_nameunit(dev));
+		if (hn_bind_tx_taskq >= 0) {
+			int cpu = hn_bind_tx_taskq;
+			struct task cpuset_task;
+			cpuset_t cpu_set;
+
+			if (cpu > mp_ncpus - 1)
+				cpu = mp_ncpus - 1;
+			CPU_SETOF(cpu, &cpu_set);
+			TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task,
+			    &cpu_set);
+			taskqueue_enqueue(sc->hn_tx_taskq, &cpuset_task);
+			taskqueue_drain(sc->hn_tx_taskq, &cpuset_task);
+		}
 	} else {
 		sc->hn_tx_taskq = hn_tx_taskq;
 	}
@@ -572,6 +614,7 @@ hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
 
 	txd->flags |= HN_TXD_FLAG_ONLIST;
 
+#ifndef HN_USE_TXDESC_BUFRING
 	mtx_lock_spin(&txr->hn_txlist_spin);
 	KASSERT(txr->hn_txdesc_avail >= 0 &&
 	    txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
@@ -579,6 +622,10 @@ hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
 	txr->hn_txdesc_avail++;
 	SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
 	mtx_unlock_spin(&txr->hn_txlist_spin);
+#else
+	atomic_add_int(&txr->hn_txdesc_avail, 1);
+	buf_ring_enqueue(txr->hn_txdesc_br, txd);
+#endif
 
 	return 1;
 }
@@ -588,6 +635,7 @@ hn_txdesc_get(struct hn_tx_ring *txr)
 {
 	struct hn_txdesc *txd;
 
+#ifndef HN_USE_TXDESC_BUFRING
 	mtx_lock_spin(&txr->hn_txlist_spin);
 	txd = SLIST_FIRST(&txr->hn_txlist);
 	if (txd != NULL) {
@@ -597,8 +645,14 @@ hn_txdesc_get(struct hn_tx_ring *txr)
 		SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
 	}
 	mtx_unlock_spin(&txr->hn_txlist_spin);
+#else
+	txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
+#endif
 
 	if (txd != NULL) {
+#ifdef HN_USE_TXDESC_BUFRING
+		atomic_subtract_int(&txr->hn_txdesc_avail, 1);
+#endif
 		KASSERT(txd->m == NULL && txd->refs == 0 &&
 		    (txd->flags & HN_TXD_FLAG_ONLIST), ("invalid txd"));
 		txd->flags &= ~HN_TXD_FLAG_ONLIST;
@@ -2061,13 +2115,20 @@ hn_create_tx_ring(struct hn_softc *sc, int id)
 
 	txr->hn_sc = sc;
 
+#ifndef HN_USE_TXDESC_BUFRING
 	mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
+#endif
 	mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
 
 	txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
 	txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
 	    M_NETVSC, M_WAITOK | M_ZERO);
+#ifndef HN_USE_TXDESC_BUFRING
 	SLIST_INIT(&txr->hn_txlist);
+#else
+	txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_NETVSC,
+	    M_WAITOK, &txr->hn_tx_lock);
+#endif
 
 	txr->hn_tx_taskq = sc->hn_tx_taskq;
 	TASK_INIT(&txr->hn_start_task, 0, hn_start_taskfunc, txr);
@@ -2171,7 +2232,11 @@ hn_create_tx_ring(struct hn_softc *sc, int id)
 
 		/* All set, put it to list */
 		txd->flags |= HN_TXD_FLAG_ONLIST;
+#ifndef HN_USE_TXDESC_BUFRING
 		SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
+#else
+		buf_ring_enqueue(txr->hn_txdesc_br, txd);
+#endif
 	}
 	txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
 
@@ -2204,6 +2269,20 @@ hn_create_tx_ring(struct hn_softc *sc, int id)
 }
 
 static void
+hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
+{
+	struct hn_tx_ring *txr = txd->txr;
+
+	KASSERT(txd->m == NULL, ("still has mbuf installed"));
+	KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
+
+	bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_msg_dmap);
+	bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_msg,
+	    txd->rndis_msg_dmap);
+	bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
+}
+
+static void
 hn_destroy_tx_ring(struct hn_tx_ring *txr)
 {
 	struct hn_txdesc *txd;
@@ -2211,28 +2290,31 @@ hn_destroy_tx_ring(struct hn_tx_ring *txr)
 	if (txr->hn_txdesc == NULL)
 		return;
 
+#ifndef HN_USE_TXDESC_BUFRING
 	while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
-		KASSERT(txd->m == NULL, ("still has mbuf installed"));
-		KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0,
-		    ("still dma mapped"));
 		SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
-
-		bus_dmamap_unload(txr->hn_tx_rndis_dtag,
-		    txd->rndis_msg_dmap);
-		bus_dmamem_free(txr->hn_tx_rndis_dtag,
-		    txd->rndis_msg, txd->rndis_msg_dmap);
-
-		bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
+		hn_txdesc_dmamap_destroy(txd);
 	}
+#else
+	while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
+		hn_txdesc_dmamap_destroy(txd);
+#endif
 
 	if (txr->hn_tx_data_dtag != NULL)
 		bus_dma_tag_destroy(txr->hn_tx_data_dtag);
 	if (txr->hn_tx_rndis_dtag != NULL)
 		bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
+
+#ifdef HN_USE_TXDESC_BUFRING
+	buf_ring_free(txr->hn_txdesc_br, M_NETVSC);
+#endif
+
 	free(txr->hn_txdesc, M_NETVSC);
 	txr->hn_txdesc = NULL;
 
+#ifndef HN_USE_TXDESC_BUFRING
 	mtx_destroy(&txr->hn_txlist_spin);
+#endif
 	mtx_destroy(&txr->hn_tx_lock);
 }
 
@@ -2377,6 +2459,18 @@ hn_tx_taskq_create(void *arg __unused)
 	hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
 	    taskqueue_thread_enqueue, &hn_tx_taskq);
 	taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
+	if (hn_bind_tx_taskq >= 0) {
+		int cpu = hn_bind_tx_taskq;
+		struct task cpuset_task;
+		cpuset_t cpu_set;
+
+		if (cpu > mp_ncpus - 1)
+			cpu = mp_ncpus - 1;
+		CPU_SETOF(cpu, &cpu_set);
+		TASK_INIT(&cpuset_task, 0, hn_cpuset_setthread_task, &cpu_set);
+		taskqueue_enqueue(hn_tx_taskq, &cpuset_task);
+		taskqueue_drain(hn_tx_taskq, &cpuset_task);
+	}
 }
 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
     hn_tx_taskq_create, NULL);