IFC @r271694

7 files changed, 96 insertions, 23 deletions

diff --git a/usr.sbin/bhyve/block_if.c b/usr.sbin/bhyve/block_if.c
index 1ec0344..b038228 100644
--- a/usr.sbin/bhyve/block_if.c
+++ b/usr.sbin/bhyve/block_if.c
@@ -278,6 +278,7 @@ blockif_open(const char *optstr, const char *ident)
 
 	bc->bc_magic = BLOCKIF_SIG;
 	bc->bc_fd = fd;
+	bc->bc_rdonly = ro;
 	bc->bc_size = size;
 	bc->bc_sectsz = sectsz;
 	pthread_mutex_init(&bc->bc_mtx, NULL);
diff --git a/usr.sbin/bhyve/pci_virtio_block.c b/usr.sbin/bhyve/pci_virtio_block.c
index cf1c655..394b116 100644
--- a/usr.sbin/bhyve/pci_virtio_block.c
+++ b/usr.sbin/bhyve/pci_virtio_block.c
@@ -133,6 +133,7 @@ static struct virtio_consts vtblk_vi_consts = {
 	pci_vtblk_notify,	/* device-wide qnotify */
 	pci_vtblk_cfgread,	/* read PCI config */
 	pci_vtblk_cfgwrite,	/* write PCI config */
+	NULL,			/* apply negotiated features */
 	VTBLK_S_HOSTCAPS,	/* our capabilities */
 };
 
diff --git a/usr.sbin/bhyve/pci_virtio_net.c b/usr.sbin/bhyve/pci_virtio_net.c
index c3b8690..5ac9ecd 100644
--- a/usr.sbin/bhyve/pci_virtio_net.c
+++ b/usr.sbin/bhyve/pci_virtio_net.c
@@ -135,11 +135,14 @@ struct pci_vtnet_softc {
 	int		vsc_rx_ready;
 	volatile int	resetting;	/* set and checked outside lock */
 
-	uint32_t	vsc_features;
+	uint64_t	vsc_features;	/* negotiated features */
+	
 	struct virtio_net_config vsc_config;
 
 	pthread_mutex_t	rx_mtx;
 	int		rx_in_progress;
+	int		rx_vhdrlen;
+	int		rx_merge;	/* merged rx bufs in use */
 
 	pthread_t 	tx_tid;
 	pthread_mutex_t	tx_mtx;
@@ -151,6 +154,7 @@ static void pci_vtnet_reset(void *);
 /* static void pci_vtnet_notify(void *, struct vqueue_info *); */
 static int pci_vtnet_cfgread(void *, int, int, uint32_t *);
 static int pci_vtnet_cfgwrite(void *, int, int, uint32_t);
+static void pci_vtnet_neg_features(void *, uint64_t);
 
 static struct virtio_consts vtnet_vi_consts = {
 	"vtnet",		/* our name */
@@ -160,6 +164,7 @@ static struct virtio_consts vtnet_vi_consts = {
 	NULL,			/* device-wide qnotify -- not used */
 	pci_vtnet_cfgread,	/* read PCI config */
 	pci_vtnet_cfgwrite,	/* write PCI config */
+	pci_vtnet_neg_features,	/* apply negotiated features */
 	VTNET_S_HOSTCAPS,	/* our capabilities */
 };
 
@@ -212,6 +217,8 @@ pci_vtnet_reset(void *vsc)
 	pci_vtnet_rxwait(sc);
 
 	sc->vsc_rx_ready = 0;
+	sc->rx_merge = 1;
+	sc->rx_vhdrlen = sizeof(struct virtio_net_rxhdr);
 
 	/* now reset rings, MSI-X vectors, and negotiated capabilities */
 	vi_reset_dev(&sc->vsc_vs);
@@ -253,14 +260,34 @@ pci_vtnet_tap_tx(struct pci_vtnet_softc *sc, struct iovec *iov, int iovcnt,
  */
 static uint8_t dummybuf[2048];
 
+static __inline struct iovec *
+rx_iov_trim(struct iovec *iov, int *niov, int tlen)
+{
+	struct iovec *riov;
+
+	/* XXX short-cut: assume first segment is >= tlen */
+	assert(iov[0].iov_len >= tlen);
+
+	iov[0].iov_len -= tlen;
+	if (iov[0].iov_len == 0) {
+		assert(*niov > 1);
+		*niov -= 1;
+		riov = &iov[1];
+	} else {
+		iov[0].iov_base = (void *)((uintptr_t)iov[0].iov_base + tlen);
+		riov = &iov[0];
+	}
+
+	return (riov);
+}
+
 static void
 pci_vtnet_tap_rx(struct pci_vtnet_softc *sc)
 {
+	struct iovec iov[VTNET_MAXSEGS], *riov;
 	struct vqueue_info *vq;
-	struct virtio_net_rxhdr *vrx;
-	uint8_t *buf;
-	int len;
-	struct iovec iov;
+	void *vrx;
+	int len, n;
 
 	/*
 	 * Should never be called without a valid tap fd
@@ -296,21 +323,19 @@ pci_vtnet_tap_rx(struct pci_vtnet_softc *sc)
 
 	do {
 		/*
-		 * Get descriptor chain, which should have just
-		 * one descriptor in it.
-		 * ??? allow guests to use multiple descs?
+		 * Get descriptor chain.
 		 */
-		assert(vq_getchain(vq, &iov, 1, NULL) == 1);
+		n = vq_getchain(vq, iov, VTNET_MAXSEGS, NULL);
+		assert(n >= 1 && n <= VTNET_MAXSEGS);
 
 		/*
 		 * Get a pointer to the rx header, and use the
 		 * data immediately following it for the packet buffer.
 		 */
-		vrx = iov.iov_base;
-		buf = (uint8_t *)(vrx + 1);
+		vrx = iov[0].iov_base;
+		riov = rx_iov_trim(iov, &n, sc->rx_vhdrlen);
 
-		len = read(sc->vsc_tapfd, buf,
-			   iov.iov_len - sizeof(struct virtio_net_rxhdr));
+		len = readv(sc->vsc_tapfd, riov, n);
 
 		if (len < 0 && errno == EWOULDBLOCK) {
 			/*
@@ -323,16 +348,21 @@ pci_vtnet_tap_rx(struct pci_vtnet_softc *sc)
 
 		/*
 		 * The only valid field in the rx packet header is the
-		 * number of buffers, which is always 1 without TSO
-		 * support.
+		 * number of buffers if merged rx bufs were negotiated.
 		 */
-		memset(vrx, 0, sizeof(struct virtio_net_rxhdr));
-		vrx->vrh_bufs = 1;
+		memset(vrx, 0, sc->rx_vhdrlen);
+
+		if (sc->rx_merge) {
+			struct virtio_net_rxhdr *vrxh;
+
+			vrxh = vrx;
+			vrxh->vrh_bufs = 1;
+		}
 
 		/*
 		 * Release this chain and handle more chains.
 		 */
-		vq_relchain(vq, len + sizeof(struct virtio_net_rxhdr));
+		vq_relchain(vq, len + sc->rx_vhdrlen);
 	} while (vq_has_descs(vq));
 
 	/* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
@@ -623,6 +653,8 @@ pci_vtnet_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
 
 	sc->resetting = 0;
 
+	sc->rx_merge = 1;
+	sc->rx_vhdrlen = sizeof(struct virtio_net_rxhdr);
 	sc->rx_in_progress = 0;
 	pthread_mutex_init(&sc->rx_mtx, NULL); 
 
@@ -656,9 +688,10 @@ pci_vtnet_cfgwrite(void *vsc, int offset, int size, uint32_t value)
 		ptr = &sc->vsc_config.mac[offset];
 		memcpy(ptr, &value, size);
 	} else {
+		/* silently ignore other writes */
 		DPRINTF(("vtnet: write to readonly reg %d\n\r", offset));
-		return (1);
 	}
+
 	return (0);
 }
 
@@ -673,6 +706,20 @@ pci_vtnet_cfgread(void *vsc, int offset, int size, uint32_t *retval)
 	return (0);
 }
 
+static void
+pci_vtnet_neg_features(void *vsc, uint64_t negotiated_features)
+{
+	struct pci_vtnet_softc *sc = vsc;
+
+	sc->vsc_features = negotiated_features;
+
+	if (!(sc->vsc_features & VIRTIO_NET_F_MRG_RXBUF)) {
+		sc->rx_merge = 0;
+		/* non-merge rx header is 2 bytes shorter */
+		sc->rx_vhdrlen -= 2;
+	}
+}
+
 struct pci_devemu pci_de_vnet = {
 	.pe_emu = 	"virtio-net",
 	.pe_init =	pci_vtnet_init,
diff --git a/usr.sbin/bhyve/pci_virtio_rnd.c b/usr.sbin/bhyve/pci_virtio_rnd.c
index 4d53183..0a31080 100644
--- a/usr.sbin/bhyve/pci_virtio_rnd.c
+++ b/usr.sbin/bhyve/pci_virtio_rnd.c
@@ -80,6 +80,7 @@ static struct virtio_consts vtrnd_vi_consts = {
 	pci_vtrnd_notify,	/* device-wide qnotify */
 	NULL,			/* read virtio config */
 	NULL,			/* write virtio config */
+	NULL,			/* apply negotiated features */
 	0,			/* our capabilities */
 };
 
diff --git a/usr.sbin/bhyve/task_switch.c b/usr.sbin/bhyve/task_switch.c
index 0002da8..b939c1a 100644
--- a/usr.sbin/bhyve/task_switch.c
+++ b/usr.sbin/bhyve/task_switch.c
@@ -725,6 +725,21 @@ vmexit_task_switch(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
 	assert(paging->cpu_mode == CPU_MODE_PROTECTED);
 
 	/*
+	 * Calculate the %eip to store in the old TSS before modifying the
+	 * 'inst_length'.
+	 */
+	eip = vmexit->rip + vmexit->inst_length;
+
+	/*
+	 * Set the 'inst_length' to '0'.
+	 *
+	 * If an exception is triggered during emulation of the task switch
+	 * then the exception handler should return to the instruction that
+	 * caused the task switch as opposed to the subsequent instruction.
+	 */
+	vmexit->inst_length = 0;
+
+	/*
 	 * Section 4.6, "Access Rights" in Intel SDM Vol 3.
 	 * The following page table accesses are implicitly supervisor mode:
 	 * - accesses to GDT or LDT to load segment descriptors
@@ -839,7 +854,6 @@ vmexit_task_switch(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
 	}
 
 	/* Save processor state in old TSS */
-	eip = vmexit->rip + vmexit->inst_length;
 	tss32_save(ctx, vcpu, task_switch, eip, &oldtss, ot_iov);
 
 	/*
@@ -870,7 +884,7 @@ vmexit_task_switch(struct vmctx *ctx, struct vm_exit *vmexit, int *pvcpu)
 	 * the saved instruction pointer will belong to the new task.
 	 */
 	vmexit->rip = newtss.tss_eip;
-	vmexit->inst_length = 0;
+	assert(vmexit->inst_length == 0);
 
 	/* Load processor state from new TSS */
 	error = tss32_restore(ctx, vcpu, task_switch, ot_sel, &newtss, nt_iov);
diff --git a/usr.sbin/bhyve/virtio.c b/usr.sbin/bhyve/virtio.c
index 1f27300..19c0d47 100644
--- a/usr.sbin/bhyve/virtio.c
+++ b/usr.sbin/bhyve/virtio.c
@@ -33,6 +33,7 @@ __FBSDID("$FreeBSD$");
 #include <stdio.h>
 #include <stdint.h>
 #include <pthread.h>
+#include <pthread_np.h>
 
 #include "bhyverun.h"
 #include "pci_emul.h"
@@ -89,6 +90,9 @@ vi_reset_dev(struct virtio_softc *vs)
 	struct vqueue_info *vq;
 	int i, nvq;
 
+	if (vs->vs_mtx)
+		assert(pthread_mutex_isowned_np(vs->vs_mtx));
+
 	nvq = vs->vs_vc->vc_nvq;
 	for (vq = vs->vs_queues, i = 0; i < nvq; vq++, i++) {
 		vq->vq_flags = 0;
@@ -99,11 +103,9 @@ vi_reset_dev(struct virtio_softc *vs)
 	vs->vs_negotiated_caps = 0;
 	vs->vs_curq = 0;
 	/* vs->vs_status = 0; -- redundant */
-	VS_LOCK(vs);
 	if (vs->vs_isr)
 		pci_lintr_deassert(vs->vs_pi);
 	vs->vs_isr = 0;
-	VS_UNLOCK(vs);
 	vs->vs_msix_cfg_idx = VIRTIO_MSI_NO_VECTOR;
 }
 
@@ -137,7 +139,9 @@ vi_intr_init(struct virtio_softc *vs, int barnum, int use_msix)
 
 	if (use_msix) {
 		vs->vs_flags |= VIRTIO_USE_MSIX;
+		VS_LOCK(vs);
 		vi_reset_dev(vs); /* set all vectors to NO_VECTOR */
+		VS_UNLOCK(vs);
 		nvec = vs->vs_vc->vc_nvq + 1;
 		if (pci_emul_add_msixcap(vs->vs_pi, nvec, barnum))
 			return (1);
@@ -694,6 +698,9 @@ bad:
 	switch (offset) {
 	case VTCFG_R_GUESTCAP:
 		vs->vs_negotiated_caps = value & vc->vc_hv_caps;
+		if (vc->vc_apply_features)
+			(*vc->vc_apply_features)(DEV_SOFTC(vs),
+			    vs->vs_negotiated_caps);
 		break;
 	case VTCFG_R_PFN:
 		if (vs->vs_curq >= vc->vc_nvq)
diff --git a/usr.sbin/bhyve/virtio.h b/usr.sbin/bhyve/virtio.h
index 1f29dfa..6f655f3 100644
--- a/usr.sbin/bhyve/virtio.h
+++ b/usr.sbin/bhyve/virtio.h
@@ -352,6 +352,8 @@ struct virtio_consts {
 					/* called to read config regs */
 	int	(*vc_cfgwrite)(void *, int, int, uint32_t);
 					/* called to write config regs */
+	void    (*vc_apply_features)(void *, uint64_t);
+				/* called to apply negotiated features */
 	uint64_t vc_hv_caps;		/* hypervisor-provided capabilities */
 };