aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2008-02-04 11:00:54 -0500
committerLinus Torvalds <torvalds@linux-foundation.org>2008-02-04 11:00:54 -0500
commit93890b71a34f9490673a6edd56b61c2124215e46 (patch)
treec5d82620f2cb69f0bf43639e63f54b0c0e2eb744 /Documentation
parentf5bb3a5e9dcdb8435471562b6cada89525cf4df1 (diff)
parent6b35e40767c6c1ac783330109ae8e0c09ea6bc82 (diff)
Merge git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux-2.6-for-linus
* git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux-2.6-for-linus: (25 commits) virtio: balloon driver virtio: Use PCI revision field to indicate virtio PCI ABI version virtio: PCI device virtio_blk: implement naming for vda-vdz,vdaa-vdzz,vdaaa-vdzzz virtio_blk: Dont waste major numbers virtio_blk: provide getgeo virtio_net: parametrize the napi_weight for virtio receive queue. virtio: free transmit skbs when notified, not on next xmit. virtio: flush buffers on open virtnet: remove double ether_setup virtio: Allow virtio to be modular and used by modules virtio: Use the sg_phys convenience function. virtio: Put the virtio under the virtualization menu virtio: handle interrupts after callbacks turned off virtio: reset function virtio: populate network rings in the probe routine, not open virtio: Tweak virtio_net defines virtio: Net header needs hdr_len virtio: remove unused id field from struct virtio_blk_outhdr virtio: clarify NO_NOTIFY flag usage ...
Diffstat (limited to 'Documentation')
-rw-r--r--Documentation/lguest/lguest.c231
1 files changed, 154 insertions, 77 deletions
diff --git a/Documentation/lguest/lguest.c b/Documentation/lguest/lguest.c
index 6c8a2386cd50..0f23d67f958f 100644
--- a/Documentation/lguest/lguest.c
+++ b/Documentation/lguest/lguest.c
@@ -34,6 +34,8 @@
34#include <zlib.h> 34#include <zlib.h>
35#include <assert.h> 35#include <assert.h>
36#include <sched.h> 36#include <sched.h>
37#include <limits.h>
38#include <stddef.h>
37#include "linux/lguest_launcher.h" 39#include "linux/lguest_launcher.h"
38#include "linux/virtio_config.h" 40#include "linux/virtio_config.h"
39#include "linux/virtio_net.h" 41#include "linux/virtio_net.h"
@@ -99,13 +101,11 @@ struct device_list
99 /* The descriptor page for the devices. */ 101 /* The descriptor page for the devices. */
100 u8 *descpage; 102 u8 *descpage;
101 103
102 /* The tail of the last descriptor. */
103 unsigned int desc_used;
104
105 /* A single linked list of devices. */ 104 /* A single linked list of devices. */
106 struct device *dev; 105 struct device *dev;
107 /* ... And an end pointer so we can easily append new devices */ 106 /* And a pointer to the last device for easy append and also for
108 struct device **lastdev; 107 * configuration appending. */
108 struct device *lastdev;
109}; 109};
110 110
111/* The list of Guest devices, based on command line arguments. */ 111/* The list of Guest devices, based on command line arguments. */
@@ -191,7 +191,14 @@ static void *_convert(struct iovec *iov, size_t size, size_t align,
191#define cpu_to_le64(v64) (v64) 191#define cpu_to_le64(v64) (v64)
192#define le16_to_cpu(v16) (v16) 192#define le16_to_cpu(v16) (v16)
193#define le32_to_cpu(v32) (v32) 193#define le32_to_cpu(v32) (v32)
194#define le64_to_cpu(v32) (v64) 194#define le64_to_cpu(v64) (v64)
195
196/* The device virtqueue descriptors are followed by feature bitmasks. */
197static u8 *get_feature_bits(struct device *dev)
198{
199 return (u8 *)(dev->desc + 1)
200 + dev->desc->num_vq * sizeof(struct lguest_vqconfig);
201}
195 202
196/*L:100 The Launcher code itself takes us out into userspace, that scary place 203/*L:100 The Launcher code itself takes us out into userspace, that scary place
197 * where pointers run wild and free! Unfortunately, like most userspace 204 * where pointers run wild and free! Unfortunately, like most userspace
@@ -914,21 +921,58 @@ static void enable_fd(int fd, struct virtqueue *vq)
914 write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd)); 921 write(waker_fd, &vq->dev->fd, sizeof(vq->dev->fd));
915} 922}
916 923
924/* Resetting a device is fairly easy. */
925static void reset_device(struct device *dev)
926{
927 struct virtqueue *vq;
928
929 verbose("Resetting device %s\n", dev->name);
930 /* Clear the status. */
931 dev->desc->status = 0;
932
933 /* Clear any features they've acked. */
934 memset(get_feature_bits(dev) + dev->desc->feature_len, 0,
935 dev->desc->feature_len);
936
937 /* Zero out the virtqueues. */
938 for (vq = dev->vq; vq; vq = vq->next) {
939 memset(vq->vring.desc, 0,
940 vring_size(vq->config.num, getpagesize()));
941 vq->last_avail_idx = 0;
942 }
943}
944
917/* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */ 945/* This is the generic routine we call when the Guest uses LHCALL_NOTIFY. */
918static void handle_output(int fd, unsigned long addr) 946static void handle_output(int fd, unsigned long addr)
919{ 947{
920 struct device *i; 948 struct device *i;
921 struct virtqueue *vq; 949 struct virtqueue *vq;
922 950
923 /* Check each virtqueue. */ 951 /* Check each device and virtqueue. */
924 for (i = devices.dev; i; i = i->next) { 952 for (i = devices.dev; i; i = i->next) {
953 /* Notifications to device descriptors reset the device. */
954 if (from_guest_phys(addr) == i->desc) {
955 reset_device(i);
956 return;
957 }
958
959 /* Notifications to virtqueues mean output has occurred. */
925 for (vq = i->vq; vq; vq = vq->next) { 960 for (vq = i->vq; vq; vq = vq->next) {
926 if (vq->config.pfn == addr/getpagesize() 961 if (vq->config.pfn != addr/getpagesize())
927 && vq->handle_output) { 962 continue;
928 verbose("Output to %s\n", vq->dev->name); 963
929 vq->handle_output(fd, vq); 964 /* Guest should acknowledge (and set features!) before
965 * using the device. */
966 if (i->desc->status == 0) {
967 warnx("%s gave early output", i->name);
930 return; 968 return;
931 } 969 }
970
971 if (strcmp(vq->dev->name, "console") != 0)
972 verbose("Output to %s\n", vq->dev->name);
973 if (vq->handle_output)
974 vq->handle_output(fd, vq);
975 return;
932 } 976 }
933 } 977 }
934 978
@@ -986,54 +1030,44 @@ static void handle_input(int fd)
986 * 1030 *
987 * All devices need a descriptor so the Guest knows it exists, and a "struct 1031 * All devices need a descriptor so the Guest knows it exists, and a "struct
988 * device" so the Launcher can keep track of it. We have common helper 1032 * device" so the Launcher can keep track of it. We have common helper
989 * routines to allocate them. 1033 * routines to allocate and manage them. */
990 *
991 * This routine allocates a new "struct lguest_device_desc" from descriptor
992 * table just above the Guest's normal memory. It returns a pointer to that
993 * descriptor. */
994static struct lguest_device_desc *new_dev_desc(u16 type)
995{
996 struct lguest_device_desc *d;
997 1034
998 /* We only have one page for all the descriptors. */ 1035/* The layout of the device page is a "struct lguest_device_desc" followed by a
999 if (devices.desc_used + sizeof(*d) > getpagesize()) 1036 * number of virtqueue descriptors, then two sets of feature bits, then an
1000 errx(1, "Too many devices"); 1037 * array of configuration bytes. This routine returns the configuration
1001 1038 * pointer. */
1002 /* We don't need to set config_len or status: page is 0 already. */ 1039static u8 *device_config(const struct device *dev)
1003 d = (void *)devices.descpage + devices.desc_used; 1040{
1004 d->type = type; 1041 return (void *)(dev->desc + 1)
1005 devices.desc_used += sizeof(*d); 1042 + dev->desc->num_vq * sizeof(struct lguest_vqconfig)
1006 1043 + dev->desc->feature_len * 2;
1007 return d;
1008} 1044}
1009 1045
1010/* Each device descriptor is followed by some configuration information. 1046/* This routine allocates a new "struct lguest_device_desc" from descriptor
1011 * Each configuration field looks like: u8 type, u8 len, [... len bytes...]. 1047 * table page just above the Guest's normal memory. It returns a pointer to
1012 * 1048 * that descriptor. */
1013 * This routine adds a new field to an existing device's descriptor. It only 1049static struct lguest_device_desc *new_dev_desc(u16 type)
1014 * works for the last device, but that's OK because that's how we use it. */
1015static void add_desc_field(struct device *dev, u8 type, u8 len, const void *c)
1016{ 1050{
1017 /* This is the last descriptor, right? */ 1051 struct lguest_device_desc d = { .type = type };
1018 assert(devices.descpage + devices.desc_used 1052 void *p;
1019 == (u8 *)(dev->desc + 1) + dev->desc->config_len);
1020 1053
1021 /* We only have one page of device descriptions. */ 1054 /* Figure out where the next device config is, based on the last one. */
1022 if (devices.desc_used + 2 + len > getpagesize()) 1055 if (devices.lastdev)
1023 errx(1, "Too many devices"); 1056 p = device_config(devices.lastdev)
1057 + devices.lastdev->desc->config_len;
1058 else
1059 p = devices.descpage;
1024 1060
1025 /* Copy in the new config header: type then length. */ 1061 /* We only have one page for all the descriptors. */
1026 devices.descpage[devices.desc_used++] = type; 1062 if (p + sizeof(d) > (void *)devices.descpage + getpagesize())
1027 devices.descpage[devices.desc_used++] = len; 1063 errx(1, "Too many devices");
1028 memcpy(devices.descpage + devices.desc_used, c, len);
1029 devices.desc_used += len;
1030 1064
1031 /* Update the device descriptor length: two byte head then data. */ 1065 /* p might not be aligned, so we memcpy in. */
1032 dev->desc->config_len += 2 + len; 1066 return memcpy(p, &d, sizeof(d));
1033} 1067}
1034 1068
1035/* This routine adds a virtqueue to a device. We specify how many descriptors 1069/* Each device descriptor is followed by the description of its virtqueues. We
1036 * the virtqueue is to have. */ 1070 * specify how many descriptors the virtqueue is to have. */
1037static void add_virtqueue(struct device *dev, unsigned int num_descs, 1071static void add_virtqueue(struct device *dev, unsigned int num_descs,
1038 void (*handle_output)(int fd, struct virtqueue *me)) 1072 void (*handle_output)(int fd, struct virtqueue *me))
1039{ 1073{
@@ -1059,9 +1093,15 @@ static void add_virtqueue(struct device *dev, unsigned int num_descs,
1059 /* Initialize the vring. */ 1093 /* Initialize the vring. */
1060 vring_init(&vq->vring, num_descs, p, getpagesize()); 1094 vring_init(&vq->vring, num_descs, p, getpagesize());
1061 1095
1062 /* Add the configuration information to this device's descriptor. */ 1096 /* Append virtqueue to this device's descriptor. We use
1063 add_desc_field(dev, VIRTIO_CONFIG_F_VIRTQUEUE, 1097 * device_config() to get the end of the device's current virtqueues;
1064 sizeof(vq->config), &vq->config); 1098 * we check that we haven't added any config or feature information
1099 * yet, otherwise we'd be overwriting them. */
1100 assert(dev->desc->config_len == 0 && dev->desc->feature_len == 0);
1101 memcpy(device_config(dev), &vq->config, sizeof(vq->config));
1102 dev->desc->num_vq++;
1103
1104 verbose("Virtqueue page %#lx\n", to_guest_phys(p));
1065 1105
1066 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is 1106 /* Add to tail of list, so dev->vq is first vq, dev->vq->next is
1067 * second. */ 1107 * second. */
@@ -1072,11 +1112,41 @@ static void add_virtqueue(struct device *dev, unsigned int num_descs,
1072 * virtqueue. */ 1112 * virtqueue. */
1073 vq->handle_output = handle_output; 1113 vq->handle_output = handle_output;
1074 1114
1075 /* Set the "Don't Notify Me" flag if we don't have a handler */ 1115 /* As an optimization, set the advisory "Don't Notify Me" flag if we
1116 * don't have a handler */
1076 if (!handle_output) 1117 if (!handle_output)
1077 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY; 1118 vq->vring.used->flags = VRING_USED_F_NO_NOTIFY;
1078} 1119}
1079 1120
1121/* The first half of the feature bitmask is for us to advertise features. The
1122 * second half if for the Guest to accept features. */
1123static void add_feature(struct device *dev, unsigned bit)
1124{
1125 u8 *features = get_feature_bits(dev);
1126
1127 /* We can't extend the feature bits once we've added config bytes */
1128 if (dev->desc->feature_len <= bit / CHAR_BIT) {
1129 assert(dev->desc->config_len == 0);
1130 dev->desc->feature_len = (bit / CHAR_BIT) + 1;
1131 }
1132
1133 features[bit / CHAR_BIT] |= (1 << (bit % CHAR_BIT));
1134}
1135
1136/* This routine sets the configuration fields for an existing device's
1137 * descriptor. It only works for the last device, but that's OK because that's
1138 * how we use it. */
1139static void set_config(struct device *dev, unsigned len, const void *conf)
1140{
1141 /* Check we haven't overflowed our single page. */
1142 if (device_config(dev) + len > devices.descpage + getpagesize())
1143 errx(1, "Too many devices");
1144
1145 /* Copy in the config information, and store the length. */
1146 memcpy(device_config(dev), conf, len);
1147 dev->desc->config_len = len;
1148}
1149
1080/* This routine does all the creation and setup of a new device, including 1150/* This routine does all the creation and setup of a new device, including
1081 * calling new_dev_desc() to allocate the descriptor and device memory. */ 1151 * calling new_dev_desc() to allocate the descriptor and device memory. */
1082static struct device *new_device(const char *name, u16 type, int fd, 1152static struct device *new_device(const char *name, u16 type, int fd,
@@ -1084,14 +1154,6 @@ static struct device *new_device(const char *name, u16 type, int fd,
1084{ 1154{
1085 struct device *dev = malloc(sizeof(*dev)); 1155 struct device *dev = malloc(sizeof(*dev));
1086 1156
1087 /* Append to device list. Prepending to a single-linked list is
1088 * easier, but the user expects the devices to be arranged on the bus
1089 * in command-line order. The first network device on the command line
1090 * is eth0, the first block device /dev/vda, etc. */
1091 *devices.lastdev = dev;
1092 dev->next = NULL;
1093 devices.lastdev = &dev->next;
1094
1095 /* Now we populate the fields one at a time. */ 1157 /* Now we populate the fields one at a time. */
1096 dev->fd = fd; 1158 dev->fd = fd;
1097 /* If we have an input handler for this file descriptor, then we add it 1159 /* If we have an input handler for this file descriptor, then we add it
@@ -1102,6 +1164,17 @@ static struct device *new_device(const char *name, u16 type, int fd,
1102 dev->handle_input = handle_input; 1164 dev->handle_input = handle_input;
1103 dev->name = name; 1165 dev->name = name;
1104 dev->vq = NULL; 1166 dev->vq = NULL;
1167
1168 /* Append to device list. Prepending to a single-linked list is
1169 * easier, but the user expects the devices to be arranged on the bus
1170 * in command-line order. The first network device on the command line
1171 * is eth0, the first block device /dev/vda, etc. */
1172 if (devices.lastdev)
1173 devices.lastdev->next = dev;
1174 else
1175 devices.dev = dev;
1176 devices.lastdev = dev;
1177
1105 return dev; 1178 return dev;
1106} 1179}
1107 1180
@@ -1226,7 +1299,7 @@ static void setup_tun_net(const char *arg)
1226 int netfd, ipfd; 1299 int netfd, ipfd;
1227 u32 ip; 1300 u32 ip;
1228 const char *br_name = NULL; 1301 const char *br_name = NULL;
1229 u8 hwaddr[6]; 1302 struct virtio_net_config conf;
1230 1303
1231 /* We open the /dev/net/tun device and tell it we want a tap device. A 1304 /* We open the /dev/net/tun device and tell it we want a tap device. A
1232 * tap device is like a tun device, only somehow different. To tell 1305 * tap device is like a tun device, only somehow different. To tell
@@ -1265,12 +1338,13 @@ static void setup_tun_net(const char *arg)
1265 ip = str2ip(arg); 1338 ip = str2ip(arg);
1266 1339
1267 /* Set up the tun device, and get the mac address for the interface. */ 1340 /* Set up the tun device, and get the mac address for the interface. */
1268 configure_device(ipfd, ifr.ifr_name, ip, hwaddr); 1341 configure_device(ipfd, ifr.ifr_name, ip, conf.mac);
1269 1342
1270 /* Tell Guest what MAC address to use. */ 1343 /* Tell Guest what MAC address to use. */
1271 add_desc_field(dev, VIRTIO_CONFIG_NET_MAC_F, sizeof(hwaddr), hwaddr); 1344 add_feature(dev, VIRTIO_NET_F_MAC);
1345 set_config(dev, sizeof(conf), &conf);
1272 1346
1273 /* We don't seed the socket any more; setup is done. */ 1347 /* We don't need the socket any more; setup is done. */
1274 close(ipfd); 1348 close(ipfd);
1275 1349
1276 verbose("device %u: tun net %u.%u.%u.%u\n", 1350 verbose("device %u: tun net %u.%u.%u.%u\n",
@@ -1458,8 +1532,7 @@ static void setup_block_file(const char *filename)
1458 struct device *dev; 1532 struct device *dev;
1459 struct vblk_info *vblk; 1533 struct vblk_info *vblk;
1460 void *stack; 1534 void *stack;
1461 u64 cap; 1535 struct virtio_blk_config conf;
1462 unsigned int val;
1463 1536
1464 /* This is the pipe the I/O thread will use to tell us I/O is done. */ 1537 /* This is the pipe the I/O thread will use to tell us I/O is done. */
1465 pipe(p); 1538 pipe(p);
@@ -1477,14 +1550,18 @@ static void setup_block_file(const char *filename)
1477 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE); 1550 vblk->fd = open_or_die(filename, O_RDWR|O_LARGEFILE);
1478 vblk->len = lseek64(vblk->fd, 0, SEEK_END); 1551 vblk->len = lseek64(vblk->fd, 0, SEEK_END);
1479 1552
1553 /* We support barriers. */
1554 add_feature(dev, VIRTIO_BLK_F_BARRIER);
1555
1480 /* Tell Guest how many sectors this device has. */ 1556 /* Tell Guest how many sectors this device has. */
1481 cap = cpu_to_le64(vblk->len / 512); 1557 conf.capacity = cpu_to_le64(vblk->len / 512);
1482 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_CAPACITY, sizeof(cap), &cap);
1483 1558
1484 /* Tell Guest not to put in too many descriptors at once: two are used 1559 /* Tell Guest not to put in too many descriptors at once: two are used
1485 * for the in and out elements. */ 1560 * for the in and out elements. */
1486 val = cpu_to_le32(VIRTQUEUE_NUM - 2); 1561 add_feature(dev, VIRTIO_BLK_F_SEG_MAX);
1487 add_desc_field(dev, VIRTIO_CONFIG_BLK_F_SEG_MAX, sizeof(val), &val); 1562 conf.seg_max = cpu_to_le32(VIRTQUEUE_NUM - 2);
1563
1564 set_config(dev, sizeof(conf), &conf);
1488 1565
1489 /* The I/O thread writes to this end of the pipe when done. */ 1566 /* The I/O thread writes to this end of the pipe when done. */
1490 vblk->done_fd = p[1]; 1567 vblk->done_fd = p[1];
@@ -1505,7 +1582,7 @@ static void setup_block_file(const char *filename)
1505 close(vblk->workpipe[0]); 1582 close(vblk->workpipe[0]);
1506 1583
1507 verbose("device %u: virtblock %llu sectors\n", 1584 verbose("device %u: virtblock %llu sectors\n",
1508 devices.device_num, cap); 1585 devices.device_num, le64_to_cpu(conf.capacity));
1509} 1586}
1510/* That's the end of device setup. :*/ 1587/* That's the end of device setup. :*/
1511 1588
@@ -1610,12 +1687,12 @@ int main(int argc, char *argv[])
1610 /* First we initialize the device list. Since console and network 1687 /* First we initialize the device list. Since console and network
1611 * device receive input from a file descriptor, we keep an fdset 1688 * device receive input from a file descriptor, we keep an fdset
1612 * (infds) and the maximum fd number (max_infd) with the head of the 1689 * (infds) and the maximum fd number (max_infd) with the head of the
1613 * list. We also keep a pointer to the last device, for easy appending 1690 * list. We also keep a pointer to the last device. Finally, we keep
1614 * to the list. Finally, we keep the next interrupt number to hand out 1691 * the next interrupt number to hand out (1: remember that 0 is used by
1615 * (1: remember that 0 is used by the timer). */ 1692 * the timer). */
1616 FD_ZERO(&devices.infds); 1693 FD_ZERO(&devices.infds);
1617 devices.max_infd = -1; 1694 devices.max_infd = -1;
1618 devices.lastdev = &devices.dev; 1695 devices.lastdev = NULL;
1619 devices.next_irq = 1; 1696 devices.next_irq = 1;
1620 1697
1621 cpu_id = 0; 1698 cpu_id = 0;