32
33#include <sys/param.h>
34#include <sys/systm.h>
35#include <sys/malloc.h>
36#include <sys/kernel.h>
37#include <vm/vm.h>
38#include <vm/pmap.h>
39
40#include <sys/bio.h>
41#include <sys/bus.h>
42#include <sys/conf.h>
43#include <sys/module.h>
44#include <sys/sysctl.h>
45
46#include <machine/bus.h>
47#include <sys/rman.h>
48#include <machine/resource.h>
49#include <machine/intr_machdep.h>
50#include <machine/vmparam.h>
51#include <sys/bus_dma.h>
52
53#include <machine/_inttypes.h>
54#include <machine/xen/xen-os.h>
55#include <machine/xen/xenvar.h>
56#include <machine/xen/xenfunc.h>
57
58#include <xen/hypervisor.h>
59#include <xen/xen_intr.h>
60#include <xen/evtchn.h>
61#include <xen/gnttab.h>
62#include <xen/interface/grant_table.h>
63#include <xen/interface/io/protocols.h>
64#include <xen/xenbus/xenbusvar.h>
65
66#include <geom/geom_disk.h>
67
68#include <dev/xen/blkfront/block.h>
69
70#include "xenbus_if.h"
71
72/* prototypes */
73static void xb_free_command(struct xb_command *cm);
74static void xb_startio(struct xb_softc *sc);
75static void blkfront_connect(struct xb_softc *);
76static void blkfront_closing(device_t);
77static int blkfront_detach(device_t);
78static int setup_blkring(struct xb_softc *);
79static void blkif_int(void *);
80static void blkfront_initialize(struct xb_softc *);
81static int blkif_completion(struct xb_command *);
82static void blkif_free(struct xb_softc *);
83static void blkif_queue_cb(void *, bus_dma_segment_t *, int, int);
84
85static MALLOC_DEFINE(M_XENBLOCKFRONT, "xbd", "Xen Block Front driver data");
86
87#define GRANT_INVALID_REF 0
88
89/* Control whether runtime update of vbds is enabled. */
90#define ENABLE_VBD_UPDATE 0
91
92#if ENABLE_VBD_UPDATE
93static void vbd_update(void);
94#endif
95
96#define BLKIF_STATE_DISCONNECTED 0
97#define BLKIF_STATE_CONNECTED 1
98#define BLKIF_STATE_SUSPENDED 2
99
100#ifdef notyet
101static char *blkif_state_name[] = {
102 [BLKIF_STATE_DISCONNECTED] = "disconnected",
103 [BLKIF_STATE_CONNECTED] = "connected",
104 [BLKIF_STATE_SUSPENDED] = "closed",
105};
106
107static char * blkif_status_name[] = {
108 [BLKIF_INTERFACE_STATUS_CLOSED] = "closed",
109 [BLKIF_INTERFACE_STATUS_DISCONNECTED] = "disconnected",
110 [BLKIF_INTERFACE_STATUS_CONNECTED] = "connected",
111 [BLKIF_INTERFACE_STATUS_CHANGED] = "changed",
112};
113#endif
114
115#if 0
116#define DPRINTK(fmt, args...) printf("[XEN] %s:%d: " fmt ".\n", __func__, __LINE__, ##args)
117#else
118#define DPRINTK(fmt, args...)
119#endif
120
121static int blkif_open(struct disk *dp);
122static int blkif_close(struct disk *dp);
123static int blkif_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td);
124static int blkif_queue_request(struct xb_softc *sc, struct xb_command *cm);
125static void xb_strategy(struct bio *bp);
126
127// In order to quiesce the device during kernel dumps, outstanding requests to
128// DOM0 for disk reads/writes need to be accounted for.
129static int xb_dump(void *, void *, vm_offset_t, off_t, size_t);
130
131/* XXX move to xb_vbd.c when VBD update support is added */
132#define MAX_VBDS 64
133
134#define XBD_SECTOR_SIZE 512 /* XXX: assume for now */
135#define XBD_SECTOR_SHFT 9
136
137/*
138 * Translate Linux major/minor to an appropriate name and unit
139 * number. For HVM guests, this allows us to use the same drive names
140 * with blkfront as the emulated drives, easing transition slightly.
141 */
142static void
143blkfront_vdevice_to_unit(uint32_t vdevice, int *unit, const char **name)
144{
145 static struct vdev_info {
146 int major;
147 int shift;
148 int base;
149 const char *name;
150 } info[] = {
151 {3, 6, 0, "ad"}, /* ide0 */
152 {22, 6, 2, "ad"}, /* ide1 */
153 {33, 6, 4, "ad"}, /* ide2 */
154 {34, 6, 6, "ad"}, /* ide3 */
155 {56, 6, 8, "ad"}, /* ide4 */
156 {57, 6, 10, "ad"}, /* ide5 */
157 {88, 6, 12, "ad"}, /* ide6 */
158 {89, 6, 14, "ad"}, /* ide7 */
159 {90, 6, 16, "ad"}, /* ide8 */
160 {91, 6, 18, "ad"}, /* ide9 */
161
162 {8, 4, 0, "da"}, /* scsi disk0 */
163 {65, 4, 16, "da"}, /* scsi disk1 */
164 {66, 4, 32, "da"}, /* scsi disk2 */
165 {67, 4, 48, "da"}, /* scsi disk3 */
166 {68, 4, 64, "da"}, /* scsi disk4 */
167 {69, 4, 80, "da"}, /* scsi disk5 */
168 {70, 4, 96, "da"}, /* scsi disk6 */
169 {71, 4, 112, "da"}, /* scsi disk7 */
170 {128, 4, 128, "da"}, /* scsi disk8 */
171 {129, 4, 144, "da"}, /* scsi disk9 */
172 {130, 4, 160, "da"}, /* scsi disk10 */
173 {131, 4, 176, "da"}, /* scsi disk11 */
174 {132, 4, 192, "da"}, /* scsi disk12 */
175 {133, 4, 208, "da"}, /* scsi disk13 */
176 {134, 4, 224, "da"}, /* scsi disk14 */
177 {135, 4, 240, "da"}, /* scsi disk15 */
178
179 {202, 4, 0, "xbd"}, /* xbd */
180
181 {0, 0, 0, NULL},
182 };
183 int major = vdevice >> 8;
184 int minor = vdevice & 0xff;
185 int i;
186
187 if (vdevice & (1 << 28)) {
188 *unit = (vdevice & ((1 << 28) - 1)) >> 8;
189 *name = "xbd";
190 return;
191 }
192
193 for (i = 0; info[i].major; i++) {
194 if (info[i].major == major) {
195 *unit = info[i].base + (minor >> info[i].shift);
196 *name = info[i].name;
197 return;
198 }
199 }
200
201 *unit = minor >> 4;
202 *name = "xbd";
203}
204
205int
206xlvbd_add(struct xb_softc *sc, blkif_sector_t sectors,
207 int vdevice, uint16_t vdisk_info, unsigned long sector_size)
208{
209 int unit, error = 0;
210 const char *name;
211
212 blkfront_vdevice_to_unit(vdevice, &unit, &name);
213
214 sc->xb_unit = unit;
215
216 if (strcmp(name, "xbd"))
217 device_printf(sc->xb_dev, "attaching as %s%d\n", name, unit);
218
219 sc->xb_disk = disk_alloc();
220 sc->xb_disk->d_unit = sc->xb_unit;
221 sc->xb_disk->d_open = blkif_open;
222 sc->xb_disk->d_close = blkif_close;
223 sc->xb_disk->d_ioctl = blkif_ioctl;
224 sc->xb_disk->d_strategy = xb_strategy;
225 sc->xb_disk->d_dump = xb_dump;
226 sc->xb_disk->d_name = name;
227 sc->xb_disk->d_drv1 = sc;
228 sc->xb_disk->d_sectorsize = sector_size;
229
230 sc->xb_disk->d_mediasize = sectors * sector_size;
231 sc->xb_disk->d_maxsize = sc->max_request_size;
232 sc->xb_disk->d_flags = 0;
| 32
33#include <sys/param.h>
34#include <sys/systm.h>
35#include <sys/malloc.h>
36#include <sys/kernel.h>
37#include <vm/vm.h>
38#include <vm/pmap.h>
39
40#include <sys/bio.h>
41#include <sys/bus.h>
42#include <sys/conf.h>
43#include <sys/module.h>
44#include <sys/sysctl.h>
45
46#include <machine/bus.h>
47#include <sys/rman.h>
48#include <machine/resource.h>
49#include <machine/intr_machdep.h>
50#include <machine/vmparam.h>
51#include <sys/bus_dma.h>
52
53#include <machine/_inttypes.h>
54#include <machine/xen/xen-os.h>
55#include <machine/xen/xenvar.h>
56#include <machine/xen/xenfunc.h>
57
58#include <xen/hypervisor.h>
59#include <xen/xen_intr.h>
60#include <xen/evtchn.h>
61#include <xen/gnttab.h>
62#include <xen/interface/grant_table.h>
63#include <xen/interface/io/protocols.h>
64#include <xen/xenbus/xenbusvar.h>
65
66#include <geom/geom_disk.h>
67
68#include <dev/xen/blkfront/block.h>
69
70#include "xenbus_if.h"
71
72/* prototypes */
73static void xb_free_command(struct xb_command *cm);
74static void xb_startio(struct xb_softc *sc);
75static void blkfront_connect(struct xb_softc *);
76static void blkfront_closing(device_t);
77static int blkfront_detach(device_t);
78static int setup_blkring(struct xb_softc *);
79static void blkif_int(void *);
80static void blkfront_initialize(struct xb_softc *);
81static int blkif_completion(struct xb_command *);
82static void blkif_free(struct xb_softc *);
83static void blkif_queue_cb(void *, bus_dma_segment_t *, int, int);
84
85static MALLOC_DEFINE(M_XENBLOCKFRONT, "xbd", "Xen Block Front driver data");
86
87#define GRANT_INVALID_REF 0
88
89/* Control whether runtime update of vbds is enabled. */
90#define ENABLE_VBD_UPDATE 0
91
92#if ENABLE_VBD_UPDATE
93static void vbd_update(void);
94#endif
95
96#define BLKIF_STATE_DISCONNECTED 0
97#define BLKIF_STATE_CONNECTED 1
98#define BLKIF_STATE_SUSPENDED 2
99
100#ifdef notyet
101static char *blkif_state_name[] = {
102 [BLKIF_STATE_DISCONNECTED] = "disconnected",
103 [BLKIF_STATE_CONNECTED] = "connected",
104 [BLKIF_STATE_SUSPENDED] = "closed",
105};
106
107static char * blkif_status_name[] = {
108 [BLKIF_INTERFACE_STATUS_CLOSED] = "closed",
109 [BLKIF_INTERFACE_STATUS_DISCONNECTED] = "disconnected",
110 [BLKIF_INTERFACE_STATUS_CONNECTED] = "connected",
111 [BLKIF_INTERFACE_STATUS_CHANGED] = "changed",
112};
113#endif
114
115#if 0
116#define DPRINTK(fmt, args...) printf("[XEN] %s:%d: " fmt ".\n", __func__, __LINE__, ##args)
117#else
118#define DPRINTK(fmt, args...)
119#endif
120
121static int blkif_open(struct disk *dp);
122static int blkif_close(struct disk *dp);
123static int blkif_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td);
124static int blkif_queue_request(struct xb_softc *sc, struct xb_command *cm);
125static void xb_strategy(struct bio *bp);
126
127// In order to quiesce the device during kernel dumps, outstanding requests to
128// DOM0 for disk reads/writes need to be accounted for.
129static int xb_dump(void *, void *, vm_offset_t, off_t, size_t);
130
131/* XXX move to xb_vbd.c when VBD update support is added */
132#define MAX_VBDS 64
133
134#define XBD_SECTOR_SIZE 512 /* XXX: assume for now */
135#define XBD_SECTOR_SHFT 9
136
137/*
138 * Translate Linux major/minor to an appropriate name and unit
139 * number. For HVM guests, this allows us to use the same drive names
140 * with blkfront as the emulated drives, easing transition slightly.
141 */
142static void
143blkfront_vdevice_to_unit(uint32_t vdevice, int *unit, const char **name)
144{
145 static struct vdev_info {
146 int major;
147 int shift;
148 int base;
149 const char *name;
150 } info[] = {
151 {3, 6, 0, "ad"}, /* ide0 */
152 {22, 6, 2, "ad"}, /* ide1 */
153 {33, 6, 4, "ad"}, /* ide2 */
154 {34, 6, 6, "ad"}, /* ide3 */
155 {56, 6, 8, "ad"}, /* ide4 */
156 {57, 6, 10, "ad"}, /* ide5 */
157 {88, 6, 12, "ad"}, /* ide6 */
158 {89, 6, 14, "ad"}, /* ide7 */
159 {90, 6, 16, "ad"}, /* ide8 */
160 {91, 6, 18, "ad"}, /* ide9 */
161
162 {8, 4, 0, "da"}, /* scsi disk0 */
163 {65, 4, 16, "da"}, /* scsi disk1 */
164 {66, 4, 32, "da"}, /* scsi disk2 */
165 {67, 4, 48, "da"}, /* scsi disk3 */
166 {68, 4, 64, "da"}, /* scsi disk4 */
167 {69, 4, 80, "da"}, /* scsi disk5 */
168 {70, 4, 96, "da"}, /* scsi disk6 */
169 {71, 4, 112, "da"}, /* scsi disk7 */
170 {128, 4, 128, "da"}, /* scsi disk8 */
171 {129, 4, 144, "da"}, /* scsi disk9 */
172 {130, 4, 160, "da"}, /* scsi disk10 */
173 {131, 4, 176, "da"}, /* scsi disk11 */
174 {132, 4, 192, "da"}, /* scsi disk12 */
175 {133, 4, 208, "da"}, /* scsi disk13 */
176 {134, 4, 224, "da"}, /* scsi disk14 */
177 {135, 4, 240, "da"}, /* scsi disk15 */
178
179 {202, 4, 0, "xbd"}, /* xbd */
180
181 {0, 0, 0, NULL},
182 };
183 int major = vdevice >> 8;
184 int minor = vdevice & 0xff;
185 int i;
186
187 if (vdevice & (1 << 28)) {
188 *unit = (vdevice & ((1 << 28) - 1)) >> 8;
189 *name = "xbd";
190 return;
191 }
192
193 for (i = 0; info[i].major; i++) {
194 if (info[i].major == major) {
195 *unit = info[i].base + (minor >> info[i].shift);
196 *name = info[i].name;
197 return;
198 }
199 }
200
201 *unit = minor >> 4;
202 *name = "xbd";
203}
204
205int
206xlvbd_add(struct xb_softc *sc, blkif_sector_t sectors,
207 int vdevice, uint16_t vdisk_info, unsigned long sector_size)
208{
209 int unit, error = 0;
210 const char *name;
211
212 blkfront_vdevice_to_unit(vdevice, &unit, &name);
213
214 sc->xb_unit = unit;
215
216 if (strcmp(name, "xbd"))
217 device_printf(sc->xb_dev, "attaching as %s%d\n", name, unit);
218
219 sc->xb_disk = disk_alloc();
220 sc->xb_disk->d_unit = sc->xb_unit;
221 sc->xb_disk->d_open = blkif_open;
222 sc->xb_disk->d_close = blkif_close;
223 sc->xb_disk->d_ioctl = blkif_ioctl;
224 sc->xb_disk->d_strategy = xb_strategy;
225 sc->xb_disk->d_dump = xb_dump;
226 sc->xb_disk->d_name = name;
227 sc->xb_disk->d_drv1 = sc;
228 sc->xb_disk->d_sectorsize = sector_size;
229
230 sc->xb_disk->d_mediasize = sectors * sector_size;
231 sc->xb_disk->d_maxsize = sc->max_request_size;
232 sc->xb_disk->d_flags = 0;
|
234
235 return error;
236}
237
238/************************ end VBD support *****************/
239
240/*
241 * Read/write routine for a buffer. Finds the proper unit, place it on
242 * the sortq and kick the controller.
243 */
244static void
245xb_strategy(struct bio *bp)
246{
247 struct xb_softc *sc = (struct xb_softc *)bp->bio_disk->d_drv1;
248
249 /* bogus disk? */
250 if (sc == NULL) {
251 bp->bio_error = EINVAL;
252 bp->bio_flags |= BIO_ERROR;
253 bp->bio_resid = bp->bio_bcount;
254 biodone(bp);
255 return;
256 }
257
258 /*
259 * Place it in the queue of disk activities for this disk
260 */
261 mtx_lock(&sc->xb_io_lock);
262
263 xb_enqueue_bio(sc, bp);
264 xb_startio(sc);
265
266 mtx_unlock(&sc->xb_io_lock);
267 return;
268}
269
270static void
271xb_bio_complete(struct xb_softc *sc, struct xb_command *cm)
272{
273 struct bio *bp;
274
275 bp = cm->bp;
276
277 if ( unlikely(cm->status != BLKIF_RSP_OKAY) ) {
278 disk_err(bp, "disk error" , -1, 0);
279 printf(" status: %x\n", cm->status);
280 bp->bio_flags |= BIO_ERROR;
281 }
282
283 if (bp->bio_flags & BIO_ERROR)
284 bp->bio_error = EIO;
285 else
286 bp->bio_resid = 0;
287
288 xb_free_command(cm);
289 biodone(bp);
290}
291
292// Quiesce the disk writes for a dump file before allowing the next buffer.
293static void
294xb_quiesce(struct xb_softc *sc)
295{
296 int mtd;
297
298 // While there are outstanding requests
299 while (!TAILQ_EMPTY(&sc->cm_busy)) {
300 RING_FINAL_CHECK_FOR_RESPONSES(&sc->ring, mtd);
301 if (mtd) {
302 /* Recieved request completions, update queue. */
303 blkif_int(sc);
304 }
305 if (!TAILQ_EMPTY(&sc->cm_busy)) {
306 /*
307 * Still pending requests, wait for the disk i/o
308 * to complete.
309 */
310 HYPERVISOR_yield();
311 }
312 }
313}
314
315/* Kernel dump function for a paravirtualized disk device */
316static void
317xb_dump_complete(struct xb_command *cm)
318{
319
320 xb_enqueue_complete(cm);
321}
322
323static int
324xb_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset,
325 size_t length)
326{
327 struct disk *dp = arg;
328 struct xb_softc *sc = (struct xb_softc *) dp->d_drv1;
329 struct xb_command *cm;
330 size_t chunk;
331 int sbp;
332 int rc = 0;
333
334 if (length <= 0)
335 return (rc);
336
337 xb_quiesce(sc); /* All quiet on the western front. */
338
339 /*
340 * If this lock is held, then this module is failing, and a
341 * successful kernel dump is highly unlikely anyway.
342 */
343 mtx_lock(&sc->xb_io_lock);
344
345 /* Split the 64KB block as needed */
346 for (sbp=0; length > 0; sbp++) {
347 cm = xb_dequeue_free(sc);
348 if (cm == NULL) {
349 mtx_unlock(&sc->xb_io_lock);
350 device_printf(sc->xb_dev, "dump: no more commands?\n");
351 return (EBUSY);
352 }
353
354 if (gnttab_alloc_grant_references(sc->max_request_segments,
355 &cm->gref_head) != 0) {
356 xb_free_command(cm);
357 mtx_unlock(&sc->xb_io_lock);
358 device_printf(sc->xb_dev, "no more grant allocs?\n");
359 return (EBUSY);
360 }
361
362 chunk = length > sc->max_request_size
363 ? sc->max_request_size : length;
364 cm->data = virtual;
365 cm->datalen = chunk;
366 cm->operation = BLKIF_OP_WRITE;
367 cm->sector_number = offset / dp->d_sectorsize;
368 cm->cm_complete = xb_dump_complete;
369
370 xb_enqueue_ready(cm);
371
372 length -= chunk;
373 offset += chunk;
374 virtual = (char *) virtual + chunk;
375 }
376
377 /* Tell DOM0 to do the I/O */
378 xb_startio(sc);
379 mtx_unlock(&sc->xb_io_lock);
380
381 /* Poll for the completion. */
382 xb_quiesce(sc); /* All quite on the eastern front */
383
384 /* If there were any errors, bail out... */
385 while ((cm = xb_dequeue_complete(sc)) != NULL) {
386 if (cm->status != BLKIF_RSP_OKAY) {
387 device_printf(sc->xb_dev,
388 "Dump I/O failed at sector %jd\n",
389 cm->sector_number);
390 rc = EIO;
391 }
392 xb_free_command(cm);
393 }
394
395 return (rc);
396}
397
398
399static int
400blkfront_probe(device_t dev)
401{
402
403 if (!strcmp(xenbus_get_type(dev), "vbd")) {
404 device_set_desc(dev, "Virtual Block Device");
405 device_quiet(dev);
406 return (0);
407 }
408
409 return (ENXIO);
410}
411
412static void
413xb_setup_sysctl(struct xb_softc *xb)
414{
415 struct sysctl_ctx_list *sysctl_ctx = NULL;
416 struct sysctl_oid *sysctl_tree = NULL;
417
418 sysctl_ctx = device_get_sysctl_ctx(xb->xb_dev);
419 if (sysctl_ctx == NULL)
420 return;
421
422 sysctl_tree = device_get_sysctl_tree(xb->xb_dev);
423 if (sysctl_tree == NULL)
424 return;
425
426 SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
427 "max_requests", CTLFLAG_RD, &xb->max_requests, -1,
428 "maximum outstanding requests (negotiated)");
429
430 SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
431 "max_request_segments", CTLFLAG_RD,
432 &xb->max_request_segments, 0,
433 "maximum number of pages per requests (negotiated)");
434
435 SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
436 "max_request_size", CTLFLAG_RD,
437 &xb->max_request_size, 0,
438 "maximum size in bytes of a request (negotiated)");
439
440 SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
441 "ring_pages", CTLFLAG_RD,
442 &xb->ring_pages, 0,
443 "communication channel pages (negotiated)");
444}
445
446/*
447 * Setup supplies the backend dir, virtual device. We place an event
448 * channel and shared frame entries. We watch backend to wait if it's
449 * ok.
450 */
451static int
452blkfront_attach(device_t dev)
453{
454 struct xb_softc *sc;
455 const char *name;
456 uint32_t vdevice;
457 int error;
458 int i;
459 int unit;
460
461 /* FIXME: Use dynamic device id if this is not set. */
462 error = xs_scanf(XST_NIL, xenbus_get_node(dev),
463 "virtual-device", NULL, "%" PRIu32, &vdevice);
464 if (error) {
465 xenbus_dev_fatal(dev, error, "reading virtual-device");
466 device_printf(dev, "Couldn't determine virtual device.\n");
467 return (error);
468 }
469
470 blkfront_vdevice_to_unit(vdevice, &unit, &name);
471 if (!strcmp(name, "xbd"))
472 device_set_unit(dev, unit);
473
474 sc = device_get_softc(dev);
475 mtx_init(&sc->xb_io_lock, "blkfront i/o lock", NULL, MTX_DEF);
476 xb_initq_free(sc);
477 xb_initq_busy(sc);
478 xb_initq_ready(sc);
479 xb_initq_complete(sc);
480 xb_initq_bio(sc);
481 for (i = 0; i < XBF_MAX_RING_PAGES; i++)
482 sc->ring_ref[i] = GRANT_INVALID_REF;
483
484 sc->xb_dev = dev;
485 sc->vdevice = vdevice;
486 sc->connected = BLKIF_STATE_DISCONNECTED;
487
488 xb_setup_sysctl(sc);
489
490 /* Wait for backend device to publish its protocol capabilities. */
491 xenbus_set_state(dev, XenbusStateInitialising);
492
493 return (0);
494}
495
496static int
497blkfront_suspend(device_t dev)
498{
499 struct xb_softc *sc = device_get_softc(dev);
500 int retval;
501 int saved_state;
502
503 /* Prevent new requests being issued until we fix things up. */
504 mtx_lock(&sc->xb_io_lock);
505 saved_state = sc->connected;
506 sc->connected = BLKIF_STATE_SUSPENDED;
507
508 /* Wait for outstanding I/O to drain. */
509 retval = 0;
510 while (TAILQ_EMPTY(&sc->cm_busy) == 0) {
511 if (msleep(&sc->cm_busy, &sc->xb_io_lock,
512 PRIBIO, "blkf_susp", 30 * hz) == EWOULDBLOCK) {
513 retval = EBUSY;
514 break;
515 }
516 }
517 mtx_unlock(&sc->xb_io_lock);
518
519 if (retval != 0)
520 sc->connected = saved_state;
521
522 return (retval);
523}
524
525static int
526blkfront_resume(device_t dev)
527{
528 struct xb_softc *sc = device_get_softc(dev);
529
530 DPRINTK("blkfront_resume: %s\n", xenbus_get_node(dev));
531
532 blkif_free(sc);
533 blkfront_initialize(sc);
534 return (0);
535}
536
537static void
538blkfront_initialize(struct xb_softc *sc)
539{
540 const char *otherend_path;
541 const char *node_path;
542 uint32_t max_ring_page_order;
543 int error;
544 int i;
545
546 if (xenbus_get_state(sc->xb_dev) != XenbusStateInitialising) {
547 /* Initialization has already been performed. */
548 return;
549 }
550
551 /*
552 * Protocol defaults valid even if negotiation for a
553 * setting fails.
554 */
555 max_ring_page_order = 0;
556 sc->ring_pages = 1;
557 sc->max_request_segments = BLKIF_MAX_SEGMENTS_PER_HEADER_BLOCK;
558 sc->max_request_size = XBF_SEGS_TO_SIZE(sc->max_request_segments);
559 sc->max_request_blocks = BLKIF_SEGS_TO_BLOCKS(sc->max_request_segments);
560
561 /*
562 * Protocol negotiation.
563 *
564 * \note xs_gather() returns on the first encountered error, so
565 * we must use independant calls in order to guarantee
566 * we don't miss information in a sparsly populated back-end
567 * tree.
568 *
569 * \note xs_scanf() does not update variables for unmatched
570 * fields.
571 */
572 otherend_path = xenbus_get_otherend_path(sc->xb_dev);
573 node_path = xenbus_get_node(sc->xb_dev);
574
575 /* Support both backend schemes for relaying ring page limits. */
576 (void)xs_scanf(XST_NIL, otherend_path,
577 "max-ring-page-order", NULL, "%" PRIu32,
578 &max_ring_page_order);
579 sc->ring_pages = 1 << max_ring_page_order;
580 (void)xs_scanf(XST_NIL, otherend_path,
581 "max-ring-pages", NULL, "%" PRIu32,
582 &sc->ring_pages);
583 if (sc->ring_pages < 1)
584 sc->ring_pages = 1;
585
586 sc->max_requests = BLKIF_MAX_RING_REQUESTS(sc->ring_pages * PAGE_SIZE);
587 (void)xs_scanf(XST_NIL, otherend_path,
588 "max-requests", NULL, "%" PRIu32,
589 &sc->max_requests);
590
591 (void)xs_scanf(XST_NIL, otherend_path,
592 "max-request-segments", NULL, "%" PRIu32,
593 &sc->max_request_segments);
594
595 (void)xs_scanf(XST_NIL, otherend_path,
596 "max-request-size", NULL, "%" PRIu32,
597 &sc->max_request_size);
598
599 if (sc->ring_pages > XBF_MAX_RING_PAGES) {
600 device_printf(sc->xb_dev, "Back-end specified ring-pages of "
601 "%u limited to front-end limit of %zu.\n",
602 sc->ring_pages, XBF_MAX_RING_PAGES);
603 sc->ring_pages = XBF_MAX_RING_PAGES;
604 }
605
606 if (powerof2(sc->ring_pages) == 0) {
607 uint32_t new_page_limit;
608
609 new_page_limit = 0x01 << (fls(sc->ring_pages) - 1);
610 device_printf(sc->xb_dev, "Back-end specified ring-pages of "
611 "%u is not a power of 2. Limited to %u.\n",
612 sc->ring_pages, new_page_limit);
613 sc->ring_pages = new_page_limit;
614 }
615
616 if (sc->max_requests > XBF_MAX_REQUESTS) {
617 device_printf(sc->xb_dev, "Back-end specified max_requests of "
618 "%u limited to front-end limit of %u.\n",
619 sc->max_requests, XBF_MAX_REQUESTS);
620 sc->max_requests = XBF_MAX_REQUESTS;
621 }
622
623 if (sc->max_request_segments > XBF_MAX_SEGMENTS_PER_REQUEST) {
624 device_printf(sc->xb_dev, "Back-end specified "
625 "max_request_segments of %u limited to "
626 "front-end limit of %u.\n",
627 sc->max_request_segments,
628 XBF_MAX_SEGMENTS_PER_REQUEST);
629 sc->max_request_segments = XBF_MAX_SEGMENTS_PER_REQUEST;
630 }
631
632 if (sc->max_request_size > XBF_MAX_REQUEST_SIZE) {
633 device_printf(sc->xb_dev, "Back-end specified "
634 "max_request_size of %u limited to front-end "
635 "limit of %u.\n", sc->max_request_size,
636 XBF_MAX_REQUEST_SIZE);
637 sc->max_request_size = XBF_MAX_REQUEST_SIZE;
638 }
639
640 if (sc->max_request_size > XBF_SEGS_TO_SIZE(sc->max_request_segments)) {
641 device_printf(sc->xb_dev, "Back-end specified "
642 "max_request_size of %u limited to front-end "
643 "limit of %u. (Too few segments.)\n",
644 sc->max_request_size,
645 XBF_SEGS_TO_SIZE(sc->max_request_segments));
646 sc->max_request_size =
647 XBF_SEGS_TO_SIZE(sc->max_request_segments);
648 }
649
650 sc->max_request_blocks = BLKIF_SEGS_TO_BLOCKS(sc->max_request_segments);
651
652 /* Allocate datastructures based on negotiated values. */
653 error = bus_dma_tag_create(bus_get_dma_tag(sc->xb_dev), /* parent */
654 512, PAGE_SIZE, /* algnmnt, boundary */
655 BUS_SPACE_MAXADDR, /* lowaddr */
656 BUS_SPACE_MAXADDR, /* highaddr */
657 NULL, NULL, /* filter, filterarg */
658 sc->max_request_size,
659 sc->max_request_segments,
660 PAGE_SIZE, /* maxsegsize */
661 BUS_DMA_ALLOCNOW, /* flags */
662 busdma_lock_mutex, /* lockfunc */
663 &sc->xb_io_lock, /* lockarg */
664 &sc->xb_io_dmat);
665 if (error != 0) {
666 xenbus_dev_fatal(sc->xb_dev, error,
667 "Cannot allocate parent DMA tag\n");
668 return;
669 }
670
671 /* Per-transaction data allocation. */
672 sc->shadow = malloc(sizeof(*sc->shadow) * sc->max_requests,
673 M_XENBLOCKFRONT, M_NOWAIT|M_ZERO);
674 if (sc->shadow == NULL) {
675 bus_dma_tag_destroy(sc->xb_io_dmat);
676 xenbus_dev_fatal(sc->xb_dev, error,
677 "Cannot allocate request structures\n");
678 return;
679 }
680
681 for (i = 0; i < sc->max_requests; i++) {
682 struct xb_command *cm;
683
684 cm = &sc->shadow[i];
685 cm->sg_refs = malloc(sizeof(grant_ref_t)
686 * sc->max_request_segments,
687 M_XENBLOCKFRONT, M_NOWAIT);
688 if (cm->sg_refs == NULL)
689 break;
690 cm->id = i;
691 cm->cm_sc = sc;
692 if (bus_dmamap_create(sc->xb_io_dmat, 0, &cm->map) != 0)
693 break;
694 xb_free_command(cm);
695 }
696
697 if (setup_blkring(sc) != 0)
698 return;
699
700 /* Support both backend schemes for relaying ring page limits. */
701 if (sc->ring_pages > 1) {
702 error = xs_printf(XST_NIL, node_path,
703 "num-ring-pages","%u", sc->ring_pages);
704 if (error) {
705 xenbus_dev_fatal(sc->xb_dev, error,
706 "writing %s/num-ring-pages",
707 node_path);
708 return;
709 }
710
711 error = xs_printf(XST_NIL, node_path,
712 "ring-page-order", "%u",
713 fls(sc->ring_pages) - 1);
714 if (error) {
715 xenbus_dev_fatal(sc->xb_dev, error,
716 "writing %s/ring-page-order",
717 node_path);
718 return;
719 }
720 }
721
722 error = xs_printf(XST_NIL, node_path,
723 "max-requests","%u", sc->max_requests);
724 if (error) {
725 xenbus_dev_fatal(sc->xb_dev, error,
726 "writing %s/max-requests",
727 node_path);
728 return;
729 }
730
731 error = xs_printf(XST_NIL, node_path,
732 "max-request-segments","%u", sc->max_request_segments);
733 if (error) {
734 xenbus_dev_fatal(sc->xb_dev, error,
735 "writing %s/max-request-segments",
736 node_path);
737 return;
738 }
739
740 error = xs_printf(XST_NIL, node_path,
741 "max-request-size","%u", sc->max_request_size);
742 if (error) {
743 xenbus_dev_fatal(sc->xb_dev, error,
744 "writing %s/max-request-size",
745 node_path);
746 return;
747 }
748
749 error = xs_printf(XST_NIL, node_path, "event-channel",
750 "%u", irq_to_evtchn_port(sc->irq));
751 if (error) {
752 xenbus_dev_fatal(sc->xb_dev, error,
753 "writing %s/event-channel",
754 node_path);
755 return;
756 }
757
758 error = xs_printf(XST_NIL, node_path,
759 "protocol", "%s", XEN_IO_PROTO_ABI_NATIVE);
760 if (error) {
761 xenbus_dev_fatal(sc->xb_dev, error,
762 "writing %s/protocol",
763 node_path);
764 return;
765 }
766
767 xenbus_set_state(sc->xb_dev, XenbusStateInitialised);
768}
769
770static int
771setup_blkring(struct xb_softc *sc)
772{
773 blkif_sring_t *sring;
774 uintptr_t sring_page_addr;
775 int error;
776 int i;
777
778 sring = malloc(sc->ring_pages * PAGE_SIZE, M_XENBLOCKFRONT,
779 M_NOWAIT|M_ZERO);
780 if (sring == NULL) {
781 xenbus_dev_fatal(sc->xb_dev, ENOMEM, "allocating shared ring");
782 return (ENOMEM);
783 }
784 SHARED_RING_INIT(sring);
785 FRONT_RING_INIT(&sc->ring, sring, sc->ring_pages * PAGE_SIZE);
786
787 for (i = 0, sring_page_addr = (uintptr_t)sring;
788 i < sc->ring_pages;
789 i++, sring_page_addr += PAGE_SIZE) {
790
791 error = xenbus_grant_ring(sc->xb_dev,
792 (vtomach(sring_page_addr) >> PAGE_SHIFT), &sc->ring_ref[i]);
793 if (error) {
794 xenbus_dev_fatal(sc->xb_dev, error,
795 "granting ring_ref(%d)", i);
796 return (error);
797 }
798 }
799 if (sc->ring_pages == 1) {
800 error = xs_printf(XST_NIL, xenbus_get_node(sc->xb_dev),
801 "ring-ref", "%u", sc->ring_ref[0]);
802 if (error) {
803 xenbus_dev_fatal(sc->xb_dev, error,
804 "writing %s/ring-ref",
805 xenbus_get_node(sc->xb_dev));
806 return (error);
807 }
808 } else {
809 for (i = 0; i < sc->ring_pages; i++) {
810 char ring_ref_name[]= "ring_refXX";
811
812 snprintf(ring_ref_name, sizeof(ring_ref_name),
813 "ring-ref%u", i);
814 error = xs_printf(XST_NIL, xenbus_get_node(sc->xb_dev),
815 ring_ref_name, "%u", sc->ring_ref[i]);
816 if (error) {
817 xenbus_dev_fatal(sc->xb_dev, error,
818 "writing %s/%s",
819 xenbus_get_node(sc->xb_dev),
820 ring_ref_name);
821 return (error);
822 }
823 }
824 }
825
826 error = bind_listening_port_to_irqhandler(
827 xenbus_get_otherend_id(sc->xb_dev),
828 "xbd", (driver_intr_t *)blkif_int, sc,
829 INTR_TYPE_BIO | INTR_MPSAFE, &sc->irq);
830 if (error) {
831 xenbus_dev_fatal(sc->xb_dev, error,
832 "bind_evtchn_to_irqhandler failed");
833 return (error);
834 }
835
836 return (0);
837}
838
839/**
840 * Callback received when the backend's state changes.
841 */
842static void
843blkfront_backend_changed(device_t dev, XenbusState backend_state)
844{
845 struct xb_softc *sc = device_get_softc(dev);
846
847 DPRINTK("backend_state=%d\n", backend_state);
848
849 switch (backend_state) {
850 case XenbusStateUnknown:
851 case XenbusStateInitialising:
852 case XenbusStateReconfigured:
853 case XenbusStateReconfiguring:
854 case XenbusStateClosed:
855 break;
856
857 case XenbusStateInitWait:
858 case XenbusStateInitialised:
859 blkfront_initialize(sc);
860 break;
861
862 case XenbusStateConnected:
863 blkfront_initialize(sc);
864 blkfront_connect(sc);
865 break;
866
867 case XenbusStateClosing:
868 if (sc->users > 0)
869 xenbus_dev_error(dev, -EBUSY,
870 "Device in use; refusing to close");
871 else
872 blkfront_closing(dev);
873 break;
874 }
875}
876
877/*
878** Invoked when the backend is finally 'ready' (and has published
879** the details about the physical device - #sectors, size, etc).
880*/
881static void
882blkfront_connect(struct xb_softc *sc)
883{
884 device_t dev = sc->xb_dev;
885 unsigned long sectors, sector_size;
886 unsigned int binfo;
887 int err, feature_barrier;
888
889 if( (sc->connected == BLKIF_STATE_CONNECTED) ||
890 (sc->connected == BLKIF_STATE_SUSPENDED) )
891 return;
892
893 DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev));
894
895 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
896 "sectors", "%lu", §ors,
897 "info", "%u", &binfo,
898 "sector-size", "%lu", §or_size,
899 NULL);
900 if (err) {
901 xenbus_dev_fatal(dev, err,
902 "reading backend fields at %s",
903 xenbus_get_otherend_path(dev));
904 return;
905 }
906 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
907 "feature-barrier", "%lu", &feature_barrier,
908 NULL);
909 if (!err || feature_barrier)
910 sc->xb_flags |= XB_BARRIER;
911
912 if (sc->xb_disk == NULL) {
913 device_printf(dev, "%juMB <%s> at %s",
914 (uintmax_t) sectors / (1048576 / sector_size),
915 device_get_desc(dev),
916 xenbus_get_node(dev));
917 bus_print_child_footer(device_get_parent(dev), dev);
918
919 xlvbd_add(sc, sectors, sc->vdevice, binfo, sector_size);
920 }
921
922 (void)xenbus_set_state(dev, XenbusStateConnected);
923
924 /* Kick pending requests. */
925 mtx_lock(&sc->xb_io_lock);
926 sc->connected = BLKIF_STATE_CONNECTED;
927 xb_startio(sc);
928 sc->xb_flags |= XB_READY;
929 mtx_unlock(&sc->xb_io_lock);
930}
931
932/**
933 * Handle the change of state of the backend to Closing. We must delete our
934 * device-layer structures now, to ensure that writes are flushed through to
935 * the backend. Once this is done, we can switch to Closed in
936 * acknowledgement.
937 */
938static void
939blkfront_closing(device_t dev)
940{
941 struct xb_softc *sc = device_get_softc(dev);
942
943 xenbus_set_state(dev, XenbusStateClosing);
944
945 DPRINTK("blkfront_closing: %s removed\n", xenbus_get_node(dev));
946
947 if (sc->xb_disk != NULL) {
948 disk_destroy(sc->xb_disk);
949 sc->xb_disk = NULL;
950 }
951
952 xenbus_set_state(dev, XenbusStateClosed);
953}
954
955
956static int
957blkfront_detach(device_t dev)
958{
959 struct xb_softc *sc = device_get_softc(dev);
960
961 DPRINTK("blkfront_remove: %s removed\n", xenbus_get_node(dev));
962
963 blkif_free(sc);
964 mtx_destroy(&sc->xb_io_lock);
965
966 return 0;
967}
968
969
970static inline void
971flush_requests(struct xb_softc *sc)
972{
973 int notify;
974
975 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->ring, notify);
976
977 if (notify)
978 notify_remote_via_irq(sc->irq);
979}
980
981static void
982blkif_restart_queue_callback(void *arg)
983{
984 struct xb_softc *sc = arg;
985
986 mtx_lock(&sc->xb_io_lock);
987
988 xb_startio(sc);
989
990 mtx_unlock(&sc->xb_io_lock);
991}
992
993static int
994blkif_open(struct disk *dp)
995{
996 struct xb_softc *sc = (struct xb_softc *)dp->d_drv1;
997
998 if (sc == NULL) {
999 printf("xb%d: not found", sc->xb_unit);
1000 return (ENXIO);
1001 }
1002
1003 sc->xb_flags |= XB_OPEN;
1004 sc->users++;
1005 return (0);
1006}
1007
1008static int
1009blkif_close(struct disk *dp)
1010{
1011 struct xb_softc *sc = (struct xb_softc *)dp->d_drv1;
1012
1013 if (sc == NULL)
1014 return (ENXIO);
1015 sc->xb_flags &= ~XB_OPEN;
1016 if (--(sc->users) == 0) {
1017 /*
1018 * Check whether we have been instructed to close. We will
1019 * have ignored this request initially, as the device was
1020 * still mounted.
1021 */
1022 if (xenbus_get_otherend_state(sc->xb_dev) == XenbusStateClosing)
1023 blkfront_closing(sc->xb_dev);
1024 }
1025 return (0);
1026}
1027
1028static int
1029blkif_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td)
1030{
1031 struct xb_softc *sc = (struct xb_softc *)dp->d_drv1;
1032
1033 if (sc == NULL)
1034 return (ENXIO);
1035
1036 return (ENOTTY);
1037}
1038
1039static void
1040xb_free_command(struct xb_command *cm)
1041{
1042
1043 KASSERT((cm->cm_flags & XB_ON_XBQ_MASK) == 0,
1044 ("Freeing command that is still on a queue\n"));
1045
1046 cm->cm_flags = 0;
1047 cm->bp = NULL;
1048 cm->cm_complete = NULL;
1049 xb_enqueue_free(cm);
1050}
1051
1052/*
1053 * blkif_queue_request
1054 *
1055 * request block io
1056 *
1057 * id: for guest use only.
1058 * operation: BLKIF_OP_{READ,WRITE,PROBE}
1059 * buffer: buffer to read/write into. this should be a
1060 * virtual address in the guest os.
1061 */
1062static struct xb_command *
1063xb_bio_command(struct xb_softc *sc)
1064{
1065 struct xb_command *cm;
1066 struct bio *bp;
1067
1068 if (unlikely(sc->connected != BLKIF_STATE_CONNECTED))
1069 return (NULL);
1070
1071 bp = xb_dequeue_bio(sc);
1072 if (bp == NULL)
1073 return (NULL);
1074
1075 if ((cm = xb_dequeue_free(sc)) == NULL) {
1076 xb_requeue_bio(sc, bp);
1077 return (NULL);
1078 }
1079
1080 if (gnttab_alloc_grant_references(sc->max_request_segments,
1081 &cm->gref_head) != 0) {
1082 gnttab_request_free_callback(&sc->callback,
1083 blkif_restart_queue_callback, sc,
1084 sc->max_request_segments);
1085 xb_requeue_bio(sc, bp);
1086 xb_enqueue_free(cm);
1087 sc->xb_flags |= XB_FROZEN;
1088 return (NULL);
1089 }
1090
1091 cm->bp = bp;
1092 cm->data = bp->bio_data;
1093 cm->datalen = bp->bio_bcount;
1094 cm->operation = (bp->bio_cmd == BIO_READ) ? BLKIF_OP_READ :
1095 BLKIF_OP_WRITE;
1096 cm->sector_number = (blkif_sector_t)bp->bio_pblkno;
1097
1098 return (cm);
1099}
1100
1101static int
1102blkif_queue_request(struct xb_softc *sc, struct xb_command *cm)
1103{
1104 int error;
1105
1106 error = bus_dmamap_load(sc->xb_io_dmat, cm->map, cm->data, cm->datalen,
1107 blkif_queue_cb, cm, 0);
1108 if (error == EINPROGRESS) {
1109 printf("EINPROGRESS\n");
1110 sc->xb_flags |= XB_FROZEN;
1111 cm->cm_flags |= XB_CMD_FROZEN;
1112 return (0);
1113 }
1114
1115 return (error);
1116}
1117
1118static void
1119blkif_queue_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
1120{
1121 struct xb_softc *sc;
1122 struct xb_command *cm;
1123 blkif_request_t *ring_req;
1124 struct blkif_request_segment *sg;
1125 struct blkif_request_segment *last_block_sg;
1126 grant_ref_t *sg_ref;
1127 vm_paddr_t buffer_ma;
1128 uint64_t fsect, lsect;
1129 int ref;
1130 int op;
1131 int block_segs;
1132
1133 cm = arg;
1134 sc = cm->cm_sc;
1135
1136//printf("%s: Start\n", __func__);
1137 if (error) {
1138 printf("error %d in blkif_queue_cb\n", error);
1139 cm->bp->bio_error = EIO;
1140 biodone(cm->bp);
1141 xb_free_command(cm);
1142 return;
1143 }
1144
1145 /* Fill out a communications ring structure. */
1146 ring_req = RING_GET_REQUEST(&sc->ring, sc->ring.req_prod_pvt);
1147 sc->ring.req_prod_pvt++;
1148 ring_req->id = cm->id;
1149 ring_req->operation = cm->operation;
1150 ring_req->sector_number = cm->sector_number;
1151 ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xb_disk;
1152 ring_req->nr_segments = nsegs;
1153 cm->nseg = nsegs;
1154
1155 block_segs = MIN(nsegs, BLKIF_MAX_SEGMENTS_PER_HEADER_BLOCK);
1156 sg = ring_req->seg;
1157 last_block_sg = sg + block_segs;
1158 sg_ref = cm->sg_refs;
1159
1160 while (1) {
1161
1162 while (sg < last_block_sg) {
1163 buffer_ma = segs->ds_addr;
1164 fsect = (buffer_ma & PAGE_MASK) >> XBD_SECTOR_SHFT;
1165 lsect = fsect + (segs->ds_len >> XBD_SECTOR_SHFT) - 1;
1166
1167 KASSERT(lsect <= 7, ("XEN disk driver data cannot "
1168 "cross a page boundary"));
1169
1170 /* install a grant reference. */
1171 ref = gnttab_claim_grant_reference(&cm->gref_head);
1172
1173 /*
1174 * GNTTAB_LIST_END == 0xffffffff, but it is private
1175 * to gnttab.c.
1176 */
1177 KASSERT(ref != ~0, ("grant_reference failed"));
1178
1179 gnttab_grant_foreign_access_ref(
1180 ref,
1181 xenbus_get_otherend_id(sc->xb_dev),
1182 buffer_ma >> PAGE_SHIFT,
1183 ring_req->operation == BLKIF_OP_WRITE);
1184
1185 *sg_ref = ref;
1186 *sg = (struct blkif_request_segment) {
1187 .gref = ref,
1188 .first_sect = fsect,
1189 .last_sect = lsect };
1190 sg++;
1191 sg_ref++;
1192 segs++;
1193 nsegs--;
1194 }
1195 block_segs = MIN(nsegs, BLKIF_MAX_SEGMENTS_PER_SEGMENT_BLOCK);
1196 if (block_segs == 0)
1197 break;
1198
1199 sg = BLKRING_GET_SEG_BLOCK(&sc->ring, sc->ring.req_prod_pvt);
1200 sc->ring.req_prod_pvt++;
1201 last_block_sg = sg + block_segs;
1202 }
1203
1204 if (cm->operation == BLKIF_OP_READ)
1205 op = BUS_DMASYNC_PREREAD;
1206 else if (cm->operation == BLKIF_OP_WRITE)
1207 op = BUS_DMASYNC_PREWRITE;
1208 else
1209 op = 0;
1210 bus_dmamap_sync(sc->xb_io_dmat, cm->map, op);
1211
1212 gnttab_free_grant_references(cm->gref_head);
1213
1214 xb_enqueue_busy(cm);
1215
1216 /*
1217 * This flag means that we're probably executing in the busdma swi
1218 * instead of in the startio context, so an explicit flush is needed.
1219 */
1220 if (cm->cm_flags & XB_CMD_FROZEN)
1221 flush_requests(sc);
1222
1223//printf("%s: Done\n", __func__);
1224 return;
1225}
1226
1227/*
1228 * Dequeue buffers and place them in the shared communication ring.
1229 * Return when no more requests can be accepted or all buffers have
1230 * been queued.
1231 *
1232 * Signal XEN once the ring has been filled out.
1233 */
1234static void
1235xb_startio(struct xb_softc *sc)
1236{
1237 struct xb_command *cm;
1238 int error, queued = 0;
1239
1240 mtx_assert(&sc->xb_io_lock, MA_OWNED);
1241
1242 if (sc->connected != BLKIF_STATE_CONNECTED)
1243 return;
1244
1245 while (RING_FREE_REQUESTS(&sc->ring) >= sc->max_request_blocks) {
1246 if (sc->xb_flags & XB_FROZEN)
1247 break;
1248
1249 cm = xb_dequeue_ready(sc);
1250
1251 if (cm == NULL)
1252 cm = xb_bio_command(sc);
1253
1254 if (cm == NULL)
1255 break;
1256
1257 if ((error = blkif_queue_request(sc, cm)) != 0) {
1258 printf("blkif_queue_request returned %d\n", error);
1259 break;
1260 }
1261 queued++;
1262 }
1263
1264 if (queued != 0)
1265 flush_requests(sc);
1266}
1267
1268static void
1269blkif_int(void *xsc)
1270{
1271 struct xb_softc *sc = xsc;
1272 struct xb_command *cm;
1273 blkif_response_t *bret;
1274 RING_IDX i, rp;
1275 int op;
1276
1277 mtx_lock(&sc->xb_io_lock);
1278
1279 if (unlikely(sc->connected == BLKIF_STATE_DISCONNECTED)) {
1280 mtx_unlock(&sc->xb_io_lock);
1281 return;
1282 }
1283
1284 again:
1285 rp = sc->ring.sring->rsp_prod;
1286 rmb(); /* Ensure we see queued responses up to 'rp'. */
1287
1288 for (i = sc->ring.rsp_cons; i != rp;) {
1289 bret = RING_GET_RESPONSE(&sc->ring, i);
1290 cm = &sc->shadow[bret->id];
1291
1292 xb_remove_busy(cm);
1293 i += blkif_completion(cm);
1294
1295 if (cm->operation == BLKIF_OP_READ)
1296 op = BUS_DMASYNC_POSTREAD;
1297 else if (cm->operation == BLKIF_OP_WRITE)
1298 op = BUS_DMASYNC_POSTWRITE;
1299 else
1300 op = 0;
1301 bus_dmamap_sync(sc->xb_io_dmat, cm->map, op);
1302 bus_dmamap_unload(sc->xb_io_dmat, cm->map);
1303
1304 /*
1305 * If commands are completing then resources are probably
1306 * being freed as well. It's a cheap assumption even when
1307 * wrong.
1308 */
1309 sc->xb_flags &= ~XB_FROZEN;
1310
1311 /*
1312 * Directly call the i/o complete routine to save an
1313 * an indirection in the common case.
1314 */
1315 cm->status = bret->status;
1316 if (cm->bp)
1317 xb_bio_complete(sc, cm);
1318 else if (cm->cm_complete)
1319 (cm->cm_complete)(cm);
1320 else
1321 xb_free_command(cm);
1322 }
1323
1324 sc->ring.rsp_cons = i;
1325
1326 if (i != sc->ring.req_prod_pvt) {
1327 int more_to_do;
1328 RING_FINAL_CHECK_FOR_RESPONSES(&sc->ring, more_to_do);
1329 if (more_to_do)
1330 goto again;
1331 } else {
1332 sc->ring.sring->rsp_event = i + 1;
1333 }
1334
1335 xb_startio(sc);
1336
1337 if (unlikely(sc->connected == BLKIF_STATE_SUSPENDED))
1338 wakeup(&sc->cm_busy);
1339
1340 mtx_unlock(&sc->xb_io_lock);
1341}
1342
1343static void
1344blkif_free(struct xb_softc *sc)
1345{
1346 uint8_t *sring_page_ptr;
1347 int i;
1348
1349 /* Prevent new requests being issued until we fix things up. */
1350 mtx_lock(&sc->xb_io_lock);
1351 sc->connected = BLKIF_STATE_DISCONNECTED;
1352 mtx_unlock(&sc->xb_io_lock);
1353
1354 /* Free resources associated with old device channel. */
1355 if (sc->ring.sring != NULL) {
1356 sring_page_ptr = (uint8_t *)sc->ring.sring;
1357 for (i = 0; i < sc->ring_pages; i++) {
1358 if (sc->ring_ref[i] != GRANT_INVALID_REF) {
1359 gnttab_end_foreign_access_ref(sc->ring_ref[i]);
1360 sc->ring_ref[i] = GRANT_INVALID_REF;
1361 }
1362 sring_page_ptr += PAGE_SIZE;
1363 }
1364 free(sc->ring.sring, M_XENBLOCKFRONT);
1365 sc->ring.sring = NULL;
1366 }
1367
1368 if (sc->shadow) {
1369
1370 for (i = 0; i < sc->max_requests; i++) {
1371 struct xb_command *cm;
1372
1373 cm = &sc->shadow[i];
1374 if (cm->sg_refs != NULL) {
1375 free(cm->sg_refs, M_XENBLOCKFRONT);
1376 cm->sg_refs = NULL;
1377 }
1378
1379 bus_dmamap_destroy(sc->xb_io_dmat, cm->map);
1380 }
1381 free(sc->shadow, M_XENBLOCKFRONT);
1382 sc->shadow = NULL;
1383
1384 bus_dma_tag_destroy(sc->xb_io_dmat);
1385
1386 xb_initq_free(sc);
1387 xb_initq_ready(sc);
1388 xb_initq_complete(sc);
1389 }
1390
1391 if (sc->irq) {
1392 unbind_from_irqhandler(sc->irq);
1393 sc->irq = 0;
1394 }
1395}
1396
1397static int
1398blkif_completion(struct xb_command *s)
1399{
1400//printf("%s: Req %p(%d)\n", __func__, s, s->nseg);
1401 gnttab_end_foreign_access_references(s->nseg, s->sg_refs);
1402 return (BLKIF_SEGS_TO_BLOCKS(s->nseg));
1403}
1404
1405/* ** Driver registration ** */
1406static device_method_t blkfront_methods[] = {
1407 /* Device interface */
1408 DEVMETHOD(device_probe, blkfront_probe),
1409 DEVMETHOD(device_attach, blkfront_attach),
1410 DEVMETHOD(device_detach, blkfront_detach),
1411 DEVMETHOD(device_shutdown, bus_generic_shutdown),
1412 DEVMETHOD(device_suspend, blkfront_suspend),
1413 DEVMETHOD(device_resume, blkfront_resume),
1414
1415 /* Xenbus interface */
1416 DEVMETHOD(xenbus_otherend_changed, blkfront_backend_changed),
1417
1418 { 0, 0 }
1419};
1420
1421static driver_t blkfront_driver = {
1422 "xbd",
1423 blkfront_methods,
1424 sizeof(struct xb_softc),
1425};
1426devclass_t blkfront_devclass;
1427
1428DRIVER_MODULE(xbd, xenbusb_front, blkfront_driver, blkfront_devclass, 0, 0);
| 234
235 return error;
236}
237
238/************************ end VBD support *****************/
239
240/*
241 * Read/write routine for a buffer. Finds the proper unit, place it on
242 * the sortq and kick the controller.
243 */
244static void
245xb_strategy(struct bio *bp)
246{
247 struct xb_softc *sc = (struct xb_softc *)bp->bio_disk->d_drv1;
248
249 /* bogus disk? */
250 if (sc == NULL) {
251 bp->bio_error = EINVAL;
252 bp->bio_flags |= BIO_ERROR;
253 bp->bio_resid = bp->bio_bcount;
254 biodone(bp);
255 return;
256 }
257
258 /*
259 * Place it in the queue of disk activities for this disk
260 */
261 mtx_lock(&sc->xb_io_lock);
262
263 xb_enqueue_bio(sc, bp);
264 xb_startio(sc);
265
266 mtx_unlock(&sc->xb_io_lock);
267 return;
268}
269
270static void
271xb_bio_complete(struct xb_softc *sc, struct xb_command *cm)
272{
273 struct bio *bp;
274
275 bp = cm->bp;
276
277 if ( unlikely(cm->status != BLKIF_RSP_OKAY) ) {
278 disk_err(bp, "disk error" , -1, 0);
279 printf(" status: %x\n", cm->status);
280 bp->bio_flags |= BIO_ERROR;
281 }
282
283 if (bp->bio_flags & BIO_ERROR)
284 bp->bio_error = EIO;
285 else
286 bp->bio_resid = 0;
287
288 xb_free_command(cm);
289 biodone(bp);
290}
291
292// Quiesce the disk writes for a dump file before allowing the next buffer.
293static void
294xb_quiesce(struct xb_softc *sc)
295{
296 int mtd;
297
298 // While there are outstanding requests
299 while (!TAILQ_EMPTY(&sc->cm_busy)) {
300 RING_FINAL_CHECK_FOR_RESPONSES(&sc->ring, mtd);
301 if (mtd) {
302 /* Recieved request completions, update queue. */
303 blkif_int(sc);
304 }
305 if (!TAILQ_EMPTY(&sc->cm_busy)) {
306 /*
307 * Still pending requests, wait for the disk i/o
308 * to complete.
309 */
310 HYPERVISOR_yield();
311 }
312 }
313}
314
315/* Kernel dump function for a paravirtualized disk device */
316static void
317xb_dump_complete(struct xb_command *cm)
318{
319
320 xb_enqueue_complete(cm);
321}
322
323static int
324xb_dump(void *arg, void *virtual, vm_offset_t physical, off_t offset,
325 size_t length)
326{
327 struct disk *dp = arg;
328 struct xb_softc *sc = (struct xb_softc *) dp->d_drv1;
329 struct xb_command *cm;
330 size_t chunk;
331 int sbp;
332 int rc = 0;
333
334 if (length <= 0)
335 return (rc);
336
337 xb_quiesce(sc); /* All quiet on the western front. */
338
339 /*
340 * If this lock is held, then this module is failing, and a
341 * successful kernel dump is highly unlikely anyway.
342 */
343 mtx_lock(&sc->xb_io_lock);
344
345 /* Split the 64KB block as needed */
346 for (sbp=0; length > 0; sbp++) {
347 cm = xb_dequeue_free(sc);
348 if (cm == NULL) {
349 mtx_unlock(&sc->xb_io_lock);
350 device_printf(sc->xb_dev, "dump: no more commands?\n");
351 return (EBUSY);
352 }
353
354 if (gnttab_alloc_grant_references(sc->max_request_segments,
355 &cm->gref_head) != 0) {
356 xb_free_command(cm);
357 mtx_unlock(&sc->xb_io_lock);
358 device_printf(sc->xb_dev, "no more grant allocs?\n");
359 return (EBUSY);
360 }
361
362 chunk = length > sc->max_request_size
363 ? sc->max_request_size : length;
364 cm->data = virtual;
365 cm->datalen = chunk;
366 cm->operation = BLKIF_OP_WRITE;
367 cm->sector_number = offset / dp->d_sectorsize;
368 cm->cm_complete = xb_dump_complete;
369
370 xb_enqueue_ready(cm);
371
372 length -= chunk;
373 offset += chunk;
374 virtual = (char *) virtual + chunk;
375 }
376
377 /* Tell DOM0 to do the I/O */
378 xb_startio(sc);
379 mtx_unlock(&sc->xb_io_lock);
380
381 /* Poll for the completion. */
382 xb_quiesce(sc); /* All quite on the eastern front */
383
384 /* If there were any errors, bail out... */
385 while ((cm = xb_dequeue_complete(sc)) != NULL) {
386 if (cm->status != BLKIF_RSP_OKAY) {
387 device_printf(sc->xb_dev,
388 "Dump I/O failed at sector %jd\n",
389 cm->sector_number);
390 rc = EIO;
391 }
392 xb_free_command(cm);
393 }
394
395 return (rc);
396}
397
398
399static int
400blkfront_probe(device_t dev)
401{
402
403 if (!strcmp(xenbus_get_type(dev), "vbd")) {
404 device_set_desc(dev, "Virtual Block Device");
405 device_quiet(dev);
406 return (0);
407 }
408
409 return (ENXIO);
410}
411
412static void
413xb_setup_sysctl(struct xb_softc *xb)
414{
415 struct sysctl_ctx_list *sysctl_ctx = NULL;
416 struct sysctl_oid *sysctl_tree = NULL;
417
418 sysctl_ctx = device_get_sysctl_ctx(xb->xb_dev);
419 if (sysctl_ctx == NULL)
420 return;
421
422 sysctl_tree = device_get_sysctl_tree(xb->xb_dev);
423 if (sysctl_tree == NULL)
424 return;
425
426 SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
427 "max_requests", CTLFLAG_RD, &xb->max_requests, -1,
428 "maximum outstanding requests (negotiated)");
429
430 SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
431 "max_request_segments", CTLFLAG_RD,
432 &xb->max_request_segments, 0,
433 "maximum number of pages per requests (negotiated)");
434
435 SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
436 "max_request_size", CTLFLAG_RD,
437 &xb->max_request_size, 0,
438 "maximum size in bytes of a request (negotiated)");
439
440 SYSCTL_ADD_UINT(sysctl_ctx, SYSCTL_CHILDREN(sysctl_tree), OID_AUTO,
441 "ring_pages", CTLFLAG_RD,
442 &xb->ring_pages, 0,
443 "communication channel pages (negotiated)");
444}
445
446/*
447 * Setup supplies the backend dir, virtual device. We place an event
448 * channel and shared frame entries. We watch backend to wait if it's
449 * ok.
450 */
451static int
452blkfront_attach(device_t dev)
453{
454 struct xb_softc *sc;
455 const char *name;
456 uint32_t vdevice;
457 int error;
458 int i;
459 int unit;
460
461 /* FIXME: Use dynamic device id if this is not set. */
462 error = xs_scanf(XST_NIL, xenbus_get_node(dev),
463 "virtual-device", NULL, "%" PRIu32, &vdevice);
464 if (error) {
465 xenbus_dev_fatal(dev, error, "reading virtual-device");
466 device_printf(dev, "Couldn't determine virtual device.\n");
467 return (error);
468 }
469
470 blkfront_vdevice_to_unit(vdevice, &unit, &name);
471 if (!strcmp(name, "xbd"))
472 device_set_unit(dev, unit);
473
474 sc = device_get_softc(dev);
475 mtx_init(&sc->xb_io_lock, "blkfront i/o lock", NULL, MTX_DEF);
476 xb_initq_free(sc);
477 xb_initq_busy(sc);
478 xb_initq_ready(sc);
479 xb_initq_complete(sc);
480 xb_initq_bio(sc);
481 for (i = 0; i < XBF_MAX_RING_PAGES; i++)
482 sc->ring_ref[i] = GRANT_INVALID_REF;
483
484 sc->xb_dev = dev;
485 sc->vdevice = vdevice;
486 sc->connected = BLKIF_STATE_DISCONNECTED;
487
488 xb_setup_sysctl(sc);
489
490 /* Wait for backend device to publish its protocol capabilities. */
491 xenbus_set_state(dev, XenbusStateInitialising);
492
493 return (0);
494}
495
496static int
497blkfront_suspend(device_t dev)
498{
499 struct xb_softc *sc = device_get_softc(dev);
500 int retval;
501 int saved_state;
502
503 /* Prevent new requests being issued until we fix things up. */
504 mtx_lock(&sc->xb_io_lock);
505 saved_state = sc->connected;
506 sc->connected = BLKIF_STATE_SUSPENDED;
507
508 /* Wait for outstanding I/O to drain. */
509 retval = 0;
510 while (TAILQ_EMPTY(&sc->cm_busy) == 0) {
511 if (msleep(&sc->cm_busy, &sc->xb_io_lock,
512 PRIBIO, "blkf_susp", 30 * hz) == EWOULDBLOCK) {
513 retval = EBUSY;
514 break;
515 }
516 }
517 mtx_unlock(&sc->xb_io_lock);
518
519 if (retval != 0)
520 sc->connected = saved_state;
521
522 return (retval);
523}
524
525static int
526blkfront_resume(device_t dev)
527{
528 struct xb_softc *sc = device_get_softc(dev);
529
530 DPRINTK("blkfront_resume: %s\n", xenbus_get_node(dev));
531
532 blkif_free(sc);
533 blkfront_initialize(sc);
534 return (0);
535}
536
537static void
538blkfront_initialize(struct xb_softc *sc)
539{
540 const char *otherend_path;
541 const char *node_path;
542 uint32_t max_ring_page_order;
543 int error;
544 int i;
545
546 if (xenbus_get_state(sc->xb_dev) != XenbusStateInitialising) {
547 /* Initialization has already been performed. */
548 return;
549 }
550
551 /*
552 * Protocol defaults valid even if negotiation for a
553 * setting fails.
554 */
555 max_ring_page_order = 0;
556 sc->ring_pages = 1;
557 sc->max_request_segments = BLKIF_MAX_SEGMENTS_PER_HEADER_BLOCK;
558 sc->max_request_size = XBF_SEGS_TO_SIZE(sc->max_request_segments);
559 sc->max_request_blocks = BLKIF_SEGS_TO_BLOCKS(sc->max_request_segments);
560
561 /*
562 * Protocol negotiation.
563 *
564 * \note xs_gather() returns on the first encountered error, so
565 * we must use independant calls in order to guarantee
566 * we don't miss information in a sparsly populated back-end
567 * tree.
568 *
569 * \note xs_scanf() does not update variables for unmatched
570 * fields.
571 */
572 otherend_path = xenbus_get_otherend_path(sc->xb_dev);
573 node_path = xenbus_get_node(sc->xb_dev);
574
575 /* Support both backend schemes for relaying ring page limits. */
576 (void)xs_scanf(XST_NIL, otherend_path,
577 "max-ring-page-order", NULL, "%" PRIu32,
578 &max_ring_page_order);
579 sc->ring_pages = 1 << max_ring_page_order;
580 (void)xs_scanf(XST_NIL, otherend_path,
581 "max-ring-pages", NULL, "%" PRIu32,
582 &sc->ring_pages);
583 if (sc->ring_pages < 1)
584 sc->ring_pages = 1;
585
586 sc->max_requests = BLKIF_MAX_RING_REQUESTS(sc->ring_pages * PAGE_SIZE);
587 (void)xs_scanf(XST_NIL, otherend_path,
588 "max-requests", NULL, "%" PRIu32,
589 &sc->max_requests);
590
591 (void)xs_scanf(XST_NIL, otherend_path,
592 "max-request-segments", NULL, "%" PRIu32,
593 &sc->max_request_segments);
594
595 (void)xs_scanf(XST_NIL, otherend_path,
596 "max-request-size", NULL, "%" PRIu32,
597 &sc->max_request_size);
598
599 if (sc->ring_pages > XBF_MAX_RING_PAGES) {
600 device_printf(sc->xb_dev, "Back-end specified ring-pages of "
601 "%u limited to front-end limit of %zu.\n",
602 sc->ring_pages, XBF_MAX_RING_PAGES);
603 sc->ring_pages = XBF_MAX_RING_PAGES;
604 }
605
606 if (powerof2(sc->ring_pages) == 0) {
607 uint32_t new_page_limit;
608
609 new_page_limit = 0x01 << (fls(sc->ring_pages) - 1);
610 device_printf(sc->xb_dev, "Back-end specified ring-pages of "
611 "%u is not a power of 2. Limited to %u.\n",
612 sc->ring_pages, new_page_limit);
613 sc->ring_pages = new_page_limit;
614 }
615
616 if (sc->max_requests > XBF_MAX_REQUESTS) {
617 device_printf(sc->xb_dev, "Back-end specified max_requests of "
618 "%u limited to front-end limit of %u.\n",
619 sc->max_requests, XBF_MAX_REQUESTS);
620 sc->max_requests = XBF_MAX_REQUESTS;
621 }
622
623 if (sc->max_request_segments > XBF_MAX_SEGMENTS_PER_REQUEST) {
624 device_printf(sc->xb_dev, "Back-end specified "
625 "max_request_segments of %u limited to "
626 "front-end limit of %u.\n",
627 sc->max_request_segments,
628 XBF_MAX_SEGMENTS_PER_REQUEST);
629 sc->max_request_segments = XBF_MAX_SEGMENTS_PER_REQUEST;
630 }
631
632 if (sc->max_request_size > XBF_MAX_REQUEST_SIZE) {
633 device_printf(sc->xb_dev, "Back-end specified "
634 "max_request_size of %u limited to front-end "
635 "limit of %u.\n", sc->max_request_size,
636 XBF_MAX_REQUEST_SIZE);
637 sc->max_request_size = XBF_MAX_REQUEST_SIZE;
638 }
639
640 if (sc->max_request_size > XBF_SEGS_TO_SIZE(sc->max_request_segments)) {
641 device_printf(sc->xb_dev, "Back-end specified "
642 "max_request_size of %u limited to front-end "
643 "limit of %u. (Too few segments.)\n",
644 sc->max_request_size,
645 XBF_SEGS_TO_SIZE(sc->max_request_segments));
646 sc->max_request_size =
647 XBF_SEGS_TO_SIZE(sc->max_request_segments);
648 }
649
650 sc->max_request_blocks = BLKIF_SEGS_TO_BLOCKS(sc->max_request_segments);
651
652 /* Allocate datastructures based on negotiated values. */
653 error = bus_dma_tag_create(bus_get_dma_tag(sc->xb_dev), /* parent */
654 512, PAGE_SIZE, /* algnmnt, boundary */
655 BUS_SPACE_MAXADDR, /* lowaddr */
656 BUS_SPACE_MAXADDR, /* highaddr */
657 NULL, NULL, /* filter, filterarg */
658 sc->max_request_size,
659 sc->max_request_segments,
660 PAGE_SIZE, /* maxsegsize */
661 BUS_DMA_ALLOCNOW, /* flags */
662 busdma_lock_mutex, /* lockfunc */
663 &sc->xb_io_lock, /* lockarg */
664 &sc->xb_io_dmat);
665 if (error != 0) {
666 xenbus_dev_fatal(sc->xb_dev, error,
667 "Cannot allocate parent DMA tag\n");
668 return;
669 }
670
671 /* Per-transaction data allocation. */
672 sc->shadow = malloc(sizeof(*sc->shadow) * sc->max_requests,
673 M_XENBLOCKFRONT, M_NOWAIT|M_ZERO);
674 if (sc->shadow == NULL) {
675 bus_dma_tag_destroy(sc->xb_io_dmat);
676 xenbus_dev_fatal(sc->xb_dev, error,
677 "Cannot allocate request structures\n");
678 return;
679 }
680
681 for (i = 0; i < sc->max_requests; i++) {
682 struct xb_command *cm;
683
684 cm = &sc->shadow[i];
685 cm->sg_refs = malloc(sizeof(grant_ref_t)
686 * sc->max_request_segments,
687 M_XENBLOCKFRONT, M_NOWAIT);
688 if (cm->sg_refs == NULL)
689 break;
690 cm->id = i;
691 cm->cm_sc = sc;
692 if (bus_dmamap_create(sc->xb_io_dmat, 0, &cm->map) != 0)
693 break;
694 xb_free_command(cm);
695 }
696
697 if (setup_blkring(sc) != 0)
698 return;
699
700 /* Support both backend schemes for relaying ring page limits. */
701 if (sc->ring_pages > 1) {
702 error = xs_printf(XST_NIL, node_path,
703 "num-ring-pages","%u", sc->ring_pages);
704 if (error) {
705 xenbus_dev_fatal(sc->xb_dev, error,
706 "writing %s/num-ring-pages",
707 node_path);
708 return;
709 }
710
711 error = xs_printf(XST_NIL, node_path,
712 "ring-page-order", "%u",
713 fls(sc->ring_pages) - 1);
714 if (error) {
715 xenbus_dev_fatal(sc->xb_dev, error,
716 "writing %s/ring-page-order",
717 node_path);
718 return;
719 }
720 }
721
722 error = xs_printf(XST_NIL, node_path,
723 "max-requests","%u", sc->max_requests);
724 if (error) {
725 xenbus_dev_fatal(sc->xb_dev, error,
726 "writing %s/max-requests",
727 node_path);
728 return;
729 }
730
731 error = xs_printf(XST_NIL, node_path,
732 "max-request-segments","%u", sc->max_request_segments);
733 if (error) {
734 xenbus_dev_fatal(sc->xb_dev, error,
735 "writing %s/max-request-segments",
736 node_path);
737 return;
738 }
739
740 error = xs_printf(XST_NIL, node_path,
741 "max-request-size","%u", sc->max_request_size);
742 if (error) {
743 xenbus_dev_fatal(sc->xb_dev, error,
744 "writing %s/max-request-size",
745 node_path);
746 return;
747 }
748
749 error = xs_printf(XST_NIL, node_path, "event-channel",
750 "%u", irq_to_evtchn_port(sc->irq));
751 if (error) {
752 xenbus_dev_fatal(sc->xb_dev, error,
753 "writing %s/event-channel",
754 node_path);
755 return;
756 }
757
758 error = xs_printf(XST_NIL, node_path,
759 "protocol", "%s", XEN_IO_PROTO_ABI_NATIVE);
760 if (error) {
761 xenbus_dev_fatal(sc->xb_dev, error,
762 "writing %s/protocol",
763 node_path);
764 return;
765 }
766
767 xenbus_set_state(sc->xb_dev, XenbusStateInitialised);
768}
769
770static int
771setup_blkring(struct xb_softc *sc)
772{
773 blkif_sring_t *sring;
774 uintptr_t sring_page_addr;
775 int error;
776 int i;
777
778 sring = malloc(sc->ring_pages * PAGE_SIZE, M_XENBLOCKFRONT,
779 M_NOWAIT|M_ZERO);
780 if (sring == NULL) {
781 xenbus_dev_fatal(sc->xb_dev, ENOMEM, "allocating shared ring");
782 return (ENOMEM);
783 }
784 SHARED_RING_INIT(sring);
785 FRONT_RING_INIT(&sc->ring, sring, sc->ring_pages * PAGE_SIZE);
786
787 for (i = 0, sring_page_addr = (uintptr_t)sring;
788 i < sc->ring_pages;
789 i++, sring_page_addr += PAGE_SIZE) {
790
791 error = xenbus_grant_ring(sc->xb_dev,
792 (vtomach(sring_page_addr) >> PAGE_SHIFT), &sc->ring_ref[i]);
793 if (error) {
794 xenbus_dev_fatal(sc->xb_dev, error,
795 "granting ring_ref(%d)", i);
796 return (error);
797 }
798 }
799 if (sc->ring_pages == 1) {
800 error = xs_printf(XST_NIL, xenbus_get_node(sc->xb_dev),
801 "ring-ref", "%u", sc->ring_ref[0]);
802 if (error) {
803 xenbus_dev_fatal(sc->xb_dev, error,
804 "writing %s/ring-ref",
805 xenbus_get_node(sc->xb_dev));
806 return (error);
807 }
808 } else {
809 for (i = 0; i < sc->ring_pages; i++) {
810 char ring_ref_name[]= "ring_refXX";
811
812 snprintf(ring_ref_name, sizeof(ring_ref_name),
813 "ring-ref%u", i);
814 error = xs_printf(XST_NIL, xenbus_get_node(sc->xb_dev),
815 ring_ref_name, "%u", sc->ring_ref[i]);
816 if (error) {
817 xenbus_dev_fatal(sc->xb_dev, error,
818 "writing %s/%s",
819 xenbus_get_node(sc->xb_dev),
820 ring_ref_name);
821 return (error);
822 }
823 }
824 }
825
826 error = bind_listening_port_to_irqhandler(
827 xenbus_get_otherend_id(sc->xb_dev),
828 "xbd", (driver_intr_t *)blkif_int, sc,
829 INTR_TYPE_BIO | INTR_MPSAFE, &sc->irq);
830 if (error) {
831 xenbus_dev_fatal(sc->xb_dev, error,
832 "bind_evtchn_to_irqhandler failed");
833 return (error);
834 }
835
836 return (0);
837}
838
839/**
840 * Callback received when the backend's state changes.
841 */
842static void
843blkfront_backend_changed(device_t dev, XenbusState backend_state)
844{
845 struct xb_softc *sc = device_get_softc(dev);
846
847 DPRINTK("backend_state=%d\n", backend_state);
848
849 switch (backend_state) {
850 case XenbusStateUnknown:
851 case XenbusStateInitialising:
852 case XenbusStateReconfigured:
853 case XenbusStateReconfiguring:
854 case XenbusStateClosed:
855 break;
856
857 case XenbusStateInitWait:
858 case XenbusStateInitialised:
859 blkfront_initialize(sc);
860 break;
861
862 case XenbusStateConnected:
863 blkfront_initialize(sc);
864 blkfront_connect(sc);
865 break;
866
867 case XenbusStateClosing:
868 if (sc->users > 0)
869 xenbus_dev_error(dev, -EBUSY,
870 "Device in use; refusing to close");
871 else
872 blkfront_closing(dev);
873 break;
874 }
875}
876
877/*
878** Invoked when the backend is finally 'ready' (and has published
879** the details about the physical device - #sectors, size, etc).
880*/
881static void
882blkfront_connect(struct xb_softc *sc)
883{
884 device_t dev = sc->xb_dev;
885 unsigned long sectors, sector_size;
886 unsigned int binfo;
887 int err, feature_barrier;
888
889 if( (sc->connected == BLKIF_STATE_CONNECTED) ||
890 (sc->connected == BLKIF_STATE_SUSPENDED) )
891 return;
892
893 DPRINTK("blkfront.c:connect:%s.\n", xenbus_get_otherend_path(dev));
894
895 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
896 "sectors", "%lu", §ors,
897 "info", "%u", &binfo,
898 "sector-size", "%lu", §or_size,
899 NULL);
900 if (err) {
901 xenbus_dev_fatal(dev, err,
902 "reading backend fields at %s",
903 xenbus_get_otherend_path(dev));
904 return;
905 }
906 err = xs_gather(XST_NIL, xenbus_get_otherend_path(dev),
907 "feature-barrier", "%lu", &feature_barrier,
908 NULL);
909 if (!err || feature_barrier)
910 sc->xb_flags |= XB_BARRIER;
911
912 if (sc->xb_disk == NULL) {
913 device_printf(dev, "%juMB <%s> at %s",
914 (uintmax_t) sectors / (1048576 / sector_size),
915 device_get_desc(dev),
916 xenbus_get_node(dev));
917 bus_print_child_footer(device_get_parent(dev), dev);
918
919 xlvbd_add(sc, sectors, sc->vdevice, binfo, sector_size);
920 }
921
922 (void)xenbus_set_state(dev, XenbusStateConnected);
923
924 /* Kick pending requests. */
925 mtx_lock(&sc->xb_io_lock);
926 sc->connected = BLKIF_STATE_CONNECTED;
927 xb_startio(sc);
928 sc->xb_flags |= XB_READY;
929 mtx_unlock(&sc->xb_io_lock);
930}
931
932/**
933 * Handle the change of state of the backend to Closing. We must delete our
934 * device-layer structures now, to ensure that writes are flushed through to
935 * the backend. Once this is done, we can switch to Closed in
936 * acknowledgement.
937 */
938static void
939blkfront_closing(device_t dev)
940{
941 struct xb_softc *sc = device_get_softc(dev);
942
943 xenbus_set_state(dev, XenbusStateClosing);
944
945 DPRINTK("blkfront_closing: %s removed\n", xenbus_get_node(dev));
946
947 if (sc->xb_disk != NULL) {
948 disk_destroy(sc->xb_disk);
949 sc->xb_disk = NULL;
950 }
951
952 xenbus_set_state(dev, XenbusStateClosed);
953}
954
955
956static int
957blkfront_detach(device_t dev)
958{
959 struct xb_softc *sc = device_get_softc(dev);
960
961 DPRINTK("blkfront_remove: %s removed\n", xenbus_get_node(dev));
962
963 blkif_free(sc);
964 mtx_destroy(&sc->xb_io_lock);
965
966 return 0;
967}
968
969
970static inline void
971flush_requests(struct xb_softc *sc)
972{
973 int notify;
974
975 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->ring, notify);
976
977 if (notify)
978 notify_remote_via_irq(sc->irq);
979}
980
981static void
982blkif_restart_queue_callback(void *arg)
983{
984 struct xb_softc *sc = arg;
985
986 mtx_lock(&sc->xb_io_lock);
987
988 xb_startio(sc);
989
990 mtx_unlock(&sc->xb_io_lock);
991}
992
993static int
994blkif_open(struct disk *dp)
995{
996 struct xb_softc *sc = (struct xb_softc *)dp->d_drv1;
997
998 if (sc == NULL) {
999 printf("xb%d: not found", sc->xb_unit);
1000 return (ENXIO);
1001 }
1002
1003 sc->xb_flags |= XB_OPEN;
1004 sc->users++;
1005 return (0);
1006}
1007
1008static int
1009blkif_close(struct disk *dp)
1010{
1011 struct xb_softc *sc = (struct xb_softc *)dp->d_drv1;
1012
1013 if (sc == NULL)
1014 return (ENXIO);
1015 sc->xb_flags &= ~XB_OPEN;
1016 if (--(sc->users) == 0) {
1017 /*
1018 * Check whether we have been instructed to close. We will
1019 * have ignored this request initially, as the device was
1020 * still mounted.
1021 */
1022 if (xenbus_get_otherend_state(sc->xb_dev) == XenbusStateClosing)
1023 blkfront_closing(sc->xb_dev);
1024 }
1025 return (0);
1026}
1027
1028static int
1029blkif_ioctl(struct disk *dp, u_long cmd, void *addr, int flag, struct thread *td)
1030{
1031 struct xb_softc *sc = (struct xb_softc *)dp->d_drv1;
1032
1033 if (sc == NULL)
1034 return (ENXIO);
1035
1036 return (ENOTTY);
1037}
1038
1039static void
1040xb_free_command(struct xb_command *cm)
1041{
1042
1043 KASSERT((cm->cm_flags & XB_ON_XBQ_MASK) == 0,
1044 ("Freeing command that is still on a queue\n"));
1045
1046 cm->cm_flags = 0;
1047 cm->bp = NULL;
1048 cm->cm_complete = NULL;
1049 xb_enqueue_free(cm);
1050}
1051
1052/*
1053 * blkif_queue_request
1054 *
1055 * request block io
1056 *
1057 * id: for guest use only.
1058 * operation: BLKIF_OP_{READ,WRITE,PROBE}
1059 * buffer: buffer to read/write into. this should be a
1060 * virtual address in the guest os.
1061 */
1062static struct xb_command *
1063xb_bio_command(struct xb_softc *sc)
1064{
1065 struct xb_command *cm;
1066 struct bio *bp;
1067
1068 if (unlikely(sc->connected != BLKIF_STATE_CONNECTED))
1069 return (NULL);
1070
1071 bp = xb_dequeue_bio(sc);
1072 if (bp == NULL)
1073 return (NULL);
1074
1075 if ((cm = xb_dequeue_free(sc)) == NULL) {
1076 xb_requeue_bio(sc, bp);
1077 return (NULL);
1078 }
1079
1080 if (gnttab_alloc_grant_references(sc->max_request_segments,
1081 &cm->gref_head) != 0) {
1082 gnttab_request_free_callback(&sc->callback,
1083 blkif_restart_queue_callback, sc,
1084 sc->max_request_segments);
1085 xb_requeue_bio(sc, bp);
1086 xb_enqueue_free(cm);
1087 sc->xb_flags |= XB_FROZEN;
1088 return (NULL);
1089 }
1090
1091 cm->bp = bp;
1092 cm->data = bp->bio_data;
1093 cm->datalen = bp->bio_bcount;
1094 cm->operation = (bp->bio_cmd == BIO_READ) ? BLKIF_OP_READ :
1095 BLKIF_OP_WRITE;
1096 cm->sector_number = (blkif_sector_t)bp->bio_pblkno;
1097
1098 return (cm);
1099}
1100
1101static int
1102blkif_queue_request(struct xb_softc *sc, struct xb_command *cm)
1103{
1104 int error;
1105
1106 error = bus_dmamap_load(sc->xb_io_dmat, cm->map, cm->data, cm->datalen,
1107 blkif_queue_cb, cm, 0);
1108 if (error == EINPROGRESS) {
1109 printf("EINPROGRESS\n");
1110 sc->xb_flags |= XB_FROZEN;
1111 cm->cm_flags |= XB_CMD_FROZEN;
1112 return (0);
1113 }
1114
1115 return (error);
1116}
1117
1118static void
1119blkif_queue_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
1120{
1121 struct xb_softc *sc;
1122 struct xb_command *cm;
1123 blkif_request_t *ring_req;
1124 struct blkif_request_segment *sg;
1125 struct blkif_request_segment *last_block_sg;
1126 grant_ref_t *sg_ref;
1127 vm_paddr_t buffer_ma;
1128 uint64_t fsect, lsect;
1129 int ref;
1130 int op;
1131 int block_segs;
1132
1133 cm = arg;
1134 sc = cm->cm_sc;
1135
1136//printf("%s: Start\n", __func__);
1137 if (error) {
1138 printf("error %d in blkif_queue_cb\n", error);
1139 cm->bp->bio_error = EIO;
1140 biodone(cm->bp);
1141 xb_free_command(cm);
1142 return;
1143 }
1144
1145 /* Fill out a communications ring structure. */
1146 ring_req = RING_GET_REQUEST(&sc->ring, sc->ring.req_prod_pvt);
1147 sc->ring.req_prod_pvt++;
1148 ring_req->id = cm->id;
1149 ring_req->operation = cm->operation;
1150 ring_req->sector_number = cm->sector_number;
1151 ring_req->handle = (blkif_vdev_t)(uintptr_t)sc->xb_disk;
1152 ring_req->nr_segments = nsegs;
1153 cm->nseg = nsegs;
1154
1155 block_segs = MIN(nsegs, BLKIF_MAX_SEGMENTS_PER_HEADER_BLOCK);
1156 sg = ring_req->seg;
1157 last_block_sg = sg + block_segs;
1158 sg_ref = cm->sg_refs;
1159
1160 while (1) {
1161
1162 while (sg < last_block_sg) {
1163 buffer_ma = segs->ds_addr;
1164 fsect = (buffer_ma & PAGE_MASK) >> XBD_SECTOR_SHFT;
1165 lsect = fsect + (segs->ds_len >> XBD_SECTOR_SHFT) - 1;
1166
1167 KASSERT(lsect <= 7, ("XEN disk driver data cannot "
1168 "cross a page boundary"));
1169
1170 /* install a grant reference. */
1171 ref = gnttab_claim_grant_reference(&cm->gref_head);
1172
1173 /*
1174 * GNTTAB_LIST_END == 0xffffffff, but it is private
1175 * to gnttab.c.
1176 */
1177 KASSERT(ref != ~0, ("grant_reference failed"));
1178
1179 gnttab_grant_foreign_access_ref(
1180 ref,
1181 xenbus_get_otherend_id(sc->xb_dev),
1182 buffer_ma >> PAGE_SHIFT,
1183 ring_req->operation == BLKIF_OP_WRITE);
1184
1185 *sg_ref = ref;
1186 *sg = (struct blkif_request_segment) {
1187 .gref = ref,
1188 .first_sect = fsect,
1189 .last_sect = lsect };
1190 sg++;
1191 sg_ref++;
1192 segs++;
1193 nsegs--;
1194 }
1195 block_segs = MIN(nsegs, BLKIF_MAX_SEGMENTS_PER_SEGMENT_BLOCK);
1196 if (block_segs == 0)
1197 break;
1198
1199 sg = BLKRING_GET_SEG_BLOCK(&sc->ring, sc->ring.req_prod_pvt);
1200 sc->ring.req_prod_pvt++;
1201 last_block_sg = sg + block_segs;
1202 }
1203
1204 if (cm->operation == BLKIF_OP_READ)
1205 op = BUS_DMASYNC_PREREAD;
1206 else if (cm->operation == BLKIF_OP_WRITE)
1207 op = BUS_DMASYNC_PREWRITE;
1208 else
1209 op = 0;
1210 bus_dmamap_sync(sc->xb_io_dmat, cm->map, op);
1211
1212 gnttab_free_grant_references(cm->gref_head);
1213
1214 xb_enqueue_busy(cm);
1215
1216 /*
1217 * This flag means that we're probably executing in the busdma swi
1218 * instead of in the startio context, so an explicit flush is needed.
1219 */
1220 if (cm->cm_flags & XB_CMD_FROZEN)
1221 flush_requests(sc);
1222
1223//printf("%s: Done\n", __func__);
1224 return;
1225}
1226
1227/*
1228 * Dequeue buffers and place them in the shared communication ring.
1229 * Return when no more requests can be accepted or all buffers have
1230 * been queued.
1231 *
1232 * Signal XEN once the ring has been filled out.
1233 */
1234static void
1235xb_startio(struct xb_softc *sc)
1236{
1237 struct xb_command *cm;
1238 int error, queued = 0;
1239
1240 mtx_assert(&sc->xb_io_lock, MA_OWNED);
1241
1242 if (sc->connected != BLKIF_STATE_CONNECTED)
1243 return;
1244
1245 while (RING_FREE_REQUESTS(&sc->ring) >= sc->max_request_blocks) {
1246 if (sc->xb_flags & XB_FROZEN)
1247 break;
1248
1249 cm = xb_dequeue_ready(sc);
1250
1251 if (cm == NULL)
1252 cm = xb_bio_command(sc);
1253
1254 if (cm == NULL)
1255 break;
1256
1257 if ((error = blkif_queue_request(sc, cm)) != 0) {
1258 printf("blkif_queue_request returned %d\n", error);
1259 break;
1260 }
1261 queued++;
1262 }
1263
1264 if (queued != 0)
1265 flush_requests(sc);
1266}
1267
1268static void
1269blkif_int(void *xsc)
1270{
1271 struct xb_softc *sc = xsc;
1272 struct xb_command *cm;
1273 blkif_response_t *bret;
1274 RING_IDX i, rp;
1275 int op;
1276
1277 mtx_lock(&sc->xb_io_lock);
1278
1279 if (unlikely(sc->connected == BLKIF_STATE_DISCONNECTED)) {
1280 mtx_unlock(&sc->xb_io_lock);
1281 return;
1282 }
1283
1284 again:
1285 rp = sc->ring.sring->rsp_prod;
1286 rmb(); /* Ensure we see queued responses up to 'rp'. */
1287
1288 for (i = sc->ring.rsp_cons; i != rp;) {
1289 bret = RING_GET_RESPONSE(&sc->ring, i);
1290 cm = &sc->shadow[bret->id];
1291
1292 xb_remove_busy(cm);
1293 i += blkif_completion(cm);
1294
1295 if (cm->operation == BLKIF_OP_READ)
1296 op = BUS_DMASYNC_POSTREAD;
1297 else if (cm->operation == BLKIF_OP_WRITE)
1298 op = BUS_DMASYNC_POSTWRITE;
1299 else
1300 op = 0;
1301 bus_dmamap_sync(sc->xb_io_dmat, cm->map, op);
1302 bus_dmamap_unload(sc->xb_io_dmat, cm->map);
1303
1304 /*
1305 * If commands are completing then resources are probably
1306 * being freed as well. It's a cheap assumption even when
1307 * wrong.
1308 */
1309 sc->xb_flags &= ~XB_FROZEN;
1310
1311 /*
1312 * Directly call the i/o complete routine to save an
1313 * an indirection in the common case.
1314 */
1315 cm->status = bret->status;
1316 if (cm->bp)
1317 xb_bio_complete(sc, cm);
1318 else if (cm->cm_complete)
1319 (cm->cm_complete)(cm);
1320 else
1321 xb_free_command(cm);
1322 }
1323
1324 sc->ring.rsp_cons = i;
1325
1326 if (i != sc->ring.req_prod_pvt) {
1327 int more_to_do;
1328 RING_FINAL_CHECK_FOR_RESPONSES(&sc->ring, more_to_do);
1329 if (more_to_do)
1330 goto again;
1331 } else {
1332 sc->ring.sring->rsp_event = i + 1;
1333 }
1334
1335 xb_startio(sc);
1336
1337 if (unlikely(sc->connected == BLKIF_STATE_SUSPENDED))
1338 wakeup(&sc->cm_busy);
1339
1340 mtx_unlock(&sc->xb_io_lock);
1341}
1342
1343static void
1344blkif_free(struct xb_softc *sc)
1345{
1346 uint8_t *sring_page_ptr;
1347 int i;
1348
1349 /* Prevent new requests being issued until we fix things up. */
1350 mtx_lock(&sc->xb_io_lock);
1351 sc->connected = BLKIF_STATE_DISCONNECTED;
1352 mtx_unlock(&sc->xb_io_lock);
1353
1354 /* Free resources associated with old device channel. */
1355 if (sc->ring.sring != NULL) {
1356 sring_page_ptr = (uint8_t *)sc->ring.sring;
1357 for (i = 0; i < sc->ring_pages; i++) {
1358 if (sc->ring_ref[i] != GRANT_INVALID_REF) {
1359 gnttab_end_foreign_access_ref(sc->ring_ref[i]);
1360 sc->ring_ref[i] = GRANT_INVALID_REF;
1361 }
1362 sring_page_ptr += PAGE_SIZE;
1363 }
1364 free(sc->ring.sring, M_XENBLOCKFRONT);
1365 sc->ring.sring = NULL;
1366 }
1367
1368 if (sc->shadow) {
1369
1370 for (i = 0; i < sc->max_requests; i++) {
1371 struct xb_command *cm;
1372
1373 cm = &sc->shadow[i];
1374 if (cm->sg_refs != NULL) {
1375 free(cm->sg_refs, M_XENBLOCKFRONT);
1376 cm->sg_refs = NULL;
1377 }
1378
1379 bus_dmamap_destroy(sc->xb_io_dmat, cm->map);
1380 }
1381 free(sc->shadow, M_XENBLOCKFRONT);
1382 sc->shadow = NULL;
1383
1384 bus_dma_tag_destroy(sc->xb_io_dmat);
1385
1386 xb_initq_free(sc);
1387 xb_initq_ready(sc);
1388 xb_initq_complete(sc);
1389 }
1390
1391 if (sc->irq) {
1392 unbind_from_irqhandler(sc->irq);
1393 sc->irq = 0;
1394 }
1395}
1396
1397static int
1398blkif_completion(struct xb_command *s)
1399{
1400//printf("%s: Req %p(%d)\n", __func__, s, s->nseg);
1401 gnttab_end_foreign_access_references(s->nseg, s->sg_refs);
1402 return (BLKIF_SEGS_TO_BLOCKS(s->nseg));
1403}
1404
1405/* ** Driver registration ** */
1406static device_method_t blkfront_methods[] = {
1407 /* Device interface */
1408 DEVMETHOD(device_probe, blkfront_probe),
1409 DEVMETHOD(device_attach, blkfront_attach),
1410 DEVMETHOD(device_detach, blkfront_detach),
1411 DEVMETHOD(device_shutdown, bus_generic_shutdown),
1412 DEVMETHOD(device_suspend, blkfront_suspend),
1413 DEVMETHOD(device_resume, blkfront_resume),
1414
1415 /* Xenbus interface */
1416 DEVMETHOD(xenbus_otherend_changed, blkfront_backend_changed),
1417
1418 { 0, 0 }
1419};
1420
1421static driver_t blkfront_driver = {
1422 "xbd",
1423 blkfront_methods,
1424 sizeof(struct xb_softc),
1425};
1426devclass_t blkfront_devclass;
1427
1428DRIVER_MODULE(xbd, xenbusb_front, blkfront_driver, blkfront_devclass, 0, 0);
|