]>
Commit | Line | Data |
---|---|---|
851b1642 AK |
1 | /* |
2 | * Linux driver for VMware's para-virtualized SCSI HBA. | |
3 | * | |
a2713cce | 4 | * Copyright (C) 2008-2014, VMware, Inc. All Rights Reserved. |
851b1642 AK |
5 | * |
6 | * This program is free software; you can redistribute it and/or modify it | |
7 | * under the terms of the GNU General Public License as published by the | |
8 | * Free Software Foundation; version 2 of the License and no later version. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but | |
11 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or | |
13 | * NON INFRINGEMENT. See the GNU General Public License for more | |
14 | * details. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
17 | * along with this program; if not, write to the Free Software | |
18 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
19 | * | |
a9310735 | 20 | * Maintained by: Arvind Kumar <arvindkumar@vmware.com> |
851b1642 AK |
21 | * |
22 | */ | |
23 | ||
24 | #include <linux/kernel.h> | |
25 | #include <linux/module.h> | |
26 | #include <linux/interrupt.h> | |
5a0e3ad6 | 27 | #include <linux/slab.h> |
851b1642 AK |
28 | #include <linux/workqueue.h> |
29 | #include <linux/pci.h> | |
30 | ||
31 | #include <scsi/scsi.h> | |
32 | #include <scsi/scsi_host.h> | |
33 | #include <scsi/scsi_cmnd.h> | |
34 | #include <scsi/scsi_device.h> | |
35 | ||
36 | #include "vmw_pvscsi.h" | |
37 | ||
38 | #define PVSCSI_LINUX_DRIVER_DESC "VMware PVSCSI driver" | |
39 | ||
40 | MODULE_DESCRIPTION(PVSCSI_LINUX_DRIVER_DESC); | |
41 | MODULE_AUTHOR("VMware, Inc."); | |
42 | MODULE_LICENSE("GPL"); | |
43 | MODULE_VERSION(PVSCSI_DRIVER_VERSION_STRING); | |
44 | ||
45 | #define PVSCSI_DEFAULT_NUM_PAGES_PER_RING 8 | |
46 | #define PVSCSI_DEFAULT_NUM_PAGES_MSG_RING 1 | |
47 | #define PVSCSI_DEFAULT_QUEUE_DEPTH 64 | |
48 | #define SGL_SIZE PAGE_SIZE | |
49 | ||
50 | struct pvscsi_sg_list { | |
51 | struct PVSCSISGElement sge[PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT]; | |
52 | }; | |
53 | ||
54 | struct pvscsi_ctx { | |
55 | /* | |
56 | * The index of the context in cmd_map serves as the context ID for a | |
57 | * 1-to-1 mapping completions back to requests. | |
58 | */ | |
59 | struct scsi_cmnd *cmd; | |
60 | struct pvscsi_sg_list *sgl; | |
61 | struct list_head list; | |
62 | dma_addr_t dataPA; | |
63 | dma_addr_t sensePA; | |
64 | dma_addr_t sglPA; | |
a2713cce | 65 | struct completion *abort_cmp; |
851b1642 AK |
66 | }; |
67 | ||
68 | struct pvscsi_adapter { | |
69 | char *mmioBase; | |
70 | unsigned int irq; | |
71 | u8 rev; | |
72 | bool use_msi; | |
73 | bool use_msix; | |
74 | bool use_msg; | |
75 | ||
76 | spinlock_t hw_lock; | |
77 | ||
78 | struct workqueue_struct *workqueue; | |
79 | struct work_struct work; | |
80 | ||
81 | struct PVSCSIRingReqDesc *req_ring; | |
82 | unsigned req_pages; | |
83 | unsigned req_depth; | |
84 | dma_addr_t reqRingPA; | |
85 | ||
86 | struct PVSCSIRingCmpDesc *cmp_ring; | |
87 | unsigned cmp_pages; | |
88 | dma_addr_t cmpRingPA; | |
89 | ||
90 | struct PVSCSIRingMsgDesc *msg_ring; | |
91 | unsigned msg_pages; | |
92 | dma_addr_t msgRingPA; | |
93 | ||
94 | struct PVSCSIRingsState *rings_state; | |
95 | dma_addr_t ringStatePA; | |
96 | ||
97 | struct pci_dev *dev; | |
98 | struct Scsi_Host *host; | |
99 | ||
100 | struct list_head cmd_pool; | |
101 | struct pvscsi_ctx *cmd_map; | |
102 | }; | |
103 | ||
104 | ||
105 | /* Command line parameters */ | |
106 | static int pvscsi_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_PER_RING; | |
107 | static int pvscsi_msg_ring_pages = PVSCSI_DEFAULT_NUM_PAGES_MSG_RING; | |
108 | static int pvscsi_cmd_per_lun = PVSCSI_DEFAULT_QUEUE_DEPTH; | |
109 | static bool pvscsi_disable_msi; | |
110 | static bool pvscsi_disable_msix; | |
111 | static bool pvscsi_use_msg = true; | |
112 | ||
113 | #define PVSCSI_RW (S_IRUSR | S_IWUSR) | |
114 | ||
115 | module_param_named(ring_pages, pvscsi_ring_pages, int, PVSCSI_RW); | |
116 | MODULE_PARM_DESC(ring_pages, "Number of pages per req/cmp ring - (default=" | |
117 | __stringify(PVSCSI_DEFAULT_NUM_PAGES_PER_RING) ")"); | |
118 | ||
119 | module_param_named(msg_ring_pages, pvscsi_msg_ring_pages, int, PVSCSI_RW); | |
120 | MODULE_PARM_DESC(msg_ring_pages, "Number of pages for the msg ring - (default=" | |
121 | __stringify(PVSCSI_DEFAULT_NUM_PAGES_MSG_RING) ")"); | |
122 | ||
123 | module_param_named(cmd_per_lun, pvscsi_cmd_per_lun, int, PVSCSI_RW); | |
124 | MODULE_PARM_DESC(cmd_per_lun, "Maximum commands per lun - (default=" | |
125 | __stringify(PVSCSI_MAX_REQ_QUEUE_DEPTH) ")"); | |
126 | ||
127 | module_param_named(disable_msi, pvscsi_disable_msi, bool, PVSCSI_RW); | |
128 | MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)"); | |
129 | ||
130 | module_param_named(disable_msix, pvscsi_disable_msix, bool, PVSCSI_RW); | |
131 | MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)"); | |
132 | ||
133 | module_param_named(use_msg, pvscsi_use_msg, bool, PVSCSI_RW); | |
134 | MODULE_PARM_DESC(use_msg, "Use msg ring when available - (default=1)"); | |
135 | ||
136 | static const struct pci_device_id pvscsi_pci_tbl[] = { | |
137 | { PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_PVSCSI) }, | |
138 | { 0 } | |
139 | }; | |
140 | ||
141 | MODULE_DEVICE_TABLE(pci, pvscsi_pci_tbl); | |
142 | ||
143 | static struct device * | |
144 | pvscsi_dev(const struct pvscsi_adapter *adapter) | |
145 | { | |
146 | return &(adapter->dev->dev); | |
147 | } | |
148 | ||
149 | static struct pvscsi_ctx * | |
150 | pvscsi_find_context(const struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd) | |
151 | { | |
152 | struct pvscsi_ctx *ctx, *end; | |
153 | ||
154 | end = &adapter->cmd_map[adapter->req_depth]; | |
155 | for (ctx = adapter->cmd_map; ctx < end; ctx++) | |
156 | if (ctx->cmd == cmd) | |
157 | return ctx; | |
158 | ||
159 | return NULL; | |
160 | } | |
161 | ||
162 | static struct pvscsi_ctx * | |
163 | pvscsi_acquire_context(struct pvscsi_adapter *adapter, struct scsi_cmnd *cmd) | |
164 | { | |
165 | struct pvscsi_ctx *ctx; | |
166 | ||
167 | if (list_empty(&adapter->cmd_pool)) | |
168 | return NULL; | |
169 | ||
170 | ctx = list_first_entry(&adapter->cmd_pool, struct pvscsi_ctx, list); | |
171 | ctx->cmd = cmd; | |
172 | list_del(&ctx->list); | |
173 | ||
174 | return ctx; | |
175 | } | |
176 | ||
177 | static void pvscsi_release_context(struct pvscsi_adapter *adapter, | |
178 | struct pvscsi_ctx *ctx) | |
179 | { | |
180 | ctx->cmd = NULL; | |
a2713cce | 181 | ctx->abort_cmp = NULL; |
851b1642 AK |
182 | list_add(&ctx->list, &adapter->cmd_pool); |
183 | } | |
184 | ||
185 | /* | |
186 | * Map a pvscsi_ctx struct to a context ID field value; we map to a simple | |
187 | * non-zero integer. ctx always points to an entry in cmd_map array, hence | |
188 | * the return value is always >=1. | |
189 | */ | |
190 | static u64 pvscsi_map_context(const struct pvscsi_adapter *adapter, | |
191 | const struct pvscsi_ctx *ctx) | |
192 | { | |
193 | return ctx - adapter->cmd_map + 1; | |
194 | } | |
195 | ||
196 | static struct pvscsi_ctx * | |
197 | pvscsi_get_context(const struct pvscsi_adapter *adapter, u64 context) | |
198 | { | |
199 | return &adapter->cmd_map[context - 1]; | |
200 | } | |
201 | ||
202 | static void pvscsi_reg_write(const struct pvscsi_adapter *adapter, | |
203 | u32 offset, u32 val) | |
204 | { | |
205 | writel(val, adapter->mmioBase + offset); | |
206 | } | |
207 | ||
208 | static u32 pvscsi_reg_read(const struct pvscsi_adapter *adapter, u32 offset) | |
209 | { | |
210 | return readl(adapter->mmioBase + offset); | |
211 | } | |
212 | ||
213 | static u32 pvscsi_read_intr_status(const struct pvscsi_adapter *adapter) | |
214 | { | |
215 | return pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_INTR_STATUS); | |
216 | } | |
217 | ||
218 | static void pvscsi_write_intr_status(const struct pvscsi_adapter *adapter, | |
219 | u32 val) | |
220 | { | |
221 | pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_STATUS, val); | |
222 | } | |
223 | ||
224 | static void pvscsi_unmask_intr(const struct pvscsi_adapter *adapter) | |
225 | { | |
226 | u32 intr_bits; | |
227 | ||
228 | intr_bits = PVSCSI_INTR_CMPL_MASK; | |
229 | if (adapter->use_msg) | |
230 | intr_bits |= PVSCSI_INTR_MSG_MASK; | |
231 | ||
232 | pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, intr_bits); | |
233 | } | |
234 | ||
235 | static void pvscsi_mask_intr(const struct pvscsi_adapter *adapter) | |
236 | { | |
237 | pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_INTR_MASK, 0); | |
238 | } | |
239 | ||
240 | static void pvscsi_write_cmd_desc(const struct pvscsi_adapter *adapter, | |
241 | u32 cmd, const void *desc, size_t len) | |
242 | { | |
243 | const u32 *ptr = desc; | |
244 | size_t i; | |
245 | ||
246 | len /= sizeof(*ptr); | |
247 | pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND, cmd); | |
248 | for (i = 0; i < len; i++) | |
249 | pvscsi_reg_write(adapter, | |
250 | PVSCSI_REG_OFFSET_COMMAND_DATA, ptr[i]); | |
251 | } | |
252 | ||
253 | static void pvscsi_abort_cmd(const struct pvscsi_adapter *adapter, | |
254 | const struct pvscsi_ctx *ctx) | |
255 | { | |
256 | struct PVSCSICmdDescAbortCmd cmd = { 0 }; | |
257 | ||
258 | cmd.target = ctx->cmd->device->id; | |
259 | cmd.context = pvscsi_map_context(adapter, ctx); | |
260 | ||
261 | pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ABORT_CMD, &cmd, sizeof(cmd)); | |
262 | } | |
263 | ||
264 | static void pvscsi_kick_rw_io(const struct pvscsi_adapter *adapter) | |
265 | { | |
266 | pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_RW_IO, 0); | |
267 | } | |
268 | ||
269 | static void pvscsi_process_request_ring(const struct pvscsi_adapter *adapter) | |
270 | { | |
271 | pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_KICK_NON_RW_IO, 0); | |
272 | } | |
273 | ||
274 | static int scsi_is_rw(unsigned char op) | |
275 | { | |
276 | return op == READ_6 || op == WRITE_6 || | |
277 | op == READ_10 || op == WRITE_10 || | |
278 | op == READ_12 || op == WRITE_12 || | |
279 | op == READ_16 || op == WRITE_16; | |
280 | } | |
281 | ||
282 | static void pvscsi_kick_io(const struct pvscsi_adapter *adapter, | |
283 | unsigned char op) | |
284 | { | |
285 | if (scsi_is_rw(op)) | |
286 | pvscsi_kick_rw_io(adapter); | |
287 | else | |
288 | pvscsi_process_request_ring(adapter); | |
289 | } | |
290 | ||
291 | static void ll_adapter_reset(const struct pvscsi_adapter *adapter) | |
292 | { | |
293 | dev_dbg(pvscsi_dev(adapter), "Adapter Reset on %p\n", adapter); | |
294 | ||
295 | pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_ADAPTER_RESET, NULL, 0); | |
296 | } | |
297 | ||
298 | static void ll_bus_reset(const struct pvscsi_adapter *adapter) | |
299 | { | |
59e13d48 | 300 | dev_dbg(pvscsi_dev(adapter), "Resetting bus on %p\n", adapter); |
851b1642 AK |
301 | |
302 | pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_BUS, NULL, 0); | |
303 | } | |
304 | ||
305 | static void ll_device_reset(const struct pvscsi_adapter *adapter, u32 target) | |
306 | { | |
307 | struct PVSCSICmdDescResetDevice cmd = { 0 }; | |
308 | ||
59e13d48 | 309 | dev_dbg(pvscsi_dev(adapter), "Resetting device: target=%u\n", target); |
851b1642 AK |
310 | |
311 | cmd.target = target; | |
312 | ||
313 | pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_RESET_DEVICE, | |
314 | &cmd, sizeof(cmd)); | |
315 | } | |
316 | ||
317 | static void pvscsi_create_sg(struct pvscsi_ctx *ctx, | |
318 | struct scatterlist *sg, unsigned count) | |
319 | { | |
320 | unsigned i; | |
321 | struct PVSCSISGElement *sge; | |
322 | ||
323 | BUG_ON(count > PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT); | |
324 | ||
325 | sge = &ctx->sgl->sge[0]; | |
326 | for (i = 0; i < count; i++, sg++) { | |
327 | sge[i].addr = sg_dma_address(sg); | |
328 | sge[i].length = sg_dma_len(sg); | |
329 | sge[i].flags = 0; | |
330 | } | |
331 | } | |
332 | ||
333 | /* | |
334 | * Map all data buffers for a command into PCI space and | |
335 | * setup the scatter/gather list if needed. | |
336 | */ | |
337 | static void pvscsi_map_buffers(struct pvscsi_adapter *adapter, | |
338 | struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd, | |
339 | struct PVSCSIRingReqDesc *e) | |
340 | { | |
341 | unsigned count; | |
342 | unsigned bufflen = scsi_bufflen(cmd); | |
343 | struct scatterlist *sg; | |
344 | ||
345 | e->dataLen = bufflen; | |
346 | e->dataAddr = 0; | |
347 | if (bufflen == 0) | |
348 | return; | |
349 | ||
350 | sg = scsi_sglist(cmd); | |
351 | count = scsi_sg_count(cmd); | |
352 | if (count != 0) { | |
353 | int segs = scsi_dma_map(cmd); | |
354 | if (segs > 1) { | |
355 | pvscsi_create_sg(ctx, sg, segs); | |
356 | ||
357 | e->flags |= PVSCSI_FLAG_CMD_WITH_SG_LIST; | |
358 | ctx->sglPA = pci_map_single(adapter->dev, ctx->sgl, | |
359 | SGL_SIZE, PCI_DMA_TODEVICE); | |
360 | e->dataAddr = ctx->sglPA; | |
361 | } else | |
362 | e->dataAddr = sg_dma_address(sg); | |
363 | } else { | |
364 | /* | |
365 | * In case there is no S/G list, scsi_sglist points | |
366 | * directly to the buffer. | |
367 | */ | |
368 | ctx->dataPA = pci_map_single(adapter->dev, sg, bufflen, | |
369 | cmd->sc_data_direction); | |
370 | e->dataAddr = ctx->dataPA; | |
371 | } | |
372 | } | |
373 | ||
374 | static void pvscsi_unmap_buffers(const struct pvscsi_adapter *adapter, | |
375 | struct pvscsi_ctx *ctx) | |
376 | { | |
377 | struct scsi_cmnd *cmd; | |
378 | unsigned bufflen; | |
379 | ||
380 | cmd = ctx->cmd; | |
381 | bufflen = scsi_bufflen(cmd); | |
382 | ||
383 | if (bufflen != 0) { | |
384 | unsigned count = scsi_sg_count(cmd); | |
385 | ||
386 | if (count != 0) { | |
387 | scsi_dma_unmap(cmd); | |
388 | if (ctx->sglPA) { | |
389 | pci_unmap_single(adapter->dev, ctx->sglPA, | |
390 | SGL_SIZE, PCI_DMA_TODEVICE); | |
391 | ctx->sglPA = 0; | |
392 | } | |
393 | } else | |
394 | pci_unmap_single(adapter->dev, ctx->dataPA, bufflen, | |
395 | cmd->sc_data_direction); | |
396 | } | |
397 | if (cmd->sense_buffer) | |
398 | pci_unmap_single(adapter->dev, ctx->sensePA, | |
399 | SCSI_SENSE_BUFFERSIZE, PCI_DMA_FROMDEVICE); | |
400 | } | |
401 | ||
6f039790 | 402 | static int pvscsi_allocate_rings(struct pvscsi_adapter *adapter) |
851b1642 AK |
403 | { |
404 | adapter->rings_state = pci_alloc_consistent(adapter->dev, PAGE_SIZE, | |
405 | &adapter->ringStatePA); | |
406 | if (!adapter->rings_state) | |
407 | return -ENOMEM; | |
408 | ||
409 | adapter->req_pages = min(PVSCSI_MAX_NUM_PAGES_REQ_RING, | |
410 | pvscsi_ring_pages); | |
411 | adapter->req_depth = adapter->req_pages | |
412 | * PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE; | |
413 | adapter->req_ring = pci_alloc_consistent(adapter->dev, | |
414 | adapter->req_pages * PAGE_SIZE, | |
415 | &adapter->reqRingPA); | |
416 | if (!adapter->req_ring) | |
417 | return -ENOMEM; | |
418 | ||
419 | adapter->cmp_pages = min(PVSCSI_MAX_NUM_PAGES_CMP_RING, | |
420 | pvscsi_ring_pages); | |
421 | adapter->cmp_ring = pci_alloc_consistent(adapter->dev, | |
422 | adapter->cmp_pages * PAGE_SIZE, | |
423 | &adapter->cmpRingPA); | |
424 | if (!adapter->cmp_ring) | |
425 | return -ENOMEM; | |
426 | ||
427 | BUG_ON(!IS_ALIGNED(adapter->ringStatePA, PAGE_SIZE)); | |
428 | BUG_ON(!IS_ALIGNED(adapter->reqRingPA, PAGE_SIZE)); | |
429 | BUG_ON(!IS_ALIGNED(adapter->cmpRingPA, PAGE_SIZE)); | |
430 | ||
431 | if (!adapter->use_msg) | |
432 | return 0; | |
433 | ||
434 | adapter->msg_pages = min(PVSCSI_MAX_NUM_PAGES_MSG_RING, | |
435 | pvscsi_msg_ring_pages); | |
436 | adapter->msg_ring = pci_alloc_consistent(adapter->dev, | |
437 | adapter->msg_pages * PAGE_SIZE, | |
438 | &adapter->msgRingPA); | |
439 | if (!adapter->msg_ring) | |
440 | return -ENOMEM; | |
441 | BUG_ON(!IS_ALIGNED(adapter->msgRingPA, PAGE_SIZE)); | |
442 | ||
443 | return 0; | |
444 | } | |
445 | ||
446 | static void pvscsi_setup_all_rings(const struct pvscsi_adapter *adapter) | |
447 | { | |
448 | struct PVSCSICmdDescSetupRings cmd = { 0 }; | |
449 | dma_addr_t base; | |
450 | unsigned i; | |
451 | ||
452 | cmd.ringsStatePPN = adapter->ringStatePA >> PAGE_SHIFT; | |
453 | cmd.reqRingNumPages = adapter->req_pages; | |
454 | cmd.cmpRingNumPages = adapter->cmp_pages; | |
455 | ||
456 | base = adapter->reqRingPA; | |
457 | for (i = 0; i < adapter->req_pages; i++) { | |
458 | cmd.reqRingPPNs[i] = base >> PAGE_SHIFT; | |
459 | base += PAGE_SIZE; | |
460 | } | |
461 | ||
462 | base = adapter->cmpRingPA; | |
463 | for (i = 0; i < adapter->cmp_pages; i++) { | |
464 | cmd.cmpRingPPNs[i] = base >> PAGE_SHIFT; | |
465 | base += PAGE_SIZE; | |
466 | } | |
467 | ||
468 | memset(adapter->rings_state, 0, PAGE_SIZE); | |
469 | memset(adapter->req_ring, 0, adapter->req_pages * PAGE_SIZE); | |
470 | memset(adapter->cmp_ring, 0, adapter->cmp_pages * PAGE_SIZE); | |
471 | ||
472 | pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_RINGS, | |
473 | &cmd, sizeof(cmd)); | |
474 | ||
475 | if (adapter->use_msg) { | |
476 | struct PVSCSICmdDescSetupMsgRing cmd_msg = { 0 }; | |
477 | ||
478 | cmd_msg.numPages = adapter->msg_pages; | |
479 | ||
480 | base = adapter->msgRingPA; | |
481 | for (i = 0; i < adapter->msg_pages; i++) { | |
482 | cmd_msg.ringPPNs[i] = base >> PAGE_SHIFT; | |
483 | base += PAGE_SIZE; | |
484 | } | |
485 | memset(adapter->msg_ring, 0, adapter->msg_pages * PAGE_SIZE); | |
486 | ||
487 | pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_SETUP_MSG_RING, | |
488 | &cmd_msg, sizeof(cmd_msg)); | |
489 | } | |
490 | } | |
491 | ||
492 | /* | |
493 | * Pull a completion descriptor off and pass the completion back | |
494 | * to the SCSI mid layer. | |
495 | */ | |
496 | static void pvscsi_complete_request(struct pvscsi_adapter *adapter, | |
497 | const struct PVSCSIRingCmpDesc *e) | |
498 | { | |
499 | struct pvscsi_ctx *ctx; | |
500 | struct scsi_cmnd *cmd; | |
a2713cce | 501 | struct completion *abort_cmp; |
851b1642 AK |
502 | u32 btstat = e->hostStatus; |
503 | u32 sdstat = e->scsiStatus; | |
504 | ||
505 | ctx = pvscsi_get_context(adapter, e->context); | |
506 | cmd = ctx->cmd; | |
a2713cce | 507 | abort_cmp = ctx->abort_cmp; |
851b1642 AK |
508 | pvscsi_unmap_buffers(adapter, ctx); |
509 | pvscsi_release_context(adapter, ctx); | |
a2713cce AK |
510 | if (abort_cmp) { |
511 | /* | |
512 | * The command was requested to be aborted. Just signal that | |
513 | * the request completed and swallow the actual cmd completion | |
514 | * here. The abort handler will post a completion for this | |
515 | * command indicating that it got successfully aborted. | |
516 | */ | |
517 | complete(abort_cmp); | |
518 | return; | |
519 | } | |
851b1642 | 520 | |
a2713cce | 521 | cmd->result = 0; |
851b1642 AK |
522 | if (sdstat != SAM_STAT_GOOD && |
523 | (btstat == BTSTAT_SUCCESS || | |
524 | btstat == BTSTAT_LINKED_COMMAND_COMPLETED || | |
525 | btstat == BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG)) { | |
526 | cmd->result = (DID_OK << 16) | sdstat; | |
527 | if (sdstat == SAM_STAT_CHECK_CONDITION && cmd->sense_buffer) | |
528 | cmd->result |= (DRIVER_SENSE << 24); | |
529 | } else | |
530 | switch (btstat) { | |
531 | case BTSTAT_SUCCESS: | |
532 | case BTSTAT_LINKED_COMMAND_COMPLETED: | |
533 | case BTSTAT_LINKED_COMMAND_COMPLETED_WITH_FLAG: | |
534 | /* If everything went fine, let's move on.. */ | |
535 | cmd->result = (DID_OK << 16); | |
536 | break; | |
537 | ||
538 | case BTSTAT_DATARUN: | |
539 | case BTSTAT_DATA_UNDERRUN: | |
540 | /* Report residual data in underruns */ | |
541 | scsi_set_resid(cmd, scsi_bufflen(cmd) - e->dataLen); | |
542 | cmd->result = (DID_ERROR << 16); | |
543 | break; | |
544 | ||
545 | case BTSTAT_SELTIMEO: | |
546 | /* Our emulation returns this for non-connected devs */ | |
547 | cmd->result = (DID_BAD_TARGET << 16); | |
548 | break; | |
549 | ||
550 | case BTSTAT_LUNMISMATCH: | |
551 | case BTSTAT_TAGREJECT: | |
552 | case BTSTAT_BADMSG: | |
553 | cmd->result = (DRIVER_INVALID << 24); | |
554 | /* fall through */ | |
555 | ||
556 | case BTSTAT_HAHARDWARE: | |
557 | case BTSTAT_INVPHASE: | |
558 | case BTSTAT_HATIMEOUT: | |
559 | case BTSTAT_NORESPONSE: | |
560 | case BTSTAT_DISCONNECT: | |
561 | case BTSTAT_HASOFTWARE: | |
562 | case BTSTAT_BUSFREE: | |
563 | case BTSTAT_SENSFAILED: | |
564 | cmd->result |= (DID_ERROR << 16); | |
565 | break; | |
566 | ||
567 | case BTSTAT_SENTRST: | |
568 | case BTSTAT_RECVRST: | |
569 | case BTSTAT_BUSRESET: | |
570 | cmd->result = (DID_RESET << 16); | |
571 | break; | |
572 | ||
573 | case BTSTAT_ABORTQUEUE: | |
574 | cmd->result = (DID_ABORT << 16); | |
575 | break; | |
576 | ||
577 | case BTSTAT_SCSIPARITY: | |
578 | cmd->result = (DID_PARITY << 16); | |
579 | break; | |
580 | ||
581 | default: | |
582 | cmd->result = (DID_ERROR << 16); | |
583 | scmd_printk(KERN_DEBUG, cmd, | |
584 | "Unknown completion status: 0x%x\n", | |
585 | btstat); | |
586 | } | |
587 | ||
588 | dev_dbg(&cmd->device->sdev_gendev, | |
589 | "cmd=%p %x ctx=%p result=0x%x status=0x%x,%x\n", | |
590 | cmd, cmd->cmnd[0], ctx, cmd->result, btstat, sdstat); | |
591 | ||
592 | cmd->scsi_done(cmd); | |
593 | } | |
594 | ||
595 | /* | |
596 | * barrier usage : Since the PVSCSI device is emulated, there could be cases | |
597 | * where we may want to serialize some accesses between the driver and the | |
598 | * emulation layer. We use compiler barriers instead of the more expensive | |
599 | * memory barriers because PVSCSI is only supported on X86 which has strong | |
600 | * memory access ordering. | |
601 | */ | |
602 | static void pvscsi_process_completion_ring(struct pvscsi_adapter *adapter) | |
603 | { | |
604 | struct PVSCSIRingsState *s = adapter->rings_state; | |
605 | struct PVSCSIRingCmpDesc *ring = adapter->cmp_ring; | |
606 | u32 cmp_entries = s->cmpNumEntriesLog2; | |
607 | ||
608 | while (s->cmpConsIdx != s->cmpProdIdx) { | |
609 | struct PVSCSIRingCmpDesc *e = ring + (s->cmpConsIdx & | |
610 | MASK(cmp_entries)); | |
611 | /* | |
612 | * This barrier() ensures that *e is not dereferenced while | |
613 | * the device emulation still writes data into the slot. | |
614 | * Since the device emulation advances s->cmpProdIdx only after | |
615 | * updating the slot we want to check it first. | |
616 | */ | |
617 | barrier(); | |
618 | pvscsi_complete_request(adapter, e); | |
619 | /* | |
620 | * This barrier() ensures that compiler doesn't reorder write | |
621 | * to s->cmpConsIdx before the read of (*e) inside | |
622 | * pvscsi_complete_request. Otherwise, device emulation may | |
623 | * overwrite *e before we had a chance to read it. | |
624 | */ | |
625 | barrier(); | |
626 | s->cmpConsIdx++; | |
627 | } | |
628 | } | |
629 | ||
630 | /* | |
631 | * Translate a Linux SCSI request into a request ring entry. | |
632 | */ | |
633 | static int pvscsi_queue_ring(struct pvscsi_adapter *adapter, | |
634 | struct pvscsi_ctx *ctx, struct scsi_cmnd *cmd) | |
635 | { | |
636 | struct PVSCSIRingsState *s; | |
637 | struct PVSCSIRingReqDesc *e; | |
638 | struct scsi_device *sdev; | |
639 | u32 req_entries; | |
640 | ||
641 | s = adapter->rings_state; | |
642 | sdev = cmd->device; | |
643 | req_entries = s->reqNumEntriesLog2; | |
644 | ||
645 | /* | |
646 | * If this condition holds, we might have room on the request ring, but | |
647 | * we might not have room on the completion ring for the response. | |
648 | * However, we have already ruled out this possibility - we would not | |
649 | * have successfully allocated a context if it were true, since we only | |
650 | * have one context per request entry. Check for it anyway, since it | |
651 | * would be a serious bug. | |
652 | */ | |
653 | if (s->reqProdIdx - s->cmpConsIdx >= 1 << req_entries) { | |
654 | scmd_printk(KERN_ERR, cmd, "vmw_pvscsi: " | |
655 | "ring full: reqProdIdx=%d cmpConsIdx=%d\n", | |
656 | s->reqProdIdx, s->cmpConsIdx); | |
657 | return -1; | |
658 | } | |
659 | ||
660 | e = adapter->req_ring + (s->reqProdIdx & MASK(req_entries)); | |
661 | ||
662 | e->bus = sdev->channel; | |
663 | e->target = sdev->id; | |
664 | memset(e->lun, 0, sizeof(e->lun)); | |
665 | e->lun[1] = sdev->lun; | |
666 | ||
667 | if (cmd->sense_buffer) { | |
668 | ctx->sensePA = pci_map_single(adapter->dev, cmd->sense_buffer, | |
669 | SCSI_SENSE_BUFFERSIZE, | |
670 | PCI_DMA_FROMDEVICE); | |
671 | e->senseAddr = ctx->sensePA; | |
672 | e->senseLen = SCSI_SENSE_BUFFERSIZE; | |
673 | } else { | |
674 | e->senseLen = 0; | |
675 | e->senseAddr = 0; | |
676 | } | |
677 | e->cdbLen = cmd->cmd_len; | |
678 | e->vcpuHint = smp_processor_id(); | |
679 | memcpy(e->cdb, cmd->cmnd, e->cdbLen); | |
680 | ||
681 | e->tag = SIMPLE_QUEUE_TAG; | |
682 | if (sdev->tagged_supported && | |
683 | (cmd->tag == HEAD_OF_QUEUE_TAG || | |
684 | cmd->tag == ORDERED_QUEUE_TAG)) | |
685 | e->tag = cmd->tag; | |
686 | ||
687 | if (cmd->sc_data_direction == DMA_FROM_DEVICE) | |
688 | e->flags = PVSCSI_FLAG_CMD_DIR_TOHOST; | |
689 | else if (cmd->sc_data_direction == DMA_TO_DEVICE) | |
690 | e->flags = PVSCSI_FLAG_CMD_DIR_TODEVICE; | |
691 | else if (cmd->sc_data_direction == DMA_NONE) | |
692 | e->flags = PVSCSI_FLAG_CMD_DIR_NONE; | |
693 | else | |
694 | e->flags = 0; | |
695 | ||
696 | pvscsi_map_buffers(adapter, ctx, cmd, e); | |
697 | ||
698 | e->context = pvscsi_map_context(adapter, ctx); | |
699 | ||
700 | barrier(); | |
701 | ||
702 | s->reqProdIdx++; | |
703 | ||
704 | return 0; | |
705 | } | |
706 | ||
f281233d | 707 | static int pvscsi_queue_lck(struct scsi_cmnd *cmd, void (*done)(struct scsi_cmnd *)) |
851b1642 AK |
708 | { |
709 | struct Scsi_Host *host = cmd->device->host; | |
710 | struct pvscsi_adapter *adapter = shost_priv(host); | |
711 | struct pvscsi_ctx *ctx; | |
712 | unsigned long flags; | |
713 | ||
714 | spin_lock_irqsave(&adapter->hw_lock, flags); | |
715 | ||
716 | ctx = pvscsi_acquire_context(adapter, cmd); | |
717 | if (!ctx || pvscsi_queue_ring(adapter, ctx, cmd) != 0) { | |
718 | if (ctx) | |
719 | pvscsi_release_context(adapter, ctx); | |
720 | spin_unlock_irqrestore(&adapter->hw_lock, flags); | |
721 | return SCSI_MLQUEUE_HOST_BUSY; | |
722 | } | |
723 | ||
724 | cmd->scsi_done = done; | |
725 | ||
726 | dev_dbg(&cmd->device->sdev_gendev, | |
727 | "queued cmd %p, ctx %p, op=%x\n", cmd, ctx, cmd->cmnd[0]); | |
728 | ||
729 | spin_unlock_irqrestore(&adapter->hw_lock, flags); | |
730 | ||
731 | pvscsi_kick_io(adapter, cmd->cmnd[0]); | |
732 | ||
733 | return 0; | |
734 | } | |
735 | ||
f281233d JG |
736 | static DEF_SCSI_QCMD(pvscsi_queue) |
737 | ||
851b1642 AK |
738 | static int pvscsi_abort(struct scsi_cmnd *cmd) |
739 | { | |
740 | struct pvscsi_adapter *adapter = shost_priv(cmd->device->host); | |
741 | struct pvscsi_ctx *ctx; | |
742 | unsigned long flags; | |
a2713cce AK |
743 | int result = SUCCESS; |
744 | DECLARE_COMPLETION_ONSTACK(abort_cmp); | |
851b1642 AK |
745 | |
746 | scmd_printk(KERN_DEBUG, cmd, "task abort on host %u, %p\n", | |
747 | adapter->host->host_no, cmd); | |
748 | ||
749 | spin_lock_irqsave(&adapter->hw_lock, flags); | |
750 | ||
751 | /* | |
752 | * Poll the completion ring first - we might be trying to abort | |
753 | * a command that is waiting to be dispatched in the completion ring. | |
754 | */ | |
755 | pvscsi_process_completion_ring(adapter); | |
756 | ||
757 | /* | |
758 | * If there is no context for the command, it either already succeeded | |
759 | * or else was never properly issued. Not our problem. | |
760 | */ | |
761 | ctx = pvscsi_find_context(adapter, cmd); | |
762 | if (!ctx) { | |
763 | scmd_printk(KERN_DEBUG, cmd, "Failed to abort cmd %p\n", cmd); | |
764 | goto out; | |
765 | } | |
766 | ||
a2713cce AK |
767 | /* |
768 | * Mark that the command has been requested to be aborted and issue | |
769 | * the abort. | |
770 | */ | |
771 | ctx->abort_cmp = &abort_cmp; | |
772 | ||
851b1642 | 773 | pvscsi_abort_cmd(adapter, ctx); |
a2713cce AK |
774 | spin_unlock_irqrestore(&adapter->hw_lock, flags); |
775 | /* Wait for 2 secs for the completion. */ | |
776 | wait_for_completion_timeout(&abort_cmp, msecs_to_jiffies(2000)); | |
777 | spin_lock_irqsave(&adapter->hw_lock, flags); | |
851b1642 | 778 | |
a2713cce AK |
779 | if (!completion_done(&abort_cmp)) { |
780 | /* | |
781 | * Failed to abort the command, unmark the fact that it | |
782 | * was requested to be aborted. | |
783 | */ | |
784 | ctx->abort_cmp = NULL; | |
785 | result = FAILED; | |
786 | scmd_printk(KERN_DEBUG, cmd, | |
787 | "Failed to get completion for aborted cmd %p\n", | |
788 | cmd); | |
789 | goto out; | |
790 | } | |
791 | ||
792 | /* | |
793 | * Successfully aborted the command. | |
794 | */ | |
795 | cmd->result = (DID_ABORT << 16); | |
796 | cmd->scsi_done(cmd); | |
851b1642 AK |
797 | |
798 | out: | |
799 | spin_unlock_irqrestore(&adapter->hw_lock, flags); | |
a2713cce | 800 | return result; |
851b1642 AK |
801 | } |
802 | ||
803 | /* | |
804 | * Abort all outstanding requests. This is only safe to use if the completion | |
805 | * ring will never be walked again or the device has been reset, because it | |
806 | * destroys the 1-1 mapping between context field passed to emulation and our | |
807 | * request structure. | |
808 | */ | |
809 | static void pvscsi_reset_all(struct pvscsi_adapter *adapter) | |
810 | { | |
811 | unsigned i; | |
812 | ||
813 | for (i = 0; i < adapter->req_depth; i++) { | |
814 | struct pvscsi_ctx *ctx = &adapter->cmd_map[i]; | |
815 | struct scsi_cmnd *cmd = ctx->cmd; | |
816 | if (cmd) { | |
817 | scmd_printk(KERN_ERR, cmd, | |
818 | "Forced reset on cmd %p\n", cmd); | |
819 | pvscsi_unmap_buffers(adapter, ctx); | |
820 | pvscsi_release_context(adapter, ctx); | |
821 | cmd->result = (DID_RESET << 16); | |
822 | cmd->scsi_done(cmd); | |
823 | } | |
824 | } | |
825 | } | |
826 | ||
827 | static int pvscsi_host_reset(struct scsi_cmnd *cmd) | |
828 | { | |
829 | struct Scsi_Host *host = cmd->device->host; | |
830 | struct pvscsi_adapter *adapter = shost_priv(host); | |
831 | unsigned long flags; | |
832 | bool use_msg; | |
833 | ||
834 | scmd_printk(KERN_INFO, cmd, "SCSI Host reset\n"); | |
835 | ||
836 | spin_lock_irqsave(&adapter->hw_lock, flags); | |
837 | ||
838 | use_msg = adapter->use_msg; | |
839 | ||
840 | if (use_msg) { | |
841 | adapter->use_msg = 0; | |
842 | spin_unlock_irqrestore(&adapter->hw_lock, flags); | |
843 | ||
844 | /* | |
845 | * Now that we know that the ISR won't add more work on the | |
846 | * workqueue we can safely flush any outstanding work. | |
847 | */ | |
848 | flush_workqueue(adapter->workqueue); | |
849 | spin_lock_irqsave(&adapter->hw_lock, flags); | |
850 | } | |
851 | ||
852 | /* | |
853 | * We're going to tear down the entire ring structure and set it back | |
854 | * up, so stalling new requests until all completions are flushed and | |
855 | * the rings are back in place. | |
856 | */ | |
857 | ||
858 | pvscsi_process_request_ring(adapter); | |
859 | ||
860 | ll_adapter_reset(adapter); | |
861 | ||
862 | /* | |
863 | * Now process any completions. Note we do this AFTER adapter reset, | |
864 | * which is strange, but stops races where completions get posted | |
865 | * between processing the ring and issuing the reset. The backend will | |
866 | * not touch the ring memory after reset, so the immediately pre-reset | |
867 | * completion ring state is still valid. | |
868 | */ | |
869 | pvscsi_process_completion_ring(adapter); | |
870 | ||
871 | pvscsi_reset_all(adapter); | |
872 | adapter->use_msg = use_msg; | |
873 | pvscsi_setup_all_rings(adapter); | |
874 | pvscsi_unmask_intr(adapter); | |
875 | ||
876 | spin_unlock_irqrestore(&adapter->hw_lock, flags); | |
877 | ||
878 | return SUCCESS; | |
879 | } | |
880 | ||
881 | static int pvscsi_bus_reset(struct scsi_cmnd *cmd) | |
882 | { | |
883 | struct Scsi_Host *host = cmd->device->host; | |
884 | struct pvscsi_adapter *adapter = shost_priv(host); | |
885 | unsigned long flags; | |
886 | ||
887 | scmd_printk(KERN_INFO, cmd, "SCSI Bus reset\n"); | |
888 | ||
889 | /* | |
890 | * We don't want to queue new requests for this bus after | |
891 | * flushing all pending requests to emulation, since new | |
892 | * requests could then sneak in during this bus reset phase, | |
893 | * so take the lock now. | |
894 | */ | |
895 | spin_lock_irqsave(&adapter->hw_lock, flags); | |
896 | ||
897 | pvscsi_process_request_ring(adapter); | |
898 | ll_bus_reset(adapter); | |
899 | pvscsi_process_completion_ring(adapter); | |
900 | ||
901 | spin_unlock_irqrestore(&adapter->hw_lock, flags); | |
902 | ||
903 | return SUCCESS; | |
904 | } | |
905 | ||
906 | static int pvscsi_device_reset(struct scsi_cmnd *cmd) | |
907 | { | |
908 | struct Scsi_Host *host = cmd->device->host; | |
909 | struct pvscsi_adapter *adapter = shost_priv(host); | |
910 | unsigned long flags; | |
911 | ||
912 | scmd_printk(KERN_INFO, cmd, "SCSI device reset on scsi%u:%u\n", | |
913 | host->host_no, cmd->device->id); | |
914 | ||
915 | /* | |
916 | * We don't want to queue new requests for this device after flushing | |
917 | * all pending requests to emulation, since new requests could then | |
918 | * sneak in during this device reset phase, so take the lock now. | |
919 | */ | |
920 | spin_lock_irqsave(&adapter->hw_lock, flags); | |
921 | ||
922 | pvscsi_process_request_ring(adapter); | |
923 | ll_device_reset(adapter, cmd->device->id); | |
924 | pvscsi_process_completion_ring(adapter); | |
925 | ||
926 | spin_unlock_irqrestore(&adapter->hw_lock, flags); | |
927 | ||
928 | return SUCCESS; | |
929 | } | |
930 | ||
931 | static struct scsi_host_template pvscsi_template; | |
932 | ||
933 | static const char *pvscsi_info(struct Scsi_Host *host) | |
934 | { | |
935 | struct pvscsi_adapter *adapter = shost_priv(host); | |
936 | static char buf[256]; | |
937 | ||
938 | sprintf(buf, "VMware PVSCSI storage adapter rev %d, req/cmp/msg rings: " | |
939 | "%u/%u/%u pages, cmd_per_lun=%u", adapter->rev, | |
940 | adapter->req_pages, adapter->cmp_pages, adapter->msg_pages, | |
941 | pvscsi_template.cmd_per_lun); | |
942 | ||
943 | return buf; | |
944 | } | |
945 | ||
946 | static struct scsi_host_template pvscsi_template = { | |
947 | .module = THIS_MODULE, | |
948 | .name = "VMware PVSCSI Host Adapter", | |
949 | .proc_name = "vmw_pvscsi", | |
950 | .info = pvscsi_info, | |
951 | .queuecommand = pvscsi_queue, | |
952 | .this_id = -1, | |
953 | .sg_tablesize = PVSCSI_MAX_NUM_SG_ENTRIES_PER_SEGMENT, | |
954 | .dma_boundary = UINT_MAX, | |
955 | .max_sectors = 0xffff, | |
956 | .use_clustering = ENABLE_CLUSTERING, | |
957 | .eh_abort_handler = pvscsi_abort, | |
958 | .eh_device_reset_handler = pvscsi_device_reset, | |
959 | .eh_bus_reset_handler = pvscsi_bus_reset, | |
960 | .eh_host_reset_handler = pvscsi_host_reset, | |
961 | }; | |
962 | ||
963 | static void pvscsi_process_msg(const struct pvscsi_adapter *adapter, | |
964 | const struct PVSCSIRingMsgDesc *e) | |
965 | { | |
966 | struct PVSCSIRingsState *s = adapter->rings_state; | |
967 | struct Scsi_Host *host = adapter->host; | |
968 | struct scsi_device *sdev; | |
969 | ||
970 | printk(KERN_INFO "vmw_pvscsi: msg type: 0x%x - MSG RING: %u/%u (%u) \n", | |
971 | e->type, s->msgProdIdx, s->msgConsIdx, s->msgNumEntriesLog2); | |
972 | ||
973 | BUILD_BUG_ON(PVSCSI_MSG_LAST != 2); | |
974 | ||
975 | if (e->type == PVSCSI_MSG_DEV_ADDED) { | |
976 | struct PVSCSIMsgDescDevStatusChanged *desc; | |
977 | desc = (struct PVSCSIMsgDescDevStatusChanged *)e; | |
978 | ||
979 | printk(KERN_INFO | |
980 | "vmw_pvscsi: msg: device added at scsi%u:%u:%u\n", | |
981 | desc->bus, desc->target, desc->lun[1]); | |
982 | ||
983 | if (!scsi_host_get(host)) | |
984 | return; | |
985 | ||
986 | sdev = scsi_device_lookup(host, desc->bus, desc->target, | |
987 | desc->lun[1]); | |
988 | if (sdev) { | |
989 | printk(KERN_INFO "vmw_pvscsi: device already exists\n"); | |
990 | scsi_device_put(sdev); | |
991 | } else | |
992 | scsi_add_device(adapter->host, desc->bus, | |
993 | desc->target, desc->lun[1]); | |
994 | ||
995 | scsi_host_put(host); | |
996 | } else if (e->type == PVSCSI_MSG_DEV_REMOVED) { | |
997 | struct PVSCSIMsgDescDevStatusChanged *desc; | |
998 | desc = (struct PVSCSIMsgDescDevStatusChanged *)e; | |
999 | ||
1000 | printk(KERN_INFO | |
1001 | "vmw_pvscsi: msg: device removed at scsi%u:%u:%u\n", | |
1002 | desc->bus, desc->target, desc->lun[1]); | |
1003 | ||
1004 | if (!scsi_host_get(host)) | |
1005 | return; | |
1006 | ||
1007 | sdev = scsi_device_lookup(host, desc->bus, desc->target, | |
1008 | desc->lun[1]); | |
1009 | if (sdev) { | |
1010 | scsi_remove_device(sdev); | |
1011 | scsi_device_put(sdev); | |
1012 | } else | |
1013 | printk(KERN_INFO | |
1014 | "vmw_pvscsi: failed to lookup scsi%u:%u:%u\n", | |
1015 | desc->bus, desc->target, desc->lun[1]); | |
1016 | ||
1017 | scsi_host_put(host); | |
1018 | } | |
1019 | } | |
1020 | ||
1021 | static int pvscsi_msg_pending(const struct pvscsi_adapter *adapter) | |
1022 | { | |
1023 | struct PVSCSIRingsState *s = adapter->rings_state; | |
1024 | ||
1025 | return s->msgProdIdx != s->msgConsIdx; | |
1026 | } | |
1027 | ||
1028 | static void pvscsi_process_msg_ring(const struct pvscsi_adapter *adapter) | |
1029 | { | |
1030 | struct PVSCSIRingsState *s = adapter->rings_state; | |
1031 | struct PVSCSIRingMsgDesc *ring = adapter->msg_ring; | |
1032 | u32 msg_entries = s->msgNumEntriesLog2; | |
1033 | ||
1034 | while (pvscsi_msg_pending(adapter)) { | |
1035 | struct PVSCSIRingMsgDesc *e = ring + (s->msgConsIdx & | |
1036 | MASK(msg_entries)); | |
1037 | ||
1038 | barrier(); | |
1039 | pvscsi_process_msg(adapter, e); | |
1040 | barrier(); | |
1041 | s->msgConsIdx++; | |
1042 | } | |
1043 | } | |
1044 | ||
1045 | static void pvscsi_msg_workqueue_handler(struct work_struct *data) | |
1046 | { | |
1047 | struct pvscsi_adapter *adapter; | |
1048 | ||
1049 | adapter = container_of(data, struct pvscsi_adapter, work); | |
1050 | ||
1051 | pvscsi_process_msg_ring(adapter); | |
1052 | } | |
1053 | ||
1054 | static int pvscsi_setup_msg_workqueue(struct pvscsi_adapter *adapter) | |
1055 | { | |
1056 | char name[32]; | |
1057 | ||
1058 | if (!pvscsi_use_msg) | |
1059 | return 0; | |
1060 | ||
1061 | pvscsi_reg_write(adapter, PVSCSI_REG_OFFSET_COMMAND, | |
1062 | PVSCSI_CMD_SETUP_MSG_RING); | |
1063 | ||
1064 | if (pvscsi_reg_read(adapter, PVSCSI_REG_OFFSET_COMMAND_STATUS) == -1) | |
1065 | return 0; | |
1066 | ||
1067 | snprintf(name, sizeof(name), | |
1068 | "vmw_pvscsi_wq_%u", adapter->host->host_no); | |
1069 | ||
1070 | adapter->workqueue = create_singlethread_workqueue(name); | |
1071 | if (!adapter->workqueue) { | |
1072 | printk(KERN_ERR "vmw_pvscsi: failed to create work queue\n"); | |
1073 | return 0; | |
1074 | } | |
1075 | INIT_WORK(&adapter->work, pvscsi_msg_workqueue_handler); | |
1076 | ||
1077 | return 1; | |
1078 | } | |
1079 | ||
1080 | static irqreturn_t pvscsi_isr(int irq, void *devp) | |
1081 | { | |
1082 | struct pvscsi_adapter *adapter = devp; | |
1083 | int handled; | |
1084 | ||
1085 | if (adapter->use_msi || adapter->use_msix) | |
1086 | handled = true; | |
1087 | else { | |
1088 | u32 val = pvscsi_read_intr_status(adapter); | |
1089 | handled = (val & PVSCSI_INTR_ALL_SUPPORTED) != 0; | |
1090 | if (handled) | |
1091 | pvscsi_write_intr_status(devp, val); | |
1092 | } | |
1093 | ||
1094 | if (handled) { | |
1095 | unsigned long flags; | |
1096 | ||
1097 | spin_lock_irqsave(&adapter->hw_lock, flags); | |
1098 | ||
1099 | pvscsi_process_completion_ring(adapter); | |
1100 | if (adapter->use_msg && pvscsi_msg_pending(adapter)) | |
1101 | queue_work(adapter->workqueue, &adapter->work); | |
1102 | ||
1103 | spin_unlock_irqrestore(&adapter->hw_lock, flags); | |
1104 | } | |
1105 | ||
1106 | return IRQ_RETVAL(handled); | |
1107 | } | |
1108 | ||
1109 | static void pvscsi_free_sgls(const struct pvscsi_adapter *adapter) | |
1110 | { | |
1111 | struct pvscsi_ctx *ctx = adapter->cmd_map; | |
1112 | unsigned i; | |
1113 | ||
1114 | for (i = 0; i < adapter->req_depth; ++i, ++ctx) | |
1115 | free_pages((unsigned long)ctx->sgl, get_order(SGL_SIZE)); | |
1116 | } | |
1117 | ||
d0e2ddff DT |
1118 | static int pvscsi_setup_msix(const struct pvscsi_adapter *adapter, |
1119 | unsigned int *irq) | |
851b1642 AK |
1120 | { |
1121 | struct msix_entry entry = { 0, PVSCSI_VECTOR_COMPLETION }; | |
1122 | int ret; | |
1123 | ||
1124 | ret = pci_enable_msix(adapter->dev, &entry, 1); | |
1125 | if (ret) | |
1126 | return ret; | |
1127 | ||
1128 | *irq = entry.vector; | |
1129 | ||
1130 | return 0; | |
1131 | } | |
1132 | ||
1133 | static void pvscsi_shutdown_intr(struct pvscsi_adapter *adapter) | |
1134 | { | |
1135 | if (adapter->irq) { | |
1136 | free_irq(adapter->irq, adapter); | |
1137 | adapter->irq = 0; | |
1138 | } | |
1139 | if (adapter->use_msi) { | |
1140 | pci_disable_msi(adapter->dev); | |
1141 | adapter->use_msi = 0; | |
1142 | } else if (adapter->use_msix) { | |
1143 | pci_disable_msix(adapter->dev); | |
1144 | adapter->use_msix = 0; | |
1145 | } | |
1146 | } | |
1147 | ||
1148 | static void pvscsi_release_resources(struct pvscsi_adapter *adapter) | |
1149 | { | |
1150 | pvscsi_shutdown_intr(adapter); | |
1151 | ||
1152 | if (adapter->workqueue) | |
1153 | destroy_workqueue(adapter->workqueue); | |
1154 | ||
1155 | if (adapter->mmioBase) | |
1156 | pci_iounmap(adapter->dev, adapter->mmioBase); | |
1157 | ||
1158 | pci_release_regions(adapter->dev); | |
1159 | ||
1160 | if (adapter->cmd_map) { | |
1161 | pvscsi_free_sgls(adapter); | |
1162 | kfree(adapter->cmd_map); | |
1163 | } | |
1164 | ||
1165 | if (adapter->rings_state) | |
1166 | pci_free_consistent(adapter->dev, PAGE_SIZE, | |
1167 | adapter->rings_state, adapter->ringStatePA); | |
1168 | ||
1169 | if (adapter->req_ring) | |
1170 | pci_free_consistent(adapter->dev, | |
1171 | adapter->req_pages * PAGE_SIZE, | |
1172 | adapter->req_ring, adapter->reqRingPA); | |
1173 | ||
1174 | if (adapter->cmp_ring) | |
1175 | pci_free_consistent(adapter->dev, | |
1176 | adapter->cmp_pages * PAGE_SIZE, | |
1177 | adapter->cmp_ring, adapter->cmpRingPA); | |
1178 | ||
1179 | if (adapter->msg_ring) | |
1180 | pci_free_consistent(adapter->dev, | |
1181 | adapter->msg_pages * PAGE_SIZE, | |
1182 | adapter->msg_ring, adapter->msgRingPA); | |
1183 | } | |
1184 | ||
1185 | /* | |
1186 | * Allocate scatter gather lists. | |
1187 | * | |
1188 | * These are statically allocated. Trying to be clever was not worth it. | |
1189 | * | |
42b2aa86 | 1190 | * Dynamic allocation can fail, and we can't go deep into the memory |
851b1642 AK |
1191 | * allocator, since we're a SCSI driver, and trying too hard to allocate |
1192 | * memory might generate disk I/O. We also don't want to fail disk I/O | |
1193 | * in that case because we can't get an allocation - the I/O could be | |
1194 | * trying to swap out data to free memory. Since that is pathological, | |
1195 | * just use a statically allocated scatter list. | |
1196 | * | |
1197 | */ | |
6f039790 | 1198 | static int pvscsi_allocate_sg(struct pvscsi_adapter *adapter) |
851b1642 AK |
1199 | { |
1200 | struct pvscsi_ctx *ctx; | |
1201 | int i; | |
1202 | ||
1203 | ctx = adapter->cmd_map; | |
1204 | BUILD_BUG_ON(sizeof(struct pvscsi_sg_list) > SGL_SIZE); | |
1205 | ||
1206 | for (i = 0; i < adapter->req_depth; ++i, ++ctx) { | |
1207 | ctx->sgl = (void *)__get_free_pages(GFP_KERNEL, | |
1208 | get_order(SGL_SIZE)); | |
1209 | ctx->sglPA = 0; | |
1210 | BUG_ON(!IS_ALIGNED(((unsigned long)ctx->sgl), PAGE_SIZE)); | |
1211 | if (!ctx->sgl) { | |
1212 | for (; i >= 0; --i, --ctx) { | |
1213 | free_pages((unsigned long)ctx->sgl, | |
1214 | get_order(SGL_SIZE)); | |
1215 | ctx->sgl = NULL; | |
1216 | } | |
1217 | return -ENOMEM; | |
1218 | } | |
1219 | } | |
1220 | ||
1221 | return 0; | |
1222 | } | |
1223 | ||
a9310735 AK |
1224 | /* |
1225 | * Query the device, fetch the config info and return the | |
1226 | * maximum number of targets on the adapter. In case of | |
1227 | * failure due to any reason return default i.e. 16. | |
1228 | */ | |
1229 | static u32 pvscsi_get_max_targets(struct pvscsi_adapter *adapter) | |
1230 | { | |
1231 | struct PVSCSICmdDescConfigCmd cmd; | |
1232 | struct PVSCSIConfigPageHeader *header; | |
1233 | struct device *dev; | |
1234 | dma_addr_t configPagePA; | |
1235 | void *config_page; | |
1236 | u32 numPhys = 16; | |
1237 | ||
1238 | dev = pvscsi_dev(adapter); | |
1239 | config_page = pci_alloc_consistent(adapter->dev, PAGE_SIZE, | |
1240 | &configPagePA); | |
1241 | if (!config_page) { | |
1242 | dev_warn(dev, "vmw_pvscsi: failed to allocate memory for config page\n"); | |
1243 | goto exit; | |
1244 | } | |
1245 | BUG_ON(configPagePA & ~PAGE_MASK); | |
1246 | ||
1247 | /* Fetch config info from the device. */ | |
1248 | cmd.configPageAddress = ((u64)PVSCSI_CONFIG_CONTROLLER_ADDRESS) << 32; | |
1249 | cmd.configPageNum = PVSCSI_CONFIG_PAGE_CONTROLLER; | |
1250 | cmd.cmpAddr = configPagePA; | |
1251 | cmd._pad = 0; | |
1252 | ||
1253 | /* | |
1254 | * Mark the completion page header with error values. If the device | |
1255 | * completes the command successfully, it sets the status values to | |
1256 | * indicate success. | |
1257 | */ | |
1258 | header = config_page; | |
1259 | memset(header, 0, sizeof *header); | |
1260 | header->hostStatus = BTSTAT_INVPARAM; | |
1261 | header->scsiStatus = SDSTAT_CHECK; | |
1262 | ||
1263 | pvscsi_write_cmd_desc(adapter, PVSCSI_CMD_CONFIG, &cmd, sizeof cmd); | |
1264 | ||
1265 | if (header->hostStatus == BTSTAT_SUCCESS && | |
1266 | header->scsiStatus == SDSTAT_GOOD) { | |
1267 | struct PVSCSIConfigPageController *config; | |
1268 | ||
1269 | config = config_page; | |
1270 | numPhys = config->numPhys; | |
1271 | } else | |
1272 | dev_warn(dev, "vmw_pvscsi: PVSCSI_CMD_CONFIG failed. hostStatus = 0x%x, scsiStatus = 0x%x\n", | |
1273 | header->hostStatus, header->scsiStatus); | |
1274 | pci_free_consistent(adapter->dev, PAGE_SIZE, config_page, configPagePA); | |
1275 | exit: | |
1276 | return numPhys; | |
1277 | } | |
1278 | ||
6f039790 | 1279 | static int pvscsi_probe(struct pci_dev *pdev, const struct pci_device_id *id) |
851b1642 AK |
1280 | { |
1281 | struct pvscsi_adapter *adapter; | |
1282 | struct Scsi_Host *host; | |
a9310735 | 1283 | struct device *dev; |
851b1642 AK |
1284 | unsigned int i; |
1285 | unsigned long flags = 0; | |
1286 | int error; | |
1287 | ||
1288 | error = -ENODEV; | |
1289 | ||
1290 | if (pci_enable_device(pdev)) | |
1291 | return error; | |
1292 | ||
1293 | if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0 && | |
1294 | pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) { | |
1295 | printk(KERN_INFO "vmw_pvscsi: using 64bit dma\n"); | |
1296 | } else if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) == 0 && | |
1297 | pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)) == 0) { | |
1298 | printk(KERN_INFO "vmw_pvscsi: using 32bit dma\n"); | |
1299 | } else { | |
1300 | printk(KERN_ERR "vmw_pvscsi: failed to set DMA mask\n"); | |
1301 | goto out_disable_device; | |
1302 | } | |
1303 | ||
1304 | pvscsi_template.can_queue = | |
1305 | min(PVSCSI_MAX_NUM_PAGES_REQ_RING, pvscsi_ring_pages) * | |
1306 | PVSCSI_MAX_NUM_REQ_ENTRIES_PER_PAGE; | |
1307 | pvscsi_template.cmd_per_lun = | |
1308 | min(pvscsi_template.can_queue, pvscsi_cmd_per_lun); | |
1309 | host = scsi_host_alloc(&pvscsi_template, sizeof(struct pvscsi_adapter)); | |
1310 | if (!host) { | |
1311 | printk(KERN_ERR "vmw_pvscsi: failed to allocate host\n"); | |
1312 | goto out_disable_device; | |
1313 | } | |
1314 | ||
1315 | adapter = shost_priv(host); | |
1316 | memset(adapter, 0, sizeof(*adapter)); | |
1317 | adapter->dev = pdev; | |
1318 | adapter->host = host; | |
1319 | ||
1320 | spin_lock_init(&adapter->hw_lock); | |
1321 | ||
1322 | host->max_channel = 0; | |
1323 | host->max_id = 16; | |
1324 | host->max_lun = 1; | |
1325 | host->max_cmd_len = 16; | |
1326 | ||
1327 | adapter->rev = pdev->revision; | |
1328 | ||
1329 | if (pci_request_regions(pdev, "vmw_pvscsi")) { | |
1330 | printk(KERN_ERR "vmw_pvscsi: pci memory selection failed\n"); | |
1331 | goto out_free_host; | |
1332 | } | |
1333 | ||
1334 | for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) { | |
1335 | if ((pci_resource_flags(pdev, i) & PCI_BASE_ADDRESS_SPACE_IO)) | |
1336 | continue; | |
1337 | ||
1338 | if (pci_resource_len(pdev, i) < PVSCSI_MEM_SPACE_SIZE) | |
1339 | continue; | |
1340 | ||
1341 | break; | |
1342 | } | |
1343 | ||
1344 | if (i == DEVICE_COUNT_RESOURCE) { | |
1345 | printk(KERN_ERR | |
1346 | "vmw_pvscsi: adapter has no suitable MMIO region\n"); | |
1347 | goto out_release_resources; | |
1348 | } | |
1349 | ||
1350 | adapter->mmioBase = pci_iomap(pdev, i, PVSCSI_MEM_SPACE_SIZE); | |
1351 | ||
1352 | if (!adapter->mmioBase) { | |
1353 | printk(KERN_ERR | |
1354 | "vmw_pvscsi: can't iomap for BAR %d memsize %lu\n", | |
1355 | i, PVSCSI_MEM_SPACE_SIZE); | |
1356 | goto out_release_resources; | |
1357 | } | |
1358 | ||
1359 | pci_set_master(pdev); | |
1360 | pci_set_drvdata(pdev, host); | |
1361 | ||
1362 | ll_adapter_reset(adapter); | |
1363 | ||
1364 | adapter->use_msg = pvscsi_setup_msg_workqueue(adapter); | |
1365 | ||
1366 | error = pvscsi_allocate_rings(adapter); | |
1367 | if (error) { | |
1368 | printk(KERN_ERR "vmw_pvscsi: unable to allocate ring memory\n"); | |
1369 | goto out_release_resources; | |
1370 | } | |
1371 | ||
a9310735 AK |
1372 | /* |
1373 | * Ask the device for max number of targets. | |
1374 | */ | |
1375 | host->max_id = pvscsi_get_max_targets(adapter); | |
1376 | dev = pvscsi_dev(adapter); | |
1377 | dev_info(dev, "vmw_pvscsi: host->max_id: %u\n", host->max_id); | |
1378 | ||
851b1642 AK |
1379 | /* |
1380 | * From this point on we should reset the adapter if anything goes | |
1381 | * wrong. | |
1382 | */ | |
1383 | pvscsi_setup_all_rings(adapter); | |
1384 | ||
1385 | adapter->cmd_map = kcalloc(adapter->req_depth, | |
1386 | sizeof(struct pvscsi_ctx), GFP_KERNEL); | |
1387 | if (!adapter->cmd_map) { | |
1388 | printk(KERN_ERR "vmw_pvscsi: failed to allocate memory.\n"); | |
1389 | error = -ENOMEM; | |
1390 | goto out_reset_adapter; | |
1391 | } | |
1392 | ||
1393 | INIT_LIST_HEAD(&adapter->cmd_pool); | |
1394 | for (i = 0; i < adapter->req_depth; i++) { | |
1395 | struct pvscsi_ctx *ctx = adapter->cmd_map + i; | |
1396 | list_add(&ctx->list, &adapter->cmd_pool); | |
1397 | } | |
1398 | ||
1399 | error = pvscsi_allocate_sg(adapter); | |
1400 | if (error) { | |
1401 | printk(KERN_ERR "vmw_pvscsi: unable to allocate s/g table\n"); | |
1402 | goto out_reset_adapter; | |
1403 | } | |
1404 | ||
1405 | if (!pvscsi_disable_msix && | |
1406 | pvscsi_setup_msix(adapter, &adapter->irq) == 0) { | |
1407 | printk(KERN_INFO "vmw_pvscsi: using MSI-X\n"); | |
1408 | adapter->use_msix = 1; | |
1409 | } else if (!pvscsi_disable_msi && pci_enable_msi(pdev) == 0) { | |
1410 | printk(KERN_INFO "vmw_pvscsi: using MSI\n"); | |
1411 | adapter->use_msi = 1; | |
1412 | adapter->irq = pdev->irq; | |
1413 | } else { | |
1414 | printk(KERN_INFO "vmw_pvscsi: using INTx\n"); | |
1415 | adapter->irq = pdev->irq; | |
1416 | flags = IRQF_SHARED; | |
1417 | } | |
1418 | ||
1419 | error = request_irq(adapter->irq, pvscsi_isr, flags, | |
1420 | "vmw_pvscsi", adapter); | |
1421 | if (error) { | |
1422 | printk(KERN_ERR | |
1423 | "vmw_pvscsi: unable to request IRQ: %d\n", error); | |
1424 | adapter->irq = 0; | |
1425 | goto out_reset_adapter; | |
1426 | } | |
1427 | ||
1428 | error = scsi_add_host(host, &pdev->dev); | |
1429 | if (error) { | |
1430 | printk(KERN_ERR | |
1431 | "vmw_pvscsi: scsi_add_host failed: %d\n", error); | |
1432 | goto out_reset_adapter; | |
1433 | } | |
1434 | ||
1435 | dev_info(&pdev->dev, "VMware PVSCSI rev %d host #%u\n", | |
1436 | adapter->rev, host->host_no); | |
1437 | ||
1438 | pvscsi_unmask_intr(adapter); | |
1439 | ||
1440 | scsi_scan_host(host); | |
1441 | ||
1442 | return 0; | |
1443 | ||
1444 | out_reset_adapter: | |
1445 | ll_adapter_reset(adapter); | |
1446 | out_release_resources: | |
1447 | pvscsi_release_resources(adapter); | |
1448 | out_free_host: | |
1449 | scsi_host_put(host); | |
1450 | out_disable_device: | |
851b1642 AK |
1451 | pci_disable_device(pdev); |
1452 | ||
1453 | return error; | |
1454 | } | |
1455 | ||
1456 | static void __pvscsi_shutdown(struct pvscsi_adapter *adapter) | |
1457 | { | |
1458 | pvscsi_mask_intr(adapter); | |
1459 | ||
1460 | if (adapter->workqueue) | |
1461 | flush_workqueue(adapter->workqueue); | |
1462 | ||
1463 | pvscsi_shutdown_intr(adapter); | |
1464 | ||
1465 | pvscsi_process_request_ring(adapter); | |
1466 | pvscsi_process_completion_ring(adapter); | |
1467 | ll_adapter_reset(adapter); | |
1468 | } | |
1469 | ||
1470 | static void pvscsi_shutdown(struct pci_dev *dev) | |
1471 | { | |
1472 | struct Scsi_Host *host = pci_get_drvdata(dev); | |
1473 | struct pvscsi_adapter *adapter = shost_priv(host); | |
1474 | ||
1475 | __pvscsi_shutdown(adapter); | |
1476 | } | |
1477 | ||
1478 | static void pvscsi_remove(struct pci_dev *pdev) | |
1479 | { | |
1480 | struct Scsi_Host *host = pci_get_drvdata(pdev); | |
1481 | struct pvscsi_adapter *adapter = shost_priv(host); | |
1482 | ||
1483 | scsi_remove_host(host); | |
1484 | ||
1485 | __pvscsi_shutdown(adapter); | |
1486 | pvscsi_release_resources(adapter); | |
1487 | ||
1488 | scsi_host_put(host); | |
1489 | ||
851b1642 AK |
1490 | pci_disable_device(pdev); |
1491 | } | |
1492 | ||
1493 | static struct pci_driver pvscsi_pci_driver = { | |
1494 | .name = "vmw_pvscsi", | |
1495 | .id_table = pvscsi_pci_tbl, | |
1496 | .probe = pvscsi_probe, | |
6f039790 | 1497 | .remove = pvscsi_remove, |
851b1642 AK |
1498 | .shutdown = pvscsi_shutdown, |
1499 | }; | |
1500 | ||
1501 | static int __init pvscsi_init(void) | |
1502 | { | |
1503 | pr_info("%s - version %s\n", | |
1504 | PVSCSI_LINUX_DRIVER_DESC, PVSCSI_DRIVER_VERSION_STRING); | |
1505 | return pci_register_driver(&pvscsi_pci_driver); | |
1506 | } | |
1507 | ||
1508 | static void __exit pvscsi_exit(void) | |
1509 | { | |
1510 | pci_unregister_driver(&pvscsi_pci_driver); | |
1511 | } | |
1512 | ||
1513 | module_init(pvscsi_init); | |
1514 | module_exit(pvscsi_exit); |