1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (c) 2017, Oracle and/or its affiliates. All rights reserved.
4 */
5
6 /*
7 * Oracle Data Analytics Accelerator (DAX)
8 *
9 * DAX is a coprocessor which resides on the SPARC M7 (DAX1) and M8
10 * (DAX2) processor chips, and has direct access to the CPU's L3
11 * caches as well as physical memory. It can perform several
12 * operations on data streams with various input and output formats.
13 * The driver provides a transport mechanism only and has limited
14 * knowledge of the various opcodes and data formats. A user space
15 * library provides high level services and translates these into low
16 * level commands which are then passed into the driver and
17 * subsequently the hypervisor and the coprocessor. The library is
18 * the recommended way for applications to use the coprocessor, and
19 * the driver interface is not intended for general use.
20 *
21 * See Documentation/arch/sparc/oradax/oracle-dax.rst for more details.
22 */
23
24 #include <linux/uaccess.h>
25 #include <linux/module.h>
26 #include <linux/delay.h>
27 #include <linux/cdev.h>
28 #include <linux/slab.h>
29 #include <linux/mm.h>
30
31 #include <asm/hypervisor.h>
32 #include <asm/mdesc.h>
33 #include <asm/oradax.h>
34
35 MODULE_LICENSE("GPL");
36 MODULE_DESCRIPTION("Driver for Oracle Data Analytics Accelerator");
37
38 #define DAX_DBG_FLG_BASIC 0x01
39 #define DAX_DBG_FLG_STAT 0x02
40 #define DAX_DBG_FLG_INFO 0x04
41 #define DAX_DBG_FLG_ALL 0xff
42
43 #define dax_err(fmt, ...) pr_err("%s: " fmt "\n", __func__, ##__VA_ARGS__)
44 #define dax_info(fmt, ...) pr_info("%s: " fmt "\n", __func__, ##__VA_ARGS__)
45
46 #define dax_dbg(fmt, ...) do { \
47 if (dax_debug & DAX_DBG_FLG_BASIC)\
48 dax_info(fmt, ##__VA_ARGS__); \
49 } while (0)
50 #define dax_stat_dbg(fmt, ...) do { \
51 if (dax_debug & DAX_DBG_FLG_STAT) \
52 dax_info(fmt, ##__VA_ARGS__); \
53 } while (0)
54 #define dax_info_dbg(fmt, ...) do { \
55 if (dax_debug & DAX_DBG_FLG_INFO) \
56 dax_info(fmt, ##__VA_ARGS__); \
57 } while (0)
58
59 #define DAX1_MINOR 1
60 #define DAX1_MAJOR 1
61 #define DAX2_MINOR 0
62 #define DAX2_MAJOR 2
63
64 #define DAX1_STR "ORCL,sun4v-dax"
65 #define DAX2_STR "ORCL,sun4v-dax2"
66
67 #define DAX_CA_ELEMS (DAX_MMAP_LEN / sizeof(struct dax_cca))
68
69 #define DAX_CCB_USEC 100
70 #define DAX_CCB_RETRIES 10000
71
72 /* stream types */
73 enum {
74 OUT,
75 PRI,
76 SEC,
77 TBL,
78 NUM_STREAM_TYPES
79 };
80
81 /* completion status */
82 #define CCA_STAT_NOT_COMPLETED 0
83 #define CCA_STAT_COMPLETED 1
84 #define CCA_STAT_FAILED 2
85 #define CCA_STAT_KILLED 3
86 #define CCA_STAT_NOT_RUN 4
87 #define CCA_STAT_PIPE_OUT 5
88 #define CCA_STAT_PIPE_SRC 6
89 #define CCA_STAT_PIPE_DST 7
90
91 /* completion err */
92 #define CCA_ERR_SUCCESS 0x0 /* no error */
93 #define CCA_ERR_OVERFLOW 0x1 /* buffer overflow */
94 #define CCA_ERR_DECODE 0x2 /* CCB decode error */
95 #define CCA_ERR_PAGE_OVERFLOW 0x3 /* page overflow */
96 #define CCA_ERR_KILLED 0x7 /* command was killed */
97 #define CCA_ERR_TIMEOUT 0x8 /* Timeout */
98 #define CCA_ERR_ADI 0x9 /* ADI error */
99 #define CCA_ERR_DATA_FMT 0xA /* data format error */
100 #define CCA_ERR_OTHER_NO_RETRY 0xE /* Other error, do not retry */
101 #define CCA_ERR_OTHER_RETRY 0xF /* Other error, retry */
102 #define CCA_ERR_PARTIAL_SYMBOL 0x80 /* QP partial symbol warning */
103
104 /* CCB address types */
105 #define DAX_ADDR_TYPE_NONE 0
106 #define DAX_ADDR_TYPE_VA_ALT 1 /* secondary context */
107 #define DAX_ADDR_TYPE_RA 2 /* real address */
108 #define DAX_ADDR_TYPE_VA 3 /* virtual address */
109
110 /* dax_header_t opcode */
111 #define DAX_OP_SYNC_NOP 0x0
112 #define DAX_OP_EXTRACT 0x1
113 #define DAX_OP_SCAN_VALUE 0x2
114 #define DAX_OP_SCAN_RANGE 0x3
115 #define DAX_OP_TRANSLATE 0x4
116 #define DAX_OP_SELECT 0x5
117 #define DAX_OP_INVERT 0x10 /* OR with translate, scan opcodes */
118
119 struct dax_header {
120 u32 ccb_version:4; /* 31:28 CCB Version */
121 /* 27:24 Sync Flags */
122 u32 pipe:1; /* Pipeline */
123 u32 longccb:1; /* Longccb. Set for scan with lu2, lu3, lu4. */
124 u32 cond:1; /* Conditional */
125 u32 serial:1; /* Serial */
126 u32 opcode:8; /* 23:16 Opcode */
127 /* 15:0 Address Type. */
128 u32 reserved:3; /* 15:13 reserved */
129 u32 table_addr_type:2; /* 12:11 Huffman Table Address Type */
130 u32 out_addr_type:3; /* 10:8 Destination Address Type */
131 u32 sec_addr_type:3; /* 7:5 Secondary Source Address Type */
132 u32 pri_addr_type:3; /* 4:2 Primary Source Address Type */
133 u32 cca_addr_type:2; /* 1:0 Completion Address Type */
134 };
135
136 struct dax_control {
137 u32 pri_fmt:4; /* 31:28 Primary Input Format */
138 u32 pri_elem_size:5; /* 27:23 Primary Input Element Size(less1) */
139 u32 pri_offset:3; /* 22:20 Primary Input Starting Offset */
140 u32 sec_encoding:1; /* 19 Secondary Input Encoding */
141 /* (must be 0 for Select) */
142 u32 sec_offset:3; /* 18:16 Secondary Input Starting Offset */
143 u32 sec_elem_size:2; /* 15:14 Secondary Input Element Size */
144 /* (must be 0 for Select) */
145 u32 out_fmt:2; /* 13:12 Output Format */
146 u32 out_elem_size:2; /* 11:10 Output Element Size */
147 u32 misc:10; /* 9:0 Opcode specific info */
148 };
149
150 struct dax_data_access {
151 u64 flow_ctrl:2; /* 63:62 Flow Control Type */
152 u64 pipe_target:2; /* 61:60 Pipeline Target */
153 u64 out_buf_size:20; /* 59:40 Output Buffer Size */
154 /* (cachelines less 1) */
155 u64 unused1:8; /* 39:32 Reserved, Set to 0 */
156 u64 out_alloc:5; /* 31:27 Output Allocation */
157 u64 unused2:1; /* 26 Reserved */
158 u64 pri_len_fmt:2; /* 25:24 Input Length Format */
159 u64 pri_len:24; /* 23:0 Input Element/Byte/Bit Count */
160 /* (less 1) */
161 };
162
163 struct dax_ccb {
164 struct dax_header hdr; /* CCB Header */
165 struct dax_control ctrl;/* Control Word */
166 void *ca; /* Completion Address */
167 void *pri; /* Primary Input Address */
168 struct dax_data_access dac; /* Data Access Control */
169 void *sec; /* Secondary Input Address */
170 u64 dword5; /* depends on opcode */
171 void *out; /* Output Address */
172 void *tbl; /* Table Address or bitmap */
173 };
174
175 struct dax_cca {
176 u8 status; /* user may mwait on this address */
177 u8 err; /* user visible error notification */
178 u8 rsvd[2]; /* reserved */
179 u32 n_remaining; /* for QP partial symbol warning */
180 u32 output_sz; /* output in bytes */
181 u32 rsvd2; /* reserved */
182 u64 run_cycles; /* run time in OCND2 cycles */
183 u64 run_stats; /* nothing reported in version 1.0 */
184 u32 n_processed; /* number input elements */
185 u32 rsvd3[5]; /* reserved */
186 u64 retval; /* command return value */
187 u64 rsvd4[8]; /* reserved */
188 };
189
190 /* per thread CCB context */
191 struct dax_ctx {
192 struct dax_ccb *ccb_buf;
193 u64 ccb_buf_ra; /* cached RA of ccb_buf */
194 struct dax_cca *ca_buf;
195 u64 ca_buf_ra; /* cached RA of ca_buf */
196 struct page *pages[DAX_CA_ELEMS][NUM_STREAM_TYPES];
197 /* array of locked pages */
198 struct task_struct *owner; /* thread that owns ctx */
199 struct task_struct *client; /* requesting thread */
200 union ccb_result result;
201 u32 ccb_count;
202 u32 fail_count;
203 };
204
205 /* driver public entry points */
206 static int dax_open(struct inode *inode, struct file *file);
207 static ssize_t dax_read(struct file *filp, char __user *buf,
208 size_t count, loff_t *ppos);
209 static ssize_t dax_write(struct file *filp, const char __user *buf,
210 size_t count, loff_t *ppos);
211 static int dax_devmap(struct file *f, struct vm_area_struct *vma);
212 static int dax_close(struct inode *i, struct file *f);
213
214 static const struct file_operations dax_fops = {
215 .owner = THIS_MODULE,
216 .open = dax_open,
217 .read = dax_read,
218 .write = dax_write,
219 .mmap = dax_devmap,
220 .release = dax_close,
221 };
222
223 static int dax_ccb_exec(struct dax_ctx *ctx, const char __user *buf,
224 size_t count, loff_t *ppos);
225 static int dax_ccb_info(u64 ca, struct ccb_info_result *info);
226 static int dax_ccb_kill(u64 ca, u16 *kill_res);
227
228 static struct cdev c_dev;
229 static dev_t first;
230 static const struct class cl = {
231 .name = DAX_NAME,
232 };
233
234 static int max_ccb_version;
235 static int dax_debug;
236 module_param(dax_debug, int, 0644);
237 MODULE_PARM_DESC(dax_debug, "Debug flags");
238
dax_attach(void)239 static int __init dax_attach(void)
240 {
241 unsigned long dummy, hv_rv, major, minor, minor_requested, max_ccbs;
242 struct mdesc_handle *hp = mdesc_grab();
243 char *prop, *dax_name;
244 bool found = false;
245 int len, ret = 0;
246 u64 pn;
247
248 if (hp == NULL) {
249 dax_err("Unable to grab mdesc");
250 return -ENODEV;
251 }
252
253 mdesc_for_each_node_by_name(hp, pn, "virtual-device") {
254 prop = (char *)mdesc_get_property(hp, pn, "name", &len);
255 if (prop == NULL)
256 continue;
257 if (strncmp(prop, "dax", strlen("dax")))
258 continue;
259 dax_dbg("Found node 0x%llx = %s", pn, prop);
260
261 prop = (char *)mdesc_get_property(hp, pn, "compatible", &len);
262 if (prop == NULL)
263 continue;
264 dax_dbg("Found node 0x%llx = %s", pn, prop);
265 found = true;
266 break;
267 }
268
269 if (!found) {
270 dax_err("No DAX device found");
271 ret = -ENODEV;
272 goto done;
273 }
274
275 if (strncmp(prop, DAX2_STR, strlen(DAX2_STR)) == 0) {
276 dax_name = DAX_NAME "2";
277 major = DAX2_MAJOR;
278 minor_requested = DAX2_MINOR;
279 max_ccb_version = 1;
280 dax_dbg("MD indicates DAX2 coprocessor");
281 } else if (strncmp(prop, DAX1_STR, strlen(DAX1_STR)) == 0) {
282 dax_name = DAX_NAME "1";
283 major = DAX1_MAJOR;
284 minor_requested = DAX1_MINOR;
285 max_ccb_version = 0;
286 dax_dbg("MD indicates DAX1 coprocessor");
287 } else {
288 dax_err("Unknown dax type: %s", prop);
289 ret = -ENODEV;
290 goto done;
291 }
292
293 minor = minor_requested;
294 dax_dbg("Registering DAX HV api with major %ld minor %ld", major,
295 minor);
296 if (sun4v_hvapi_register(HV_GRP_DAX, major, &minor)) {
297 dax_err("hvapi_register failed");
298 ret = -ENODEV;
299 goto done;
300 } else {
301 dax_dbg("Max minor supported by HV = %ld (major %ld)", minor,
302 major);
303 minor = min(minor, minor_requested);
304 dax_dbg("registered DAX major %ld minor %ld", major, minor);
305 }
306
307 /* submit a zero length ccb array to query coprocessor queue size */
308 hv_rv = sun4v_ccb_submit(0, 0, HV_CCB_QUERY_CMD, 0, &max_ccbs, &dummy);
309 if (hv_rv != 0) {
310 dax_err("get_hwqueue_size failed with status=%ld and max_ccbs=%ld",
311 hv_rv, max_ccbs);
312 ret = -ENODEV;
313 goto done;
314 }
315
316 if (max_ccbs != DAX_MAX_CCBS) {
317 dax_err("HV reports unsupported max_ccbs=%ld", max_ccbs);
318 ret = -ENODEV;
319 goto done;
320 }
321
322 if (alloc_chrdev_region(&first, 0, 1, DAX_NAME) < 0) {
323 dax_err("alloc_chrdev_region failed");
324 ret = -ENXIO;
325 goto done;
326 }
327
328 ret = class_register(&cl);
329 if (ret)
330 goto class_error;
331
332 if (device_create(&cl, NULL, first, NULL, dax_name) == NULL) {
333 dax_err("device_create failed");
334 ret = -ENXIO;
335 goto device_error;
336 }
337
338 cdev_init(&c_dev, &dax_fops);
339 if (cdev_add(&c_dev, first, 1) == -1) {
340 dax_err("cdev_add failed");
341 ret = -ENXIO;
342 goto cdev_error;
343 }
344
345 pr_info("Attached DAX module\n");
346 goto done;
347
348 cdev_error:
349 device_destroy(&cl, first);
350 device_error:
351 class_unregister(&cl);
352 class_error:
353 unregister_chrdev_region(first, 1);
354 done:
355 mdesc_release(hp);
356 return ret;
357 }
358 module_init(dax_attach);
359
dax_detach(void)360 static void __exit dax_detach(void)
361 {
362 pr_info("Cleaning up DAX module\n");
363 cdev_del(&c_dev);
364 device_destroy(&cl, first);
365 class_unregister(&cl);
366 unregister_chrdev_region(first, 1);
367 }
368 module_exit(dax_detach);
369
370 /* map completion area */
dax_devmap(struct file * f,struct vm_area_struct * vma)371 static int dax_devmap(struct file *f, struct vm_area_struct *vma)
372 {
373 struct dax_ctx *ctx = (struct dax_ctx *)f->private_data;
374 size_t len = vma->vm_end - vma->vm_start;
375
376 dax_dbg("len=0x%lx, flags=0x%lx", len, vma->vm_flags);
377
378 if (ctx->owner != current) {
379 dax_dbg("devmap called from wrong thread");
380 return -EINVAL;
381 }
382
383 if (len != DAX_MMAP_LEN) {
384 dax_dbg("len(%lu) != DAX_MMAP_LEN(%d)", len, DAX_MMAP_LEN);
385 return -EINVAL;
386 }
387
388 /* completion area is mapped read-only for user */
389 if (vma->vm_flags & VM_WRITE)
390 return -EPERM;
391 vm_flags_clear(vma, VM_MAYWRITE);
392
393 if (remap_pfn_range(vma, vma->vm_start, ctx->ca_buf_ra >> PAGE_SHIFT,
394 len, vma->vm_page_prot))
395 return -EAGAIN;
396
397 dax_dbg("mmapped completion area at uva 0x%lx", vma->vm_start);
398 return 0;
399 }
400
401 /* Unlock user pages. Called during dequeue or device close */
dax_unlock_pages(struct dax_ctx * ctx,int ccb_index,int nelem)402 static void dax_unlock_pages(struct dax_ctx *ctx, int ccb_index, int nelem)
403 {
404 int i, j;
405
406 for (i = ccb_index; i < ccb_index + nelem; i++) {
407 for (j = 0; j < NUM_STREAM_TYPES; j++) {
408 struct page *p = ctx->pages[i][j];
409
410 if (p) {
411 dax_dbg("freeing page %p", p);
412 unpin_user_pages_dirty_lock(&p, 1, j == OUT);
413 ctx->pages[i][j] = NULL;
414 }
415 }
416 }
417 }
418
dax_lock_page(void * va,struct page ** p)419 static int dax_lock_page(void *va, struct page **p)
420 {
421 int ret;
422
423 dax_dbg("uva %p", va);
424
425 ret = pin_user_pages_fast((unsigned long)va, 1, FOLL_WRITE, p);
426 if (ret == 1) {
427 dax_dbg("locked page %p, for VA %p", *p, va);
428 return 0;
429 }
430
431 dax_dbg("pin_user_pages failed, va=%p, ret=%d", va, ret);
432 return -1;
433 }
434
dax_lock_pages(struct dax_ctx * ctx,int idx,int nelem,u64 * err_va)435 static int dax_lock_pages(struct dax_ctx *ctx, int idx,
436 int nelem, u64 *err_va)
437 {
438 int i;
439
440 for (i = 0; i < nelem; i++) {
441 struct dax_ccb *ccbp = &ctx->ccb_buf[i];
442
443 /*
444 * For each address in the CCB whose type is virtual,
445 * lock the page and change the type to virtual alternate
446 * context. On error, return the offending address in
447 * err_va.
448 */
449 if (ccbp->hdr.out_addr_type == DAX_ADDR_TYPE_VA) {
450 dax_dbg("output");
451 if (dax_lock_page(ccbp->out,
452 &ctx->pages[i + idx][OUT]) != 0) {
453 *err_va = (u64)ccbp->out;
454 goto error;
455 }
456 ccbp->hdr.out_addr_type = DAX_ADDR_TYPE_VA_ALT;
457 }
458
459 if (ccbp->hdr.pri_addr_type == DAX_ADDR_TYPE_VA) {
460 dax_dbg("input");
461 if (dax_lock_page(ccbp->pri,
462 &ctx->pages[i + idx][PRI]) != 0) {
463 *err_va = (u64)ccbp->pri;
464 goto error;
465 }
466 ccbp->hdr.pri_addr_type = DAX_ADDR_TYPE_VA_ALT;
467 }
468
469 if (ccbp->hdr.sec_addr_type == DAX_ADDR_TYPE_VA) {
470 dax_dbg("sec input");
471 if (dax_lock_page(ccbp->sec,
472 &ctx->pages[i + idx][SEC]) != 0) {
473 *err_va = (u64)ccbp->sec;
474 goto error;
475 }
476 ccbp->hdr.sec_addr_type = DAX_ADDR_TYPE_VA_ALT;
477 }
478
479 if (ccbp->hdr.table_addr_type == DAX_ADDR_TYPE_VA) {
480 dax_dbg("tbl");
481 if (dax_lock_page(ccbp->tbl,
482 &ctx->pages[i + idx][TBL]) != 0) {
483 *err_va = (u64)ccbp->tbl;
484 goto error;
485 }
486 ccbp->hdr.table_addr_type = DAX_ADDR_TYPE_VA_ALT;
487 }
488
489 /* skip over 2nd 64 bytes of long CCB */
490 if (ccbp->hdr.longccb)
491 i++;
492 }
493 return DAX_SUBMIT_OK;
494
495 error:
496 dax_unlock_pages(ctx, idx, nelem);
497 return DAX_SUBMIT_ERR_NOACCESS;
498 }
499
dax_ccb_wait(struct dax_ctx * ctx,int idx)500 static void dax_ccb_wait(struct dax_ctx *ctx, int idx)
501 {
502 int ret, nretries;
503 u16 kill_res;
504
505 dax_dbg("idx=%d", idx);
506
507 for (nretries = 0; nretries < DAX_CCB_RETRIES; nretries++) {
508 if (ctx->ca_buf[idx].status == CCA_STAT_NOT_COMPLETED)
509 udelay(DAX_CCB_USEC);
510 else
511 return;
512 }
513 dax_dbg("ctx (%p): CCB[%d] timed out, wait usec=%d, retries=%d. Killing ccb",
514 (void *)ctx, idx, DAX_CCB_USEC, DAX_CCB_RETRIES);
515
516 ret = dax_ccb_kill(ctx->ca_buf_ra + idx * sizeof(struct dax_cca),
517 &kill_res);
518 dax_dbg("Kill CCB[%d] %s", idx, ret ? "failed" : "succeeded");
519 }
520
dax_close(struct inode * ino,struct file * f)521 static int dax_close(struct inode *ino, struct file *f)
522 {
523 struct dax_ctx *ctx = (struct dax_ctx *)f->private_data;
524 int i;
525
526 f->private_data = NULL;
527
528 for (i = 0; i < DAX_CA_ELEMS; i++) {
529 if (ctx->ca_buf[i].status == CCA_STAT_NOT_COMPLETED) {
530 dax_dbg("CCB[%d] not completed", i);
531 dax_ccb_wait(ctx, i);
532 }
533 dax_unlock_pages(ctx, i, 1);
534 }
535
536 kfree(ctx->ccb_buf);
537 kfree(ctx->ca_buf);
538 dax_stat_dbg("CCBs: %d good, %d bad", ctx->ccb_count, ctx->fail_count);
539 kfree(ctx);
540
541 return 0;
542 }
543
dax_read(struct file * f,char __user * buf,size_t count,loff_t * ppos)544 static ssize_t dax_read(struct file *f, char __user *buf,
545 size_t count, loff_t *ppos)
546 {
547 struct dax_ctx *ctx = f->private_data;
548
549 if (ctx->client != current)
550 return -EUSERS;
551
552 ctx->client = NULL;
553
554 if (count != sizeof(union ccb_result))
555 return -EINVAL;
556 if (copy_to_user(buf, &ctx->result, sizeof(union ccb_result)))
557 return -EFAULT;
558 return count;
559 }
560
dax_write(struct file * f,const char __user * buf,size_t count,loff_t * ppos)561 static ssize_t dax_write(struct file *f, const char __user *buf,
562 size_t count, loff_t *ppos)
563 {
564 struct dax_ctx *ctx = f->private_data;
565 struct dax_command hdr;
566 unsigned long ca;
567 int i, idx, ret;
568
569 if (ctx->client != NULL)
570 return -EINVAL;
571
572 if (count == 0 || count > DAX_MAX_CCBS * sizeof(struct dax_ccb))
573 return -EINVAL;
574
575 if (count % sizeof(struct dax_ccb) == 0)
576 return dax_ccb_exec(ctx, buf, count, ppos); /* CCB EXEC */
577
578 if (count != sizeof(struct dax_command))
579 return -EINVAL;
580
581 /* immediate command */
582 if (ctx->owner != current)
583 return -EUSERS;
584
585 if (copy_from_user(&hdr, buf, sizeof(hdr)))
586 return -EFAULT;
587
588 ca = ctx->ca_buf_ra + hdr.ca_offset;
589
590 switch (hdr.command) {
591 case CCB_KILL:
592 if (hdr.ca_offset >= DAX_MMAP_LEN) {
593 dax_dbg("invalid ca_offset (%d) >= ca_buflen (%d)",
594 hdr.ca_offset, DAX_MMAP_LEN);
595 return -EINVAL;
596 }
597
598 ret = dax_ccb_kill(ca, &ctx->result.kill.action);
599 if (ret != 0) {
600 dax_dbg("dax_ccb_kill failed (ret=%d)", ret);
601 return ret;
602 }
603
604 dax_info_dbg("killed (ca_offset %d)", hdr.ca_offset);
605 idx = hdr.ca_offset / sizeof(struct dax_cca);
606 ctx->ca_buf[idx].status = CCA_STAT_KILLED;
607 ctx->ca_buf[idx].err = CCA_ERR_KILLED;
608 ctx->client = current;
609 return count;
610
611 case CCB_INFO:
612 if (hdr.ca_offset >= DAX_MMAP_LEN) {
613 dax_dbg("invalid ca_offset (%d) >= ca_buflen (%d)",
614 hdr.ca_offset, DAX_MMAP_LEN);
615 return -EINVAL;
616 }
617
618 ret = dax_ccb_info(ca, &ctx->result.info);
619 if (ret != 0) {
620 dax_dbg("dax_ccb_info failed (ret=%d)", ret);
621 return ret;
622 }
623
624 dax_info_dbg("info succeeded on ca_offset %d", hdr.ca_offset);
625 ctx->client = current;
626 return count;
627
628 case CCB_DEQUEUE:
629 for (i = 0; i < DAX_CA_ELEMS; i++) {
630 if (ctx->ca_buf[i].status !=
631 CCA_STAT_NOT_COMPLETED)
632 dax_unlock_pages(ctx, i, 1);
633 }
634 return count;
635
636 default:
637 return -EINVAL;
638 }
639 }
640
dax_open(struct inode * inode,struct file * f)641 static int dax_open(struct inode *inode, struct file *f)
642 {
643 struct dax_ctx *ctx = NULL;
644 int i;
645
646 ctx = kzalloc_obj(*ctx);
647 if (ctx == NULL)
648 goto done;
649
650 ctx->ccb_buf = kzalloc_objs(struct dax_ccb, DAX_MAX_CCBS);
651 if (ctx->ccb_buf == NULL)
652 goto done;
653
654 ctx->ccb_buf_ra = virt_to_phys(ctx->ccb_buf);
655 dax_dbg("ctx->ccb_buf=0x%p, ccb_buf_ra=0x%llx",
656 (void *)ctx->ccb_buf, ctx->ccb_buf_ra);
657
658 /* allocate CCB completion area buffer */
659 ctx->ca_buf = kzalloc(DAX_MMAP_LEN, GFP_KERNEL);
660 if (ctx->ca_buf == NULL)
661 goto alloc_error;
662 for (i = 0; i < DAX_CA_ELEMS; i++)
663 ctx->ca_buf[i].status = CCA_STAT_COMPLETED;
664
665 ctx->ca_buf_ra = virt_to_phys(ctx->ca_buf);
666 dax_dbg("ctx=0x%p, ctx->ca_buf=0x%p, ca_buf_ra=0x%llx",
667 (void *)ctx, (void *)ctx->ca_buf, ctx->ca_buf_ra);
668
669 ctx->owner = current;
670 f->private_data = ctx;
671 return 0;
672
673 alloc_error:
674 kfree(ctx->ccb_buf);
675 done:
676 kfree(ctx);
677 return -ENOMEM;
678 }
679
dax_hv_errno(unsigned long hv_ret,int * ret)680 static char *dax_hv_errno(unsigned long hv_ret, int *ret)
681 {
682 switch (hv_ret) {
683 case HV_EBADALIGN:
684 *ret = -EFAULT;
685 return "HV_EBADALIGN";
686 case HV_ENORADDR:
687 *ret = -EFAULT;
688 return "HV_ENORADDR";
689 case HV_EINVAL:
690 *ret = -EINVAL;
691 return "HV_EINVAL";
692 case HV_EWOULDBLOCK:
693 *ret = -EAGAIN;
694 return "HV_EWOULDBLOCK";
695 case HV_ENOACCESS:
696 *ret = -EPERM;
697 return "HV_ENOACCESS";
698 default:
699 break;
700 }
701
702 *ret = -EIO;
703 return "UNKNOWN";
704 }
705
dax_ccb_kill(u64 ca,u16 * kill_res)706 static int dax_ccb_kill(u64 ca, u16 *kill_res)
707 {
708 unsigned long hv_ret;
709 int count, ret = 0;
710 char *err_str;
711
712 for (count = 0; count < DAX_CCB_RETRIES; count++) {
713 dax_dbg("attempting kill on ca_ra 0x%llx", ca);
714 hv_ret = sun4v_ccb_kill(ca, kill_res);
715
716 if (hv_ret == HV_EOK) {
717 dax_info_dbg("HV_EOK (ca_ra 0x%llx): %d", ca,
718 *kill_res);
719 } else {
720 err_str = dax_hv_errno(hv_ret, &ret);
721 dax_dbg("%s (ca_ra 0x%llx)", err_str, ca);
722 }
723
724 if (ret != -EAGAIN)
725 return ret;
726 dax_info_dbg("ccb_kill count = %d", count);
727 udelay(DAX_CCB_USEC);
728 }
729
730 return -EAGAIN;
731 }
732
dax_ccb_info(u64 ca,struct ccb_info_result * info)733 static int dax_ccb_info(u64 ca, struct ccb_info_result *info)
734 {
735 unsigned long hv_ret;
736 char *err_str;
737 int ret = 0;
738
739 dax_dbg("attempting info on ca_ra 0x%llx", ca);
740 hv_ret = sun4v_ccb_info(ca, info);
741
742 if (hv_ret == HV_EOK) {
743 dax_info_dbg("HV_EOK (ca_ra 0x%llx): %d", ca, info->state);
744 if (info->state == DAX_CCB_ENQUEUED) {
745 dax_info_dbg("dax_unit %d, queue_num %d, queue_pos %d",
746 info->inst_num, info->q_num, info->q_pos);
747 }
748 } else {
749 err_str = dax_hv_errno(hv_ret, &ret);
750 dax_dbg("%s (ca_ra 0x%llx)", err_str, ca);
751 }
752
753 return ret;
754 }
755
dax_prt_ccbs(struct dax_ccb * ccb,int nelem)756 static void dax_prt_ccbs(struct dax_ccb *ccb, int nelem)
757 {
758 int i, j;
759 u64 *ccbp;
760
761 dax_dbg("ccb buffer:");
762 for (i = 0; i < nelem; i++) {
763 ccbp = (u64 *)&ccb[i];
764 dax_dbg(" %sccb[%d]", ccb[i].hdr.longccb ? "long " : "", i);
765 for (j = 0; j < 8; j++)
766 dax_dbg("\tccb[%d].dwords[%d]=0x%llx",
767 i, j, *(ccbp + j));
768 }
769 }
770
771 /*
772 * Validates user CCB content. Also sets completion address and address types
773 * for all addresses contained in CCB.
774 */
dax_preprocess_usr_ccbs(struct dax_ctx * ctx,int idx,int nelem)775 static int dax_preprocess_usr_ccbs(struct dax_ctx *ctx, int idx, int nelem)
776 {
777 int i;
778
779 /*
780 * The user is not allowed to specify real address types in
781 * the CCB header. This must be enforced by the kernel before
782 * submitting the CCBs to HV. The only allowed values for all
783 * address fields are VA or IMM
784 */
785 for (i = 0; i < nelem; i++) {
786 struct dax_ccb *ccbp = &ctx->ccb_buf[i];
787 unsigned long ca_offset;
788
789 if (ccbp->hdr.ccb_version > max_ccb_version)
790 return DAX_SUBMIT_ERR_CCB_INVAL;
791
792 switch (ccbp->hdr.opcode) {
793 case DAX_OP_SYNC_NOP:
794 case DAX_OP_EXTRACT:
795 case DAX_OP_SCAN_VALUE:
796 case DAX_OP_SCAN_RANGE:
797 case DAX_OP_TRANSLATE:
798 case DAX_OP_SCAN_VALUE | DAX_OP_INVERT:
799 case DAX_OP_SCAN_RANGE | DAX_OP_INVERT:
800 case DAX_OP_TRANSLATE | DAX_OP_INVERT:
801 case DAX_OP_SELECT:
802 break;
803 default:
804 return DAX_SUBMIT_ERR_CCB_INVAL;
805 }
806
807 if (ccbp->hdr.out_addr_type != DAX_ADDR_TYPE_VA &&
808 ccbp->hdr.out_addr_type != DAX_ADDR_TYPE_NONE) {
809 dax_dbg("invalid out_addr_type in user CCB[%d]", i);
810 return DAX_SUBMIT_ERR_CCB_INVAL;
811 }
812
813 if (ccbp->hdr.pri_addr_type != DAX_ADDR_TYPE_VA &&
814 ccbp->hdr.pri_addr_type != DAX_ADDR_TYPE_NONE) {
815 dax_dbg("invalid pri_addr_type in user CCB[%d]", i);
816 return DAX_SUBMIT_ERR_CCB_INVAL;
817 }
818
819 if (ccbp->hdr.sec_addr_type != DAX_ADDR_TYPE_VA &&
820 ccbp->hdr.sec_addr_type != DAX_ADDR_TYPE_NONE) {
821 dax_dbg("invalid sec_addr_type in user CCB[%d]", i);
822 return DAX_SUBMIT_ERR_CCB_INVAL;
823 }
824
825 if (ccbp->hdr.table_addr_type != DAX_ADDR_TYPE_VA &&
826 ccbp->hdr.table_addr_type != DAX_ADDR_TYPE_NONE) {
827 dax_dbg("invalid table_addr_type in user CCB[%d]", i);
828 return DAX_SUBMIT_ERR_CCB_INVAL;
829 }
830
831 /* set completion (real) address and address type */
832 ccbp->hdr.cca_addr_type = DAX_ADDR_TYPE_RA;
833 ca_offset = (idx + i) * sizeof(struct dax_cca);
834 ccbp->ca = (void *)ctx->ca_buf_ra + ca_offset;
835 memset(&ctx->ca_buf[idx + i], 0, sizeof(struct dax_cca));
836
837 dax_dbg("ccb[%d]=%p, ca_offset=0x%lx, compl RA=0x%llx",
838 i, ccbp, ca_offset, ctx->ca_buf_ra + ca_offset);
839
840 /* skip over 2nd 64 bytes of long CCB */
841 if (ccbp->hdr.longccb)
842 i++;
843 }
844
845 return DAX_SUBMIT_OK;
846 }
847
dax_ccb_exec(struct dax_ctx * ctx,const char __user * buf,size_t count,loff_t * ppos)848 static int dax_ccb_exec(struct dax_ctx *ctx, const char __user *buf,
849 size_t count, loff_t *ppos)
850 {
851 unsigned long accepted_len, hv_rv;
852 int i, idx, nccbs, naccepted;
853
854 ctx->client = current;
855 idx = *ppos;
856 nccbs = count / sizeof(struct dax_ccb);
857
858 if (ctx->owner != current) {
859 dax_dbg("wrong thread");
860 ctx->result.exec.status = DAX_SUBMIT_ERR_THR_INIT;
861 return 0;
862 }
863 dax_dbg("args: ccb_buf_len=%ld, idx=%d", count, idx);
864
865 /* for given index and length, verify ca_buf range exists */
866 if (idx < 0 || idx > (DAX_CA_ELEMS - nccbs)) {
867 ctx->result.exec.status = DAX_SUBMIT_ERR_NO_CA_AVAIL;
868 return 0;
869 }
870
871 /*
872 * Copy CCBs into kernel buffer to prevent modification by the
873 * user in between validation and submission.
874 */
875 if (copy_from_user(ctx->ccb_buf, buf, count)) {
876 dax_dbg("copyin of user CCB buffer failed");
877 ctx->result.exec.status = DAX_SUBMIT_ERR_CCB_ARR_MMU_MISS;
878 return 0;
879 }
880
881 /* check to see if ca_buf[idx] .. ca_buf[idx + nccbs] are available */
882 for (i = idx; i < idx + nccbs; i++) {
883 if (ctx->ca_buf[i].status == CCA_STAT_NOT_COMPLETED) {
884 dax_dbg("CA range not available, dequeue needed");
885 ctx->result.exec.status = DAX_SUBMIT_ERR_NO_CA_AVAIL;
886 return 0;
887 }
888 }
889 dax_unlock_pages(ctx, idx, nccbs);
890
891 ctx->result.exec.status = dax_preprocess_usr_ccbs(ctx, idx, nccbs);
892 if (ctx->result.exec.status != DAX_SUBMIT_OK)
893 return 0;
894
895 ctx->result.exec.status = dax_lock_pages(ctx, idx, nccbs,
896 &ctx->result.exec.status_data);
897 if (ctx->result.exec.status != DAX_SUBMIT_OK)
898 return 0;
899
900 if (dax_debug & DAX_DBG_FLG_BASIC)
901 dax_prt_ccbs(ctx->ccb_buf, nccbs);
902
903 hv_rv = sun4v_ccb_submit(ctx->ccb_buf_ra, count,
904 HV_CCB_QUERY_CMD | HV_CCB_VA_SECONDARY, 0,
905 &accepted_len, &ctx->result.exec.status_data);
906
907 switch (hv_rv) {
908 case HV_EOK:
909 /*
910 * Hcall succeeded with no errors but the accepted
911 * length may be less than the requested length. The
912 * only way the driver can resubmit the remainder is
913 * to wait for completion of the submitted CCBs since
914 * there is no way to guarantee the ordering semantics
915 * required by the client applications. Therefore we
916 * let the user library deal with resubmissions.
917 */
918 ctx->result.exec.status = DAX_SUBMIT_OK;
919 break;
920 case HV_EWOULDBLOCK:
921 /*
922 * This is a transient HV API error. The user library
923 * can retry.
924 */
925 dax_dbg("hcall returned HV_EWOULDBLOCK");
926 ctx->result.exec.status = DAX_SUBMIT_ERR_WOULDBLOCK;
927 break;
928 case HV_ENOMAP:
929 /*
930 * HV was unable to translate a VA. The VA it could
931 * not translate is returned in the status_data param.
932 */
933 dax_dbg("hcall returned HV_ENOMAP");
934 ctx->result.exec.status = DAX_SUBMIT_ERR_NOMAP;
935 break;
936 case HV_EINVAL:
937 /*
938 * This is the result of an invalid user CCB as HV is
939 * validating some of the user CCB fields. Pass this
940 * error back to the user. There is no supporting info
941 * to isolate the invalid field.
942 */
943 dax_dbg("hcall returned HV_EINVAL");
944 ctx->result.exec.status = DAX_SUBMIT_ERR_CCB_INVAL;
945 break;
946 case HV_ENOACCESS:
947 /*
948 * HV found a VA that did not have the appropriate
949 * permissions (such as the w bit). The VA in question
950 * is returned in status_data param.
951 */
952 dax_dbg("hcall returned HV_ENOACCESS");
953 ctx->result.exec.status = DAX_SUBMIT_ERR_NOACCESS;
954 break;
955 case HV_EUNAVAILABLE:
956 /*
957 * The requested CCB operation could not be performed
958 * at this time. Return the specific unavailable code
959 * in the status_data field.
960 */
961 dax_dbg("hcall returned HV_EUNAVAILABLE");
962 ctx->result.exec.status = DAX_SUBMIT_ERR_UNAVAIL;
963 break;
964 default:
965 ctx->result.exec.status = DAX_SUBMIT_ERR_INTERNAL;
966 dax_dbg("unknown hcall return value (%ld)", hv_rv);
967 break;
968 }
969
970 /* unlock pages associated with the unaccepted CCBs */
971 naccepted = accepted_len / sizeof(struct dax_ccb);
972 dax_unlock_pages(ctx, idx + naccepted, nccbs - naccepted);
973
974 /* mark unaccepted CCBs as not completed */
975 for (i = idx + naccepted; i < idx + nccbs; i++)
976 ctx->ca_buf[i].status = CCA_STAT_COMPLETED;
977
978 ctx->ccb_count += naccepted;
979 ctx->fail_count += nccbs - naccepted;
980
981 dax_dbg("hcall rv=%ld, accepted_len=%ld, status_data=0x%llx, ret status=%d",
982 hv_rv, accepted_len, ctx->result.exec.status_data,
983 ctx->result.exec.status);
984
985 if (count == accepted_len)
986 ctx->client = NULL; /* no read needed to complete protocol */
987 return accepted_len;
988 }
989