xref: /openbmc/linux/drivers/sbus/char/oradax.c (revision 234489ac)
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 struct class *cl;
230 static dev_t first;
231 
232 static int max_ccb_version;
233 static int dax_debug;
234 module_param(dax_debug, int, 0644);
235 MODULE_PARM_DESC(dax_debug, "Debug flags");
236 
237 static int __init dax_attach(void)
238 {
239 	unsigned long dummy, hv_rv, major, minor, minor_requested, max_ccbs;
240 	struct mdesc_handle *hp = mdesc_grab();
241 	char *prop, *dax_name;
242 	bool found = false;
243 	int len, ret = 0;
244 	u64 pn;
245 
246 	if (hp == NULL) {
247 		dax_err("Unable to grab mdesc");
248 		return -ENODEV;
249 	}
250 
251 	mdesc_for_each_node_by_name(hp, pn, "virtual-device") {
252 		prop = (char *)mdesc_get_property(hp, pn, "name", &len);
253 		if (prop == NULL)
254 			continue;
255 		if (strncmp(prop, "dax", strlen("dax")))
256 			continue;
257 		dax_dbg("Found node 0x%llx = %s", pn, prop);
258 
259 		prop = (char *)mdesc_get_property(hp, pn, "compatible", &len);
260 		if (prop == NULL)
261 			continue;
262 		dax_dbg("Found node 0x%llx = %s", pn, prop);
263 		found = true;
264 		break;
265 	}
266 
267 	if (!found) {
268 		dax_err("No DAX device found");
269 		ret = -ENODEV;
270 		goto done;
271 	}
272 
273 	if (strncmp(prop, DAX2_STR, strlen(DAX2_STR)) == 0) {
274 		dax_name = DAX_NAME "2";
275 		major = DAX2_MAJOR;
276 		minor_requested = DAX2_MINOR;
277 		max_ccb_version = 1;
278 		dax_dbg("MD indicates DAX2 coprocessor");
279 	} else if (strncmp(prop, DAX1_STR, strlen(DAX1_STR)) == 0) {
280 		dax_name = DAX_NAME "1";
281 		major = DAX1_MAJOR;
282 		minor_requested = DAX1_MINOR;
283 		max_ccb_version = 0;
284 		dax_dbg("MD indicates DAX1 coprocessor");
285 	} else {
286 		dax_err("Unknown dax type: %s", prop);
287 		ret = -ENODEV;
288 		goto done;
289 	}
290 
291 	minor = minor_requested;
292 	dax_dbg("Registering DAX HV api with major %ld minor %ld", major,
293 		minor);
294 	if (sun4v_hvapi_register(HV_GRP_DAX, major, &minor)) {
295 		dax_err("hvapi_register failed");
296 		ret = -ENODEV;
297 		goto done;
298 	} else {
299 		dax_dbg("Max minor supported by HV = %ld (major %ld)", minor,
300 			major);
301 		minor = min(minor, minor_requested);
302 		dax_dbg("registered DAX major %ld minor %ld", major, minor);
303 	}
304 
305 	/* submit a zero length ccb array to query coprocessor queue size */
306 	hv_rv = sun4v_ccb_submit(0, 0, HV_CCB_QUERY_CMD, 0, &max_ccbs, &dummy);
307 	if (hv_rv != 0) {
308 		dax_err("get_hwqueue_size failed with status=%ld and max_ccbs=%ld",
309 			hv_rv, max_ccbs);
310 		ret = -ENODEV;
311 		goto done;
312 	}
313 
314 	if (max_ccbs != DAX_MAX_CCBS) {
315 		dax_err("HV reports unsupported max_ccbs=%ld", max_ccbs);
316 		ret = -ENODEV;
317 		goto done;
318 	}
319 
320 	if (alloc_chrdev_region(&first, 0, 1, DAX_NAME) < 0) {
321 		dax_err("alloc_chrdev_region failed");
322 		ret = -ENXIO;
323 		goto done;
324 	}
325 
326 	cl = class_create(DAX_NAME);
327 	if (IS_ERR(cl)) {
328 		dax_err("class_create failed");
329 		ret = PTR_ERR(cl);
330 		goto class_error;
331 	}
332 
333 	if (device_create(cl, NULL, first, NULL, dax_name) == NULL) {
334 		dax_err("device_create failed");
335 		ret = -ENXIO;
336 		goto device_error;
337 	}
338 
339 	cdev_init(&c_dev, &dax_fops);
340 	if (cdev_add(&c_dev, first, 1) == -1) {
341 		dax_err("cdev_add failed");
342 		ret = -ENXIO;
343 		goto cdev_error;
344 	}
345 
346 	pr_info("Attached DAX module\n");
347 	goto done;
348 
349 cdev_error:
350 	device_destroy(cl, first);
351 device_error:
352 	class_destroy(cl);
353 class_error:
354 	unregister_chrdev_region(first, 1);
355 done:
356 	mdesc_release(hp);
357 	return ret;
358 }
359 module_init(dax_attach);
360 
361 static void __exit dax_detach(void)
362 {
363 	pr_info("Cleaning up DAX module\n");
364 	cdev_del(&c_dev);
365 	device_destroy(cl, first);
366 	class_destroy(cl);
367 	unregister_chrdev_region(first, 1);
368 }
369 module_exit(dax_detach);
370 
371 /* map completion area */
372 static int dax_devmap(struct file *f, struct vm_area_struct *vma)
373 {
374 	struct dax_ctx *ctx = (struct dax_ctx *)f->private_data;
375 	size_t len = vma->vm_end - vma->vm_start;
376 
377 	dax_dbg("len=0x%lx, flags=0x%lx", len, vma->vm_flags);
378 
379 	if (ctx->owner != current) {
380 		dax_dbg("devmap called from wrong thread");
381 		return -EINVAL;
382 	}
383 
384 	if (len != DAX_MMAP_LEN) {
385 		dax_dbg("len(%lu) != DAX_MMAP_LEN(%d)", len, DAX_MMAP_LEN);
386 		return -EINVAL;
387 	}
388 
389 	/* completion area is mapped read-only for user */
390 	if (vma->vm_flags & VM_WRITE)
391 		return -EPERM;
392 	vm_flags_clear(vma, VM_MAYWRITE);
393 
394 	if (remap_pfn_range(vma, vma->vm_start, ctx->ca_buf_ra >> PAGE_SHIFT,
395 			    len, vma->vm_page_prot))
396 		return -EAGAIN;
397 
398 	dax_dbg("mmapped completion area at uva 0x%lx", vma->vm_start);
399 	return 0;
400 }
401 
402 /* Unlock user pages. Called during dequeue or device close */
403 static void dax_unlock_pages(struct dax_ctx *ctx, int ccb_index, int nelem)
404 {
405 	int i, j;
406 
407 	for (i = ccb_index; i < ccb_index + nelem; i++) {
408 		for (j = 0; j < NUM_STREAM_TYPES; j++) {
409 			struct page *p = ctx->pages[i][j];
410 
411 			if (p) {
412 				dax_dbg("freeing page %p", p);
413 				unpin_user_pages_dirty_lock(&p, 1, j == OUT);
414 				ctx->pages[i][j] = NULL;
415 			}
416 		}
417 	}
418 }
419 
420 static int dax_lock_page(void *va, struct page **p)
421 {
422 	int ret;
423 
424 	dax_dbg("uva %p", va);
425 
426 	ret = pin_user_pages_fast((unsigned long)va, 1, FOLL_WRITE, p);
427 	if (ret == 1) {
428 		dax_dbg("locked page %p, for VA %p", *p, va);
429 		return 0;
430 	}
431 
432 	dax_dbg("pin_user_pages failed, va=%p, ret=%d", va, ret);
433 	return -1;
434 }
435 
436 static int dax_lock_pages(struct dax_ctx *ctx, int idx,
437 			  int nelem, u64 *err_va)
438 {
439 	int i;
440 
441 	for (i = 0; i < nelem; i++) {
442 		struct dax_ccb *ccbp = &ctx->ccb_buf[i];
443 
444 		/*
445 		 * For each address in the CCB whose type is virtual,
446 		 * lock the page and change the type to virtual alternate
447 		 * context. On error, return the offending address in
448 		 * err_va.
449 		 */
450 		if (ccbp->hdr.out_addr_type == DAX_ADDR_TYPE_VA) {
451 			dax_dbg("output");
452 			if (dax_lock_page(ccbp->out,
453 					  &ctx->pages[i + idx][OUT]) != 0) {
454 				*err_va = (u64)ccbp->out;
455 				goto error;
456 			}
457 			ccbp->hdr.out_addr_type = DAX_ADDR_TYPE_VA_ALT;
458 		}
459 
460 		if (ccbp->hdr.pri_addr_type == DAX_ADDR_TYPE_VA) {
461 			dax_dbg("input");
462 			if (dax_lock_page(ccbp->pri,
463 					  &ctx->pages[i + idx][PRI]) != 0) {
464 				*err_va = (u64)ccbp->pri;
465 				goto error;
466 			}
467 			ccbp->hdr.pri_addr_type = DAX_ADDR_TYPE_VA_ALT;
468 		}
469 
470 		if (ccbp->hdr.sec_addr_type == DAX_ADDR_TYPE_VA) {
471 			dax_dbg("sec input");
472 			if (dax_lock_page(ccbp->sec,
473 					  &ctx->pages[i + idx][SEC]) != 0) {
474 				*err_va = (u64)ccbp->sec;
475 				goto error;
476 			}
477 			ccbp->hdr.sec_addr_type = DAX_ADDR_TYPE_VA_ALT;
478 		}
479 
480 		if (ccbp->hdr.table_addr_type == DAX_ADDR_TYPE_VA) {
481 			dax_dbg("tbl");
482 			if (dax_lock_page(ccbp->tbl,
483 					  &ctx->pages[i + idx][TBL]) != 0) {
484 				*err_va = (u64)ccbp->tbl;
485 				goto error;
486 			}
487 			ccbp->hdr.table_addr_type = DAX_ADDR_TYPE_VA_ALT;
488 		}
489 
490 		/* skip over 2nd 64 bytes of long CCB */
491 		if (ccbp->hdr.longccb)
492 			i++;
493 	}
494 	return DAX_SUBMIT_OK;
495 
496 error:
497 	dax_unlock_pages(ctx, idx, nelem);
498 	return DAX_SUBMIT_ERR_NOACCESS;
499 }
500 
501 static void dax_ccb_wait(struct dax_ctx *ctx, int idx)
502 {
503 	int ret, nretries;
504 	u16 kill_res;
505 
506 	dax_dbg("idx=%d", idx);
507 
508 	for (nretries = 0; nretries < DAX_CCB_RETRIES; nretries++) {
509 		if (ctx->ca_buf[idx].status == CCA_STAT_NOT_COMPLETED)
510 			udelay(DAX_CCB_USEC);
511 		else
512 			return;
513 	}
514 	dax_dbg("ctx (%p): CCB[%d] timed out, wait usec=%d, retries=%d. Killing ccb",
515 		(void *)ctx, idx, DAX_CCB_USEC, DAX_CCB_RETRIES);
516 
517 	ret = dax_ccb_kill(ctx->ca_buf_ra + idx * sizeof(struct dax_cca),
518 			   &kill_res);
519 	dax_dbg("Kill CCB[%d] %s", idx, ret ? "failed" : "succeeded");
520 }
521 
522 static int dax_close(struct inode *ino, struct file *f)
523 {
524 	struct dax_ctx *ctx = (struct dax_ctx *)f->private_data;
525 	int i;
526 
527 	f->private_data = NULL;
528 
529 	for (i = 0; i < DAX_CA_ELEMS; i++) {
530 		if (ctx->ca_buf[i].status == CCA_STAT_NOT_COMPLETED) {
531 			dax_dbg("CCB[%d] not completed", i);
532 			dax_ccb_wait(ctx, i);
533 		}
534 		dax_unlock_pages(ctx, i, 1);
535 	}
536 
537 	kfree(ctx->ccb_buf);
538 	kfree(ctx->ca_buf);
539 	dax_stat_dbg("CCBs: %d good, %d bad", ctx->ccb_count, ctx->fail_count);
540 	kfree(ctx);
541 
542 	return 0;
543 }
544 
545 static ssize_t dax_read(struct file *f, char __user *buf,
546 			size_t count, loff_t *ppos)
547 {
548 	struct dax_ctx *ctx = f->private_data;
549 
550 	if (ctx->client != current)
551 		return -EUSERS;
552 
553 	ctx->client = NULL;
554 
555 	if (count != sizeof(union ccb_result))
556 		return -EINVAL;
557 	if (copy_to_user(buf, &ctx->result, sizeof(union ccb_result)))
558 		return -EFAULT;
559 	return count;
560 }
561 
562 static ssize_t dax_write(struct file *f, const char __user *buf,
563 			 size_t count, loff_t *ppos)
564 {
565 	struct dax_ctx *ctx = f->private_data;
566 	struct dax_command hdr;
567 	unsigned long ca;
568 	int i, idx, ret;
569 
570 	if (ctx->client != NULL)
571 		return -EINVAL;
572 
573 	if (count == 0 || count > DAX_MAX_CCBS * sizeof(struct dax_ccb))
574 		return -EINVAL;
575 
576 	if (count % sizeof(struct dax_ccb) == 0)
577 		return dax_ccb_exec(ctx, buf, count, ppos); /* CCB EXEC */
578 
579 	if (count != sizeof(struct dax_command))
580 		return -EINVAL;
581 
582 	/* immediate command */
583 	if (ctx->owner != current)
584 		return -EUSERS;
585 
586 	if (copy_from_user(&hdr, buf, sizeof(hdr)))
587 		return -EFAULT;
588 
589 	ca = ctx->ca_buf_ra + hdr.ca_offset;
590 
591 	switch (hdr.command) {
592 	case CCB_KILL:
593 		if (hdr.ca_offset >= DAX_MMAP_LEN) {
594 			dax_dbg("invalid ca_offset (%d) >= ca_buflen (%d)",
595 				hdr.ca_offset, DAX_MMAP_LEN);
596 			return -EINVAL;
597 		}
598 
599 		ret = dax_ccb_kill(ca, &ctx->result.kill.action);
600 		if (ret != 0) {
601 			dax_dbg("dax_ccb_kill failed (ret=%d)", ret);
602 			return ret;
603 		}
604 
605 		dax_info_dbg("killed (ca_offset %d)", hdr.ca_offset);
606 		idx = hdr.ca_offset / sizeof(struct dax_cca);
607 		ctx->ca_buf[idx].status = CCA_STAT_KILLED;
608 		ctx->ca_buf[idx].err = CCA_ERR_KILLED;
609 		ctx->client = current;
610 		return count;
611 
612 	case CCB_INFO:
613 		if (hdr.ca_offset >= DAX_MMAP_LEN) {
614 			dax_dbg("invalid ca_offset (%d) >= ca_buflen (%d)",
615 				hdr.ca_offset, DAX_MMAP_LEN);
616 			return -EINVAL;
617 		}
618 
619 		ret = dax_ccb_info(ca, &ctx->result.info);
620 		if (ret != 0) {
621 			dax_dbg("dax_ccb_info failed (ret=%d)", ret);
622 			return ret;
623 		}
624 
625 		dax_info_dbg("info succeeded on ca_offset %d", hdr.ca_offset);
626 		ctx->client = current;
627 		return count;
628 
629 	case CCB_DEQUEUE:
630 		for (i = 0; i < DAX_CA_ELEMS; i++) {
631 			if (ctx->ca_buf[i].status !=
632 			    CCA_STAT_NOT_COMPLETED)
633 				dax_unlock_pages(ctx, i, 1);
634 		}
635 		return count;
636 
637 	default:
638 		return -EINVAL;
639 	}
640 }
641 
642 static int dax_open(struct inode *inode, struct file *f)
643 {
644 	struct dax_ctx *ctx = NULL;
645 	int i;
646 
647 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
648 	if (ctx == NULL)
649 		goto done;
650 
651 	ctx->ccb_buf = kcalloc(DAX_MAX_CCBS, sizeof(struct dax_ccb),
652 			       GFP_KERNEL);
653 	if (ctx->ccb_buf == NULL)
654 		goto done;
655 
656 	ctx->ccb_buf_ra = virt_to_phys(ctx->ccb_buf);
657 	dax_dbg("ctx->ccb_buf=0x%p, ccb_buf_ra=0x%llx",
658 		(void *)ctx->ccb_buf, ctx->ccb_buf_ra);
659 
660 	/* allocate CCB completion area buffer */
661 	ctx->ca_buf = kzalloc(DAX_MMAP_LEN, GFP_KERNEL);
662 	if (ctx->ca_buf == NULL)
663 		goto alloc_error;
664 	for (i = 0; i < DAX_CA_ELEMS; i++)
665 		ctx->ca_buf[i].status = CCA_STAT_COMPLETED;
666 
667 	ctx->ca_buf_ra = virt_to_phys(ctx->ca_buf);
668 	dax_dbg("ctx=0x%p, ctx->ca_buf=0x%p, ca_buf_ra=0x%llx",
669 		(void *)ctx, (void *)ctx->ca_buf, ctx->ca_buf_ra);
670 
671 	ctx->owner = current;
672 	f->private_data = ctx;
673 	return 0;
674 
675 alloc_error:
676 	kfree(ctx->ccb_buf);
677 done:
678 	kfree(ctx);
679 	return -ENOMEM;
680 }
681 
682 static char *dax_hv_errno(unsigned long hv_ret, int *ret)
683 {
684 	switch (hv_ret) {
685 	case HV_EBADALIGN:
686 		*ret = -EFAULT;
687 		return "HV_EBADALIGN";
688 	case HV_ENORADDR:
689 		*ret = -EFAULT;
690 		return "HV_ENORADDR";
691 	case HV_EINVAL:
692 		*ret = -EINVAL;
693 		return "HV_EINVAL";
694 	case HV_EWOULDBLOCK:
695 		*ret = -EAGAIN;
696 		return "HV_EWOULDBLOCK";
697 	case HV_ENOACCESS:
698 		*ret = -EPERM;
699 		return "HV_ENOACCESS";
700 	default:
701 		break;
702 	}
703 
704 	*ret = -EIO;
705 	return "UNKNOWN";
706 }
707 
708 static int dax_ccb_kill(u64 ca, u16 *kill_res)
709 {
710 	unsigned long hv_ret;
711 	int count, ret = 0;
712 	char *err_str;
713 
714 	for (count = 0; count < DAX_CCB_RETRIES; count++) {
715 		dax_dbg("attempting kill on ca_ra 0x%llx", ca);
716 		hv_ret = sun4v_ccb_kill(ca, kill_res);
717 
718 		if (hv_ret == HV_EOK) {
719 			dax_info_dbg("HV_EOK (ca_ra 0x%llx): %d", ca,
720 				     *kill_res);
721 		} else {
722 			err_str = dax_hv_errno(hv_ret, &ret);
723 			dax_dbg("%s (ca_ra 0x%llx)", err_str, ca);
724 		}
725 
726 		if (ret != -EAGAIN)
727 			return ret;
728 		dax_info_dbg("ccb_kill count = %d", count);
729 		udelay(DAX_CCB_USEC);
730 	}
731 
732 	return -EAGAIN;
733 }
734 
735 static int dax_ccb_info(u64 ca, struct ccb_info_result *info)
736 {
737 	unsigned long hv_ret;
738 	char *err_str;
739 	int ret = 0;
740 
741 	dax_dbg("attempting info on ca_ra 0x%llx", ca);
742 	hv_ret = sun4v_ccb_info(ca, info);
743 
744 	if (hv_ret == HV_EOK) {
745 		dax_info_dbg("HV_EOK (ca_ra 0x%llx): %d", ca, info->state);
746 		if (info->state == DAX_CCB_ENQUEUED) {
747 			dax_info_dbg("dax_unit %d, queue_num %d, queue_pos %d",
748 				     info->inst_num, info->q_num, info->q_pos);
749 		}
750 	} else {
751 		err_str = dax_hv_errno(hv_ret, &ret);
752 		dax_dbg("%s (ca_ra 0x%llx)", err_str, ca);
753 	}
754 
755 	return ret;
756 }
757 
758 static void dax_prt_ccbs(struct dax_ccb *ccb, int nelem)
759 {
760 	int i, j;
761 	u64 *ccbp;
762 
763 	dax_dbg("ccb buffer:");
764 	for (i = 0; i < nelem; i++) {
765 		ccbp = (u64 *)&ccb[i];
766 		dax_dbg(" %sccb[%d]", ccb[i].hdr.longccb ? "long " : "",  i);
767 		for (j = 0; j < 8; j++)
768 			dax_dbg("\tccb[%d].dwords[%d]=0x%llx",
769 				i, j, *(ccbp + j));
770 	}
771 }
772 
773 /*
774  * Validates user CCB content.  Also sets completion address and address types
775  * for all addresses contained in CCB.
776  */
777 static int dax_preprocess_usr_ccbs(struct dax_ctx *ctx, int idx, int nelem)
778 {
779 	int i;
780 
781 	/*
782 	 * The user is not allowed to specify real address types in
783 	 * the CCB header.  This must be enforced by the kernel before
784 	 * submitting the CCBs to HV.  The only allowed values for all
785 	 * address fields are VA or IMM
786 	 */
787 	for (i = 0; i < nelem; i++) {
788 		struct dax_ccb *ccbp = &ctx->ccb_buf[i];
789 		unsigned long ca_offset;
790 
791 		if (ccbp->hdr.ccb_version > max_ccb_version)
792 			return DAX_SUBMIT_ERR_CCB_INVAL;
793 
794 		switch (ccbp->hdr.opcode) {
795 		case DAX_OP_SYNC_NOP:
796 		case DAX_OP_EXTRACT:
797 		case DAX_OP_SCAN_VALUE:
798 		case DAX_OP_SCAN_RANGE:
799 		case DAX_OP_TRANSLATE:
800 		case DAX_OP_SCAN_VALUE | DAX_OP_INVERT:
801 		case DAX_OP_SCAN_RANGE | DAX_OP_INVERT:
802 		case DAX_OP_TRANSLATE | DAX_OP_INVERT:
803 		case DAX_OP_SELECT:
804 			break;
805 		default:
806 			return DAX_SUBMIT_ERR_CCB_INVAL;
807 		}
808 
809 		if (ccbp->hdr.out_addr_type != DAX_ADDR_TYPE_VA &&
810 		    ccbp->hdr.out_addr_type != DAX_ADDR_TYPE_NONE) {
811 			dax_dbg("invalid out_addr_type in user CCB[%d]", i);
812 			return DAX_SUBMIT_ERR_CCB_INVAL;
813 		}
814 
815 		if (ccbp->hdr.pri_addr_type != DAX_ADDR_TYPE_VA &&
816 		    ccbp->hdr.pri_addr_type != DAX_ADDR_TYPE_NONE) {
817 			dax_dbg("invalid pri_addr_type in user CCB[%d]", i);
818 			return DAX_SUBMIT_ERR_CCB_INVAL;
819 		}
820 
821 		if (ccbp->hdr.sec_addr_type != DAX_ADDR_TYPE_VA &&
822 		    ccbp->hdr.sec_addr_type != DAX_ADDR_TYPE_NONE) {
823 			dax_dbg("invalid sec_addr_type in user CCB[%d]", i);
824 			return DAX_SUBMIT_ERR_CCB_INVAL;
825 		}
826 
827 		if (ccbp->hdr.table_addr_type != DAX_ADDR_TYPE_VA &&
828 		    ccbp->hdr.table_addr_type != DAX_ADDR_TYPE_NONE) {
829 			dax_dbg("invalid table_addr_type in user CCB[%d]", i);
830 			return DAX_SUBMIT_ERR_CCB_INVAL;
831 		}
832 
833 		/* set completion (real) address and address type */
834 		ccbp->hdr.cca_addr_type = DAX_ADDR_TYPE_RA;
835 		ca_offset = (idx + i) * sizeof(struct dax_cca);
836 		ccbp->ca = (void *)ctx->ca_buf_ra + ca_offset;
837 		memset(&ctx->ca_buf[idx + i], 0, sizeof(struct dax_cca));
838 
839 		dax_dbg("ccb[%d]=%p, ca_offset=0x%lx, compl RA=0x%llx",
840 			i, ccbp, ca_offset, ctx->ca_buf_ra + ca_offset);
841 
842 		/* skip over 2nd 64 bytes of long CCB */
843 		if (ccbp->hdr.longccb)
844 			i++;
845 	}
846 
847 	return DAX_SUBMIT_OK;
848 }
849 
850 static int dax_ccb_exec(struct dax_ctx *ctx, const char __user *buf,
851 			size_t count, loff_t *ppos)
852 {
853 	unsigned long accepted_len, hv_rv;
854 	int i, idx, nccbs, naccepted;
855 
856 	ctx->client = current;
857 	idx = *ppos;
858 	nccbs = count / sizeof(struct dax_ccb);
859 
860 	if (ctx->owner != current) {
861 		dax_dbg("wrong thread");
862 		ctx->result.exec.status = DAX_SUBMIT_ERR_THR_INIT;
863 		return 0;
864 	}
865 	dax_dbg("args: ccb_buf_len=%ld, idx=%d", count, idx);
866 
867 	/* for given index and length, verify ca_buf range exists */
868 	if (idx < 0 || idx > (DAX_CA_ELEMS - nccbs)) {
869 		ctx->result.exec.status = DAX_SUBMIT_ERR_NO_CA_AVAIL;
870 		return 0;
871 	}
872 
873 	/*
874 	 * Copy CCBs into kernel buffer to prevent modification by the
875 	 * user in between validation and submission.
876 	 */
877 	if (copy_from_user(ctx->ccb_buf, buf, count)) {
878 		dax_dbg("copyin of user CCB buffer failed");
879 		ctx->result.exec.status = DAX_SUBMIT_ERR_CCB_ARR_MMU_MISS;
880 		return 0;
881 	}
882 
883 	/* check to see if ca_buf[idx] .. ca_buf[idx + nccbs] are available */
884 	for (i = idx; i < idx + nccbs; i++) {
885 		if (ctx->ca_buf[i].status == CCA_STAT_NOT_COMPLETED) {
886 			dax_dbg("CA range not available, dequeue needed");
887 			ctx->result.exec.status = DAX_SUBMIT_ERR_NO_CA_AVAIL;
888 			return 0;
889 		}
890 	}
891 	dax_unlock_pages(ctx, idx, nccbs);
892 
893 	ctx->result.exec.status = dax_preprocess_usr_ccbs(ctx, idx, nccbs);
894 	if (ctx->result.exec.status != DAX_SUBMIT_OK)
895 		return 0;
896 
897 	ctx->result.exec.status = dax_lock_pages(ctx, idx, nccbs,
898 						 &ctx->result.exec.status_data);
899 	if (ctx->result.exec.status != DAX_SUBMIT_OK)
900 		return 0;
901 
902 	if (dax_debug & DAX_DBG_FLG_BASIC)
903 		dax_prt_ccbs(ctx->ccb_buf, nccbs);
904 
905 	hv_rv = sun4v_ccb_submit(ctx->ccb_buf_ra, count,
906 				 HV_CCB_QUERY_CMD | HV_CCB_VA_SECONDARY, 0,
907 				 &accepted_len, &ctx->result.exec.status_data);
908 
909 	switch (hv_rv) {
910 	case HV_EOK:
911 		/*
912 		 * Hcall succeeded with no errors but the accepted
913 		 * length may be less than the requested length.  The
914 		 * only way the driver can resubmit the remainder is
915 		 * to wait for completion of the submitted CCBs since
916 		 * there is no way to guarantee the ordering semantics
917 		 * required by the client applications.  Therefore we
918 		 * let the user library deal with resubmissions.
919 		 */
920 		ctx->result.exec.status = DAX_SUBMIT_OK;
921 		break;
922 	case HV_EWOULDBLOCK:
923 		/*
924 		 * This is a transient HV API error. The user library
925 		 * can retry.
926 		 */
927 		dax_dbg("hcall returned HV_EWOULDBLOCK");
928 		ctx->result.exec.status = DAX_SUBMIT_ERR_WOULDBLOCK;
929 		break;
930 	case HV_ENOMAP:
931 		/*
932 		 * HV was unable to translate a VA. The VA it could
933 		 * not translate is returned in the status_data param.
934 		 */
935 		dax_dbg("hcall returned HV_ENOMAP");
936 		ctx->result.exec.status = DAX_SUBMIT_ERR_NOMAP;
937 		break;
938 	case HV_EINVAL:
939 		/*
940 		 * This is the result of an invalid user CCB as HV is
941 		 * validating some of the user CCB fields.  Pass this
942 		 * error back to the user. There is no supporting info
943 		 * to isolate the invalid field.
944 		 */
945 		dax_dbg("hcall returned HV_EINVAL");
946 		ctx->result.exec.status = DAX_SUBMIT_ERR_CCB_INVAL;
947 		break;
948 	case HV_ENOACCESS:
949 		/*
950 		 * HV found a VA that did not have the appropriate
951 		 * permissions (such as the w bit). The VA in question
952 		 * is returned in status_data param.
953 		 */
954 		dax_dbg("hcall returned HV_ENOACCESS");
955 		ctx->result.exec.status = DAX_SUBMIT_ERR_NOACCESS;
956 		break;
957 	case HV_EUNAVAILABLE:
958 		/*
959 		 * The requested CCB operation could not be performed
960 		 * at this time. Return the specific unavailable code
961 		 * in the status_data field.
962 		 */
963 		dax_dbg("hcall returned HV_EUNAVAILABLE");
964 		ctx->result.exec.status = DAX_SUBMIT_ERR_UNAVAIL;
965 		break;
966 	default:
967 		ctx->result.exec.status = DAX_SUBMIT_ERR_INTERNAL;
968 		dax_dbg("unknown hcall return value (%ld)", hv_rv);
969 		break;
970 	}
971 
972 	/* unlock pages associated with the unaccepted CCBs */
973 	naccepted = accepted_len / sizeof(struct dax_ccb);
974 	dax_unlock_pages(ctx, idx + naccepted, nccbs - naccepted);
975 
976 	/* mark unaccepted CCBs as not completed */
977 	for (i = idx + naccepted; i < idx + nccbs; i++)
978 		ctx->ca_buf[i].status = CCA_STAT_COMPLETED;
979 
980 	ctx->ccb_count += naccepted;
981 	ctx->fail_count += nccbs - naccepted;
982 
983 	dax_dbg("hcall rv=%ld, accepted_len=%ld, status_data=0x%llx, ret status=%d",
984 		hv_rv, accepted_len, ctx->result.exec.status_data,
985 		ctx->result.exec.status);
986 
987 	if (count == accepted_len)
988 		ctx->client = NULL; /* no read needed to complete protocol */
989 	return accepted_len;
990 }
991