Merge tag 'for-linus' of git://github.com/openrisc/linux
[linux-2.6-microblaze.git] / drivers / dma / dmatest.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * DMA Engine test module
4  *
5  * Copyright (C) 2007 Atmel Corporation
6  * Copyright (C) 2013 Intel Corporation
7  */
8 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
9
10 #include <linux/delay.h>
11 #include <linux/dma-mapping.h>
12 #include <linux/dmaengine.h>
13 #include <linux/freezer.h>
14 #include <linux/init.h>
15 #include <linux/kthread.h>
16 #include <linux/sched/task.h>
17 #include <linux/module.h>
18 #include <linux/moduleparam.h>
19 #include <linux/random.h>
20 #include <linux/slab.h>
21 #include <linux/wait.h>
22
23 static unsigned int test_buf_size = 16384;
24 module_param(test_buf_size, uint, S_IRUGO | S_IWUSR);
25 MODULE_PARM_DESC(test_buf_size, "Size of the memcpy test buffer");
26
27 static char test_device[32];
28 module_param_string(device, test_device, sizeof(test_device),
29                 S_IRUGO | S_IWUSR);
30 MODULE_PARM_DESC(device, "Bus ID of the DMA Engine to test (default: any)");
31
32 static unsigned int threads_per_chan = 1;
33 module_param(threads_per_chan, uint, S_IRUGO | S_IWUSR);
34 MODULE_PARM_DESC(threads_per_chan,
35                 "Number of threads to start per channel (default: 1)");
36
37 static unsigned int max_channels;
38 module_param(max_channels, uint, S_IRUGO | S_IWUSR);
39 MODULE_PARM_DESC(max_channels,
40                 "Maximum number of channels to use (default: all)");
41
42 static unsigned int iterations;
43 module_param(iterations, uint, S_IRUGO | S_IWUSR);
44 MODULE_PARM_DESC(iterations,
45                 "Iterations before stopping test (default: infinite)");
46
47 static unsigned int dmatest;
48 module_param(dmatest, uint, S_IRUGO | S_IWUSR);
49 MODULE_PARM_DESC(dmatest,
50                 "dmatest 0-memcpy 1-memset (default: 0)");
51
52 static unsigned int xor_sources = 3;
53 module_param(xor_sources, uint, S_IRUGO | S_IWUSR);
54 MODULE_PARM_DESC(xor_sources,
55                 "Number of xor source buffers (default: 3)");
56
57 static unsigned int pq_sources = 3;
58 module_param(pq_sources, uint, S_IRUGO | S_IWUSR);
59 MODULE_PARM_DESC(pq_sources,
60                 "Number of p+q source buffers (default: 3)");
61
62 static int timeout = 3000;
63 module_param(timeout, int, S_IRUGO | S_IWUSR);
64 MODULE_PARM_DESC(timeout, "Transfer Timeout in msec (default: 3000), "
65                  "Pass -1 for infinite timeout");
66
67 static bool noverify;
68 module_param(noverify, bool, S_IRUGO | S_IWUSR);
69 MODULE_PARM_DESC(noverify, "Disable data verification (default: verify)");
70
71 static bool norandom;
72 module_param(norandom, bool, 0644);
73 MODULE_PARM_DESC(norandom, "Disable random offset setup (default: random)");
74
75 static bool verbose;
76 module_param(verbose, bool, S_IRUGO | S_IWUSR);
77 MODULE_PARM_DESC(verbose, "Enable \"success\" result messages (default: off)");
78
79 static int alignment = -1;
80 module_param(alignment, int, 0644);
81 MODULE_PARM_DESC(alignment, "Custom data address alignment taken as 2^(alignment) (default: not used (-1))");
82
83 static unsigned int transfer_size;
84 module_param(transfer_size, uint, 0644);
85 MODULE_PARM_DESC(transfer_size, "Optional custom transfer size in bytes (default: not used (0))");
86
87 static bool polled;
88 module_param(polled, bool, S_IRUGO | S_IWUSR);
89 MODULE_PARM_DESC(polled, "Use polling for completion instead of interrupts");
90
91 /**
92  * struct dmatest_params - test parameters.
93  * @buf_size:           size of the memcpy test buffer
94  * @channel:            bus ID of the channel to test
95  * @device:             bus ID of the DMA Engine to test
96  * @threads_per_chan:   number of threads to start per channel
97  * @max_channels:       maximum number of channels to use
98  * @iterations:         iterations before stopping test
99  * @xor_sources:        number of xor source buffers
100  * @pq_sources:         number of p+q source buffers
101  * @timeout:            transfer timeout in msec, -1 for infinite timeout
102  * @noverify:           disable data verification
103  * @norandom:           disable random offset setup
104  * @alignment:          custom data address alignment taken as 2^alignment
105  * @transfer_size:      custom transfer size in bytes
106  * @polled:             use polling for completion instead of interrupts
107  */
108 struct dmatest_params {
109         unsigned int    buf_size;
110         char            channel[20];
111         char            device[32];
112         unsigned int    threads_per_chan;
113         unsigned int    max_channels;
114         unsigned int    iterations;
115         unsigned int    xor_sources;
116         unsigned int    pq_sources;
117         int             timeout;
118         bool            noverify;
119         bool            norandom;
120         int             alignment;
121         unsigned int    transfer_size;
122         bool            polled;
123 };
124
125 /**
126  * struct dmatest_info - test information.
127  * @params:             test parameters
128  * @channels:           channels under test
129  * @nr_channels:        number of channels under test
130  * @lock:               access protection to the fields of this structure
131  * @did_init:           module has been initialized completely
132  */
133 static struct dmatest_info {
134         /* Test parameters */
135         struct dmatest_params   params;
136
137         /* Internal state */
138         struct list_head        channels;
139         unsigned int            nr_channels;
140         struct mutex            lock;
141         bool                    did_init;
142 } test_info = {
143         .channels = LIST_HEAD_INIT(test_info.channels),
144         .lock = __MUTEX_INITIALIZER(test_info.lock),
145 };
146
147 static int dmatest_run_set(const char *val, const struct kernel_param *kp);
148 static int dmatest_run_get(char *val, const struct kernel_param *kp);
149 static const struct kernel_param_ops run_ops = {
150         .set = dmatest_run_set,
151         .get = dmatest_run_get,
152 };
153 static bool dmatest_run;
154 module_param_cb(run, &run_ops, &dmatest_run, S_IRUGO | S_IWUSR);
155 MODULE_PARM_DESC(run, "Run the test (default: false)");
156
157 static int dmatest_chan_set(const char *val, const struct kernel_param *kp);
158 static int dmatest_chan_get(char *val, const struct kernel_param *kp);
159 static const struct kernel_param_ops multi_chan_ops = {
160         .set = dmatest_chan_set,
161         .get = dmatest_chan_get,
162 };
163
164 static char test_channel[20];
165 static struct kparam_string newchan_kps = {
166         .string = test_channel,
167         .maxlen = 20,
168 };
169 module_param_cb(channel, &multi_chan_ops, &newchan_kps, 0644);
170 MODULE_PARM_DESC(channel, "Bus ID of the channel to test (default: any)");
171
172 static int dmatest_test_list_get(char *val, const struct kernel_param *kp);
173 static const struct kernel_param_ops test_list_ops = {
174         .get = dmatest_test_list_get,
175 };
176 module_param_cb(test_list, &test_list_ops, NULL, 0444);
177 MODULE_PARM_DESC(test_list, "Print current test list");
178
179 /* Maximum amount of mismatched bytes in buffer to print */
180 #define MAX_ERROR_COUNT         32
181
182 /*
183  * Initialization patterns. All bytes in the source buffer has bit 7
184  * set, all bytes in the destination buffer has bit 7 cleared.
185  *
186  * Bit 6 is set for all bytes which are to be copied by the DMA
187  * engine. Bit 5 is set for all bytes which are to be overwritten by
188  * the DMA engine.
189  *
190  * The remaining bits are the inverse of a counter which increments by
191  * one for each byte address.
192  */
193 #define PATTERN_SRC             0x80
194 #define PATTERN_DST             0x00
195 #define PATTERN_COPY            0x40
196 #define PATTERN_OVERWRITE       0x20
197 #define PATTERN_COUNT_MASK      0x1f
198 #define PATTERN_MEMSET_IDX      0x01
199
200 /* Fixed point arithmetic ops */
201 #define FIXPT_SHIFT             8
202 #define FIXPNT_MASK             0xFF
203 #define FIXPT_TO_INT(a) ((a) >> FIXPT_SHIFT)
204 #define INT_TO_FIXPT(a) ((a) << FIXPT_SHIFT)
205 #define FIXPT_GET_FRAC(a)       ((((a) & FIXPNT_MASK) * 100) >> FIXPT_SHIFT)
206
207 /* poor man's completion - we want to use wait_event_freezable() on it */
208 struct dmatest_done {
209         bool                    done;
210         wait_queue_head_t       *wait;
211 };
212
213 struct dmatest_data {
214         u8              **raw;
215         u8              **aligned;
216         unsigned int    cnt;
217         unsigned int    off;
218 };
219
220 struct dmatest_thread {
221         struct list_head        node;
222         struct dmatest_info     *info;
223         struct task_struct      *task;
224         struct dma_chan         *chan;
225         struct dmatest_data     src;
226         struct dmatest_data     dst;
227         enum dma_transaction_type type;
228         wait_queue_head_t done_wait;
229         struct dmatest_done test_done;
230         bool                    done;
231         bool                    pending;
232 };
233
234 struct dmatest_chan {
235         struct list_head        node;
236         struct dma_chan         *chan;
237         struct list_head        threads;
238 };
239
240 static DECLARE_WAIT_QUEUE_HEAD(thread_wait);
241 static bool wait;
242
243 static bool is_threaded_test_run(struct dmatest_info *info)
244 {
245         struct dmatest_chan *dtc;
246
247         list_for_each_entry(dtc, &info->channels, node) {
248                 struct dmatest_thread *thread;
249
250                 list_for_each_entry(thread, &dtc->threads, node) {
251                         if (!thread->done && !thread->pending)
252                                 return true;
253                 }
254         }
255
256         return false;
257 }
258
259 static bool is_threaded_test_pending(struct dmatest_info *info)
260 {
261         struct dmatest_chan *dtc;
262
263         list_for_each_entry(dtc, &info->channels, node) {
264                 struct dmatest_thread *thread;
265
266                 list_for_each_entry(thread, &dtc->threads, node) {
267                         if (thread->pending)
268                                 return true;
269                 }
270         }
271
272         return false;
273 }
274
275 static int dmatest_wait_get(char *val, const struct kernel_param *kp)
276 {
277         struct dmatest_info *info = &test_info;
278         struct dmatest_params *params = &info->params;
279
280         if (params->iterations)
281                 wait_event(thread_wait, !is_threaded_test_run(info));
282         wait = true;
283         return param_get_bool(val, kp);
284 }
285
286 static const struct kernel_param_ops wait_ops = {
287         .get = dmatest_wait_get,
288         .set = param_set_bool,
289 };
290 module_param_cb(wait, &wait_ops, &wait, S_IRUGO);
291 MODULE_PARM_DESC(wait, "Wait for tests to complete (default: false)");
292
293 static bool dmatest_match_channel(struct dmatest_params *params,
294                 struct dma_chan *chan)
295 {
296         if (params->channel[0] == '\0')
297                 return true;
298         return strcmp(dma_chan_name(chan), params->channel) == 0;
299 }
300
301 static bool dmatest_match_device(struct dmatest_params *params,
302                 struct dma_device *device)
303 {
304         if (params->device[0] == '\0')
305                 return true;
306         return strcmp(dev_name(device->dev), params->device) == 0;
307 }
308
309 static unsigned long dmatest_random(void)
310 {
311         unsigned long buf;
312
313         prandom_bytes(&buf, sizeof(buf));
314         return buf;
315 }
316
317 static inline u8 gen_inv_idx(u8 index, bool is_memset)
318 {
319         u8 val = is_memset ? PATTERN_MEMSET_IDX : index;
320
321         return ~val & PATTERN_COUNT_MASK;
322 }
323
324 static inline u8 gen_src_value(u8 index, bool is_memset)
325 {
326         return PATTERN_SRC | gen_inv_idx(index, is_memset);
327 }
328
329 static inline u8 gen_dst_value(u8 index, bool is_memset)
330 {
331         return PATTERN_DST | gen_inv_idx(index, is_memset);
332 }
333
334 static void dmatest_init_srcs(u8 **bufs, unsigned int start, unsigned int len,
335                 unsigned int buf_size, bool is_memset)
336 {
337         unsigned int i;
338         u8 *buf;
339
340         for (; (buf = *bufs); bufs++) {
341                 for (i = 0; i < start; i++)
342                         buf[i] = gen_src_value(i, is_memset);
343                 for ( ; i < start + len; i++)
344                         buf[i] = gen_src_value(i, is_memset) | PATTERN_COPY;
345                 for ( ; i < buf_size; i++)
346                         buf[i] = gen_src_value(i, is_memset);
347                 buf++;
348         }
349 }
350
351 static void dmatest_init_dsts(u8 **bufs, unsigned int start, unsigned int len,
352                 unsigned int buf_size, bool is_memset)
353 {
354         unsigned int i;
355         u8 *buf;
356
357         for (; (buf = *bufs); bufs++) {
358                 for (i = 0; i < start; i++)
359                         buf[i] = gen_dst_value(i, is_memset);
360                 for ( ; i < start + len; i++)
361                         buf[i] = gen_dst_value(i, is_memset) |
362                                                 PATTERN_OVERWRITE;
363                 for ( ; i < buf_size; i++)
364                         buf[i] = gen_dst_value(i, is_memset);
365         }
366 }
367
368 static void dmatest_mismatch(u8 actual, u8 pattern, unsigned int index,
369                 unsigned int counter, bool is_srcbuf, bool is_memset)
370 {
371         u8              diff = actual ^ pattern;
372         u8              expected = pattern | gen_inv_idx(counter, is_memset);
373         const char      *thread_name = current->comm;
374
375         if (is_srcbuf)
376                 pr_warn("%s: srcbuf[0x%x] overwritten! Expected %02x, got %02x\n",
377                         thread_name, index, expected, actual);
378         else if ((pattern & PATTERN_COPY)
379                         && (diff & (PATTERN_COPY | PATTERN_OVERWRITE)))
380                 pr_warn("%s: dstbuf[0x%x] not copied! Expected %02x, got %02x\n",
381                         thread_name, index, expected, actual);
382         else if (diff & PATTERN_SRC)
383                 pr_warn("%s: dstbuf[0x%x] was copied! Expected %02x, got %02x\n",
384                         thread_name, index, expected, actual);
385         else
386                 pr_warn("%s: dstbuf[0x%x] mismatch! Expected %02x, got %02x\n",
387                         thread_name, index, expected, actual);
388 }
389
390 static unsigned int dmatest_verify(u8 **bufs, unsigned int start,
391                 unsigned int end, unsigned int counter, u8 pattern,
392                 bool is_srcbuf, bool is_memset)
393 {
394         unsigned int i;
395         unsigned int error_count = 0;
396         u8 actual;
397         u8 expected;
398         u8 *buf;
399         unsigned int counter_orig = counter;
400
401         for (; (buf = *bufs); bufs++) {
402                 counter = counter_orig;
403                 for (i = start; i < end; i++) {
404                         actual = buf[i];
405                         expected = pattern | gen_inv_idx(counter, is_memset);
406                         if (actual != expected) {
407                                 if (error_count < MAX_ERROR_COUNT)
408                                         dmatest_mismatch(actual, pattern, i,
409                                                          counter, is_srcbuf,
410                                                          is_memset);
411                                 error_count++;
412                         }
413                         counter++;
414                 }
415         }
416
417         if (error_count > MAX_ERROR_COUNT)
418                 pr_warn("%s: %u errors suppressed\n",
419                         current->comm, error_count - MAX_ERROR_COUNT);
420
421         return error_count;
422 }
423
424
425 static void dmatest_callback(void *arg)
426 {
427         struct dmatest_done *done = arg;
428         struct dmatest_thread *thread =
429                 container_of(done, struct dmatest_thread, test_done);
430         if (!thread->done) {
431                 done->done = true;
432                 wake_up_all(done->wait);
433         } else {
434                 /*
435                  * If thread->done, it means that this callback occurred
436                  * after the parent thread has cleaned up. This can
437                  * happen in the case that driver doesn't implement
438                  * the terminate_all() functionality and a dma operation
439                  * did not occur within the timeout period
440                  */
441                 WARN(1, "dmatest: Kernel memory may be corrupted!!\n");
442         }
443 }
444
445 static unsigned int min_odd(unsigned int x, unsigned int y)
446 {
447         unsigned int val = min(x, y);
448
449         return val % 2 ? val : val - 1;
450 }
451
452 static void result(const char *err, unsigned int n, unsigned int src_off,
453                    unsigned int dst_off, unsigned int len, unsigned long data)
454 {
455         pr_info("%s: result #%u: '%s' with src_off=0x%x dst_off=0x%x len=0x%x (%lu)\n",
456                 current->comm, n, err, src_off, dst_off, len, data);
457 }
458
459 static void dbg_result(const char *err, unsigned int n, unsigned int src_off,
460                        unsigned int dst_off, unsigned int len,
461                        unsigned long data)
462 {
463         pr_debug("%s: result #%u: '%s' with src_off=0x%x dst_off=0x%x len=0x%x (%lu)\n",
464                  current->comm, n, err, src_off, dst_off, len, data);
465 }
466
467 #define verbose_result(err, n, src_off, dst_off, len, data) ({  \
468         if (verbose)                                            \
469                 result(err, n, src_off, dst_off, len, data);    \
470         else                                                    \
471                 dbg_result(err, n, src_off, dst_off, len, data);\
472 })
473
474 static unsigned long long dmatest_persec(s64 runtime, unsigned int val)
475 {
476         unsigned long long per_sec = 1000000;
477
478         if (runtime <= 0)
479                 return 0;
480
481         /* drop precision until runtime is 32-bits */
482         while (runtime > UINT_MAX) {
483                 runtime >>= 1;
484                 per_sec <<= 1;
485         }
486
487         per_sec *= val;
488         per_sec = INT_TO_FIXPT(per_sec);
489         do_div(per_sec, runtime);
490
491         return per_sec;
492 }
493
494 static unsigned long long dmatest_KBs(s64 runtime, unsigned long long len)
495 {
496         return FIXPT_TO_INT(dmatest_persec(runtime, len >> 10));
497 }
498
499 static void __dmatest_free_test_data(struct dmatest_data *d, unsigned int cnt)
500 {
501         unsigned int i;
502
503         for (i = 0; i < cnt; i++)
504                 kfree(d->raw[i]);
505
506         kfree(d->aligned);
507         kfree(d->raw);
508 }
509
510 static void dmatest_free_test_data(struct dmatest_data *d)
511 {
512         __dmatest_free_test_data(d, d->cnt);
513 }
514
515 static int dmatest_alloc_test_data(struct dmatest_data *d,
516                 unsigned int buf_size, u8 align)
517 {
518         unsigned int i = 0;
519
520         d->raw = kcalloc(d->cnt + 1, sizeof(u8 *), GFP_KERNEL);
521         if (!d->raw)
522                 return -ENOMEM;
523
524         d->aligned = kcalloc(d->cnt + 1, sizeof(u8 *), GFP_KERNEL);
525         if (!d->aligned)
526                 goto err;
527
528         for (i = 0; i < d->cnt; i++) {
529                 d->raw[i] = kmalloc(buf_size + align, GFP_KERNEL);
530                 if (!d->raw[i])
531                         goto err;
532
533                 /* align to alignment restriction */
534                 if (align)
535                         d->aligned[i] = PTR_ALIGN(d->raw[i], align);
536                 else
537                         d->aligned[i] = d->raw[i];
538         }
539
540         return 0;
541 err:
542         __dmatest_free_test_data(d, i);
543         return -ENOMEM;
544 }
545
546 /*
547  * This function repeatedly tests DMA transfers of various lengths and
548  * offsets for a given operation type until it is told to exit by
549  * kthread_stop(). There may be multiple threads running this function
550  * in parallel for a single channel, and there may be multiple channels
551  * being tested in parallel.
552  *
553  * Before each test, the source and destination buffer is initialized
554  * with a known pattern. This pattern is different depending on
555  * whether it's in an area which is supposed to be copied or
556  * overwritten, and different in the source and destination buffers.
557  * So if the DMA engine doesn't copy exactly what we tell it to copy,
558  * we'll notice.
559  */
560 static int dmatest_func(void *data)
561 {
562         struct dmatest_thread   *thread = data;
563         struct dmatest_done     *done = &thread->test_done;
564         struct dmatest_info     *info;
565         struct dmatest_params   *params;
566         struct dma_chan         *chan;
567         struct dma_device       *dev;
568         unsigned int            error_count;
569         unsigned int            failed_tests = 0;
570         unsigned int            total_tests = 0;
571         dma_cookie_t            cookie;
572         enum dma_status         status;
573         enum dma_ctrl_flags     flags;
574         u8                      *pq_coefs = NULL;
575         int                     ret;
576         unsigned int            buf_size;
577         struct dmatest_data     *src;
578         struct dmatest_data     *dst;
579         int                     i;
580         ktime_t                 ktime, start, diff;
581         ktime_t                 filltime = 0;
582         ktime_t                 comparetime = 0;
583         s64                     runtime = 0;
584         unsigned long long      total_len = 0;
585         unsigned long long      iops = 0;
586         u8                      align = 0;
587         bool                    is_memset = false;
588         dma_addr_t              *srcs;
589         dma_addr_t              *dma_pq;
590
591         set_freezable();
592
593         ret = -ENOMEM;
594
595         smp_rmb();
596         thread->pending = false;
597         info = thread->info;
598         params = &info->params;
599         chan = thread->chan;
600         dev = chan->device;
601         src = &thread->src;
602         dst = &thread->dst;
603         if (thread->type == DMA_MEMCPY) {
604                 align = params->alignment < 0 ? dev->copy_align :
605                                                 params->alignment;
606                 src->cnt = dst->cnt = 1;
607         } else if (thread->type == DMA_MEMSET) {
608                 align = params->alignment < 0 ? dev->fill_align :
609                                                 params->alignment;
610                 src->cnt = dst->cnt = 1;
611                 is_memset = true;
612         } else if (thread->type == DMA_XOR) {
613                 /* force odd to ensure dst = src */
614                 src->cnt = min_odd(params->xor_sources | 1, dev->max_xor);
615                 dst->cnt = 1;
616                 align = params->alignment < 0 ? dev->xor_align :
617                                                 params->alignment;
618         } else if (thread->type == DMA_PQ) {
619                 /* force odd to ensure dst = src */
620                 src->cnt = min_odd(params->pq_sources | 1, dma_maxpq(dev, 0));
621                 dst->cnt = 2;
622                 align = params->alignment < 0 ? dev->pq_align :
623                                                 params->alignment;
624
625                 pq_coefs = kmalloc(params->pq_sources + 1, GFP_KERNEL);
626                 if (!pq_coefs)
627                         goto err_thread_type;
628
629                 for (i = 0; i < src->cnt; i++)
630                         pq_coefs[i] = 1;
631         } else
632                 goto err_thread_type;
633
634         /* Check if buffer count fits into map count variable (u8) */
635         if ((src->cnt + dst->cnt) >= 255) {
636                 pr_err("too many buffers (%d of 255 supported)\n",
637                        src->cnt + dst->cnt);
638                 goto err_free_coefs;
639         }
640
641         buf_size = params->buf_size;
642         if (1 << align > buf_size) {
643                 pr_err("%u-byte buffer too small for %d-byte alignment\n",
644                        buf_size, 1 << align);
645                 goto err_free_coefs;
646         }
647
648         if (dmatest_alloc_test_data(src, buf_size, align) < 0)
649                 goto err_free_coefs;
650
651         if (dmatest_alloc_test_data(dst, buf_size, align) < 0)
652                 goto err_src;
653
654         set_user_nice(current, 10);
655
656         srcs = kcalloc(src->cnt, sizeof(dma_addr_t), GFP_KERNEL);
657         if (!srcs)
658                 goto err_dst;
659
660         dma_pq = kcalloc(dst->cnt, sizeof(dma_addr_t), GFP_KERNEL);
661         if (!dma_pq)
662                 goto err_srcs_array;
663
664         /*
665          * src and dst buffers are freed by ourselves below
666          */
667         if (params->polled)
668                 flags = DMA_CTRL_ACK;
669         else
670                 flags = DMA_CTRL_ACK | DMA_PREP_INTERRUPT;
671
672         ktime = ktime_get();
673         while (!(kthread_should_stop() ||
674                (params->iterations && total_tests >= params->iterations))) {
675                 struct dma_async_tx_descriptor *tx = NULL;
676                 struct dmaengine_unmap_data *um;
677                 dma_addr_t *dsts;
678                 unsigned int len;
679
680                 total_tests++;
681
682                 if (params->transfer_size) {
683                         if (params->transfer_size >= buf_size) {
684                                 pr_err("%u-byte transfer size must be lower than %u-buffer size\n",
685                                        params->transfer_size, buf_size);
686                                 break;
687                         }
688                         len = params->transfer_size;
689                 } else if (params->norandom) {
690                         len = buf_size;
691                 } else {
692                         len = dmatest_random() % buf_size + 1;
693                 }
694
695                 /* Do not alter transfer size explicitly defined by user */
696                 if (!params->transfer_size) {
697                         len = (len >> align) << align;
698                         if (!len)
699                                 len = 1 << align;
700                 }
701                 total_len += len;
702
703                 if (params->norandom) {
704                         src->off = 0;
705                         dst->off = 0;
706                 } else {
707                         src->off = dmatest_random() % (buf_size - len + 1);
708                         dst->off = dmatest_random() % (buf_size - len + 1);
709
710                         src->off = (src->off >> align) << align;
711                         dst->off = (dst->off >> align) << align;
712                 }
713
714                 if (!params->noverify) {
715                         start = ktime_get();
716                         dmatest_init_srcs(src->aligned, src->off, len,
717                                           buf_size, is_memset);
718                         dmatest_init_dsts(dst->aligned, dst->off, len,
719                                           buf_size, is_memset);
720
721                         diff = ktime_sub(ktime_get(), start);
722                         filltime = ktime_add(filltime, diff);
723                 }
724
725                 um = dmaengine_get_unmap_data(dev->dev, src->cnt + dst->cnt,
726                                               GFP_KERNEL);
727                 if (!um) {
728                         failed_tests++;
729                         result("unmap data NULL", total_tests,
730                                src->off, dst->off, len, ret);
731                         continue;
732                 }
733
734                 um->len = buf_size;
735                 for (i = 0; i < src->cnt; i++) {
736                         void *buf = src->aligned[i];
737                         struct page *pg = virt_to_page(buf);
738                         unsigned long pg_off = offset_in_page(buf);
739
740                         um->addr[i] = dma_map_page(dev->dev, pg, pg_off,
741                                                    um->len, DMA_TO_DEVICE);
742                         srcs[i] = um->addr[i] + src->off;
743                         ret = dma_mapping_error(dev->dev, um->addr[i]);
744                         if (ret) {
745                                 result("src mapping error", total_tests,
746                                        src->off, dst->off, len, ret);
747                                 goto error_unmap_continue;
748                         }
749                         um->to_cnt++;
750                 }
751                 /* map with DMA_BIDIRECTIONAL to force writeback/invalidate */
752                 dsts = &um->addr[src->cnt];
753                 for (i = 0; i < dst->cnt; i++) {
754                         void *buf = dst->aligned[i];
755                         struct page *pg = virt_to_page(buf);
756                         unsigned long pg_off = offset_in_page(buf);
757
758                         dsts[i] = dma_map_page(dev->dev, pg, pg_off, um->len,
759                                                DMA_BIDIRECTIONAL);
760                         ret = dma_mapping_error(dev->dev, dsts[i]);
761                         if (ret) {
762                                 result("dst mapping error", total_tests,
763                                        src->off, dst->off, len, ret);
764                                 goto error_unmap_continue;
765                         }
766                         um->bidi_cnt++;
767                 }
768
769                 if (thread->type == DMA_MEMCPY)
770                         tx = dev->device_prep_dma_memcpy(chan,
771                                                          dsts[0] + dst->off,
772                                                          srcs[0], len, flags);
773                 else if (thread->type == DMA_MEMSET)
774                         tx = dev->device_prep_dma_memset(chan,
775                                                 dsts[0] + dst->off,
776                                                 *(src->aligned[0] + src->off),
777                                                 len, flags);
778                 else if (thread->type == DMA_XOR)
779                         tx = dev->device_prep_dma_xor(chan,
780                                                       dsts[0] + dst->off,
781                                                       srcs, src->cnt,
782                                                       len, flags);
783                 else if (thread->type == DMA_PQ) {
784                         for (i = 0; i < dst->cnt; i++)
785                                 dma_pq[i] = dsts[i] + dst->off;
786                         tx = dev->device_prep_dma_pq(chan, dma_pq, srcs,
787                                                      src->cnt, pq_coefs,
788                                                      len, flags);
789                 }
790
791                 if (!tx) {
792                         result("prep error", total_tests, src->off,
793                                dst->off, len, ret);
794                         msleep(100);
795                         goto error_unmap_continue;
796                 }
797
798                 done->done = false;
799                 if (!params->polled) {
800                         tx->callback = dmatest_callback;
801                         tx->callback_param = done;
802                 }
803                 cookie = tx->tx_submit(tx);
804
805                 if (dma_submit_error(cookie)) {
806                         result("submit error", total_tests, src->off,
807                                dst->off, len, ret);
808                         msleep(100);
809                         goto error_unmap_continue;
810                 }
811
812                 if (params->polled) {
813                         status = dma_sync_wait(chan, cookie);
814                         dmaengine_terminate_sync(chan);
815                         if (status == DMA_COMPLETE)
816                                 done->done = true;
817                 } else {
818                         dma_async_issue_pending(chan);
819
820                         wait_event_freezable_timeout(thread->done_wait,
821                                         done->done,
822                                         msecs_to_jiffies(params->timeout));
823
824                         status = dma_async_is_tx_complete(chan, cookie, NULL,
825                                                           NULL);
826                 }
827
828                 if (!done->done) {
829                         result("test timed out", total_tests, src->off, dst->off,
830                                len, 0);
831                         goto error_unmap_continue;
832                 } else if (status != DMA_COMPLETE) {
833                         result(status == DMA_ERROR ?
834                                "completion error status" :
835                                "completion busy status", total_tests, src->off,
836                                dst->off, len, ret);
837                         goto error_unmap_continue;
838                 }
839
840                 dmaengine_unmap_put(um);
841
842                 if (params->noverify) {
843                         verbose_result("test passed", total_tests, src->off,
844                                        dst->off, len, 0);
845                         continue;
846                 }
847
848                 start = ktime_get();
849                 pr_debug("%s: verifying source buffer...\n", current->comm);
850                 error_count = dmatest_verify(src->aligned, 0, src->off,
851                                 0, PATTERN_SRC, true, is_memset);
852                 error_count += dmatest_verify(src->aligned, src->off,
853                                 src->off + len, src->off,
854                                 PATTERN_SRC | PATTERN_COPY, true, is_memset);
855                 error_count += dmatest_verify(src->aligned, src->off + len,
856                                 buf_size, src->off + len,
857                                 PATTERN_SRC, true, is_memset);
858
859                 pr_debug("%s: verifying dest buffer...\n", current->comm);
860                 error_count += dmatest_verify(dst->aligned, 0, dst->off,
861                                 0, PATTERN_DST, false, is_memset);
862
863                 error_count += dmatest_verify(dst->aligned, dst->off,
864                                 dst->off + len, src->off,
865                                 PATTERN_SRC | PATTERN_COPY, false, is_memset);
866
867                 error_count += dmatest_verify(dst->aligned, dst->off + len,
868                                 buf_size, dst->off + len,
869                                 PATTERN_DST, false, is_memset);
870
871                 diff = ktime_sub(ktime_get(), start);
872                 comparetime = ktime_add(comparetime, diff);
873
874                 if (error_count) {
875                         result("data error", total_tests, src->off, dst->off,
876                                len, error_count);
877                         failed_tests++;
878                 } else {
879                         verbose_result("test passed", total_tests, src->off,
880                                        dst->off, len, 0);
881                 }
882
883                 continue;
884
885 error_unmap_continue:
886                 dmaengine_unmap_put(um);
887                 failed_tests++;
888         }
889         ktime = ktime_sub(ktime_get(), ktime);
890         ktime = ktime_sub(ktime, comparetime);
891         ktime = ktime_sub(ktime, filltime);
892         runtime = ktime_to_us(ktime);
893
894         ret = 0;
895         kfree(dma_pq);
896 err_srcs_array:
897         kfree(srcs);
898 err_dst:
899         dmatest_free_test_data(dst);
900 err_src:
901         dmatest_free_test_data(src);
902 err_free_coefs:
903         kfree(pq_coefs);
904 err_thread_type:
905         iops = dmatest_persec(runtime, total_tests);
906         pr_info("%s: summary %u tests, %u failures %llu.%02llu iops %llu KB/s (%d)\n",
907                 current->comm, total_tests, failed_tests,
908                 FIXPT_TO_INT(iops), FIXPT_GET_FRAC(iops),
909                 dmatest_KBs(runtime, total_len), ret);
910
911         /* terminate all transfers on specified channels */
912         if (ret || failed_tests)
913                 dmaengine_terminate_sync(chan);
914
915         thread->done = true;
916         wake_up(&thread_wait);
917
918         return ret;
919 }
920
921 static void dmatest_cleanup_channel(struct dmatest_chan *dtc)
922 {
923         struct dmatest_thread   *thread;
924         struct dmatest_thread   *_thread;
925         int                     ret;
926
927         list_for_each_entry_safe(thread, _thread, &dtc->threads, node) {
928                 ret = kthread_stop(thread->task);
929                 pr_debug("thread %s exited with status %d\n",
930                          thread->task->comm, ret);
931                 list_del(&thread->node);
932                 put_task_struct(thread->task);
933                 kfree(thread);
934         }
935
936         /* terminate all transfers on specified channels */
937         dmaengine_terminate_sync(dtc->chan);
938
939         kfree(dtc);
940 }
941
942 static int dmatest_add_threads(struct dmatest_info *info,
943                 struct dmatest_chan *dtc, enum dma_transaction_type type)
944 {
945         struct dmatest_params *params = &info->params;
946         struct dmatest_thread *thread;
947         struct dma_chan *chan = dtc->chan;
948         char *op;
949         unsigned int i;
950
951         if (type == DMA_MEMCPY)
952                 op = "copy";
953         else if (type == DMA_MEMSET)
954                 op = "set";
955         else if (type == DMA_XOR)
956                 op = "xor";
957         else if (type == DMA_PQ)
958                 op = "pq";
959         else
960                 return -EINVAL;
961
962         for (i = 0; i < params->threads_per_chan; i++) {
963                 thread = kzalloc(sizeof(struct dmatest_thread), GFP_KERNEL);
964                 if (!thread) {
965                         pr_warn("No memory for %s-%s%u\n",
966                                 dma_chan_name(chan), op, i);
967                         break;
968                 }
969                 thread->info = info;
970                 thread->chan = dtc->chan;
971                 thread->type = type;
972                 thread->test_done.wait = &thread->done_wait;
973                 init_waitqueue_head(&thread->done_wait);
974                 smp_wmb();
975                 thread->task = kthread_create(dmatest_func, thread, "%s-%s%u",
976                                 dma_chan_name(chan), op, i);
977                 if (IS_ERR(thread->task)) {
978                         pr_warn("Failed to create thread %s-%s%u\n",
979                                 dma_chan_name(chan), op, i);
980                         kfree(thread);
981                         break;
982                 }
983
984                 /* srcbuf and dstbuf are allocated by the thread itself */
985                 get_task_struct(thread->task);
986                 list_add_tail(&thread->node, &dtc->threads);
987                 thread->pending = true;
988         }
989
990         return i;
991 }
992
993 static int dmatest_add_channel(struct dmatest_info *info,
994                 struct dma_chan *chan)
995 {
996         struct dmatest_chan     *dtc;
997         struct dma_device       *dma_dev = chan->device;
998         unsigned int            thread_count = 0;
999         int cnt;
1000
1001         dtc = kmalloc(sizeof(struct dmatest_chan), GFP_KERNEL);
1002         if (!dtc) {
1003                 pr_warn("No memory for %s\n", dma_chan_name(chan));
1004                 return -ENOMEM;
1005         }
1006
1007         dtc->chan = chan;
1008         INIT_LIST_HEAD(&dtc->threads);
1009
1010         if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1011                 if (dmatest == 0) {
1012                         cnt = dmatest_add_threads(info, dtc, DMA_MEMCPY);
1013                         thread_count += cnt > 0 ? cnt : 0;
1014                 }
1015         }
1016
1017         if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
1018                 if (dmatest == 1) {
1019                         cnt = dmatest_add_threads(info, dtc, DMA_MEMSET);
1020                         thread_count += cnt > 0 ? cnt : 0;
1021                 }
1022         }
1023
1024         if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1025                 cnt = dmatest_add_threads(info, dtc, DMA_XOR);
1026                 thread_count += cnt > 0 ? cnt : 0;
1027         }
1028         if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1029                 cnt = dmatest_add_threads(info, dtc, DMA_PQ);
1030                 thread_count += cnt > 0 ? cnt : 0;
1031         }
1032
1033         pr_info("Added %u threads using %s\n",
1034                 thread_count, dma_chan_name(chan));
1035
1036         list_add_tail(&dtc->node, &info->channels);
1037         info->nr_channels++;
1038
1039         return 0;
1040 }
1041
1042 static bool filter(struct dma_chan *chan, void *param)
1043 {
1044         struct dmatest_params *params = param;
1045
1046         if (!dmatest_match_channel(params, chan) ||
1047             !dmatest_match_device(params, chan->device))
1048                 return false;
1049         else
1050                 return true;
1051 }
1052
1053 static void request_channels(struct dmatest_info *info,
1054                              enum dma_transaction_type type)
1055 {
1056         dma_cap_mask_t mask;
1057
1058         dma_cap_zero(mask);
1059         dma_cap_set(type, mask);
1060         for (;;) {
1061                 struct dmatest_params *params = &info->params;
1062                 struct dma_chan *chan;
1063
1064                 chan = dma_request_channel(mask, filter, params);
1065                 if (chan) {
1066                         if (dmatest_add_channel(info, chan)) {
1067                                 dma_release_channel(chan);
1068                                 break; /* add_channel failed, punt */
1069                         }
1070                 } else
1071                         break; /* no more channels available */
1072                 if (params->max_channels &&
1073                     info->nr_channels >= params->max_channels)
1074                         break; /* we have all we need */
1075         }
1076 }
1077
1078 static void add_threaded_test(struct dmatest_info *info)
1079 {
1080         struct dmatest_params *params = &info->params;
1081
1082         /* Copy test parameters */
1083         params->buf_size = test_buf_size;
1084         strlcpy(params->channel, strim(test_channel), sizeof(params->channel));
1085         strlcpy(params->device, strim(test_device), sizeof(params->device));
1086         params->threads_per_chan = threads_per_chan;
1087         params->max_channels = max_channels;
1088         params->iterations = iterations;
1089         params->xor_sources = xor_sources;
1090         params->pq_sources = pq_sources;
1091         params->timeout = timeout;
1092         params->noverify = noverify;
1093         params->norandom = norandom;
1094         params->alignment = alignment;
1095         params->transfer_size = transfer_size;
1096         params->polled = polled;
1097
1098         request_channels(info, DMA_MEMCPY);
1099         request_channels(info, DMA_MEMSET);
1100         request_channels(info, DMA_XOR);
1101         request_channels(info, DMA_PQ);
1102 }
1103
1104 static void run_pending_tests(struct dmatest_info *info)
1105 {
1106         struct dmatest_chan *dtc;
1107         unsigned int thread_count = 0;
1108
1109         list_for_each_entry(dtc, &info->channels, node) {
1110                 struct dmatest_thread *thread;
1111
1112                 thread_count = 0;
1113                 list_for_each_entry(thread, &dtc->threads, node) {
1114                         wake_up_process(thread->task);
1115                         thread_count++;
1116                 }
1117                 pr_info("Started %u threads using %s\n",
1118                         thread_count, dma_chan_name(dtc->chan));
1119         }
1120 }
1121
1122 static void stop_threaded_test(struct dmatest_info *info)
1123 {
1124         struct dmatest_chan *dtc, *_dtc;
1125         struct dma_chan *chan;
1126
1127         list_for_each_entry_safe(dtc, _dtc, &info->channels, node) {
1128                 list_del(&dtc->node);
1129                 chan = dtc->chan;
1130                 dmatest_cleanup_channel(dtc);
1131                 pr_debug("dropped channel %s\n", dma_chan_name(chan));
1132                 dma_release_channel(chan);
1133         }
1134
1135         info->nr_channels = 0;
1136 }
1137
1138 static void start_threaded_tests(struct dmatest_info *info)
1139 {
1140         /* we might be called early to set run=, defer running until all
1141          * parameters have been evaluated
1142          */
1143         if (!info->did_init)
1144                 return;
1145
1146         run_pending_tests(info);
1147 }
1148
1149 static int dmatest_run_get(char *val, const struct kernel_param *kp)
1150 {
1151         struct dmatest_info *info = &test_info;
1152
1153         mutex_lock(&info->lock);
1154         if (is_threaded_test_run(info)) {
1155                 dmatest_run = true;
1156         } else {
1157                 if (!is_threaded_test_pending(info))
1158                         stop_threaded_test(info);
1159                 dmatest_run = false;
1160         }
1161         mutex_unlock(&info->lock);
1162
1163         return param_get_bool(val, kp);
1164 }
1165
1166 static int dmatest_run_set(const char *val, const struct kernel_param *kp)
1167 {
1168         struct dmatest_info *info = &test_info;
1169         int ret;
1170
1171         mutex_lock(&info->lock);
1172         ret = param_set_bool(val, kp);
1173         if (ret) {
1174                 mutex_unlock(&info->lock);
1175                 return ret;
1176         } else if (dmatest_run) {
1177                 if (!is_threaded_test_pending(info)) {
1178                         pr_info("No channels configured, continue with any\n");
1179                         add_threaded_test(info);
1180                 }
1181                 start_threaded_tests(info);
1182         } else {
1183                 stop_threaded_test(info);
1184         }
1185
1186         mutex_unlock(&info->lock);
1187
1188         return ret;
1189 }
1190
1191 static int dmatest_chan_set(const char *val, const struct kernel_param *kp)
1192 {
1193         struct dmatest_info *info = &test_info;
1194         struct dmatest_chan *dtc;
1195         char chan_reset_val[20];
1196         int ret = 0;
1197
1198         mutex_lock(&info->lock);
1199         ret = param_set_copystring(val, kp);
1200         if (ret) {
1201                 mutex_unlock(&info->lock);
1202                 return ret;
1203         }
1204         /*Clear any previously run threads */
1205         if (!is_threaded_test_run(info) && !is_threaded_test_pending(info))
1206                 stop_threaded_test(info);
1207         /* Reject channels that are already registered */
1208         if (is_threaded_test_pending(info)) {
1209                 list_for_each_entry(dtc, &info->channels, node) {
1210                         if (strcmp(dma_chan_name(dtc->chan),
1211                                    strim(test_channel)) == 0) {
1212                                 dtc = list_last_entry(&info->channels,
1213                                                       struct dmatest_chan,
1214                                                       node);
1215                                 strlcpy(chan_reset_val,
1216                                         dma_chan_name(dtc->chan),
1217                                         sizeof(chan_reset_val));
1218                                 ret = -EBUSY;
1219                                 goto add_chan_err;
1220                         }
1221                 }
1222         }
1223
1224         add_threaded_test(info);
1225
1226         /* Check if channel was added successfully */
1227         dtc = list_last_entry(&info->channels, struct dmatest_chan, node);
1228
1229         if (dtc->chan) {
1230                 /*
1231                  * if new channel was not successfully added, revert the
1232                  * "test_channel" string to the name of the last successfully
1233                  * added channel. exception for when users issues empty string
1234                  * to channel parameter.
1235                  */
1236                 if ((strcmp(dma_chan_name(dtc->chan), strim(test_channel)) != 0)
1237                     && (strcmp("", strim(test_channel)) != 0)) {
1238                         ret = -EINVAL;
1239                         strlcpy(chan_reset_val, dma_chan_name(dtc->chan),
1240                                 sizeof(chan_reset_val));
1241                         goto add_chan_err;
1242                 }
1243
1244         } else {
1245                 /* Clear test_channel if no channels were added successfully */
1246                 strlcpy(chan_reset_val, "", sizeof(chan_reset_val));
1247                 ret = -EBUSY;
1248                 goto add_chan_err;
1249         }
1250
1251         mutex_unlock(&info->lock);
1252
1253         return ret;
1254
1255 add_chan_err:
1256         param_set_copystring(chan_reset_val, kp);
1257         mutex_unlock(&info->lock);
1258
1259         return ret;
1260 }
1261
1262 static int dmatest_chan_get(char *val, const struct kernel_param *kp)
1263 {
1264         struct dmatest_info *info = &test_info;
1265
1266         mutex_lock(&info->lock);
1267         if (!is_threaded_test_run(info) && !is_threaded_test_pending(info)) {
1268                 stop_threaded_test(info);
1269                 strlcpy(test_channel, "", sizeof(test_channel));
1270         }
1271         mutex_unlock(&info->lock);
1272
1273         return param_get_string(val, kp);
1274 }
1275
1276 static int dmatest_test_list_get(char *val, const struct kernel_param *kp)
1277 {
1278         struct dmatest_info *info = &test_info;
1279         struct dmatest_chan *dtc;
1280         unsigned int thread_count = 0;
1281
1282         list_for_each_entry(dtc, &info->channels, node) {
1283                 struct dmatest_thread *thread;
1284
1285                 thread_count = 0;
1286                 list_for_each_entry(thread, &dtc->threads, node) {
1287                         thread_count++;
1288                 }
1289                 pr_info("%u threads using %s\n",
1290                         thread_count, dma_chan_name(dtc->chan));
1291         }
1292
1293         return 0;
1294 }
1295
1296 static int __init dmatest_init(void)
1297 {
1298         struct dmatest_info *info = &test_info;
1299         struct dmatest_params *params = &info->params;
1300
1301         if (dmatest_run) {
1302                 mutex_lock(&info->lock);
1303                 add_threaded_test(info);
1304                 run_pending_tests(info);
1305                 mutex_unlock(&info->lock);
1306         }
1307
1308         if (params->iterations && wait)
1309                 wait_event(thread_wait, !is_threaded_test_run(info));
1310
1311         /* module parameters are stable, inittime tests are started,
1312          * let userspace take over 'run' control
1313          */
1314         info->did_init = true;
1315
1316         return 0;
1317 }
1318 /* when compiled-in wait for drivers to load first */
1319 late_initcall(dmatest_init);
1320
1321 static void __exit dmatest_exit(void)
1322 {
1323         struct dmatest_info *info = &test_info;
1324
1325         mutex_lock(&info->lock);
1326         stop_threaded_test(info);
1327         mutex_unlock(&info->lock);
1328 }
1329 module_exit(dmatest_exit);
1330
1331 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
1332 MODULE_LICENSE("GPL v2");