Merge tag 'block-6.1-2022-12-08' of git://git.kernel.dk/linux
[platform/kernel/linux-rpi.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/err.h>
11 #include <linux/delay.h>
12 #include <linux/dma-mapping.h>
13 #include <linux/dmaengine.h>
14 #include <linux/freezer.h>
15 #include <linux/init.h>
16 #include <linux/kthread.h>
17 #include <linux/sched/task.h>
18 #include <linux/module.h>
19 #include <linux/moduleparam.h>
20 #include <linux/random.h>
21 #include <linux/slab.h>
22 #include <linux/wait.h>
23
24 static unsigned int test_buf_size = 16384;
25 module_param(test_buf_size, uint, 0644);
26 MODULE_PARM_DESC(test_buf_size, "Size of the memcpy test buffer");
27
28 static char test_device[32];
29 module_param_string(device, test_device, sizeof(test_device), 0644);
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, 0644);
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, 0644);
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, 0644);
44 MODULE_PARM_DESC(iterations,
45                 "Iterations before stopping test (default: infinite)");
46
47 static unsigned int dmatest;
48 module_param(dmatest, uint, 0644);
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, 0644);
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, 0644);
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, 0644);
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, 0644);
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, 0644);
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, 0644);
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  * @last_error:         test has faced configuration issues
133  */
134 static struct dmatest_info {
135         /* Test parameters */
136         struct dmatest_params   params;
137
138         /* Internal state */
139         struct list_head        channels;
140         unsigned int            nr_channels;
141         int                     last_error;
142         struct mutex            lock;
143         bool                    did_init;
144 } test_info = {
145         .channels = LIST_HEAD_INIT(test_info.channels),
146         .lock = __MUTEX_INITIALIZER(test_info.lock),
147 };
148
149 static int dmatest_run_set(const char *val, const struct kernel_param *kp);
150 static int dmatest_run_get(char *val, const struct kernel_param *kp);
151 static const struct kernel_param_ops run_ops = {
152         .set = dmatest_run_set,
153         .get = dmatest_run_get,
154 };
155 static bool dmatest_run;
156 module_param_cb(run, &run_ops, &dmatest_run, 0644);
157 MODULE_PARM_DESC(run, "Run the test (default: false)");
158
159 static int dmatest_chan_set(const char *val, const struct kernel_param *kp);
160 static int dmatest_chan_get(char *val, const struct kernel_param *kp);
161 static const struct kernel_param_ops multi_chan_ops = {
162         .set = dmatest_chan_set,
163         .get = dmatest_chan_get,
164 };
165
166 static char test_channel[20];
167 static struct kparam_string newchan_kps = {
168         .string = test_channel,
169         .maxlen = 20,
170 };
171 module_param_cb(channel, &multi_chan_ops, &newchan_kps, 0644);
172 MODULE_PARM_DESC(channel, "Bus ID of the channel to test (default: any)");
173
174 static int dmatest_test_list_get(char *val, const struct kernel_param *kp);
175 static const struct kernel_param_ops test_list_ops = {
176         .get = dmatest_test_list_get,
177 };
178 module_param_cb(test_list, &test_list_ops, NULL, 0444);
179 MODULE_PARM_DESC(test_list, "Print current test list");
180
181 /* Maximum amount of mismatched bytes in buffer to print */
182 #define MAX_ERROR_COUNT         32
183
184 /*
185  * Initialization patterns. All bytes in the source buffer has bit 7
186  * set, all bytes in the destination buffer has bit 7 cleared.
187  *
188  * Bit 6 is set for all bytes which are to be copied by the DMA
189  * engine. Bit 5 is set for all bytes which are to be overwritten by
190  * the DMA engine.
191  *
192  * The remaining bits are the inverse of a counter which increments by
193  * one for each byte address.
194  */
195 #define PATTERN_SRC             0x80
196 #define PATTERN_DST             0x00
197 #define PATTERN_COPY            0x40
198 #define PATTERN_OVERWRITE       0x20
199 #define PATTERN_COUNT_MASK      0x1f
200 #define PATTERN_MEMSET_IDX      0x01
201
202 /* Fixed point arithmetic ops */
203 #define FIXPT_SHIFT             8
204 #define FIXPNT_MASK             0xFF
205 #define FIXPT_TO_INT(a) ((a) >> FIXPT_SHIFT)
206 #define INT_TO_FIXPT(a) ((a) << FIXPT_SHIFT)
207 #define FIXPT_GET_FRAC(a)       ((((a) & FIXPNT_MASK) * 100) >> FIXPT_SHIFT)
208
209 /* poor man's completion - we want to use wait_event_freezable() on it */
210 struct dmatest_done {
211         bool                    done;
212         wait_queue_head_t       *wait;
213 };
214
215 struct dmatest_data {
216         u8              **raw;
217         u8              **aligned;
218         unsigned int    cnt;
219         unsigned int    off;
220 };
221
222 struct dmatest_thread {
223         struct list_head        node;
224         struct dmatest_info     *info;
225         struct task_struct      *task;
226         struct dma_chan         *chan;
227         struct dmatest_data     src;
228         struct dmatest_data     dst;
229         enum dma_transaction_type type;
230         wait_queue_head_t done_wait;
231         struct dmatest_done test_done;
232         bool                    done;
233         bool                    pending;
234 };
235
236 struct dmatest_chan {
237         struct list_head        node;
238         struct dma_chan         *chan;
239         struct list_head        threads;
240 };
241
242 static DECLARE_WAIT_QUEUE_HEAD(thread_wait);
243 static bool wait;
244
245 static bool is_threaded_test_run(struct dmatest_info *info)
246 {
247         struct dmatest_chan *dtc;
248
249         list_for_each_entry(dtc, &info->channels, node) {
250                 struct dmatest_thread *thread;
251
252                 list_for_each_entry(thread, &dtc->threads, node) {
253                         if (!thread->done && !thread->pending)
254                                 return true;
255                 }
256         }
257
258         return false;
259 }
260
261 static bool is_threaded_test_pending(struct dmatest_info *info)
262 {
263         struct dmatest_chan *dtc;
264
265         list_for_each_entry(dtc, &info->channels, node) {
266                 struct dmatest_thread *thread;
267
268                 list_for_each_entry(thread, &dtc->threads, node) {
269                         if (thread->pending)
270                                 return true;
271                 }
272         }
273
274         return false;
275 }
276
277 static int dmatest_wait_get(char *val, const struct kernel_param *kp)
278 {
279         struct dmatest_info *info = &test_info;
280         struct dmatest_params *params = &info->params;
281
282         if (params->iterations)
283                 wait_event(thread_wait, !is_threaded_test_run(info));
284         wait = true;
285         return param_get_bool(val, kp);
286 }
287
288 static const struct kernel_param_ops wait_ops = {
289         .get = dmatest_wait_get,
290         .set = param_set_bool,
291 };
292 module_param_cb(wait, &wait_ops, &wait, 0444);
293 MODULE_PARM_DESC(wait, "Wait for tests to complete (default: false)");
294
295 static bool dmatest_match_channel(struct dmatest_params *params,
296                 struct dma_chan *chan)
297 {
298         if (params->channel[0] == '\0')
299                 return true;
300         return strcmp(dma_chan_name(chan), params->channel) == 0;
301 }
302
303 static bool dmatest_match_device(struct dmatest_params *params,
304                 struct dma_device *device)
305 {
306         if (params->device[0] == '\0')
307                 return true;
308         return strcmp(dev_name(device->dev), params->device) == 0;
309 }
310
311 static unsigned long dmatest_random(void)
312 {
313         unsigned long buf;
314
315         get_random_bytes(&buf, sizeof(buf));
316         return buf;
317 }
318
319 static inline u8 gen_inv_idx(u8 index, bool is_memset)
320 {
321         u8 val = is_memset ? PATTERN_MEMSET_IDX : index;
322
323         return ~val & PATTERN_COUNT_MASK;
324 }
325
326 static inline u8 gen_src_value(u8 index, bool is_memset)
327 {
328         return PATTERN_SRC | gen_inv_idx(index, is_memset);
329 }
330
331 static inline u8 gen_dst_value(u8 index, bool is_memset)
332 {
333         return PATTERN_DST | gen_inv_idx(index, is_memset);
334 }
335
336 static void dmatest_init_srcs(u8 **bufs, unsigned int start, unsigned int len,
337                 unsigned int buf_size, bool is_memset)
338 {
339         unsigned int i;
340         u8 *buf;
341
342         for (; (buf = *bufs); bufs++) {
343                 for (i = 0; i < start; i++)
344                         buf[i] = gen_src_value(i, is_memset);
345                 for ( ; i < start + len; i++)
346                         buf[i] = gen_src_value(i, is_memset) | PATTERN_COPY;
347                 for ( ; i < buf_size; i++)
348                         buf[i] = gen_src_value(i, is_memset);
349                 buf++;
350         }
351 }
352
353 static void dmatest_init_dsts(u8 **bufs, unsigned int start, unsigned int len,
354                 unsigned int buf_size, bool is_memset)
355 {
356         unsigned int i;
357         u8 *buf;
358
359         for (; (buf = *bufs); bufs++) {
360                 for (i = 0; i < start; i++)
361                         buf[i] = gen_dst_value(i, is_memset);
362                 for ( ; i < start + len; i++)
363                         buf[i] = gen_dst_value(i, is_memset) |
364                                                 PATTERN_OVERWRITE;
365                 for ( ; i < buf_size; i++)
366                         buf[i] = gen_dst_value(i, is_memset);
367         }
368 }
369
370 static void dmatest_mismatch(u8 actual, u8 pattern, unsigned int index,
371                 unsigned int counter, bool is_srcbuf, bool is_memset)
372 {
373         u8              diff = actual ^ pattern;
374         u8              expected = pattern | gen_inv_idx(counter, is_memset);
375         const char      *thread_name = current->comm;
376
377         if (is_srcbuf)
378                 pr_warn("%s: srcbuf[0x%x] overwritten! Expected %02x, got %02x\n",
379                         thread_name, index, expected, actual);
380         else if ((pattern & PATTERN_COPY)
381                         && (diff & (PATTERN_COPY | PATTERN_OVERWRITE)))
382                 pr_warn("%s: dstbuf[0x%x] not copied! Expected %02x, got %02x\n",
383                         thread_name, index, expected, actual);
384         else if (diff & PATTERN_SRC)
385                 pr_warn("%s: dstbuf[0x%x] was copied! Expected %02x, got %02x\n",
386                         thread_name, index, expected, actual);
387         else
388                 pr_warn("%s: dstbuf[0x%x] mismatch! Expected %02x, got %02x\n",
389                         thread_name, index, expected, actual);
390 }
391
392 static unsigned int dmatest_verify(u8 **bufs, unsigned int start,
393                 unsigned int end, unsigned int counter, u8 pattern,
394                 bool is_srcbuf, bool is_memset)
395 {
396         unsigned int i;
397         unsigned int error_count = 0;
398         u8 actual;
399         u8 expected;
400         u8 *buf;
401         unsigned int counter_orig = counter;
402
403         for (; (buf = *bufs); bufs++) {
404                 counter = counter_orig;
405                 for (i = start; i < end; i++) {
406                         actual = buf[i];
407                         expected = pattern | gen_inv_idx(counter, is_memset);
408                         if (actual != expected) {
409                                 if (error_count < MAX_ERROR_COUNT)
410                                         dmatest_mismatch(actual, pattern, i,
411                                                          counter, is_srcbuf,
412                                                          is_memset);
413                                 error_count++;
414                         }
415                         counter++;
416                 }
417         }
418
419         if (error_count > MAX_ERROR_COUNT)
420                 pr_warn("%s: %u errors suppressed\n",
421                         current->comm, error_count - MAX_ERROR_COUNT);
422
423         return error_count;
424 }
425
426
427 static void dmatest_callback(void *arg)
428 {
429         struct dmatest_done *done = arg;
430         struct dmatest_thread *thread =
431                 container_of(done, struct dmatest_thread, test_done);
432         if (!thread->done) {
433                 done->done = true;
434                 wake_up_all(done->wait);
435         } else {
436                 /*
437                  * If thread->done, it means that this callback occurred
438                  * after the parent thread has cleaned up. This can
439                  * happen in the case that driver doesn't implement
440                  * the terminate_all() functionality and a dma operation
441                  * did not occur within the timeout period
442                  */
443                 WARN(1, "dmatest: Kernel memory may be corrupted!!\n");
444         }
445 }
446
447 static unsigned int min_odd(unsigned int x, unsigned int y)
448 {
449         unsigned int val = min(x, y);
450
451         return val % 2 ? val : val - 1;
452 }
453
454 static void result(const char *err, unsigned int n, unsigned int src_off,
455                    unsigned int dst_off, unsigned int len, unsigned long data)
456 {
457         if (IS_ERR_VALUE(data)) {
458                 pr_info("%s: result #%u: '%s' with src_off=0x%x dst_off=0x%x len=0x%x (%ld)\n",
459                         current->comm, n, err, src_off, dst_off, len, data);
460         } else {
461                 pr_info("%s: result #%u: '%s' with src_off=0x%x dst_off=0x%x len=0x%x (%lu)\n",
462                         current->comm, n, err, src_off, dst_off, len, data);
463         }
464 }
465
466 static void dbg_result(const char *err, unsigned int n, unsigned int src_off,
467                        unsigned int dst_off, unsigned int len,
468                        unsigned long data)
469 {
470         pr_debug("%s: result #%u: '%s' with src_off=0x%x dst_off=0x%x len=0x%x (%lu)\n",
471                  current->comm, n, err, src_off, dst_off, len, data);
472 }
473
474 #define verbose_result(err, n, src_off, dst_off, len, data) ({  \
475         if (verbose)                                            \
476                 result(err, n, src_off, dst_off, len, data);    \
477         else                                                    \
478                 dbg_result(err, n, src_off, dst_off, len, data);\
479 })
480
481 static unsigned long long dmatest_persec(s64 runtime, unsigned int val)
482 {
483         unsigned long long per_sec = 1000000;
484
485         if (runtime <= 0)
486                 return 0;
487
488         /* drop precision until runtime is 32-bits */
489         while (runtime > UINT_MAX) {
490                 runtime >>= 1;
491                 per_sec <<= 1;
492         }
493
494         per_sec *= val;
495         per_sec = INT_TO_FIXPT(per_sec);
496         do_div(per_sec, runtime);
497
498         return per_sec;
499 }
500
501 static unsigned long long dmatest_KBs(s64 runtime, unsigned long long len)
502 {
503         return FIXPT_TO_INT(dmatest_persec(runtime, len >> 10));
504 }
505
506 static void __dmatest_free_test_data(struct dmatest_data *d, unsigned int cnt)
507 {
508         unsigned int i;
509
510         for (i = 0; i < cnt; i++)
511                 kfree(d->raw[i]);
512
513         kfree(d->aligned);
514         kfree(d->raw);
515 }
516
517 static void dmatest_free_test_data(struct dmatest_data *d)
518 {
519         __dmatest_free_test_data(d, d->cnt);
520 }
521
522 static int dmatest_alloc_test_data(struct dmatest_data *d,
523                 unsigned int buf_size, u8 align)
524 {
525         unsigned int i = 0;
526
527         d->raw = kcalloc(d->cnt + 1, sizeof(u8 *), GFP_KERNEL);
528         if (!d->raw)
529                 return -ENOMEM;
530
531         d->aligned = kcalloc(d->cnt + 1, sizeof(u8 *), GFP_KERNEL);
532         if (!d->aligned)
533                 goto err;
534
535         for (i = 0; i < d->cnt; i++) {
536                 d->raw[i] = kmalloc(buf_size + align, GFP_KERNEL);
537                 if (!d->raw[i])
538                         goto err;
539
540                 /* align to alignment restriction */
541                 if (align)
542                         d->aligned[i] = PTR_ALIGN(d->raw[i], align);
543                 else
544                         d->aligned[i] = d->raw[i];
545         }
546
547         return 0;
548 err:
549         __dmatest_free_test_data(d, i);
550         return -ENOMEM;
551 }
552
553 /*
554  * This function repeatedly tests DMA transfers of various lengths and
555  * offsets for a given operation type until it is told to exit by
556  * kthread_stop(). There may be multiple threads running this function
557  * in parallel for a single channel, and there may be multiple channels
558  * being tested in parallel.
559  *
560  * Before each test, the source and destination buffer is initialized
561  * with a known pattern. This pattern is different depending on
562  * whether it's in an area which is supposed to be copied or
563  * overwritten, and different in the source and destination buffers.
564  * So if the DMA engine doesn't copy exactly what we tell it to copy,
565  * we'll notice.
566  */
567 static int dmatest_func(void *data)
568 {
569         struct dmatest_thread   *thread = data;
570         struct dmatest_done     *done = &thread->test_done;
571         struct dmatest_info     *info;
572         struct dmatest_params   *params;
573         struct dma_chan         *chan;
574         struct dma_device       *dev;
575         struct device           *dma_dev;
576         unsigned int            error_count;
577         unsigned int            failed_tests = 0;
578         unsigned int            total_tests = 0;
579         dma_cookie_t            cookie;
580         enum dma_status         status;
581         enum dma_ctrl_flags     flags;
582         u8                      *pq_coefs = NULL;
583         int                     ret;
584         unsigned int            buf_size;
585         struct dmatest_data     *src;
586         struct dmatest_data     *dst;
587         int                     i;
588         ktime_t                 ktime, start, diff;
589         ktime_t                 filltime = 0;
590         ktime_t                 comparetime = 0;
591         s64                     runtime = 0;
592         unsigned long long      total_len = 0;
593         unsigned long long      iops = 0;
594         u8                      align = 0;
595         bool                    is_memset = false;
596         dma_addr_t              *srcs;
597         dma_addr_t              *dma_pq;
598
599         set_freezable();
600
601         ret = -ENOMEM;
602
603         smp_rmb();
604         thread->pending = false;
605         info = thread->info;
606         params = &info->params;
607         chan = thread->chan;
608         dev = chan->device;
609         dma_dev = dmaengine_get_dma_device(chan);
610
611         src = &thread->src;
612         dst = &thread->dst;
613         if (thread->type == DMA_MEMCPY) {
614                 align = params->alignment < 0 ? dev->copy_align :
615                                                 params->alignment;
616                 src->cnt = dst->cnt = 1;
617         } else if (thread->type == DMA_MEMSET) {
618                 align = params->alignment < 0 ? dev->fill_align :
619                                                 params->alignment;
620                 src->cnt = dst->cnt = 1;
621                 is_memset = true;
622         } else if (thread->type == DMA_XOR) {
623                 /* force odd to ensure dst = src */
624                 src->cnt = min_odd(params->xor_sources | 1, dev->max_xor);
625                 dst->cnt = 1;
626                 align = params->alignment < 0 ? dev->xor_align :
627                                                 params->alignment;
628         } else if (thread->type == DMA_PQ) {
629                 /* force odd to ensure dst = src */
630                 src->cnt = min_odd(params->pq_sources | 1, dma_maxpq(dev, 0));
631                 dst->cnt = 2;
632                 align = params->alignment < 0 ? dev->pq_align :
633                                                 params->alignment;
634
635                 pq_coefs = kmalloc(params->pq_sources + 1, GFP_KERNEL);
636                 if (!pq_coefs)
637                         goto err_thread_type;
638
639                 for (i = 0; i < src->cnt; i++)
640                         pq_coefs[i] = 1;
641         } else
642                 goto err_thread_type;
643
644         /* Check if buffer count fits into map count variable (u8) */
645         if ((src->cnt + dst->cnt) >= 255) {
646                 pr_err("too many buffers (%d of 255 supported)\n",
647                        src->cnt + dst->cnt);
648                 goto err_free_coefs;
649         }
650
651         buf_size = params->buf_size;
652         if (1 << align > buf_size) {
653                 pr_err("%u-byte buffer too small for %d-byte alignment\n",
654                        buf_size, 1 << align);
655                 goto err_free_coefs;
656         }
657
658         if (dmatest_alloc_test_data(src, buf_size, align) < 0)
659                 goto err_free_coefs;
660
661         if (dmatest_alloc_test_data(dst, buf_size, align) < 0)
662                 goto err_src;
663
664         set_user_nice(current, 10);
665
666         srcs = kcalloc(src->cnt, sizeof(dma_addr_t), GFP_KERNEL);
667         if (!srcs)
668                 goto err_dst;
669
670         dma_pq = kcalloc(dst->cnt, sizeof(dma_addr_t), GFP_KERNEL);
671         if (!dma_pq)
672                 goto err_srcs_array;
673
674         /*
675          * src and dst buffers are freed by ourselves below
676          */
677         if (params->polled)
678                 flags = DMA_CTRL_ACK;
679         else
680                 flags = DMA_CTRL_ACK | DMA_PREP_INTERRUPT;
681
682         ktime = ktime_get();
683         while (!(kthread_should_stop() ||
684                (params->iterations && total_tests >= params->iterations))) {
685                 struct dma_async_tx_descriptor *tx = NULL;
686                 struct dmaengine_unmap_data *um;
687                 dma_addr_t *dsts;
688                 unsigned int len;
689
690                 total_tests++;
691
692                 if (params->transfer_size) {
693                         if (params->transfer_size >= buf_size) {
694                                 pr_err("%u-byte transfer size must be lower than %u-buffer size\n",
695                                        params->transfer_size, buf_size);
696                                 break;
697                         }
698                         len = params->transfer_size;
699                 } else if (params->norandom) {
700                         len = buf_size;
701                 } else {
702                         len = dmatest_random() % buf_size + 1;
703                 }
704
705                 /* Do not alter transfer size explicitly defined by user */
706                 if (!params->transfer_size) {
707                         len = (len >> align) << align;
708                         if (!len)
709                                 len = 1 << align;
710                 }
711                 total_len += len;
712
713                 if (params->norandom) {
714                         src->off = 0;
715                         dst->off = 0;
716                 } else {
717                         src->off = dmatest_random() % (buf_size - len + 1);
718                         dst->off = dmatest_random() % (buf_size - len + 1);
719
720                         src->off = (src->off >> align) << align;
721                         dst->off = (dst->off >> align) << align;
722                 }
723
724                 if (!params->noverify) {
725                         start = ktime_get();
726                         dmatest_init_srcs(src->aligned, src->off, len,
727                                           buf_size, is_memset);
728                         dmatest_init_dsts(dst->aligned, dst->off, len,
729                                           buf_size, is_memset);
730
731                         diff = ktime_sub(ktime_get(), start);
732                         filltime = ktime_add(filltime, diff);
733                 }
734
735                 um = dmaengine_get_unmap_data(dma_dev, src->cnt + dst->cnt,
736                                               GFP_KERNEL);
737                 if (!um) {
738                         failed_tests++;
739                         result("unmap data NULL", total_tests,
740                                src->off, dst->off, len, ret);
741                         continue;
742                 }
743
744                 um->len = buf_size;
745                 for (i = 0; i < src->cnt; i++) {
746                         void *buf = src->aligned[i];
747                         struct page *pg = virt_to_page(buf);
748                         unsigned long pg_off = offset_in_page(buf);
749
750                         um->addr[i] = dma_map_page(dma_dev, pg, pg_off,
751                                                    um->len, DMA_TO_DEVICE);
752                         srcs[i] = um->addr[i] + src->off;
753                         ret = dma_mapping_error(dma_dev, um->addr[i]);
754                         if (ret) {
755                                 result("src mapping error", total_tests,
756                                        src->off, dst->off, len, ret);
757                                 goto error_unmap_continue;
758                         }
759                         um->to_cnt++;
760                 }
761                 /* map with DMA_BIDIRECTIONAL to force writeback/invalidate */
762                 dsts = &um->addr[src->cnt];
763                 for (i = 0; i < dst->cnt; i++) {
764                         void *buf = dst->aligned[i];
765                         struct page *pg = virt_to_page(buf);
766                         unsigned long pg_off = offset_in_page(buf);
767
768                         dsts[i] = dma_map_page(dma_dev, pg, pg_off, um->len,
769                                                DMA_BIDIRECTIONAL);
770                         ret = dma_mapping_error(dma_dev, dsts[i]);
771                         if (ret) {
772                                 result("dst mapping error", total_tests,
773                                        src->off, dst->off, len, ret);
774                                 goto error_unmap_continue;
775                         }
776                         um->bidi_cnt++;
777                 }
778
779                 if (thread->type == DMA_MEMCPY)
780                         tx = dev->device_prep_dma_memcpy(chan,
781                                                          dsts[0] + dst->off,
782                                                          srcs[0], len, flags);
783                 else if (thread->type == DMA_MEMSET)
784                         tx = dev->device_prep_dma_memset(chan,
785                                                 dsts[0] + dst->off,
786                                                 *(src->aligned[0] + src->off),
787                                                 len, flags);
788                 else if (thread->type == DMA_XOR)
789                         tx = dev->device_prep_dma_xor(chan,
790                                                       dsts[0] + dst->off,
791                                                       srcs, src->cnt,
792                                                       len, flags);
793                 else if (thread->type == DMA_PQ) {
794                         for (i = 0; i < dst->cnt; i++)
795                                 dma_pq[i] = dsts[i] + dst->off;
796                         tx = dev->device_prep_dma_pq(chan, dma_pq, srcs,
797                                                      src->cnt, pq_coefs,
798                                                      len, flags);
799                 }
800
801                 if (!tx) {
802                         result("prep error", total_tests, src->off,
803                                dst->off, len, ret);
804                         msleep(100);
805                         goto error_unmap_continue;
806                 }
807
808                 done->done = false;
809                 if (!params->polled) {
810                         tx->callback = dmatest_callback;
811                         tx->callback_param = done;
812                 }
813                 cookie = tx->tx_submit(tx);
814
815                 if (dma_submit_error(cookie)) {
816                         result("submit error", total_tests, src->off,
817                                dst->off, len, ret);
818                         msleep(100);
819                         goto error_unmap_continue;
820                 }
821
822                 if (params->polled) {
823                         status = dma_sync_wait(chan, cookie);
824                         dmaengine_terminate_sync(chan);
825                         if (status == DMA_COMPLETE)
826                                 done->done = true;
827                 } else {
828                         dma_async_issue_pending(chan);
829
830                         wait_event_freezable_timeout(thread->done_wait,
831                                         done->done,
832                                         msecs_to_jiffies(params->timeout));
833
834                         status = dma_async_is_tx_complete(chan, cookie, NULL,
835                                                           NULL);
836                 }
837
838                 if (!done->done) {
839                         result("test timed out", total_tests, src->off, dst->off,
840                                len, 0);
841                         goto error_unmap_continue;
842                 } else if (status != DMA_COMPLETE &&
843                            !(dma_has_cap(DMA_COMPLETION_NO_ORDER,
844                                          dev->cap_mask) &&
845                              status == DMA_OUT_OF_ORDER)) {
846                         result(status == DMA_ERROR ?
847                                "completion error status" :
848                                "completion busy status", total_tests, src->off,
849                                dst->off, len, ret);
850                         goto error_unmap_continue;
851                 }
852
853                 dmaengine_unmap_put(um);
854
855                 if (params->noverify) {
856                         verbose_result("test passed", total_tests, src->off,
857                                        dst->off, len, 0);
858                         continue;
859                 }
860
861                 start = ktime_get();
862                 pr_debug("%s: verifying source buffer...\n", current->comm);
863                 error_count = dmatest_verify(src->aligned, 0, src->off,
864                                 0, PATTERN_SRC, true, is_memset);
865                 error_count += dmatest_verify(src->aligned, src->off,
866                                 src->off + len, src->off,
867                                 PATTERN_SRC | PATTERN_COPY, true, is_memset);
868                 error_count += dmatest_verify(src->aligned, src->off + len,
869                                 buf_size, src->off + len,
870                                 PATTERN_SRC, true, is_memset);
871
872                 pr_debug("%s: verifying dest buffer...\n", current->comm);
873                 error_count += dmatest_verify(dst->aligned, 0, dst->off,
874                                 0, PATTERN_DST, false, is_memset);
875
876                 error_count += dmatest_verify(dst->aligned, dst->off,
877                                 dst->off + len, src->off,
878                                 PATTERN_SRC | PATTERN_COPY, false, is_memset);
879
880                 error_count += dmatest_verify(dst->aligned, dst->off + len,
881                                 buf_size, dst->off + len,
882                                 PATTERN_DST, false, is_memset);
883
884                 diff = ktime_sub(ktime_get(), start);
885                 comparetime = ktime_add(comparetime, diff);
886
887                 if (error_count) {
888                         result("data error", total_tests, src->off, dst->off,
889                                len, error_count);
890                         failed_tests++;
891                 } else {
892                         verbose_result("test passed", total_tests, src->off,
893                                        dst->off, len, 0);
894                 }
895
896                 continue;
897
898 error_unmap_continue:
899                 dmaengine_unmap_put(um);
900                 failed_tests++;
901         }
902         ktime = ktime_sub(ktime_get(), ktime);
903         ktime = ktime_sub(ktime, comparetime);
904         ktime = ktime_sub(ktime, filltime);
905         runtime = ktime_to_us(ktime);
906
907         ret = 0;
908         kfree(dma_pq);
909 err_srcs_array:
910         kfree(srcs);
911 err_dst:
912         dmatest_free_test_data(dst);
913 err_src:
914         dmatest_free_test_data(src);
915 err_free_coefs:
916         kfree(pq_coefs);
917 err_thread_type:
918         iops = dmatest_persec(runtime, total_tests);
919         pr_info("%s: summary %u tests, %u failures %llu.%02llu iops %llu KB/s (%d)\n",
920                 current->comm, total_tests, failed_tests,
921                 FIXPT_TO_INT(iops), FIXPT_GET_FRAC(iops),
922                 dmatest_KBs(runtime, total_len), ret);
923
924         /* terminate all transfers on specified channels */
925         if (ret || failed_tests)
926                 dmaengine_terminate_sync(chan);
927
928         thread->done = true;
929         wake_up(&thread_wait);
930
931         return ret;
932 }
933
934 static void dmatest_cleanup_channel(struct dmatest_chan *dtc)
935 {
936         struct dmatest_thread   *thread;
937         struct dmatest_thread   *_thread;
938         int                     ret;
939
940         list_for_each_entry_safe(thread, _thread, &dtc->threads, node) {
941                 ret = kthread_stop(thread->task);
942                 pr_debug("thread %s exited with status %d\n",
943                          thread->task->comm, ret);
944                 list_del(&thread->node);
945                 put_task_struct(thread->task);
946                 kfree(thread);
947         }
948
949         /* terminate all transfers on specified channels */
950         dmaengine_terminate_sync(dtc->chan);
951
952         kfree(dtc);
953 }
954
955 static int dmatest_add_threads(struct dmatest_info *info,
956                 struct dmatest_chan *dtc, enum dma_transaction_type type)
957 {
958         struct dmatest_params *params = &info->params;
959         struct dmatest_thread *thread;
960         struct dma_chan *chan = dtc->chan;
961         char *op;
962         unsigned int i;
963
964         if (type == DMA_MEMCPY)
965                 op = "copy";
966         else if (type == DMA_MEMSET)
967                 op = "set";
968         else if (type == DMA_XOR)
969                 op = "xor";
970         else if (type == DMA_PQ)
971                 op = "pq";
972         else
973                 return -EINVAL;
974
975         for (i = 0; i < params->threads_per_chan; i++) {
976                 thread = kzalloc(sizeof(struct dmatest_thread), GFP_KERNEL);
977                 if (!thread) {
978                         pr_warn("No memory for %s-%s%u\n",
979                                 dma_chan_name(chan), op, i);
980                         break;
981                 }
982                 thread->info = info;
983                 thread->chan = dtc->chan;
984                 thread->type = type;
985                 thread->test_done.wait = &thread->done_wait;
986                 init_waitqueue_head(&thread->done_wait);
987                 smp_wmb();
988                 thread->task = kthread_create(dmatest_func, thread, "%s-%s%u",
989                                 dma_chan_name(chan), op, i);
990                 if (IS_ERR(thread->task)) {
991                         pr_warn("Failed to create thread %s-%s%u\n",
992                                 dma_chan_name(chan), op, i);
993                         kfree(thread);
994                         break;
995                 }
996
997                 /* srcbuf and dstbuf are allocated by the thread itself */
998                 get_task_struct(thread->task);
999                 list_add_tail(&thread->node, &dtc->threads);
1000                 thread->pending = true;
1001         }
1002
1003         return i;
1004 }
1005
1006 static int dmatest_add_channel(struct dmatest_info *info,
1007                 struct dma_chan *chan)
1008 {
1009         struct dmatest_chan     *dtc;
1010         struct dma_device       *dma_dev = chan->device;
1011         unsigned int            thread_count = 0;
1012         int cnt;
1013
1014         dtc = kmalloc(sizeof(struct dmatest_chan), GFP_KERNEL);
1015         if (!dtc) {
1016                 pr_warn("No memory for %s\n", dma_chan_name(chan));
1017                 return -ENOMEM;
1018         }
1019
1020         dtc->chan = chan;
1021         INIT_LIST_HEAD(&dtc->threads);
1022
1023         if (dma_has_cap(DMA_COMPLETION_NO_ORDER, dma_dev->cap_mask) &&
1024             info->params.polled) {
1025                 info->params.polled = false;
1026                 pr_warn("DMA_COMPLETION_NO_ORDER, polled disabled\n");
1027         }
1028
1029         if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
1030                 if (dmatest == 0) {
1031                         cnt = dmatest_add_threads(info, dtc, DMA_MEMCPY);
1032                         thread_count += cnt > 0 ? cnt : 0;
1033                 }
1034         }
1035
1036         if (dma_has_cap(DMA_MEMSET, dma_dev->cap_mask)) {
1037                 if (dmatest == 1) {
1038                         cnt = dmatest_add_threads(info, dtc, DMA_MEMSET);
1039                         thread_count += cnt > 0 ? cnt : 0;
1040                 }
1041         }
1042
1043         if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
1044                 cnt = dmatest_add_threads(info, dtc, DMA_XOR);
1045                 thread_count += cnt > 0 ? cnt : 0;
1046         }
1047         if (dma_has_cap(DMA_PQ, dma_dev->cap_mask)) {
1048                 cnt = dmatest_add_threads(info, dtc, DMA_PQ);
1049                 thread_count += cnt > 0 ? cnt : 0;
1050         }
1051
1052         pr_info("Added %u threads using %s\n",
1053                 thread_count, dma_chan_name(chan));
1054
1055         list_add_tail(&dtc->node, &info->channels);
1056         info->nr_channels++;
1057
1058         return 0;
1059 }
1060
1061 static bool filter(struct dma_chan *chan, void *param)
1062 {
1063         return dmatest_match_channel(param, chan) && dmatest_match_device(param, chan->device);
1064 }
1065
1066 static void request_channels(struct dmatest_info *info,
1067                              enum dma_transaction_type type)
1068 {
1069         dma_cap_mask_t mask;
1070
1071         dma_cap_zero(mask);
1072         dma_cap_set(type, mask);
1073         for (;;) {
1074                 struct dmatest_params *params = &info->params;
1075                 struct dma_chan *chan;
1076
1077                 chan = dma_request_channel(mask, filter, params);
1078                 if (chan) {
1079                         if (dmatest_add_channel(info, chan)) {
1080                                 dma_release_channel(chan);
1081                                 break; /* add_channel failed, punt */
1082                         }
1083                 } else
1084                         break; /* no more channels available */
1085                 if (params->max_channels &&
1086                     info->nr_channels >= params->max_channels)
1087                         break; /* we have all we need */
1088         }
1089 }
1090
1091 static void add_threaded_test(struct dmatest_info *info)
1092 {
1093         struct dmatest_params *params = &info->params;
1094
1095         /* Copy test parameters */
1096         params->buf_size = test_buf_size;
1097         strscpy(params->channel, strim(test_channel), sizeof(params->channel));
1098         strscpy(params->device, strim(test_device), sizeof(params->device));
1099         params->threads_per_chan = threads_per_chan;
1100         params->max_channels = max_channels;
1101         params->iterations = iterations;
1102         params->xor_sources = xor_sources;
1103         params->pq_sources = pq_sources;
1104         params->timeout = timeout;
1105         params->noverify = noverify;
1106         params->norandom = norandom;
1107         params->alignment = alignment;
1108         params->transfer_size = transfer_size;
1109         params->polled = polled;
1110
1111         request_channels(info, DMA_MEMCPY);
1112         request_channels(info, DMA_MEMSET);
1113         request_channels(info, DMA_XOR);
1114         request_channels(info, DMA_PQ);
1115 }
1116
1117 static void run_pending_tests(struct dmatest_info *info)
1118 {
1119         struct dmatest_chan *dtc;
1120         unsigned int thread_count = 0;
1121
1122         list_for_each_entry(dtc, &info->channels, node) {
1123                 struct dmatest_thread *thread;
1124
1125                 thread_count = 0;
1126                 list_for_each_entry(thread, &dtc->threads, node) {
1127                         wake_up_process(thread->task);
1128                         thread_count++;
1129                 }
1130                 pr_info("Started %u threads using %s\n",
1131                         thread_count, dma_chan_name(dtc->chan));
1132         }
1133 }
1134
1135 static void stop_threaded_test(struct dmatest_info *info)
1136 {
1137         struct dmatest_chan *dtc, *_dtc;
1138         struct dma_chan *chan;
1139
1140         list_for_each_entry_safe(dtc, _dtc, &info->channels, node) {
1141                 list_del(&dtc->node);
1142                 chan = dtc->chan;
1143                 dmatest_cleanup_channel(dtc);
1144                 pr_debug("dropped channel %s\n", dma_chan_name(chan));
1145                 dma_release_channel(chan);
1146         }
1147
1148         info->nr_channels = 0;
1149 }
1150
1151 static void start_threaded_tests(struct dmatest_info *info)
1152 {
1153         /* we might be called early to set run=, defer running until all
1154          * parameters have been evaluated
1155          */
1156         if (!info->did_init)
1157                 return;
1158
1159         run_pending_tests(info);
1160 }
1161
1162 static int dmatest_run_get(char *val, const struct kernel_param *kp)
1163 {
1164         struct dmatest_info *info = &test_info;
1165
1166         mutex_lock(&info->lock);
1167         if (is_threaded_test_run(info)) {
1168                 dmatest_run = true;
1169         } else {
1170                 if (!is_threaded_test_pending(info))
1171                         stop_threaded_test(info);
1172                 dmatest_run = false;
1173         }
1174         mutex_unlock(&info->lock);
1175
1176         return param_get_bool(val, kp);
1177 }
1178
1179 static int dmatest_run_set(const char *val, const struct kernel_param *kp)
1180 {
1181         struct dmatest_info *info = &test_info;
1182         int ret;
1183
1184         mutex_lock(&info->lock);
1185         ret = param_set_bool(val, kp);
1186         if (ret) {
1187                 mutex_unlock(&info->lock);
1188                 return ret;
1189         } else if (dmatest_run) {
1190                 if (!is_threaded_test_pending(info)) {
1191                         /*
1192                          * We have nothing to run. This can be due to:
1193                          */
1194                         ret = info->last_error;
1195                         if (ret) {
1196                                 /* 1) Misconfiguration */
1197                                 pr_err("Channel misconfigured, can't continue\n");
1198                                 mutex_unlock(&info->lock);
1199                                 return ret;
1200                         } else {
1201                                 /* 2) We rely on defaults */
1202                                 pr_info("No channels configured, continue with any\n");
1203                                 if (!is_threaded_test_run(info))
1204                                         stop_threaded_test(info);
1205                                 add_threaded_test(info);
1206                         }
1207                 }
1208                 start_threaded_tests(info);
1209         } else {
1210                 stop_threaded_test(info);
1211         }
1212
1213         mutex_unlock(&info->lock);
1214
1215         return ret;
1216 }
1217
1218 static int dmatest_chan_set(const char *val, const struct kernel_param *kp)
1219 {
1220         struct dmatest_info *info = &test_info;
1221         struct dmatest_chan *dtc;
1222         char chan_reset_val[20];
1223         int ret;
1224
1225         mutex_lock(&info->lock);
1226         ret = param_set_copystring(val, kp);
1227         if (ret) {
1228                 mutex_unlock(&info->lock);
1229                 return ret;
1230         }
1231         /*Clear any previously run threads */
1232         if (!is_threaded_test_run(info) && !is_threaded_test_pending(info))
1233                 stop_threaded_test(info);
1234         /* Reject channels that are already registered */
1235         if (is_threaded_test_pending(info)) {
1236                 list_for_each_entry(dtc, &info->channels, node) {
1237                         if (strcmp(dma_chan_name(dtc->chan),
1238                                    strim(test_channel)) == 0) {
1239                                 dtc = list_last_entry(&info->channels,
1240                                                       struct dmatest_chan,
1241                                                       node);
1242                                 strscpy(chan_reset_val,
1243                                         dma_chan_name(dtc->chan),
1244                                         sizeof(chan_reset_val));
1245                                 ret = -EBUSY;
1246                                 goto add_chan_err;
1247                         }
1248                 }
1249         }
1250
1251         add_threaded_test(info);
1252
1253         /* Check if channel was added successfully */
1254         if (!list_empty(&info->channels)) {
1255                 /*
1256                  * if new channel was not successfully added, revert the
1257                  * "test_channel" string to the name of the last successfully
1258                  * added channel. exception for when users issues empty string
1259                  * to channel parameter.
1260                  */
1261                 dtc = list_last_entry(&info->channels, struct dmatest_chan, node);
1262                 if ((strcmp(dma_chan_name(dtc->chan), strim(test_channel)) != 0)
1263                     && (strcmp("", strim(test_channel)) != 0)) {
1264                         ret = -EINVAL;
1265                         strscpy(chan_reset_val, dma_chan_name(dtc->chan),
1266                                 sizeof(chan_reset_val));
1267                         goto add_chan_err;
1268                 }
1269
1270         } else {
1271                 /* Clear test_channel if no channels were added successfully */
1272                 strscpy(chan_reset_val, "", sizeof(chan_reset_val));
1273                 ret = -EBUSY;
1274                 goto add_chan_err;
1275         }
1276
1277         info->last_error = ret;
1278         mutex_unlock(&info->lock);
1279
1280         return ret;
1281
1282 add_chan_err:
1283         param_set_copystring(chan_reset_val, kp);
1284         info->last_error = ret;
1285         mutex_unlock(&info->lock);
1286
1287         return ret;
1288 }
1289
1290 static int dmatest_chan_get(char *val, const struct kernel_param *kp)
1291 {
1292         struct dmatest_info *info = &test_info;
1293
1294         mutex_lock(&info->lock);
1295         if (!is_threaded_test_run(info) && !is_threaded_test_pending(info)) {
1296                 stop_threaded_test(info);
1297                 strscpy(test_channel, "", sizeof(test_channel));
1298         }
1299         mutex_unlock(&info->lock);
1300
1301         return param_get_string(val, kp);
1302 }
1303
1304 static int dmatest_test_list_get(char *val, const struct kernel_param *kp)
1305 {
1306         struct dmatest_info *info = &test_info;
1307         struct dmatest_chan *dtc;
1308         unsigned int thread_count = 0;
1309
1310         list_for_each_entry(dtc, &info->channels, node) {
1311                 struct dmatest_thread *thread;
1312
1313                 thread_count = 0;
1314                 list_for_each_entry(thread, &dtc->threads, node) {
1315                         thread_count++;
1316                 }
1317                 pr_info("%u threads using %s\n",
1318                         thread_count, dma_chan_name(dtc->chan));
1319         }
1320
1321         return 0;
1322 }
1323
1324 static int __init dmatest_init(void)
1325 {
1326         struct dmatest_info *info = &test_info;
1327         struct dmatest_params *params = &info->params;
1328
1329         if (dmatest_run) {
1330                 mutex_lock(&info->lock);
1331                 add_threaded_test(info);
1332                 run_pending_tests(info);
1333                 mutex_unlock(&info->lock);
1334         }
1335
1336         if (params->iterations && wait)
1337                 wait_event(thread_wait, !is_threaded_test_run(info));
1338
1339         /* module parameters are stable, inittime tests are started,
1340          * let userspace take over 'run' control
1341          */
1342         info->did_init = true;
1343
1344         return 0;
1345 }
1346 /* when compiled-in wait for drivers to load first */
1347 late_initcall(dmatest_init);
1348
1349 static void __exit dmatest_exit(void)
1350 {
1351         struct dmatest_info *info = &test_info;
1352
1353         mutex_lock(&info->lock);
1354         stop_threaded_test(info);
1355         mutex_unlock(&info->lock);
1356 }
1357 module_exit(dmatest_exit);
1358
1359 MODULE_AUTHOR("Haavard Skinnemoen (Atmel)");
1360 MODULE_LICENSE("GPL v2");