drivers/gpu/host1x/cdma.c

   1 /*
   2  * Tegra host1x Command DMA
   3  *
   4  * Copyright (c) 2010-2013, NVIDIA Corporation.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms and conditions of the GNU General Public License,
   8  * version 2, as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope it will be useful, but WITHOUT
  11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  * more details.
  14  *
  15  * You should have received a copy of the GNU General Public License
  16  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  17  */
  18
  19
  20 #include <asm/cacheflush.h>
  21 #include <linux/device.h>
  22 #include <linux/dma-mapping.h>
  23 #include <linux/host1x.h>
  24 #include <linux/interrupt.h>
  25 #include <linux/kernel.h>
  26 #include <linux/kfifo.h>
  27 #include <linux/slab.h>
  28 #include <trace/events/host1x.h>
  29
  30 #include "cdma.h"
  31 #include "channel.h"
  32 #include "dev.h"
  33 #include "debug.h"
  34 #include "job.h"
  35
  36 /*
  37  * push_buffer
  38  *
  39  * The push buffer is a circular array of words to be fetched by command DMA.
  40  * Note that it works slightly differently to the sync queue; fence == pos
  41  * means that the push buffer is full, not empty.
  42  */
  43
  44 #define HOST1X_PUSHBUFFER_SLOTS 512
  45
  46 /*
  47  * Clean up push buffer resources
  48  */
  49 static void host1x_pushbuffer_destroy(struct push_buffer *pb)
  50 {
  51         struct host1x_cdma *cdma = pb_to_cdma(pb);
  52         struct host1x *host1x = cdma_to_host1x(cdma);
  53
  54         if (!pb->mapped)
  55                 return;
  56
  57         if (host1x->domain) {
  58                 iommu_unmap(host1x->domain, pb->dma, pb->alloc_size);
  59                 free_iova(&host1x->iova, iova_pfn(&host1x->iova, pb->dma));
  60         }
  61
  62         dma_free_wc(host1x->dev, pb->alloc_size, pb->mapped, pb->phys);
  63
  64         pb->mapped = NULL;
  65         pb->phys = 0;
  66 }
  67
  68 /*
  69  * Init push buffer resources
  70  */
  71 static int host1x_pushbuffer_init(struct push_buffer *pb)
  72 {
  73         struct host1x_cdma *cdma = pb_to_cdma(pb);
  74         struct host1x *host1x = cdma_to_host1x(cdma);
  75         struct iova *alloc;
  76         u32 size;
  77         int err;
  78
  79         pb->mapped = NULL;
  80         pb->phys = 0;
  81         pb->size = HOST1X_PUSHBUFFER_SLOTS * 8;
  82
  83         size = pb->size + 4;
  84
  85         /* initialize buffer pointers */
  86         pb->fence = pb->size - 8;
  87         pb->pos = 0;
  88
  89         if (host1x->domain) {
  90                 unsigned long shift;
  91
  92                 size = iova_align(&host1x->iova, size);
  93
  94                 pb->mapped = dma_alloc_wc(host1x->dev, size, &pb->phys,
  95                                           GFP_KERNEL);
  96                 if (!pb->mapped)
  97                         return -ENOMEM;
  98
  99                 shift = iova_shift(&host1x->iova);
 100                 alloc = alloc_iova(&host1x->iova, size >> shift,
 101                                    host1x->iova_end >> shift, true);
 102                 if (!alloc) {
 103                         err = -ENOMEM;
 104                         goto iommu_free_mem;
 105                 }
 106
 107                 pb->dma = iova_dma_addr(&host1x->iova, alloc);
 108                 err = iommu_map(host1x->domain, pb->dma, pb->phys, size,
 109                                 IOMMU_READ);
 110                 if (err)
 111                         goto iommu_free_iova;
 112         } else {
 113                 pb->mapped = dma_alloc_wc(host1x->dev, size, &pb->phys,
 114                                           GFP_KERNEL);
 115                 if (!pb->mapped)
 116                         return -ENOMEM;
 117
 118                 pb->dma = pb->phys;
 119         }
 120
 121         pb->alloc_size = size;
 122
 123         host1x_hw_pushbuffer_init(host1x, pb);
 124
 125         return 0;
 126
 127 iommu_free_iova:
 128         __free_iova(&host1x->iova, alloc);
 129 iommu_free_mem:
 130         dma_free_wc(host1x->dev, size, pb->mapped, pb->phys);
 131
 132         return err;
 133 }
 134
 135 /*
 136  * Push two words to the push buffer
 137  * Caller must ensure push buffer is not full
 138  */
 139 static void host1x_pushbuffer_push(struct push_buffer *pb, u32 op1, u32 op2)
 140 {
 141         u32 *p = (u32 *)((void *)pb->mapped + pb->pos);
 142
 143         WARN_ON(pb->pos == pb->fence);
 144         *(p++) = op1;
 145         *(p++) = op2;
 146         pb->pos = (pb->pos + 8) & (pb->size - 1);
 147 }
 148
 149 /*
 150  * Pop a number of two word slots from the push buffer
 151  * Caller must ensure push buffer is not empty
 152  */
 153 static void host1x_pushbuffer_pop(struct push_buffer *pb, unsigned int slots)
 154 {
 155         /* Advance the next write position */
 156         pb->fence = (pb->fence + slots * 8) & (pb->size - 1);
 157 }
 158
 159 /*
 160  * Return the number of two word slots free in the push buffer
 161  */
 162 static u32 host1x_pushbuffer_space(struct push_buffer *pb)
 163 {
 164         return ((pb->fence - pb->pos) & (pb->size - 1)) / 8;
 165 }
 166
 167 /*
 168  * Sleep (if necessary) until the requested event happens
 169  *   - CDMA_EVENT_SYNC_QUEUE_EMPTY : sync queue is completely empty.
 170  *     - Returns 1
 171  *   - CDMA_EVENT_PUSH_BUFFER_SPACE : there is space in the push buffer
 172  *     - Return the amount of space (> 0)
 173  * Must be called with the cdma lock held.
 174  */
 175 unsigned int host1x_cdma_wait_locked(struct host1x_cdma *cdma,
 176                                      enum cdma_event event)
 177 {
 178         for (;;) {
 179                 struct push_buffer *pb = &cdma->push_buffer;
 180                 unsigned int space;
 181
 182                 switch (event) {
 183                 case CDMA_EVENT_SYNC_QUEUE_EMPTY:
 184                         space = list_empty(&cdma->sync_queue) ? 1 : 0;
 185                         break;
 186
 187                 case CDMA_EVENT_PUSH_BUFFER_SPACE:
 188                         space = host1x_pushbuffer_space(pb);
 189                         break;
 190
 191                 default:
 192                         WARN_ON(1);
 193                         return -EINVAL;
 194                 }
 195
 196                 if (space)
 197                         return space;
 198
 199                 trace_host1x_wait_cdma(dev_name(cdma_to_channel(cdma)->dev),
 200                                        event);
 201
 202                 /* If somebody has managed to already start waiting, yield */
 203                 if (cdma->event != CDMA_EVENT_NONE) {
 204                         mutex_unlock(&cdma->lock);
 205                         schedule();
 206                         mutex_lock(&cdma->lock);
 207                         continue;
 208                 }
 209
 210                 cdma->event = event;
 211
 212                 mutex_unlock(&cdma->lock);
 213                 down(&cdma->sem);
 214                 mutex_lock(&cdma->lock);
 215         }
 216
 217         return 0;
 218 }
 219
 220 /*
 221  * Start timer that tracks the time spent by the job.
 222  * Must be called with the cdma lock held.
 223  */
 224 static void cdma_start_timer_locked(struct host1x_cdma *cdma,
 225                                     struct host1x_job *job)
 226 {
 227         struct host1x *host = cdma_to_host1x(cdma);
 228
 229         if (cdma->timeout.client) {
 230                 /* timer already started */
 231                 return;
 232         }
 233
 234         cdma->timeout.client = job->client;
 235         cdma->timeout.syncpt = host1x_syncpt_get(host, job->syncpt_id);
 236         cdma->timeout.syncpt_val = job->syncpt_end;
 237         cdma->timeout.start_ktime = ktime_get();
 238
 239         schedule_delayed_work(&cdma->timeout.wq,
 240                               msecs_to_jiffies(job->timeout));
 241 }
 242
 243 /*
 244  * Stop timer when a buffer submission completes.
 245  * Must be called with the cdma lock held.
 246  */
 247 static void stop_cdma_timer_locked(struct host1x_cdma *cdma)
 248 {
 249         cancel_delayed_work(&cdma->timeout.wq);
 250         cdma->timeout.client = NULL;
 251 }
 252
 253 /*
 254  * For all sync queue entries that have already finished according to the
 255  * current sync point registers:
 256  *  - unpin & unref their mems
 257  *  - pop their push buffer slots
 258  *  - remove them from the sync queue
 259  * This is normally called from the host code's worker thread, but can be
 260  * called manually if necessary.
 261  * Must be called with the cdma lock held.
 262  */
 263 static void update_cdma_locked(struct host1x_cdma *cdma)
 264 {
 265         bool signal = false;
 266         struct host1x *host1x = cdma_to_host1x(cdma);
 267         struct host1x_job *job, *n;
 268
 269         /* If CDMA is stopped, queue is cleared and we can return */
 270         if (!cdma->running)
 271                 return;
 272
 273         /*
 274          * Walk the sync queue, reading the sync point registers as necessary,
 275          * to consume as many sync queue entries as possible without blocking
 276          */
 277         list_for_each_entry_safe(job, n, &cdma->sync_queue, list) {
 278                 struct host1x_syncpt *sp =
 279                         host1x_syncpt_get(host1x, job->syncpt_id);
 280
 281                 /* Check whether this syncpt has completed, and bail if not */
 282                 if (!host1x_syncpt_is_expired(sp, job->syncpt_end)) {
 283                         /* Start timer on next pending syncpt */
 284                         if (job->timeout)
 285                                 cdma_start_timer_locked(cdma, job);
 286
 287                         break;
 288                 }
 289
 290                 /* Cancel timeout, when a buffer completes */
 291                 if (cdma->timeout.client)
 292                         stop_cdma_timer_locked(cdma);
 293
 294                 /* Unpin the memory */
 295                 host1x_job_unpin(job);
 296
 297                 /* Pop push buffer slots */
 298                 if (job->num_slots) {
 299                         struct push_buffer *pb = &cdma->push_buffer;
 300
 301                         host1x_pushbuffer_pop(pb, job->num_slots);
 302
 303                         if (cdma->event == CDMA_EVENT_PUSH_BUFFER_SPACE)
 304                                 signal = true;
 305                 }
 306
 307                 list_del(&job->list);
 308                 host1x_job_put(job);
 309         }
 310
 311         if (cdma->event == CDMA_EVENT_SYNC_QUEUE_EMPTY &&
 312             list_empty(&cdma->sync_queue))
 313                 signal = true;
 314
 315         if (signal) {
 316                 cdma->event = CDMA_EVENT_NONE;
 317                 up(&cdma->sem);
 318         }
 319 }
 320
 321 void host1x_cdma_update_sync_queue(struct host1x_cdma *cdma,
 322                                    struct device *dev)
 323 {
 324         struct host1x *host1x = cdma_to_host1x(cdma);
 325         u32 restart_addr, syncpt_incrs, syncpt_val;
 326         struct host1x_job *job = NULL;
 327
 328         syncpt_val = host1x_syncpt_load(cdma->timeout.syncpt);
 329
 330         dev_dbg(dev, "%s: starting cleanup (thresh %d)\n",
 331                 __func__, syncpt_val);
 332
 333         /*
 334          * Move the sync_queue read pointer to the first entry that hasn't
 335          * completed based on the current HW syncpt value. It's likely there
 336          * won't be any (i.e. we're still at the head), but covers the case
 337          * where a syncpt incr happens just prior/during the teardown.
 338          */
 339
 340         dev_dbg(dev, "%s: skip completed buffers still in sync_queue\n",
 341                 __func__);
 342
 343         list_for_each_entry(job, &cdma->sync_queue, list) {
 344                 if (syncpt_val < job->syncpt_end)
 345                         break;
 346
 347                 host1x_job_dump(dev, job);
 348         }
 349
 350         /*
 351          * Walk the sync_queue, first incrementing with the CPU syncpts that
 352          * are partially executed (the first buffer) or fully skipped while
 353          * still in the current context (slots are also NOP-ed).
 354          *
 355          * At the point contexts are interleaved, syncpt increments must be
 356          * done inline with the pushbuffer from a GATHER buffer to maintain
 357          * the order (slots are modified to be a GATHER of syncpt incrs).
 358          *
 359          * Note: save in restart_addr the location where the timed out buffer
 360          * started in the PB, so we can start the refetch from there (with the
 361          * modified NOP-ed PB slots). This lets things appear to have completed
 362          * properly for this buffer and resources are freed.
 363          */
 364
 365         dev_dbg(dev, "%s: perform CPU incr on pending same ctx buffers\n",
 366                 __func__);
 367
 368         if (!list_empty(&cdma->sync_queue))
 369                 restart_addr = job->first_get;
 370         else
 371                 restart_addr = cdma->last_pos;
 372
 373         /* do CPU increments as long as this context continues */
 374         list_for_each_entry_from(job, &cdma->sync_queue, list) {
 375                 /* different context, gets us out of this loop */
 376                 if (job->client != cdma->timeout.client)
 377                         break;
 378
 379                 /* won't need a timeout when replayed */
 380                 job->timeout = 0;
 381
 382                 syncpt_incrs = job->syncpt_end - syncpt_val;
 383                 dev_dbg(dev, "%s: CPU incr (%d)\n", __func__, syncpt_incrs);
 384
 385                 host1x_job_dump(dev, job);
 386
 387                 /* safe to use CPU to incr syncpts */
 388                 host1x_hw_cdma_timeout_cpu_incr(host1x, cdma, job->first_get,
 389                                                 syncpt_incrs, job->syncpt_end,
 390                                                 job->num_slots);
 391
 392                 syncpt_val += syncpt_incrs;
 393         }
 394
 395         /*
 396          * The following sumbits from the same client may be dependent on the
 397          * failed submit and therefore they may fail. Force a small timeout
 398          * to make the queue cleanup faster.
 399          */
 400
 401         list_for_each_entry_from(job, &cdma->sync_queue, list)
 402                 if (job->client == cdma->timeout.client)
 403                         job->timeout = min_t(unsigned int, job->timeout, 500);
 404
 405         dev_dbg(dev, "%s: finished sync_queue modification\n", __func__);
 406
 407         /* roll back DMAGET and start up channel again */
 408         host1x_hw_cdma_resume(host1x, cdma, restart_addr);
 409 }
 410
 411 /*
 412  * Create a cdma
 413  */
 414 int host1x_cdma_init(struct host1x_cdma *cdma)
 415 {
 416         int err;
 417
 418         mutex_init(&cdma->lock);
 419         sema_init(&cdma->sem, 0);
 420
 421         INIT_LIST_HEAD(&cdma->sync_queue);
 422
 423         cdma->event = CDMA_EVENT_NONE;
 424         cdma->running = false;
 425         cdma->torndown = false;
 426
 427         err = host1x_pushbuffer_init(&cdma->push_buffer);
 428         if (err)
 429                 return err;
 430
 431         return 0;
 432 }
 433
 434 /*
 435  * Destroy a cdma
 436  */
 437 int host1x_cdma_deinit(struct host1x_cdma *cdma)
 438 {
 439         struct push_buffer *pb = &cdma->push_buffer;
 440         struct host1x *host1x = cdma_to_host1x(cdma);
 441
 442         if (cdma->running) {
 443                 pr_warn("%s: CDMA still running\n", __func__);
 444                 return -EBUSY;
 445         }
 446
 447         host1x_pushbuffer_destroy(pb);
 448         host1x_hw_cdma_timeout_destroy(host1x, cdma);
 449
 450         return 0;
 451 }
 452
 453 /*
 454  * Begin a cdma submit
 455  */
 456 int host1x_cdma_begin(struct host1x_cdma *cdma, struct host1x_job *job)
 457 {
 458         struct host1x *host1x = cdma_to_host1x(cdma);
 459
 460         mutex_lock(&cdma->lock);
 461
 462         if (job->timeout) {
 463                 /* init state on first submit with timeout value */
 464                 if (!cdma->timeout.initialized) {
 465                         int err;
 466
 467                         err = host1x_hw_cdma_timeout_init(host1x, cdma,
 468                                                           job->syncpt_id);
 469                         if (err) {
 470                                 mutex_unlock(&cdma->lock);
 471                                 return err;
 472                         }
 473                 }
 474         }
 475
 476         if (!cdma->running)
 477                 host1x_hw_cdma_start(host1x, cdma);
 478
 479         cdma->slots_free = 0;
 480         cdma->slots_used = 0;
 481         cdma->first_get = cdma->push_buffer.pos;
 482
 483         trace_host1x_cdma_begin(dev_name(job->channel->dev));
 484         return 0;
 485 }
 486
 487 /*
 488  * Push two words into a push buffer slot
 489  * Blocks as necessary if the push buffer is full.
 490  */
 491 void host1x_cdma_push(struct host1x_cdma *cdma, u32 op1, u32 op2)
 492 {
 493         struct host1x *host1x = cdma_to_host1x(cdma);
 494         struct push_buffer *pb = &cdma->push_buffer;
 495         u32 slots_free = cdma->slots_free;
 496
 497         if (host1x_debug_trace_cmdbuf)
 498                 trace_host1x_cdma_push(dev_name(cdma_to_channel(cdma)->dev),
 499                                        op1, op2);
 500
 501         if (slots_free == 0) {
 502                 host1x_hw_cdma_flush(host1x, cdma);
 503                 slots_free = host1x_cdma_wait_locked(cdma,
 504                                                 CDMA_EVENT_PUSH_BUFFER_SPACE);
 505         }
 506
 507         cdma->slots_free = slots_free - 1;
 508         cdma->slots_used++;
 509         host1x_pushbuffer_push(pb, op1, op2);
 510 }
 511
 512 /*
 513  * End a cdma submit
 514  * Kick off DMA, add job to the sync queue, and a number of slots to be freed
 515  * from the pushbuffer. The handles for a submit must all be pinned at the same
 516  * time, but they can be unpinned in smaller chunks.
 517  */
 518 void host1x_cdma_end(struct host1x_cdma *cdma,
 519                      struct host1x_job *job)
 520 {
 521         struct host1x *host1x = cdma_to_host1x(cdma);
 522         bool idle = list_empty(&cdma->sync_queue);
 523
 524         host1x_hw_cdma_flush(host1x, cdma);
 525
 526         job->first_get = cdma->first_get;
 527         job->num_slots = cdma->slots_used;
 528         host1x_job_get(job);
 529         list_add_tail(&job->list, &cdma->sync_queue);
 530
 531         /* start timer on idle -> active transitions */
 532         if (job->timeout && idle)
 533                 cdma_start_timer_locked(cdma, job);
 534
 535         trace_host1x_cdma_end(dev_name(job->channel->dev));
 536         mutex_unlock(&cdma->lock);
 537 }
 538
 539 /*
 540  * Update cdma state according to current sync point values
 541  */
 542 void host1x_cdma_update(struct host1x_cdma *cdma)
 543 {
 544         mutex_lock(&cdma->lock);
 545         update_cdma_locked(cdma);
 546         mutex_unlock(&cdma->lock);
 547 }