Home | History | Annotate | Download | only in nouveau
      1 /*
      2  * Copyright 2012 Red Hat Inc.
      3  *
      4  * Permission is hereby granted, free of charge, to any person obtaining a
      5  * copy of this software and associated documentation files (the "Software"),
      6  * to deal in the Software without restriction, including without limitation
      7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
      8  * and/or sell copies of the Software, and to permit persons to whom the
      9  * Software is furnished to do so, subject to the following conditions:
     10  *
     11  * The above copyright notice and this permission notice shall be included in
     12  * all copies or substantial portions of the Software.
     13  *
     14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
     15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
     16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
     17  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
     18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
     19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
     20  * OTHER DEALINGS IN THE SOFTWARE.
     21  *
     22  * Authors: Ben Skeggs
     23  */
     24 
     25 #ifdef HAVE_CONFIG_H
     26 #include <config.h>
     27 #endif
     28 
     29 #include <stdio.h>
     30 #include <stdlib.h>
     31 #include <stdint.h>
     32 #include <stdbool.h>
     33 #include <string.h>
     34 #include <assert.h>
     35 #include <errno.h>
     36 
     37 #include <xf86drm.h>
     38 #include <xf86atomic.h>
     39 #include "libdrm_lists.h"
     40 #include "nouveau_drm.h"
     41 
     42 #include "nouveau.h"
     43 #include "private.h"
     44 
     45 struct nouveau_pushbuf_krec {
     46 	struct nouveau_pushbuf_krec *next;
     47 	struct drm_nouveau_gem_pushbuf_bo buffer[NOUVEAU_GEM_MAX_BUFFERS];
     48 	struct drm_nouveau_gem_pushbuf_reloc reloc[NOUVEAU_GEM_MAX_RELOCS];
     49 	struct drm_nouveau_gem_pushbuf_push push[NOUVEAU_GEM_MAX_PUSH];
     50 	int nr_buffer;
     51 	int nr_reloc;
     52 	int nr_push;
     53 	uint64_t vram_used;
     54 	uint64_t gart_used;
     55 };
     56 
     57 struct nouveau_pushbuf_priv {
     58 	struct nouveau_pushbuf base;
     59 	struct nouveau_pushbuf_krec *list;
     60 	struct nouveau_pushbuf_krec *krec;
     61 	struct nouveau_list bctx_list;
     62 	struct nouveau_bo *bo;
     63 	uint32_t type;
     64 	uint32_t suffix0;
     65 	uint32_t suffix1;
     66 	uint32_t *ptr;
     67 	uint32_t *bgn;
     68 	int bo_next;
     69 	int bo_nr;
     70 	struct nouveau_bo *bos[];
     71 };
     72 
     73 static inline struct nouveau_pushbuf_priv *
     74 nouveau_pushbuf(struct nouveau_pushbuf *push)
     75 {
     76 	return (struct nouveau_pushbuf_priv *)push;
     77 }
     78 
     79 static int pushbuf_validate(struct nouveau_pushbuf *, bool);
     80 static int pushbuf_flush(struct nouveau_pushbuf *);
     81 
     82 static bool
     83 pushbuf_kref_fits(struct nouveau_pushbuf *push, struct nouveau_bo *bo,
     84 		  uint32_t *domains)
     85 {
     86 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
     87 	struct nouveau_pushbuf_krec *krec = nvpb->krec;
     88 	struct nouveau_device *dev = push->client->device;
     89 	struct nouveau_bo *kbo;
     90 	struct drm_nouveau_gem_pushbuf_bo *kref;
     91 	int i;
     92 
     93 	/* VRAM is the only valid domain.  GART and VRAM|GART buffers
     94 	 * are all accounted to GART, so if this doesn't fit in VRAM
     95 	 * straight up, a flush is needed.
     96 	 */
     97 	if (*domains == NOUVEAU_GEM_DOMAIN_VRAM) {
     98 		if (krec->vram_used + bo->size > dev->vram_limit)
     99 			return false;
    100 		krec->vram_used += bo->size;
    101 		return true;
    102 	}
    103 
    104 	/* GART or VRAM|GART buffer.  Account both of these buffer types
    105 	 * to GART only for the moment, which simplifies things.  If the
    106 	 * buffer can fit already, we're done here.
    107 	 */
    108 	if (krec->gart_used + bo->size <= dev->gart_limit) {
    109 		krec->gart_used += bo->size;
    110 		return true;
    111 	}
    112 
    113 	/* Ran out of GART space, if it's a VRAM|GART buffer and it'll
    114 	 * fit into available VRAM, turn it into a VRAM buffer
    115 	 */
    116 	if ((*domains & NOUVEAU_GEM_DOMAIN_VRAM) &&
    117 	    krec->vram_used + bo->size <= dev->vram_limit) {
    118 		*domains &= NOUVEAU_GEM_DOMAIN_VRAM;
    119 		krec->vram_used += bo->size;
    120 		return true;
    121 	}
    122 
    123 	/* Still couldn't fit the buffer in anywhere, so as a last resort;
    124 	 * scan the buffer list for VRAM|GART buffers and turn them into
    125 	 * VRAM buffers until we have enough space in GART for this one
    126 	 */
    127 	kref = krec->buffer;
    128 	for (i = 0; i < krec->nr_buffer; i++, kref++) {
    129 		if (!(kref->valid_domains & NOUVEAU_GEM_DOMAIN_GART))
    130 			continue;
    131 
    132 		kbo = (void *)(unsigned long)kref->user_priv;
    133 		if (!(kref->valid_domains & NOUVEAU_GEM_DOMAIN_VRAM) ||
    134 		    krec->vram_used + kbo->size > dev->vram_limit)
    135 			continue;
    136 
    137 		kref->valid_domains &= NOUVEAU_GEM_DOMAIN_VRAM;
    138 		krec->gart_used -= kbo->size;
    139 		krec->vram_used += kbo->size;
    140 		if (krec->gart_used + bo->size <= dev->gart_limit) {
    141 			krec->gart_used += bo->size;
    142 			return true;
    143 		}
    144 	}
    145 
    146 	/* Couldn't resolve a placement, need to force a flush */
    147 	return false;
    148 }
    149 
    150 static struct drm_nouveau_gem_pushbuf_bo *
    151 pushbuf_kref(struct nouveau_pushbuf *push, struct nouveau_bo *bo,
    152 	     uint32_t flags)
    153 {
    154 	struct nouveau_device *dev = push->client->device;
    155 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
    156 	struct nouveau_pushbuf_krec *krec = nvpb->krec;
    157 	struct nouveau_pushbuf *fpush;
    158 	struct drm_nouveau_gem_pushbuf_bo *kref;
    159 	uint32_t domains, domains_wr, domains_rd;
    160 
    161 	domains = 0;
    162 	if (flags & NOUVEAU_BO_VRAM)
    163 		domains |= NOUVEAU_GEM_DOMAIN_VRAM;
    164 	if (flags & NOUVEAU_BO_GART)
    165 		domains |= NOUVEAU_GEM_DOMAIN_GART;
    166 	domains_wr = domains * !!(flags & NOUVEAU_BO_WR);
    167 	domains_rd = domains * !!(flags & NOUVEAU_BO_RD);
    168 
    169 	/* if buffer is referenced on another pushbuf that is owned by the
    170 	 * same client, we need to flush the other pushbuf first to ensure
    171 	 * the correct ordering of commands
    172 	 */
    173 	fpush = cli_push_get(push->client, bo);
    174 	if (fpush && fpush != push)
    175 		pushbuf_flush(fpush);
    176 
    177 	kref = cli_kref_get(push->client, bo);
    178 	if (kref) {
    179 		/* possible conflict in memory types - flush and retry */
    180 		if (!(kref->valid_domains & domains))
    181 			return NULL;
    182 
    183 		/* VRAM|GART buffer turning into a VRAM buffer.  Make sure
    184 		 * it'll fit in VRAM and force a flush if not.
    185 		 */
    186 		if ((kref->valid_domains  & NOUVEAU_GEM_DOMAIN_GART) &&
    187 		    (            domains == NOUVEAU_GEM_DOMAIN_VRAM)) {
    188 			if (krec->vram_used + bo->size > dev->vram_limit)
    189 				return NULL;
    190 			krec->vram_used += bo->size;
    191 			krec->gart_used -= bo->size;
    192 		}
    193 
    194 		kref->valid_domains &= domains;
    195 		kref->write_domains |= domains_wr;
    196 		kref->read_domains  |= domains_rd;
    197 	} else {
    198 		if (krec->nr_buffer == NOUVEAU_GEM_MAX_BUFFERS ||
    199 		    !pushbuf_kref_fits(push, bo, &domains))
    200 			return NULL;
    201 
    202 		kref = &krec->buffer[krec->nr_buffer++];
    203 		kref->user_priv = (unsigned long)bo;
    204 		kref->handle = bo->handle;
    205 		kref->valid_domains = domains;
    206 		kref->write_domains = domains_wr;
    207 		kref->read_domains = domains_rd;
    208 		kref->presumed.valid = 1;
    209 		kref->presumed.offset = bo->offset;
    210 		if (bo->flags & NOUVEAU_BO_VRAM)
    211 			kref->presumed.domain = NOUVEAU_GEM_DOMAIN_VRAM;
    212 		else
    213 			kref->presumed.domain = NOUVEAU_GEM_DOMAIN_GART;
    214 
    215 		cli_kref_set(push->client, bo, kref, push);
    216 		atomic_inc(&nouveau_bo(bo)->refcnt);
    217 	}
    218 
    219 	return kref;
    220 }
    221 
    222 static uint32_t
    223 pushbuf_krel(struct nouveau_pushbuf *push, struct nouveau_bo *bo,
    224 	     uint32_t data, uint32_t flags, uint32_t vor, uint32_t tor)
    225 {
    226 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
    227 	struct nouveau_pushbuf_krec *krec = nvpb->krec;
    228 	struct drm_nouveau_gem_pushbuf_reloc *krel;
    229 	struct drm_nouveau_gem_pushbuf_bo *pkref;
    230 	struct drm_nouveau_gem_pushbuf_bo *bkref;
    231 	uint32_t reloc = data;
    232 
    233 	pkref = cli_kref_get(push->client, nvpb->bo);
    234 	bkref = cli_kref_get(push->client, bo);
    235 	krel  = &krec->reloc[krec->nr_reloc++];
    236 
    237 	krel->reloc_bo_index = pkref - krec->buffer;
    238 	krel->reloc_bo_offset = (push->cur - nvpb->ptr) * 4;
    239 	krel->bo_index = bkref - krec->buffer;
    240 	krel->flags = 0;
    241 	krel->data = data;
    242 	krel->vor = vor;
    243 	krel->tor = tor;
    244 
    245 	if (flags & NOUVEAU_BO_LOW) {
    246 		reloc = (bkref->presumed.offset + data);
    247 		krel->flags |= NOUVEAU_GEM_RELOC_LOW;
    248 	} else
    249 	if (flags & NOUVEAU_BO_HIGH) {
    250 		reloc = (bkref->presumed.offset + data) >> 32;
    251 		krel->flags |= NOUVEAU_GEM_RELOC_HIGH;
    252 	}
    253 	if (flags & NOUVEAU_BO_OR) {
    254 		if (bkref->presumed.domain & NOUVEAU_GEM_DOMAIN_VRAM)
    255 			reloc |= vor;
    256 		else
    257 			reloc |= tor;
    258 		krel->flags |= NOUVEAU_GEM_RELOC_OR;
    259 	}
    260 
    261 	return reloc;
    262 }
    263 
    264 static void
    265 pushbuf_dump(struct nouveau_pushbuf_krec *krec, int krec_id, int chid)
    266 {
    267 	struct drm_nouveau_gem_pushbuf_reloc *krel;
    268 	struct drm_nouveau_gem_pushbuf_push *kpsh;
    269 	struct drm_nouveau_gem_pushbuf_bo *kref;
    270 	struct nouveau_bo *bo;
    271 	uint32_t *bgn, *end;
    272 	int i;
    273 
    274 	err("ch%d: krec %d pushes %d bufs %d relocs %d\n", chid,
    275 	    krec_id, krec->nr_push, krec->nr_buffer, krec->nr_reloc);
    276 
    277 	kref = krec->buffer;
    278 	for (i = 0; i < krec->nr_buffer; i++, kref++) {
    279 		err("ch%d: buf %08x %08x %08x %08x %08x\n", chid, i,
    280 		    kref->handle, kref->valid_domains,
    281 		    kref->read_domains, kref->write_domains);
    282 	}
    283 
    284 	krel = krec->reloc;
    285 	for (i = 0; i < krec->nr_reloc; i++, krel++) {
    286 		err("ch%d: rel %08x %08x %08x %08x %08x %08x %08x\n",
    287 		    chid, krel->reloc_bo_index, krel->reloc_bo_offset,
    288 		    krel->bo_index, krel->flags, krel->data,
    289 		    krel->vor, krel->tor);
    290 	}
    291 
    292 	kpsh = krec->push;
    293 	for (i = 0; i < krec->nr_push; i++, kpsh++) {
    294 		kref = krec->buffer + kpsh->bo_index;
    295 		bo = (void *)(unsigned long)kref->user_priv;
    296 		bgn = (uint32_t *)((char *)bo->map + kpsh->offset);
    297 		end = bgn + (kpsh->length /4);
    298 
    299 		err("ch%d: psh %08x %010llx %010llx\n", chid, kpsh->bo_index,
    300 		    (unsigned long long)kpsh->offset,
    301 		    (unsigned long long)(kpsh->offset + kpsh->length));
    302 		while (bgn < end)
    303 			err("\t0x%08x\n", *bgn++);
    304 	}
    305 }
    306 
    307 static int
    308 pushbuf_submit(struct nouveau_pushbuf *push, struct nouveau_object *chan)
    309 {
    310 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
    311 	struct nouveau_pushbuf_krec *krec = nvpb->list;
    312 	struct nouveau_device *dev = push->client->device;
    313 	struct drm_nouveau_gem_pushbuf_bo_presumed *info;
    314 	struct drm_nouveau_gem_pushbuf_bo *kref;
    315 	struct drm_nouveau_gem_pushbuf req;
    316 	struct nouveau_fifo *fifo = chan->data;
    317 	struct nouveau_bo *bo;
    318 	int krec_id = 0;
    319 	int ret = 0, i;
    320 
    321 	if (chan->oclass != NOUVEAU_FIFO_CHANNEL_CLASS)
    322 		return -EINVAL;
    323 
    324 	if (push->kick_notify)
    325 		push->kick_notify(push);
    326 
    327 	nouveau_pushbuf_data(push, NULL, 0, 0);
    328 
    329 	while (krec && krec->nr_push) {
    330 		req.channel = fifo->channel;
    331 		req.nr_buffers = krec->nr_buffer;
    332 		req.buffers = (uint64_t)(unsigned long)krec->buffer;
    333 		req.nr_relocs = krec->nr_reloc;
    334 		req.nr_push = krec->nr_push;
    335 		req.relocs = (uint64_t)(unsigned long)krec->reloc;
    336 		req.push = (uint64_t)(unsigned long)krec->push;
    337 		req.suffix0 = nvpb->suffix0;
    338 		req.suffix1 = nvpb->suffix1;
    339 		req.vram_available = 0; /* for valgrind */
    340 		req.gart_available = 0;
    341 
    342 		if (dbg_on(0))
    343 			pushbuf_dump(krec, krec_id++, fifo->channel);
    344 
    345 #ifndef SIMULATE
    346 		ret = drmCommandWriteRead(dev->fd, DRM_NOUVEAU_GEM_PUSHBUF,
    347 					  &req, sizeof(req));
    348 		nvpb->suffix0 = req.suffix0;
    349 		nvpb->suffix1 = req.suffix1;
    350 		dev->vram_limit = (req.vram_available *
    351 				nouveau_device(dev)->vram_limit_percent) / 100;
    352 		dev->gart_limit = (req.gart_available *
    353 				nouveau_device(dev)->gart_limit_percent) / 100;
    354 #else
    355 		if (dbg_on(31))
    356 			ret = -EINVAL;
    357 #endif
    358 
    359 		if (ret) {
    360 			err("kernel rejected pushbuf: %s\n", strerror(-ret));
    361 			pushbuf_dump(krec, krec_id++, fifo->channel);
    362 			break;
    363 		}
    364 
    365 		kref = krec->buffer;
    366 		for (i = 0; i < krec->nr_buffer; i++, kref++) {
    367 			bo = (void *)(unsigned long)kref->user_priv;
    368 
    369 			info = &kref->presumed;
    370 			if (!info->valid) {
    371 				bo->flags &= ~NOUVEAU_BO_APER;
    372 				if (info->domain == NOUVEAU_GEM_DOMAIN_VRAM)
    373 					bo->flags |= NOUVEAU_BO_VRAM;
    374 				else
    375 					bo->flags |= NOUVEAU_BO_GART;
    376 				bo->offset = info->offset;
    377 			}
    378 
    379 			if (kref->write_domains)
    380 				nouveau_bo(bo)->access |= NOUVEAU_BO_WR;
    381 			if (kref->read_domains)
    382 				nouveau_bo(bo)->access |= NOUVEAU_BO_RD;
    383 		}
    384 
    385 		krec = krec->next;
    386 	}
    387 
    388 	return ret;
    389 }
    390 
    391 static int
    392 pushbuf_flush(struct nouveau_pushbuf *push)
    393 {
    394 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
    395 	struct nouveau_pushbuf_krec *krec = nvpb->krec;
    396 	struct drm_nouveau_gem_pushbuf_bo *kref;
    397 	struct nouveau_bufctx *bctx, *btmp;
    398 	struct nouveau_bo *bo;
    399 	int ret = 0, i;
    400 
    401 	if (push->channel) {
    402 		ret = pushbuf_submit(push, push->channel);
    403 	} else {
    404 		nouveau_pushbuf_data(push, NULL, 0, 0);
    405 		krec->next = malloc(sizeof(*krec));
    406 		nvpb->krec = krec->next;
    407 	}
    408 
    409 	kref = krec->buffer;
    410 	for (i = 0; i < krec->nr_buffer; i++, kref++) {
    411 		bo = (void *)(unsigned long)kref->user_priv;
    412 		cli_kref_set(push->client, bo, NULL, NULL);
    413 		if (push->channel)
    414 			nouveau_bo_ref(NULL, &bo);
    415 	}
    416 
    417 	krec = nvpb->krec;
    418 	krec->vram_used = 0;
    419 	krec->gart_used = 0;
    420 	krec->nr_buffer = 0;
    421 	krec->nr_reloc = 0;
    422 	krec->nr_push = 0;
    423 
    424 	DRMLISTFOREACHENTRYSAFE(bctx, btmp, &nvpb->bctx_list, head) {
    425 		DRMLISTJOIN(&bctx->current, &bctx->pending);
    426 		DRMINITLISTHEAD(&bctx->current);
    427 		DRMLISTDELINIT(&bctx->head);
    428 	}
    429 
    430 	return ret;
    431 }
    432 
    433 static void
    434 pushbuf_refn_fail(struct nouveau_pushbuf *push, int sref, int srel)
    435 {
    436 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
    437 	struct nouveau_pushbuf_krec *krec = nvpb->krec;
    438 	struct drm_nouveau_gem_pushbuf_bo *kref;
    439 
    440 	kref = krec->buffer + sref;
    441 	while (krec->nr_buffer-- > sref) {
    442 		struct nouveau_bo *bo = (void *)(unsigned long)kref->user_priv;
    443 		cli_kref_set(push->client, bo, NULL, NULL);
    444 		nouveau_bo_ref(NULL, &bo);
    445 		kref++;
    446 	}
    447 	krec->nr_buffer = sref;
    448 	krec->nr_reloc = srel;
    449 }
    450 
    451 static int
    452 pushbuf_refn(struct nouveau_pushbuf *push, bool retry,
    453 	     struct nouveau_pushbuf_refn *refs, int nr)
    454 {
    455 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
    456 	struct nouveau_pushbuf_krec *krec = nvpb->krec;
    457 	struct drm_nouveau_gem_pushbuf_bo *kref;
    458 	int sref = krec->nr_buffer;
    459 	int ret = 0, i;
    460 
    461 	for (i = 0; i < nr; i++) {
    462 		kref = pushbuf_kref(push, refs[i].bo, refs[i].flags);
    463 		if (!kref) {
    464 			ret = -ENOSPC;
    465 			break;
    466 		}
    467 	}
    468 
    469 	if (ret) {
    470 		pushbuf_refn_fail(push, sref, krec->nr_reloc);
    471 		if (retry) {
    472 			pushbuf_flush(push);
    473 			nouveau_pushbuf_space(push, 0, 0, 0);
    474 			return pushbuf_refn(push, false, refs, nr);
    475 		}
    476 	}
    477 
    478 	return ret;
    479 }
    480 
    481 static int
    482 pushbuf_validate(struct nouveau_pushbuf *push, bool retry)
    483 {
    484 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
    485 	struct nouveau_pushbuf_krec *krec = nvpb->krec;
    486 	struct drm_nouveau_gem_pushbuf_bo *kref;
    487 	struct nouveau_bufctx *bctx = push->bufctx;
    488 	struct nouveau_bufref *bref;
    489 	int relocs = bctx ? bctx->relocs * 2: 0;
    490 	int sref, srel, ret;
    491 
    492 	ret = nouveau_pushbuf_space(push, relocs, relocs, 0);
    493 	if (ret || bctx == NULL)
    494 		return ret;
    495 
    496 	sref = krec->nr_buffer;
    497 	srel = krec->nr_reloc;
    498 
    499 	DRMLISTDEL(&bctx->head);
    500 	DRMLISTADD(&bctx->head, &nvpb->bctx_list);
    501 
    502 	DRMLISTFOREACHENTRY(bref, &bctx->pending, thead) {
    503 		kref = pushbuf_kref(push, bref->bo, bref->flags);
    504 		if (!kref) {
    505 			ret = -ENOSPC;
    506 			break;
    507 		}
    508 
    509 		if (bref->packet) {
    510 			pushbuf_krel(push, bref->bo, bref->packet, 0, 0, 0);
    511 			*push->cur++ = 0;
    512 			pushbuf_krel(push, bref->bo, bref->data, bref->flags,
    513 					   bref->vor, bref->tor);
    514 			*push->cur++ = 0;
    515 		}
    516 	}
    517 
    518 	DRMLISTJOIN(&bctx->pending, &bctx->current);
    519 	DRMINITLISTHEAD(&bctx->pending);
    520 
    521 	if (ret) {
    522 		pushbuf_refn_fail(push, sref, srel);
    523 		if (retry) {
    524 			pushbuf_flush(push);
    525 			return pushbuf_validate(push, false);
    526 		}
    527 	}
    528 
    529 	return ret;
    530 }
    531 
    532 drm_public int
    533 nouveau_pushbuf_new(struct nouveau_client *client, struct nouveau_object *chan,
    534 		    int nr, uint32_t size, bool immediate,
    535 		    struct nouveau_pushbuf **ppush)
    536 {
    537 	struct nouveau_device *dev = client->device;
    538 	struct nouveau_fifo *fifo = chan->data;
    539 	struct nouveau_pushbuf_priv *nvpb;
    540 	struct nouveau_pushbuf *push;
    541 	struct drm_nouveau_gem_pushbuf req = {};
    542 	int ret;
    543 
    544 	if (chan->oclass != NOUVEAU_FIFO_CHANNEL_CLASS)
    545 		return -EINVAL;
    546 
    547 	/* nop pushbuf call, to get the current "return to main" sequence
    548 	 * we need to append to the pushbuf on early chipsets
    549 	 */
    550 	req.channel = fifo->channel;
    551 	req.nr_push = 0;
    552 	ret = drmCommandWriteRead(dev->fd, DRM_NOUVEAU_GEM_PUSHBUF,
    553 				  &req, sizeof(req));
    554 	if (ret)
    555 		return ret;
    556 
    557 	nvpb = calloc(1, sizeof(*nvpb) + nr * sizeof(*nvpb->bos));
    558 	if (!nvpb)
    559 		return -ENOMEM;
    560 
    561 #ifndef SIMULATE
    562 	nvpb->suffix0 = req.suffix0;
    563 	nvpb->suffix1 = req.suffix1;
    564 #else
    565 	nvpb->suffix0 = 0xffffffff;
    566 	nvpb->suffix1 = 0xffffffff;
    567 #endif
    568 	nvpb->krec = calloc(1, sizeof(*nvpb->krec));
    569 	nvpb->list = nvpb->krec;
    570 	if (!nvpb->krec) {
    571 		free(nvpb);
    572 		return -ENOMEM;
    573 	}
    574 
    575 	push = &nvpb->base;
    576 	push->client = client;
    577 	push->channel = immediate ? chan : NULL;
    578 	push->flags = NOUVEAU_BO_RD;
    579 	if (fifo->pushbuf & NOUVEAU_GEM_DOMAIN_GART) {
    580 		push->flags |= NOUVEAU_BO_GART;
    581 		nvpb->type   = NOUVEAU_BO_GART;
    582 	} else
    583 	if (fifo->pushbuf & NOUVEAU_GEM_DOMAIN_VRAM) {
    584 		push->flags |= NOUVEAU_BO_VRAM;
    585 		nvpb->type   = NOUVEAU_BO_VRAM;
    586 	}
    587 	nvpb->type |= NOUVEAU_BO_MAP;
    588 
    589 	for (nvpb->bo_nr = 0; nvpb->bo_nr < nr; nvpb->bo_nr++) {
    590 		ret = nouveau_bo_new(client->device, nvpb->type, 0, size,
    591 				     NULL, &nvpb->bos[nvpb->bo_nr]);
    592 		if (ret) {
    593 			nouveau_pushbuf_del(&push);
    594 			return ret;
    595 		}
    596 	}
    597 
    598 	DRMINITLISTHEAD(&nvpb->bctx_list);
    599 	*ppush = push;
    600 	return 0;
    601 }
    602 
    603 drm_public void
    604 nouveau_pushbuf_del(struct nouveau_pushbuf **ppush)
    605 {
    606 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(*ppush);
    607 	if (nvpb) {
    608 		struct drm_nouveau_gem_pushbuf_bo *kref;
    609 		struct nouveau_pushbuf_krec *krec;
    610 		while ((krec = nvpb->list)) {
    611 			kref = krec->buffer;
    612 			while (krec->nr_buffer--) {
    613 				unsigned long priv = kref++->user_priv;
    614 				struct nouveau_bo *bo = (void *)priv;
    615 				cli_kref_set(nvpb->base.client, bo, NULL, NULL);
    616 				nouveau_bo_ref(NULL, &bo);
    617 			}
    618 			nvpb->list = krec->next;
    619 			free(krec);
    620 		}
    621 		while (nvpb->bo_nr--)
    622 			nouveau_bo_ref(NULL, &nvpb->bos[nvpb->bo_nr]);
    623 		nouveau_bo_ref(NULL, &nvpb->bo);
    624 		free(nvpb);
    625 	}
    626 	*ppush = NULL;
    627 }
    628 
    629 drm_public struct nouveau_bufctx *
    630 nouveau_pushbuf_bufctx(struct nouveau_pushbuf *push, struct nouveau_bufctx *ctx)
    631 {
    632 	struct nouveau_bufctx *prev = push->bufctx;
    633 	push->bufctx = ctx;
    634 	return prev;
    635 }
    636 
    637 drm_public int
    638 nouveau_pushbuf_space(struct nouveau_pushbuf *push,
    639 		      uint32_t dwords, uint32_t relocs, uint32_t pushes)
    640 {
    641 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
    642 	struct nouveau_pushbuf_krec *krec = nvpb->krec;
    643 	struct nouveau_client *client = push->client;
    644 	struct nouveau_bo *bo = NULL;
    645 	bool flushed = false;
    646 	int ret = 0;
    647 
    648 	/* switch to next buffer if insufficient space in the current one */
    649 	if (push->cur + dwords >= push->end) {
    650 		if (nvpb->bo_next < nvpb->bo_nr) {
    651 			nouveau_bo_ref(nvpb->bos[nvpb->bo_next++], &bo);
    652 			if (nvpb->bo_next == nvpb->bo_nr && push->channel)
    653 				nvpb->bo_next = 0;
    654 		} else {
    655 			ret = nouveau_bo_new(client->device, nvpb->type, 0,
    656 					     nvpb->bos[0]->size, NULL, &bo);
    657 			if (ret)
    658 				return ret;
    659 		}
    660 	}
    661 
    662 	/* make sure there's always enough space to queue up the pending
    663 	 * data in the pushbuf proper
    664 	 */
    665 	pushes++;
    666 
    667 	/* need to flush if we've run out of space on an immediate pushbuf,
    668 	 * if the new buffer won't fit, or if the kernel push/reloc limits
    669 	 * have been hit
    670 	 */
    671 	if ((bo && ( push->channel ||
    672 		    !pushbuf_kref(push, bo, push->flags))) ||
    673 	    krec->nr_reloc + relocs >= NOUVEAU_GEM_MAX_RELOCS ||
    674 	    krec->nr_push + pushes >= NOUVEAU_GEM_MAX_PUSH) {
    675 		if (nvpb->bo && krec->nr_buffer)
    676 			pushbuf_flush(push);
    677 		flushed = true;
    678 	}
    679 
    680 	/* if necessary, switch to new buffer */
    681 	if (bo) {
    682 		ret = nouveau_bo_map(bo, NOUVEAU_BO_WR, push->client);
    683 		if (ret)
    684 			return ret;
    685 
    686 		nouveau_pushbuf_data(push, NULL, 0, 0);
    687 		nouveau_bo_ref(bo, &nvpb->bo);
    688 		nouveau_bo_ref(NULL, &bo);
    689 
    690 		nvpb->bgn = nvpb->bo->map;
    691 		nvpb->ptr = nvpb->bgn;
    692 		push->cur = nvpb->bgn;
    693 		push->end = push->cur + (nvpb->bo->size / 4);
    694 		push->end -= 2 + push->rsvd_kick; /* space for suffix */
    695 	}
    696 
    697 	pushbuf_kref(push, nvpb->bo, push->flags);
    698 	return flushed ? pushbuf_validate(push, false) : 0;
    699 }
    700 
    701 drm_public void
    702 nouveau_pushbuf_data(struct nouveau_pushbuf *push, struct nouveau_bo *bo,
    703 		     uint64_t offset, uint64_t length)
    704 {
    705 	struct nouveau_pushbuf_priv *nvpb = nouveau_pushbuf(push);
    706 	struct nouveau_pushbuf_krec *krec = nvpb->krec;
    707 	struct drm_nouveau_gem_pushbuf_push *kpsh;
    708 	struct drm_nouveau_gem_pushbuf_bo *kref;
    709 
    710 	if (bo != nvpb->bo && nvpb->bgn != push->cur) {
    711 		if (nvpb->suffix0 || nvpb->suffix1) {
    712 			*push->cur++ = nvpb->suffix0;
    713 			*push->cur++ = nvpb->suffix1;
    714 		}
    715 
    716 		nouveau_pushbuf_data(push, nvpb->bo,
    717 				     (nvpb->bgn - nvpb->ptr) * 4,
    718 				     (push->cur - nvpb->bgn) * 4);
    719 		nvpb->bgn = push->cur;
    720 	}
    721 
    722 	if (bo) {
    723 		kref = cli_kref_get(push->client, bo);
    724 		kpsh = &krec->push[krec->nr_push++];
    725 		kpsh->bo_index = kref - krec->buffer;
    726 		kpsh->offset   = offset;
    727 		kpsh->length   = length;
    728 	}
    729 }
    730 
    731 drm_public int
    732 nouveau_pushbuf_refn(struct nouveau_pushbuf *push,
    733 		     struct nouveau_pushbuf_refn *refs, int nr)
    734 {
    735 	return pushbuf_refn(push, true, refs, nr);
    736 }
    737 
    738 drm_public void
    739 nouveau_pushbuf_reloc(struct nouveau_pushbuf *push, struct nouveau_bo *bo,
    740 		      uint32_t data, uint32_t flags, uint32_t vor, uint32_t tor)
    741 {
    742 	*push->cur = pushbuf_krel(push, bo, data, flags, vor, tor);
    743 	push->cur++;
    744 }
    745 
    746 drm_public int
    747 nouveau_pushbuf_validate(struct nouveau_pushbuf *push)
    748 {
    749 	return pushbuf_validate(push, true);
    750 }
    751 
    752 drm_public uint32_t
    753 nouveau_pushbuf_refd(struct nouveau_pushbuf *push, struct nouveau_bo *bo)
    754 {
    755 	struct drm_nouveau_gem_pushbuf_bo *kref;
    756 	uint32_t flags = 0;
    757 
    758 	if (cli_push_get(push->client, bo) == push) {
    759 		kref = cli_kref_get(push->client, bo);
    760 		if (kref->read_domains)
    761 			flags |= NOUVEAU_BO_RD;
    762 		if (kref->write_domains)
    763 			flags |= NOUVEAU_BO_WR;
    764 	}
    765 
    766 	return flags;
    767 }
    768 
    769 drm_public int
    770 nouveau_pushbuf_kick(struct nouveau_pushbuf *push, struct nouveau_object *chan)
    771 {
    772 	if (!push->channel)
    773 		return pushbuf_submit(push, chan);
    774 	pushbuf_flush(push);
    775 	return pushbuf_validate(push, false);
    776 }
    777