18c2ecf20Sopenharmony_ci/*
28c2ecf20Sopenharmony_ci * Copyright 2014 Advanced Micro Devices, Inc.
38c2ecf20Sopenharmony_ci * All Rights Reserved.
48c2ecf20Sopenharmony_ci *
58c2ecf20Sopenharmony_ci * Permission is hereby granted, free of charge, to any person obtaining a
68c2ecf20Sopenharmony_ci * copy of this software and associated documentation files (the
78c2ecf20Sopenharmony_ci * "Software"), to deal in the Software without restriction, including
88c2ecf20Sopenharmony_ci * without limitation the rights to use, copy, modify, merge, publish,
98c2ecf20Sopenharmony_ci * distribute, sub license, and/or sell copies of the Software, and to
108c2ecf20Sopenharmony_ci * permit persons to whom the Software is furnished to do so, subject to
118c2ecf20Sopenharmony_ci * the following conditions:
128c2ecf20Sopenharmony_ci *
138c2ecf20Sopenharmony_ci * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
148c2ecf20Sopenharmony_ci * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
158c2ecf20Sopenharmony_ci * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
168c2ecf20Sopenharmony_ci * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
178c2ecf20Sopenharmony_ci * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
188c2ecf20Sopenharmony_ci * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
198c2ecf20Sopenharmony_ci * USE OR OTHER DEALINGS IN THE SOFTWARE.
208c2ecf20Sopenharmony_ci *
218c2ecf20Sopenharmony_ci * The above copyright notice and this permission notice (including the
228c2ecf20Sopenharmony_ci * next paragraph) shall be included in all copies or substantial portions
238c2ecf20Sopenharmony_ci * of the Software.
248c2ecf20Sopenharmony_ci *
258c2ecf20Sopenharmony_ci */
268c2ecf20Sopenharmony_ci/*
278c2ecf20Sopenharmony_ci * Authors:
288c2ecf20Sopenharmony_ci *    Christian König <christian.koenig@amd.com>
298c2ecf20Sopenharmony_ci */
308c2ecf20Sopenharmony_ci
318c2ecf20Sopenharmony_ci#include "radeon.h"
328c2ecf20Sopenharmony_ci#include "radeon_trace.h"
338c2ecf20Sopenharmony_ci
348c2ecf20Sopenharmony_ci/**
358c2ecf20Sopenharmony_ci * radeon_sync_create - zero init sync object
368c2ecf20Sopenharmony_ci *
378c2ecf20Sopenharmony_ci * @sync: sync object to initialize
388c2ecf20Sopenharmony_ci *
398c2ecf20Sopenharmony_ci * Just clear the sync object for now.
408c2ecf20Sopenharmony_ci */
418c2ecf20Sopenharmony_civoid radeon_sync_create(struct radeon_sync *sync)
428c2ecf20Sopenharmony_ci{
438c2ecf20Sopenharmony_ci	unsigned i;
448c2ecf20Sopenharmony_ci
458c2ecf20Sopenharmony_ci	for (i = 0; i < RADEON_NUM_SYNCS; ++i)
468c2ecf20Sopenharmony_ci		sync->semaphores[i] = NULL;
478c2ecf20Sopenharmony_ci
488c2ecf20Sopenharmony_ci	for (i = 0; i < RADEON_NUM_RINGS; ++i)
498c2ecf20Sopenharmony_ci		sync->sync_to[i] = NULL;
508c2ecf20Sopenharmony_ci
518c2ecf20Sopenharmony_ci	sync->last_vm_update = NULL;
528c2ecf20Sopenharmony_ci}
538c2ecf20Sopenharmony_ci
548c2ecf20Sopenharmony_ci/**
558c2ecf20Sopenharmony_ci * radeon_sync_fence - use the semaphore to sync to a fence
568c2ecf20Sopenharmony_ci *
578c2ecf20Sopenharmony_ci * @sync: sync object to add fence to
588c2ecf20Sopenharmony_ci * @fence: fence to sync to
598c2ecf20Sopenharmony_ci *
608c2ecf20Sopenharmony_ci * Sync to the fence using the semaphore objects
618c2ecf20Sopenharmony_ci */
628c2ecf20Sopenharmony_civoid radeon_sync_fence(struct radeon_sync *sync,
638c2ecf20Sopenharmony_ci		       struct radeon_fence *fence)
648c2ecf20Sopenharmony_ci{
658c2ecf20Sopenharmony_ci	struct radeon_fence *other;
668c2ecf20Sopenharmony_ci
678c2ecf20Sopenharmony_ci	if (!fence)
688c2ecf20Sopenharmony_ci		return;
698c2ecf20Sopenharmony_ci
708c2ecf20Sopenharmony_ci	other = sync->sync_to[fence->ring];
718c2ecf20Sopenharmony_ci	sync->sync_to[fence->ring] = radeon_fence_later(fence, other);
728c2ecf20Sopenharmony_ci
738c2ecf20Sopenharmony_ci	if (fence->is_vm_update) {
748c2ecf20Sopenharmony_ci		other = sync->last_vm_update;
758c2ecf20Sopenharmony_ci		sync->last_vm_update = radeon_fence_later(fence, other);
768c2ecf20Sopenharmony_ci	}
778c2ecf20Sopenharmony_ci}
788c2ecf20Sopenharmony_ci
798c2ecf20Sopenharmony_ci/**
808c2ecf20Sopenharmony_ci * radeon_sync_resv - use the semaphores to sync to a reservation object
818c2ecf20Sopenharmony_ci *
828c2ecf20Sopenharmony_ci * @sync: sync object to add fences from reservation object to
838c2ecf20Sopenharmony_ci * @resv: reservation object with embedded fence
848c2ecf20Sopenharmony_ci * @shared: true if we should only sync to the exclusive fence
858c2ecf20Sopenharmony_ci *
868c2ecf20Sopenharmony_ci * Sync to the fence using the semaphore objects
878c2ecf20Sopenharmony_ci */
888c2ecf20Sopenharmony_ciint radeon_sync_resv(struct radeon_device *rdev,
898c2ecf20Sopenharmony_ci		     struct radeon_sync *sync,
908c2ecf20Sopenharmony_ci		     struct dma_resv *resv,
918c2ecf20Sopenharmony_ci		     bool shared)
928c2ecf20Sopenharmony_ci{
938c2ecf20Sopenharmony_ci	struct dma_resv_list *flist;
948c2ecf20Sopenharmony_ci	struct dma_fence *f;
958c2ecf20Sopenharmony_ci	struct radeon_fence *fence;
968c2ecf20Sopenharmony_ci	unsigned i;
978c2ecf20Sopenharmony_ci	int r = 0;
988c2ecf20Sopenharmony_ci
998c2ecf20Sopenharmony_ci	/* always sync to the exclusive fence */
1008c2ecf20Sopenharmony_ci	f = dma_resv_get_excl(resv);
1018c2ecf20Sopenharmony_ci	fence = f ? to_radeon_fence(f) : NULL;
1028c2ecf20Sopenharmony_ci	if (fence && fence->rdev == rdev)
1038c2ecf20Sopenharmony_ci		radeon_sync_fence(sync, fence);
1048c2ecf20Sopenharmony_ci	else if (f)
1058c2ecf20Sopenharmony_ci		r = dma_fence_wait(f, true);
1068c2ecf20Sopenharmony_ci
1078c2ecf20Sopenharmony_ci	flist = dma_resv_get_list(resv);
1088c2ecf20Sopenharmony_ci	if (shared || !flist || r)
1098c2ecf20Sopenharmony_ci		return r;
1108c2ecf20Sopenharmony_ci
1118c2ecf20Sopenharmony_ci	for (i = 0; i < flist->shared_count; ++i) {
1128c2ecf20Sopenharmony_ci		f = rcu_dereference_protected(flist->shared[i],
1138c2ecf20Sopenharmony_ci					      dma_resv_held(resv));
1148c2ecf20Sopenharmony_ci		fence = to_radeon_fence(f);
1158c2ecf20Sopenharmony_ci		if (fence && fence->rdev == rdev)
1168c2ecf20Sopenharmony_ci			radeon_sync_fence(sync, fence);
1178c2ecf20Sopenharmony_ci		else
1188c2ecf20Sopenharmony_ci			r = dma_fence_wait(f, true);
1198c2ecf20Sopenharmony_ci
1208c2ecf20Sopenharmony_ci		if (r)
1218c2ecf20Sopenharmony_ci			break;
1228c2ecf20Sopenharmony_ci	}
1238c2ecf20Sopenharmony_ci	return r;
1248c2ecf20Sopenharmony_ci}
1258c2ecf20Sopenharmony_ci
1268c2ecf20Sopenharmony_ci/**
1278c2ecf20Sopenharmony_ci * radeon_sync_rings - sync ring to all registered fences
1288c2ecf20Sopenharmony_ci *
1298c2ecf20Sopenharmony_ci * @rdev: radeon_device pointer
1308c2ecf20Sopenharmony_ci * @sync: sync object to use
1318c2ecf20Sopenharmony_ci * @ring: ring that needs sync
1328c2ecf20Sopenharmony_ci *
1338c2ecf20Sopenharmony_ci * Ensure that all registered fences are signaled before letting
1348c2ecf20Sopenharmony_ci * the ring continue. The caller must hold the ring lock.
1358c2ecf20Sopenharmony_ci */
1368c2ecf20Sopenharmony_ciint radeon_sync_rings(struct radeon_device *rdev,
1378c2ecf20Sopenharmony_ci		      struct radeon_sync *sync,
1388c2ecf20Sopenharmony_ci		      int ring)
1398c2ecf20Sopenharmony_ci{
1408c2ecf20Sopenharmony_ci	unsigned count = 0;
1418c2ecf20Sopenharmony_ci	int i, r;
1428c2ecf20Sopenharmony_ci
1438c2ecf20Sopenharmony_ci	for (i = 0; i < RADEON_NUM_RINGS; ++i) {
1448c2ecf20Sopenharmony_ci		struct radeon_fence *fence = sync->sync_to[i];
1458c2ecf20Sopenharmony_ci		struct radeon_semaphore *semaphore;
1468c2ecf20Sopenharmony_ci
1478c2ecf20Sopenharmony_ci		/* check if we really need to sync */
1488c2ecf20Sopenharmony_ci		if (!radeon_fence_need_sync(fence, ring))
1498c2ecf20Sopenharmony_ci			continue;
1508c2ecf20Sopenharmony_ci
1518c2ecf20Sopenharmony_ci		/* prevent GPU deadlocks */
1528c2ecf20Sopenharmony_ci		if (!rdev->ring[i].ready) {
1538c2ecf20Sopenharmony_ci			dev_err(rdev->dev, "Syncing to a disabled ring!");
1548c2ecf20Sopenharmony_ci			return -EINVAL;
1558c2ecf20Sopenharmony_ci		}
1568c2ecf20Sopenharmony_ci
1578c2ecf20Sopenharmony_ci		if (count >= RADEON_NUM_SYNCS) {
1588c2ecf20Sopenharmony_ci			/* not enough room, wait manually */
1598c2ecf20Sopenharmony_ci			r = radeon_fence_wait(fence, false);
1608c2ecf20Sopenharmony_ci			if (r)
1618c2ecf20Sopenharmony_ci				return r;
1628c2ecf20Sopenharmony_ci			continue;
1638c2ecf20Sopenharmony_ci		}
1648c2ecf20Sopenharmony_ci		r = radeon_semaphore_create(rdev, &semaphore);
1658c2ecf20Sopenharmony_ci		if (r)
1668c2ecf20Sopenharmony_ci			return r;
1678c2ecf20Sopenharmony_ci
1688c2ecf20Sopenharmony_ci		sync->semaphores[count++] = semaphore;
1698c2ecf20Sopenharmony_ci
1708c2ecf20Sopenharmony_ci		/* allocate enough space for sync command */
1718c2ecf20Sopenharmony_ci		r = radeon_ring_alloc(rdev, &rdev->ring[i], 16);
1728c2ecf20Sopenharmony_ci		if (r)
1738c2ecf20Sopenharmony_ci			return r;
1748c2ecf20Sopenharmony_ci
1758c2ecf20Sopenharmony_ci		/* emit the signal semaphore */
1768c2ecf20Sopenharmony_ci		if (!radeon_semaphore_emit_signal(rdev, i, semaphore)) {
1778c2ecf20Sopenharmony_ci			/* signaling wasn't successful wait manually */
1788c2ecf20Sopenharmony_ci			radeon_ring_undo(&rdev->ring[i]);
1798c2ecf20Sopenharmony_ci			r = radeon_fence_wait(fence, false);
1808c2ecf20Sopenharmony_ci			if (r)
1818c2ecf20Sopenharmony_ci				return r;
1828c2ecf20Sopenharmony_ci			continue;
1838c2ecf20Sopenharmony_ci		}
1848c2ecf20Sopenharmony_ci
1858c2ecf20Sopenharmony_ci		/* we assume caller has already allocated space on waiters ring */
1868c2ecf20Sopenharmony_ci		if (!radeon_semaphore_emit_wait(rdev, ring, semaphore)) {
1878c2ecf20Sopenharmony_ci			/* waiting wasn't successful wait manually */
1888c2ecf20Sopenharmony_ci			radeon_ring_undo(&rdev->ring[i]);
1898c2ecf20Sopenharmony_ci			r = radeon_fence_wait(fence, false);
1908c2ecf20Sopenharmony_ci			if (r)
1918c2ecf20Sopenharmony_ci				return r;
1928c2ecf20Sopenharmony_ci			continue;
1938c2ecf20Sopenharmony_ci		}
1948c2ecf20Sopenharmony_ci
1958c2ecf20Sopenharmony_ci		radeon_ring_commit(rdev, &rdev->ring[i], false);
1968c2ecf20Sopenharmony_ci		radeon_fence_note_sync(fence, ring);
1978c2ecf20Sopenharmony_ci	}
1988c2ecf20Sopenharmony_ci
1998c2ecf20Sopenharmony_ci	return 0;
2008c2ecf20Sopenharmony_ci}
2018c2ecf20Sopenharmony_ci
2028c2ecf20Sopenharmony_ci/**
2038c2ecf20Sopenharmony_ci * radeon_sync_free - free the sync object
2048c2ecf20Sopenharmony_ci *
2058c2ecf20Sopenharmony_ci * @rdev: radeon_device pointer
2068c2ecf20Sopenharmony_ci * @sync: sync object to use
2078c2ecf20Sopenharmony_ci * @fence: fence to use for the free
2088c2ecf20Sopenharmony_ci *
2098c2ecf20Sopenharmony_ci * Free the sync object by freeing all semaphores in it.
2108c2ecf20Sopenharmony_ci */
2118c2ecf20Sopenharmony_civoid radeon_sync_free(struct radeon_device *rdev,
2128c2ecf20Sopenharmony_ci		      struct radeon_sync *sync,
2138c2ecf20Sopenharmony_ci		      struct radeon_fence *fence)
2148c2ecf20Sopenharmony_ci{
2158c2ecf20Sopenharmony_ci	unsigned i;
2168c2ecf20Sopenharmony_ci
2178c2ecf20Sopenharmony_ci	for (i = 0; i < RADEON_NUM_SYNCS; ++i)
2188c2ecf20Sopenharmony_ci		radeon_semaphore_free(rdev, &sync->semaphores[i], fence);
2198c2ecf20Sopenharmony_ci}
220