From 4dfd6e43cfe303c8b23421ef32738db2ee52e291 Mon Sep 17 00:00:00 2001 From: Debarshi Dutta Date: Wed, 18 Apr 2018 11:03:02 +0530 Subject: gpu: nvgpu: create a wrapper over sync_fences This patch constructs an abstraction to hide the sync_fence functionality from the common code. struct nvgpu_os_fence acts as an abstraction for struct sync_fence. struct nvgpu_os_fence consists of an ops structure named nvgpu_os_fence_ops which contains an API to do pushbuffer programming to generate wait commands for the fence. The current implementation of nvgpu only allows for wait method on a sync_fence which was generated using a similar backend(i.e. either Nvhost Syncpoints or Semaphores). In this patch, a generic API is introduced which will decide the type of the underlying implementation of the struct nvgpu_os_fence at runtime and run the corresponding wait implementation on it. This patch changes the channel_sync_gk20a's semaphore specific implementation to use the abstract API. A subsequent patch will make the changes for the nvhost_syncpoint based implementations as well. JIRA NVGPU-66 Change-Id: If6675bfde5885c3d15d2ca380bb6c7c0e240e734 Signed-off-by: Debarshi Dutta Reviewed-on: https://git-master.nvidia.com/r/1667218 Reviewed-by: mobile promotions Tested-by: mobile promotions --- drivers/gpu/nvgpu/Makefile | 4 +- drivers/gpu/nvgpu/common/linux/os_fence_android.c | 71 +++++++++++++ .../gpu/nvgpu/common/linux/os_fence_android_sema.c | 107 +++++++++++++++++++ drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c | 117 +++++++-------------- drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.h | 5 + .../nvgpu/include/nvgpu/linux/os_fence_android.h | 42 ++++++++ drivers/gpu/nvgpu/include/nvgpu/os_fence.h | 111 +++++++++++++++++++ 7 files changed, 379 insertions(+), 78 deletions(-) create mode 100644 drivers/gpu/nvgpu/common/linux/os_fence_android.c create mode 100644 drivers/gpu/nvgpu/common/linux/os_fence_android_sema.c create mode 100644 drivers/gpu/nvgpu/include/nvgpu/linux/os_fence_android.h create mode 100644 drivers/gpu/nvgpu/include/nvgpu/os_fence.h diff --git a/drivers/gpu/nvgpu/Makefile b/drivers/gpu/nvgpu/Makefile index 32166f31..9a46f12c 100644 --- a/drivers/gpu/nvgpu/Makefile +++ b/drivers/gpu/nvgpu/Makefile @@ -160,7 +160,9 @@ nvgpu-$(CONFIG_TEGRA_GK20A) += \ common/linux/platform_gp10b_tegra.o \ common/linux/platform_gv11b_tegra.o -nvgpu-$(CONFIG_SYNC) += gk20a/sync_gk20a.o +nvgpu-$(CONFIG_SYNC) += gk20a/sync_gk20a.o \ + common/linux/os_fence_android.o \ + common/linux/os_fence_android_sema.o nvgpu-$(CONFIG_GK20A_PCI) += common/linux/pci.o \ common/linux/pci_usermode.o \ diff --git a/drivers/gpu/nvgpu/common/linux/os_fence_android.c b/drivers/gpu/nvgpu/common/linux/os_fence_android.c new file mode 100644 index 00000000..d689a2a8 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/os_fence_android.c @@ -0,0 +1,71 @@ +/* + * Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see . + */ +#include +#include +#include + +#include "gk20a/gk20a.h" + +#include "../drivers/staging/android/sync.h" + +inline struct sync_fence *nvgpu_get_sync_fence(struct nvgpu_os_fence *s) +{ + struct sync_fence *fence = (struct sync_fence *)s->priv; + return fence; +} + +static void nvgpu_os_fence_clear(struct nvgpu_os_fence *fence_out) +{ + fence_out->priv = NULL; + fence_out->g = NULL; + fence_out->ops = NULL; +} + +void nvgpu_os_fence_init(struct nvgpu_os_fence *fence_out, + struct gk20a *g, const struct nvgpu_os_fence_ops *fops, + struct sync_fence *fence) +{ + fence_out->g = g; + fence_out->ops = fops; + fence_out->priv = (void *)fence; +} + +void nvgpu_os_fence_android_drop_ref(struct nvgpu_os_fence *s) +{ + struct sync_fence *fence = nvgpu_get_sync_fence(s); + + sync_fence_put(fence); + + nvgpu_os_fence_clear(s); +} + +int nvgpu_os_fence_fdget(struct nvgpu_os_fence *fence_out, + struct channel_gk20a *c, int fd) +{ + int err; + + err = nvgpu_os_fence_sema_fdget(fence_out, c, fd); + + /* TO-DO + * check if fence is empty and if CONFIG_TEGRA_GK20A_NVHOST + * is enabled, try to get a sync_fence using + * corresponding nvhost method. + */ + if (err) + nvgpu_err(c->g, "error obtaining fence from fd %d", fd); + + return err; +} diff --git a/drivers/gpu/nvgpu/common/linux/os_fence_android_sema.c b/drivers/gpu/nvgpu/common/linux/os_fence_android_sema.c new file mode 100644 index 00000000..b61bd893 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/os_fence_android_sema.c @@ -0,0 +1,107 @@ +/* + * Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. + * + * This program is free software; you can redistribute it and/or modify it + * under the terms and conditions of the GNU General Public License, + * version 2, as published by the Free Software Foundation. + * + * This program is distributed in the hope it will be useful, but WITHOUT + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for + * more details. + * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see . + */ + +#include + +#include +#include +#include +#include + +#include "gk20a/sync_gk20a.h" +#include "gk20a/channel_sync_gk20a.h" +#include "gk20a/mm_gk20a.h" + +#include "../drivers/staging/android/sync.h" + +int nvgpu_os_fence_sema_wait_gen_cmd(struct nvgpu_os_fence *s, + struct priv_cmd_entry *wait_cmd, + struct channel_gk20a *c, + int max_wait_cmds) +{ + int err; + const int wait_cmd_size = 8; + int num_wait_cmds; + int i; + struct nvgpu_semaphore *sema; + struct sync_fence *sync_fence = nvgpu_get_sync_fence(s); + + num_wait_cmds = sync_fence->num_fences; + if (num_wait_cmds == 0) + return 0; + + if (max_wait_cmds && num_wait_cmds > max_wait_cmds) + return -EINVAL; + + err = gk20a_channel_alloc_priv_cmdbuf(c, + wait_cmd_size * num_wait_cmds, + wait_cmd); + if (err) { + nvgpu_err(c->g, "not enough priv cmd buffer space"); + return err; + } + + for (i = 0; i < num_wait_cmds; i++) { + struct fence *f = sync_fence->cbs[i].sync_pt; + struct sync_pt *pt = sync_pt_from_fence(f); + + sema = gk20a_sync_pt_sema(pt); + gk20a_channel_gen_sema_wait_cmd(c, sema, wait_cmd, + wait_cmd_size, i); + } + + return 0; +} + +static const struct nvgpu_os_fence_ops sema_ops = { + .program_waits = nvgpu_os_fence_sema_wait_gen_cmd, + .drop_ref = nvgpu_os_fence_android_drop_ref, +}; + +int nvgpu_os_fence_sema_create( + struct nvgpu_os_fence *fence_out, + struct channel_gk20a *c, + struct nvgpu_semaphore *sema) +{ + struct sync_fence *fence; + + fence = gk20a_sync_fence_create(c, sema, "f-gk20a-0x%04x", + nvgpu_semaphore_gpu_ro_va(sema)); + + if (!fence) { + nvgpu_err(c->g, "error constructing new fence: f-gk20a-0x%04x", + (u32)nvgpu_semaphore_gpu_ro_va(sema)); + + return -ENOMEM; + } + + nvgpu_os_fence_init(fence_out, c->g, &sema_ops, fence); + + return 0; +} + +int nvgpu_os_fence_sema_fdget(struct nvgpu_os_fence *fence_out, + struct channel_gk20a *c, int fd) +{ + struct sync_fence *fence = gk20a_sync_fence_fdget(fd); + + if (!fence) + return -EINVAL; + + nvgpu_os_fence_init(fence_out, c->g, &sema_ops, fence); + + return 0; +} diff --git a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c index c0e035ea..a0b0ac1e 100644 --- a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c +++ b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.c @@ -29,6 +29,7 @@ #include #include #include +#include #include "channel_sync_gk20a.h" #include "gk20a.h" @@ -472,6 +473,23 @@ static void add_sema_cmd(struct gk20a *g, struct channel_gk20a *c, va, cmd->gva, cmd->mem->gpu_va, ob); } +void gk20a_channel_gen_sema_wait_cmd(struct channel_gk20a *c, + struct nvgpu_semaphore *sema, struct priv_cmd_entry *wait_cmd, + u32 wait_cmd_size, int pos) +{ + if (!sema) { + /* expired */ + nvgpu_memset(c->g, wait_cmd->mem, + (wait_cmd->off + pos * wait_cmd_size) * sizeof(u32), + 0, wait_cmd_size * sizeof(u32)); + } else { + WARN_ON(!sema->incremented); + add_sema_cmd(c->g, c, sema, wait_cmd, + pos * wait_cmd_size, true, false); + nvgpu_semaphore_put(sema); + } +} + static int gk20a_channel_semaphore_wait_syncpt( struct gk20a_channel_sync *s, u32 id, u32 thresh, struct priv_cmd_entry *entry) @@ -483,64 +501,6 @@ static int gk20a_channel_semaphore_wait_syncpt( return -ENODEV; } -#ifdef CONFIG_SYNC -static int semaphore_wait_fd_native(struct channel_gk20a *c, int fd, - struct priv_cmd_entry *wait_cmd, int max_wait_cmds) -{ - struct sync_fence *sync_fence; - int err; - const int wait_cmd_size = 8; - int num_wait_cmds; - int i; - - sync_fence = gk20a_sync_fence_fdget(fd); - if (!sync_fence) - return -EINVAL; - - num_wait_cmds = sync_fence->num_fences; - if (num_wait_cmds == 0) { - err = 0; - goto put_fence; - } - - if (max_wait_cmds && sync_fence->num_fences > max_wait_cmds) { - err = -EINVAL; - goto put_fence; - } - - err = gk20a_channel_alloc_priv_cmdbuf(c, - wait_cmd_size * num_wait_cmds, - wait_cmd); - if (err) { - nvgpu_err(c->g, "not enough priv cmd buffer space"); - goto put_fence; - } - - for (i = 0; i < sync_fence->num_fences; i++) { - struct fence *f = sync_fence->cbs[i].sync_pt; - struct sync_pt *pt = sync_pt_from_fence(f); - struct nvgpu_semaphore *sema; - - sema = gk20a_sync_pt_sema(pt); - if (!sema) { - /* expired */ - nvgpu_memset(c->g, wait_cmd->mem, - (wait_cmd->off + i * wait_cmd_size) * sizeof(u32), - 0, wait_cmd_size * sizeof(u32)); - } else { - WARN_ON(!sema->incremented); - add_sema_cmd(c->g, c, sema, wait_cmd, - i * wait_cmd_size, true, false); - nvgpu_semaphore_put(sema); - } - } - -put_fence: - sync_fence_put(sync_fence); - return err; -} -#endif - static int gk20a_channel_semaphore_wait_fd( struct gk20a_channel_sync *s, int fd, struct priv_cmd_entry *entry, int max_wait_cmds) @@ -548,13 +508,20 @@ static int gk20a_channel_semaphore_wait_fd( struct gk20a_channel_semaphore *sema = container_of(s, struct gk20a_channel_semaphore, ops); struct channel_gk20a *c = sema->c; -#ifdef CONFIG_SYNC - return semaphore_wait_fd_native(c, fd, entry, max_wait_cmds); -#else - nvgpu_err(c->g, - "trying to use sync fds with CONFIG_SYNC disabled"); - return -ENODEV; -#endif + + struct nvgpu_os_fence os_fence = {0}; + int err; + + err = nvgpu_os_fence_fdget(&os_fence, c, fd); + if (err) + return err; + + err = os_fence.ops->program_waits(&os_fence, + entry, c, max_wait_cmds); + + os_fence.ops->drop_ref(&os_fence); + + return err; } static int __gk20a_channel_semaphore_incr( @@ -570,6 +537,7 @@ static int __gk20a_channel_semaphore_incr( struct nvgpu_semaphore *semaphore; int err = 0; struct sync_fence *sync_fence = NULL; + struct nvgpu_os_fence os_fence = {0}; semaphore = nvgpu_semaphore_alloc(c); if (!semaphore) { @@ -589,18 +557,15 @@ static int __gk20a_channel_semaphore_incr( /* Release the completion semaphore. */ add_sema_cmd(c->g, c, semaphore, incr_cmd, 0, false, wfi_cmd); -#ifdef CONFIG_SYNC if (need_sync_fence) { - sync_fence = gk20a_sync_fence_create(c, - semaphore, "f-gk20a-0x%04x", - nvgpu_semaphore_gpu_ro_va(semaphore)); + err = nvgpu_os_fence_sema_create(&os_fence, c, + semaphore); - if (!sync_fence) { - err = -ENOMEM; + if (err) goto clean_up_sema; - } + + sync_fence = (struct sync_fence *)os_fence.priv; } -#endif err = gk20a_fence_from_semaphore(fence, semaphore, @@ -608,10 +573,8 @@ static int __gk20a_channel_semaphore_incr( sync_fence); if (err) { -#ifdef CONFIG_SYNC - if (sync_fence) - sync_fence_put(sync_fence); -#endif + if (nvgpu_os_fence_is_initialized(&os_fence)) + os_fence.ops->drop_ref(&os_fence); goto clean_up_sema; } diff --git a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.h b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.h index d63b358f..565b4f86 100644 --- a/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.h +++ b/drivers/gpu/nvgpu/gk20a/channel_sync_gk20a.h @@ -32,6 +32,7 @@ struct priv_cmd_entry; struct channel_gk20a; struct gk20a_fence; struct gk20a; +struct nvgpu_semaphore; struct gk20a_channel_sync { nvgpu_atomic_t refcount; @@ -103,6 +104,10 @@ struct gk20a_channel_sync { void (*destroy)(struct gk20a_channel_sync *s); }; +void gk20a_channel_gen_sema_wait_cmd(struct channel_gk20a *c, + struct nvgpu_semaphore *sema, struct priv_cmd_entry *wait_cmd, + u32 wait_cmd_size, int pos); + void gk20a_channel_sync_destroy(struct gk20a_channel_sync *sync, bool set_safe_state); struct gk20a_channel_sync *gk20a_channel_sync_create(struct channel_gk20a *c, diff --git a/drivers/gpu/nvgpu/include/nvgpu/linux/os_fence_android.h b/drivers/gpu/nvgpu/include/nvgpu/linux/os_fence_android.h new file mode 100644 index 00000000..79cc51ea --- /dev/null +++ b/drivers/gpu/nvgpu/include/nvgpu/linux/os_fence_android.h @@ -0,0 +1,42 @@ +/* + * Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#ifndef __NVGPU_OS_FENCE_ANDROID_H__ +#define __NVGPU_OS_FENCE_ANDROID_H__ + +struct gk20a; +struct nvgpu_os_fence; +struct sync_fence; +struct channel_gk20a; + +struct sync_fence *nvgpu_get_sync_fence(struct nvgpu_os_fence *s); + +void nvgpu_os_fence_android_drop_ref(struct nvgpu_os_fence *s); + +int nvgpu_os_fence_sema_fdget(struct nvgpu_os_fence *fence_out, + struct channel_gk20a *c, int fd); + +void nvgpu_os_fence_init(struct nvgpu_os_fence *fence_out, + struct gk20a *g, const struct nvgpu_os_fence_ops *fops, + struct sync_fence *fence); + +#endif \ No newline at end of file diff --git a/drivers/gpu/nvgpu/include/nvgpu/os_fence.h b/drivers/gpu/nvgpu/include/nvgpu/os_fence.h new file mode 100644 index 00000000..c8d24fc2 --- /dev/null +++ b/drivers/gpu/nvgpu/include/nvgpu/os_fence.h @@ -0,0 +1,111 @@ +/* + * nvgpu os fence + * + * Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved. + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER + * DEALINGS IN THE SOFTWARE. + */ + +#ifndef __NVGPU_OS_FENCE__ +#define __NVGPU_OS_FENCE__ + +struct nvgpu_semaphore; +struct channel_gk20a; +struct priv_cmd_entry; + +/* + * struct nvgpu_os_fence adds an abstraction to the earlier Android Sync + * Framework, specifically the sync-fence mechanism and the newer DMA sync + * APIs from linux-4.9. This abstraction provides the high-level definition + * as well as APIs that can be used by other OSes in future to have their own + * alternatives for the sync-framework. + */ +struct nvgpu_os_fence; + +/* + * struct nvgpu_os_fence depends on the following ops structure + */ +struct nvgpu_os_fence_ops { + /* + * This API is used to iterate through multiple fence points within the + * fence and program the pushbuffer method for wait command. + */ + int (*program_waits)(struct nvgpu_os_fence *s, + struct priv_cmd_entry *wait_cmd, + struct channel_gk20a *c, + int max_wait_cmds); + + /* + * This should be the last operation on the OS fence. The + * OS fence acts as a place-holder for the underlying fence + * implementation e.g. sync_fences. For each construct/fdget call + * there needs to be a drop_ref call. This reduces a reference count + * for the underlying sync_fence. + */ + void (*drop_ref)(struct nvgpu_os_fence *s); +}; + +/* + * The priv structure here is used to contain the struct sync_fence + * for LINUX_VERSION <= 4.9 and dma_fence for LINUX_VERSION > 4.9 + */ +struct nvgpu_os_fence { + void *priv; + struct gk20a *g; + const struct nvgpu_os_fence_ops *ops; +}; + +/* + * This API is used to validate the nvgpu_os_fence + */ +static inline int nvgpu_os_fence_is_initialized(struct nvgpu_os_fence *fence) +{ + return (fence->ops != NULL); +} + +#ifdef CONFIG_SYNC + +int nvgpu_os_fence_sema_create( + struct nvgpu_os_fence *fence_out, + struct channel_gk20a *c, + struct nvgpu_semaphore *sema); + +int nvgpu_os_fence_fdget( + struct nvgpu_os_fence *fence_out, + struct channel_gk20a *c, int fd); + +#else + +static inline int nvgpu_os_fence_sema_create( + struct nvgpu_os_fence *fence_out, + struct channel_gk20a *c, + struct nvgpu_semaphore *sema) +{ + return -ENOSYS; +} +static inline int nvgpu_os_fence_fdget( + struct nvgpu_os_fence *fence_out, + struct channel_gk20a *c, int fd) +{ + return -ENOSYS; +} + +#endif /* CONFIG_SYNC */ + +#endif /* __NVGPU_OS_FENCE__ */ -- cgit v1.2.2