From b3446bc0b6fca6cb992667f80a95f8503b6a652a Mon Sep 17 00:00:00 2001 From: Alex Waterman Date: Thu, 7 Sep 2017 15:27:55 -0700 Subject: gpu: nvgpu: Move dma_buf usage from mm_gk20a.c Move most of the dma_buf usage present in the mm_gk20a.c code out to Linux specific code and some commom/mm code. There's two primary groups of code: 1. dma_buf priv field code (for holding comptag data) 2. Comptag usage that relies on dma_buf pointers For (1) the dma_buf code was simply moved to common/linux/dmabuf.c since most of this code is clearly Linux specific. The comptag code was a bit more complicated since there is two parts to the comptag code. Firstly there's the code that manages the comptag memory. This is essentially a simple allocator. This was moved to common/mm/comptags.c since it can be shared across all chips. The second set of code is moved to common/linux/comptags.c since it is the interface between dma_bufs and the comptag memory. Two other fixes were done as well: - Add struct gk20a to the comptag allocator init so that the proper nvgpu_vzalloc() function could be used. - Add necessary includes to common/linux/vm_priv.h. JIRA NVGPU-30 JIRA NVGPU-138 Change-Id: I96c57f2763e5ebe18a2f2ee4b33e0e1a2597848c Signed-off-by: Alex Waterman Reviewed-on: https://git-master.nvidia.com/r/1566628 Reviewed-by: mobile promotions Tested-by: mobile promotions --- drivers/gpu/nvgpu/common/linux/cde.c | 1 + drivers/gpu/nvgpu/common/linux/comptags.c | 70 +++++++++ drivers/gpu/nvgpu/common/linux/dmabuf.c | 247 ++++++++++++++++++++++++++++++ drivers/gpu/nvgpu/common/linux/dmabuf.h | 73 +++++++++ drivers/gpu/nvgpu/common/linux/vm.c | 29 +--- drivers/gpu/nvgpu/common/linux/vm_priv.h | 5 - drivers/gpu/nvgpu/common/mm/comptags.c | 95 ++++++++++++ 7 files changed, 487 insertions(+), 33 deletions(-) create mode 100644 drivers/gpu/nvgpu/common/linux/comptags.c create mode 100644 drivers/gpu/nvgpu/common/linux/dmabuf.c create mode 100644 drivers/gpu/nvgpu/common/linux/dmabuf.h create mode 100644 drivers/gpu/nvgpu/common/mm/comptags.c (limited to 'drivers/gpu/nvgpu/common') diff --git a/drivers/gpu/nvgpu/common/linux/cde.c b/drivers/gpu/nvgpu/common/linux/cde.c index c3a9b770..577d86e8 100644 --- a/drivers/gpu/nvgpu/common/linux/cde.c +++ b/drivers/gpu/nvgpu/common/linux/cde.c @@ -39,6 +39,7 @@ #include "cde.h" #include "os_linux.h" +#include "dmabuf.h" #include #include diff --git a/drivers/gpu/nvgpu/common/linux/comptags.c b/drivers/gpu/nvgpu/common/linux/comptags.c new file mode 100644 index 00000000..517429d8 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/comptags.c @@ -0,0 +1,70 @@ +/* +* Copyright (c) 2017, 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 "dmabuf.h" + +void gk20a_get_comptags(struct device *dev, struct dma_buf *dmabuf, + struct gk20a_comptags *comptags) +{ + struct gk20a_dmabuf_priv *priv = dma_buf_get_drvdata(dmabuf, dev); + + if (!comptags) + return; + + if (!priv) { + memset(comptags, 0, sizeof(*comptags)); + return; + } + + *comptags = priv->comptags; +} + +int gk20a_alloc_comptags(struct gk20a *g, + struct device *dev, + struct dma_buf *dmabuf, + struct gk20a_comptag_allocator *allocator, + u32 lines) +{ + struct gk20a_dmabuf_priv *priv = dma_buf_get_drvdata(dmabuf, dev); + u32 ctaglines_allocsize; + u32 offset; + int err; + + if (!priv) + return -ENOSYS; + + if (!lines) + return -EINVAL; + + ctaglines_allocsize = lines; + + /* store the allocator so we can use it when we free the ctags */ + priv->comptag_allocator = allocator; + err = gk20a_comptaglines_alloc(allocator, &offset, + ctaglines_allocsize); + if (err) + return err; + + priv->comptags.offset = offset; + priv->comptags.lines = lines; + priv->comptags.allocated_lines = ctaglines_allocsize; + + return 0; +} diff --git a/drivers/gpu/nvgpu/common/linux/dmabuf.c b/drivers/gpu/nvgpu/common/linux/dmabuf.c new file mode 100644 index 00000000..0b07b255 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/dmabuf.c @@ -0,0 +1,247 @@ +/* +* Copyright (c) 2017, 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 + +#include "gk20a/gk20a.h" +#include "gk20a/platform_gk20a.h" + +#include "dmabuf.h" +#include "vm_priv.h" +#include "os_linux.h" + +static void gk20a_mm_delete_priv(void *_priv) +{ + struct gk20a_buffer_state *s, *s_tmp; + struct gk20a_dmabuf_priv *priv = _priv; + struct gk20a *g; + + if (!priv) + return; + + g = priv->g; + + if (priv->comptags.lines) { + BUG_ON(!priv->comptag_allocator); + gk20a_comptaglines_free(priv->comptag_allocator, + priv->comptags.offset, + priv->comptags.allocated_lines); + } + + /* Free buffer states */ + nvgpu_list_for_each_entry_safe(s, s_tmp, &priv->states, + gk20a_buffer_state, list) { + gk20a_fence_put(s->fence); + nvgpu_list_del(&s->list); + nvgpu_kfree(g, s); + } + + nvgpu_kfree(g, priv); +} + +enum nvgpu_aperture gk20a_dmabuf_aperture(struct gk20a *g, + struct dma_buf *dmabuf) +{ + struct gk20a *buf_owner = nvgpu_vidmem_buf_owner(dmabuf); + bool unified_memory = nvgpu_is_enabled(g, NVGPU_MM_UNIFIED_MEMORY); + + if (buf_owner == NULL) { + /* Not nvgpu-allocated, assume system memory */ + return APERTURE_SYSMEM; + } else if (WARN_ON(buf_owner == g && unified_memory)) { + /* Looks like our video memory, but this gpu doesn't support + * it. Warn about a bug and bail out */ + nvgpu_warn(g, + "dmabuf is our vidmem but we don't have local vidmem"); + return APERTURE_INVALID; + } else if (buf_owner != g) { + /* Someone else's vidmem */ + return APERTURE_INVALID; + } else { + /* Yay, buf_owner == g */ + return APERTURE_VIDMEM; + } +} + +struct sg_table *gk20a_mm_pin(struct device *dev, struct dma_buf *dmabuf) +{ + struct gk20a_dmabuf_priv *priv; + + priv = dma_buf_get_drvdata(dmabuf, dev); + if (WARN_ON(!priv)) + return ERR_PTR(-EINVAL); + + nvgpu_mutex_acquire(&priv->lock); + + if (priv->pin_count == 0) { + priv->attach = dma_buf_attach(dmabuf, dev); + if (IS_ERR(priv->attach)) { + nvgpu_mutex_release(&priv->lock); + return (struct sg_table *)priv->attach; + } + + priv->sgt = dma_buf_map_attachment(priv->attach, + DMA_BIDIRECTIONAL); + if (IS_ERR(priv->sgt)) { + dma_buf_detach(dmabuf, priv->attach); + nvgpu_mutex_release(&priv->lock); + return priv->sgt; + } + } + + priv->pin_count++; + nvgpu_mutex_release(&priv->lock); + return priv->sgt; +} + +void gk20a_mm_unpin(struct device *dev, struct dma_buf *dmabuf, + struct sg_table *sgt) +{ + struct gk20a_dmabuf_priv *priv = dma_buf_get_drvdata(dmabuf, dev); + dma_addr_t dma_addr; + + if (IS_ERR(priv) || !priv) + return; + + nvgpu_mutex_acquire(&priv->lock); + WARN_ON(priv->sgt != sgt); + priv->pin_count--; + WARN_ON(priv->pin_count < 0); + dma_addr = sg_dma_address(priv->sgt->sgl); + if (priv->pin_count == 0) { + dma_buf_unmap_attachment(priv->attach, priv->sgt, + DMA_BIDIRECTIONAL); + dma_buf_detach(dmabuf, priv->attach); + } + nvgpu_mutex_release(&priv->lock); +} + +int gk20a_dmabuf_alloc_drvdata(struct dma_buf *dmabuf, struct device *dev) +{ + struct gk20a *g = gk20a_get_platform(dev)->g; + struct gk20a_dmabuf_priv *priv; + static u64 priv_count = 0; + + priv = dma_buf_get_drvdata(dmabuf, dev); + if (likely(priv)) + return 0; + + nvgpu_mutex_acquire(&g->mm.priv_lock); + priv = dma_buf_get_drvdata(dmabuf, dev); + if (priv) + goto priv_exist_or_err; + + priv = nvgpu_kzalloc(g, sizeof(*priv)); + if (!priv) { + priv = ERR_PTR(-ENOMEM); + goto priv_exist_or_err; + } + + nvgpu_mutex_init(&priv->lock); + nvgpu_init_list_node(&priv->states); + priv->buffer_id = ++priv_count; + priv->g = g; + dma_buf_set_drvdata(dmabuf, dev, priv, gk20a_mm_delete_priv); + +priv_exist_or_err: + nvgpu_mutex_release(&g->mm.priv_lock); + if (IS_ERR(priv)) + return -ENOMEM; + + return 0; +} + +int gk20a_dmabuf_get_state(struct dma_buf *dmabuf, struct gk20a *g, + u64 offset, struct gk20a_buffer_state **state) +{ + int err = 0; + struct gk20a_dmabuf_priv *priv; + struct gk20a_buffer_state *s; + struct device *dev = dev_from_gk20a(g); + + if (WARN_ON(offset >= (u64)dmabuf->size)) + return -EINVAL; + + err = gk20a_dmabuf_alloc_drvdata(dmabuf, dev); + if (err) + return err; + + priv = dma_buf_get_drvdata(dmabuf, dev); + if (WARN_ON(!priv)) + return -ENOSYS; + + nvgpu_mutex_acquire(&priv->lock); + + nvgpu_list_for_each_entry(s, &priv->states, gk20a_buffer_state, list) + if (s->offset == offset) + goto out; + + /* State not found, create state. */ + s = nvgpu_kzalloc(g, sizeof(*s)); + if (!s) { + err = -ENOMEM; + goto out; + } + + s->offset = offset; + nvgpu_init_list_node(&s->list); + nvgpu_mutex_init(&s->lock); + nvgpu_list_add_tail(&s->list, &priv->states); + +out: + nvgpu_mutex_release(&priv->lock); + if (!err) + *state = s; + return err; +} + +int gk20a_mm_get_buffer_info(struct device *dev, int dmabuf_fd, + u64 *buffer_id, u64 *buffer_len) +{ + struct dma_buf *dmabuf; + struct gk20a_dmabuf_priv *priv; + int err = 0; + + dmabuf = dma_buf_get(dmabuf_fd); + if (IS_ERR(dmabuf)) { + dev_warn(dev, "%s: fd %d is not a dmabuf", __func__, dmabuf_fd); + return PTR_ERR(dmabuf); + } + + err = gk20a_dmabuf_alloc_drvdata(dmabuf, dev); + if (err) { + dev_warn(dev, "Failed to allocate dmabuf drvdata (err = %d)", + err); + goto clean_up; + } + + priv = dma_buf_get_drvdata(dmabuf, dev); + if (likely(priv)) { + *buffer_id = priv->buffer_id; + *buffer_len = dmabuf->size; + } + +clean_up: + dma_buf_put(dmabuf); + return err; +} diff --git a/drivers/gpu/nvgpu/common/linux/dmabuf.h b/drivers/gpu/nvgpu/common/linux/dmabuf.h new file mode 100644 index 00000000..718386c5 --- /dev/null +++ b/drivers/gpu/nvgpu/common/linux/dmabuf.h @@ -0,0 +1,73 @@ +/* +* Copyright (c) 2017, 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 . + */ + +#ifndef __COMMON_LINUX_DMABUF_H__ +#define __COMMON_LINUX_DMABUF_H__ + +#include +#include +#include +#include + +struct sg_table; +struct dma_buf; +struct dma_buf_attachment; +struct device; + +struct gk20a; +struct gk20a_buffer_state; + +struct gk20a_dmabuf_priv { + struct nvgpu_mutex lock; + + struct gk20a *g; + + struct gk20a_comptag_allocator *comptag_allocator; + struct gk20a_comptags comptags; + + struct dma_buf_attachment *attach; + struct sg_table *sgt; + + int pin_count; + + struct nvgpu_list_node states; + + u64 buffer_id; +}; + +/* + * These are implemented in common/linux/comptags.c - these are dmabuf related + * functions though so they are defined here. They cannot be defined in + * since that file must be OS agnostic. + */ +int gk20a_alloc_comptags(struct gk20a *g, + struct device *dev, + struct dma_buf *dmabuf, + struct gk20a_comptag_allocator *allocator, + u32 lines); +void gk20a_get_comptags(struct device *dev, struct dma_buf *dmabuf, + struct gk20a_comptags *comptags); + +struct sg_table *gk20a_mm_pin(struct device *dev, struct dma_buf *dmabuf); +void gk20a_mm_unpin(struct device *dev, struct dma_buf *dmabuf, + struct sg_table *sgt); + +int gk20a_dmabuf_alloc_drvdata(struct dma_buf *dmabuf, struct device *dev); + +int gk20a_dmabuf_get_state(struct dma_buf *dmabuf, struct gk20a *g, + u64 offset, struct gk20a_buffer_state **state); + +#endif diff --git a/drivers/gpu/nvgpu/common/linux/vm.c b/drivers/gpu/nvgpu/common/linux/vm.c index 45058321..b686d616 100644 --- a/drivers/gpu/nvgpu/common/linux/vm.c +++ b/drivers/gpu/nvgpu/common/linux/vm.c @@ -24,7 +24,6 @@ #include #include #include -#include #include #include @@ -36,33 +35,7 @@ #include "vm_priv.h" #include "os_linux.h" - -/* - * Temporary location for this code until a dmabuf.c file exists. - */ -enum nvgpu_aperture gk20a_dmabuf_aperture(struct gk20a *g, - struct dma_buf *dmabuf) -{ - struct gk20a *buf_owner = nvgpu_vidmem_buf_owner(dmabuf); - bool unified_memory = nvgpu_is_enabled(g, NVGPU_MM_UNIFIED_MEMORY); - - if (buf_owner == NULL) { - /* Not nvgpu-allocated, assume system memory */ - return APERTURE_SYSMEM; - } else if (WARN_ON(buf_owner == g && unified_memory)) { - /* Looks like our video memory, but this gpu doesn't support - * it. Warn about a bug and bail out */ - nvgpu_warn(g, - "dmabuf is our vidmem but we don't have local vidmem"); - return APERTURE_INVALID; - } else if (buf_owner != g) { - /* Someone else's vidmem */ - return APERTURE_INVALID; - } else { - /* Yay, buf_owner == g */ - return APERTURE_VIDMEM; - } -} +#include "dmabuf.h" static struct nvgpu_mapped_buf *__nvgpu_vm_find_mapped_buf_reverse( struct vm_gk20a *vm, struct dma_buf *dmabuf, u32 kind) diff --git a/drivers/gpu/nvgpu/common/linux/vm_priv.h b/drivers/gpu/nvgpu/common/linux/vm_priv.h index 1eadf1d0..fa173d59 100644 --- a/drivers/gpu/nvgpu/common/linux/vm_priv.h +++ b/drivers/gpu/nvgpu/common/linux/vm_priv.h @@ -102,10 +102,5 @@ int setup_buffer_kind_and_compression(struct vm_gk20a *vm, u32 flags, struct buffer_attrs *bfr, enum gmmu_pgsz_gk20a pgsz_idx); -int gk20a_alloc_comptags(struct gk20a *g, - struct device *dev, - struct dma_buf *dmabuf, - struct gk20a_comptag_allocator *allocator, - u32 lines); #endif diff --git a/drivers/gpu/nvgpu/common/mm/comptags.c b/drivers/gpu/nvgpu/common/mm/comptags.c new file mode 100644 index 00000000..01ab646a --- /dev/null +++ b/drivers/gpu/nvgpu/common/mm/comptags.c @@ -0,0 +1,95 @@ +/* + * Copyright (c) 2017, 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. + */ + +#include +#include + +#include "gk20a/gk20a.h" + +int gk20a_comptaglines_alloc(struct gk20a_comptag_allocator *allocator, + u32 *offset, u32 len) +{ + unsigned long addr; + int err = 0; + + nvgpu_mutex_acquire(&allocator->lock); + addr = bitmap_find_next_zero_area(allocator->bitmap, allocator->size, + 0, len, 0); + if (addr < allocator->size) { + /* number zero is reserved; bitmap base is 1 */ + *offset = 1 + addr; + bitmap_set(allocator->bitmap, addr, len); + } else { + err = -ENOMEM; + } + nvgpu_mutex_release(&allocator->lock); + + return err; +} + +void gk20a_comptaglines_free(struct gk20a_comptag_allocator *allocator, + u32 offset, u32 len) +{ + /* number zero is reserved; bitmap base is 1 */ + u32 addr = offset - 1; + + WARN_ON(offset == 0); + WARN_ON(addr > allocator->size); + WARN_ON(addr + len > allocator->size); + + nvgpu_mutex_acquire(&allocator->lock); + bitmap_clear(allocator->bitmap, addr, len); + nvgpu_mutex_release(&allocator->lock); +} + +int gk20a_comptag_allocator_init(struct gk20a *g, + struct gk20a_comptag_allocator *allocator, + unsigned long size) +{ + nvgpu_mutex_init(&allocator->lock); + + /* + * 0th comptag is special and is never used. The base for this bitmap + * is 1, and its size is one less than the size of comptag store. + */ + size--; + allocator->bitmap = nvgpu_vzalloc(g, + BITS_TO_LONGS(size) * sizeof(long)); + if (!allocator->bitmap) + return -ENOMEM; + + allocator->size = size; + + return 0; +} + +void gk20a_comptag_allocator_destroy(struct gk20a *g, + struct gk20a_comptag_allocator *allocator) +{ + /* + * called only when exiting the driver (gk20a_remove, or unwinding the + * init stage); no users should be active, so taking the mutex is + * unnecessary here. + */ + allocator->size = 0; + nvgpu_vfree(allocator->g, allocator->bitmap); +} -- cgit v1.2.2