gpu: nvgpu: RDMA implementation

This change adds RDMA supports for tegra iGPU.
1. Cuda Process allocates the memory and passes
   the VA and size to the custom kernel driver.
2. The custom kernel driver maps the user allocated
   buf and does the DMA to/from it.
3. Only supports iGPU + cudaHostAlloc sysmem
4. Works only for a given process.
5. Address should be sysmem page aligned and size should
   be multiple of sysmem page size.
6. The custom kernel driver must register a free_callback when get_page()
   function is called.

Bug 200438879

Signed-off-by: Preetham Chandru R <pchandru@nvidia.com>
Change-Id: I43ec45734eb46d30341d0701550206c16e051106
Reviewed-on: https://git-master.nvidia.com/r/1953780
GVS: Gerrit_Virtual_Submit
Reviewed-by: Alex Waterman <alexw@nvidia.com>
Reviewed-by: Bibek Basu <bbasu@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

4 files changed, 502 insertions, 0 deletions

diff --git a/drivers/gpu/nvgpu/Kconfig b/drivers/gpu/nvgpu/Kconfig
index bbd5f6f0..7dba61a3 100644
--- a/drivers/gpu/nvgpu/Kconfig
+++ b/drivers/gpu/nvgpu/Kconfig
@@ -40,6 +40,13 @@ config GK20A_PM_QOS
           Enable support to pass PM_QOS constraints to devfreq based
           scaling.
 
+config GK20A_RDMA
+        bool "Support GK20A RDMA"
+        depends on GK20A && MMU_NOTIFIER
+        default n
+        help
+          Say Y here to enable GK20A RDMA features.
+
 config NVGPU_TRACK_MEM_USAGE
         bool "Track the usage of system memory in nvgpu"
         depends on GK20A
diff --git a/drivers/gpu/nvgpu/Makefile b/drivers/gpu/nvgpu/Makefile
index e0fd70e6..fdfaf092 100644
--- a/drivers/gpu/nvgpu/Makefile
+++ b/drivers/gpu/nvgpu/Makefile
@@ -176,6 +176,9 @@ nvgpu-$(CONFIG_NVGPU_SUPPORT_CDE) += \
         os/linux/cde_gm20b.o \
         os/linux/cde_gp10b.o
 
+nvgpu-$(CONFIG_GK20A_RDMA) += \
+        os/linux/nvidia_p2p.o
+
 ifeq ($(CONFIG_DEBUG_FS),y)
 nvgpu-$(CONFIG_NVGPU_SUPPORT_CDE) += \
         os/linux/debug_cde.o
diff --git a/drivers/gpu/nvgpu/include/nvgpu/linux/nvidia_p2p.h b/drivers/gpu/nvgpu/include/nvgpu/linux/nvidia_p2p.h
new file mode 100644
index 00000000..c1dee7cf
--- /dev/null
+++ b/drivers/gpu/nvgpu/include/nvgpu/linux/nvidia_p2p.h
@@ -0,0 +1,196 @@
+/*
+ * 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 __NVIDIA_P2P_H__
+#define __NVIDIA_P2P_H__
+
+#include <linux/dma-mapping.h>
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/miscdevice.h>
+#include <linux/fs.h>
+#include <linux/mm.h>
+#include <linux/mman.h>
+#include <linux/mmu_notifier.h>
+#include <linux/types.h>
+#include <linux/uaccess.h>
+#include <linux/sched.h>
+#include <linux/spinlock.h>
+#include <linux/slab.h>
+#include <linux/highmem.h>
+#include <linux/mutex.h>
+#include <linux/spinlock.h>
+#include <nvgpu/linux/lock.h>
+
+#define NVIDIA_P2P_UNINITIALIZED 0x0
+#define NVIDIA_P2P_PINNED 0x1
+#define NVIDIA_P2P_MAPPED 0x2
+
+enum nvidia_p2p_page_size_type {
+        NVIDIA_P2P_PAGE_SIZE_4KB = 0,
+        NVIDIA_P2P_PAGE_SIZE_64KB,
+        NVIDIA_P2P_PAGE_SIZE_128KB,
+        NVIDIA_P2P_PAGE_SIZE_COUNT
+};
+
+struct nvidia_p2p_page_table {
+        u32 page_size;
+        u64 size;
+        u32 entries;
+        struct page **pages;
+
+        u64 vaddr;
+        u32 mapped;
+
+        struct mm_struct *mm;
+        struct mmu_notifier mn;
+        struct nvgpu_mutex lock;
+        void (*free_callback)(void *data);
+        void *data;
+};
+
+struct nvidia_p2p_dma_mapping {
+        dma_addr_t *hw_address;
+        u32 *hw_len;
+        u32 entries;
+
+        struct sg_table *sgt;
+        struct device *dev;
+        struct nvidia_p2p_page_table *page_table;
+        enum dma_data_direction direction;
+};
+
+/*
+ * @brief
+ *   Make the pages underlying a range of GPU virtual memory
+ *   accessible to a third-party device.
+ *
+ * @param[in]     vaddr
+ *   A GPU Virtual Address
+ * @param[in]     size
+ *   The size of the requested mapping.
+ *   Size must be a multiple of Page size.
+ * @param[out]    **page_table
+ *   A pointer to struct nvidia_p2p_page_table
+ * @param[in]     free_callback
+ *   A non-NULL pointer to the function to be invoked when the pages
+ *   underlying the virtual address range are freed
+ *   implicitly. Must be non NULL.
+ * @param[in]     data
+ *   A non-NULL opaque pointer to private data to be passed to the
+ *   callback function.
+ *
+ * @return
+ *    0           upon successful completion.
+ *    Negative number if any error
+ */
+int nvidia_p2p_get_pages(u64 vaddr, u64 size,
+                        struct nvidia_p2p_page_table **page_table,
+                        void (*free_callback)(void *data), void *data);
+/*
+ * @brief
+ * Release the pages previously made accessible to
+ * a third-party device.
+ *
+ * @param[in]    *page_table
+ *   A pointer to struct nvidia_p2p_page_table
+ *
+ * @return
+ *    0           upon successful completion.
+ *   -ENOMEM      if the driver failed to allocate memory or if
+ *     insufficient resources were available to complete the operation.
+ *    Negative number if any other error
+ */
+int nvidia_p2p_put_pages(struct nvidia_p2p_page_table *page_table);
+
+/*
+ * @brief
+ * Release the pages previously made accessible to
+ * a third-party device. This is called  during the
+ * execution of the free_callback().
+ *
+ * @param[in]    *page_table
+ *   A pointer to struct nvidia_p2p_page_table
+ *
+ * @return
+ *    0           upon successful completion.
+ *   -ENOMEM      if the driver failed to allocate memory or if
+ *     insufficient resources were available to complete the operation.
+ *    Negative number if any other error
+ */
+int nvidia_p2p_free_page_table(struct nvidia_p2p_page_table *page_table);
+
+/*
+ * @brief
+ *   Map the pages retrieved using nvidia_p2p_get_pages and
+ *   pass the dma address to a third-party device.
+ *
+ * @param[in]   *dev
+ *   The peer device that needs to DMA to/from the
+ *   mapping.
+ * @param[in]   *page_table
+ *   A pointer to struct nvidia_p2p_page_table
+ * @param[out]  **map
+ *   A pointer to struct nvidia_p2p_dma_mapping.
+ *   The DMA mapping containing the DMA addresses to use.
+ * @param[in]    direction
+ *   DMA direction
+ *
+ * @return
+ *    0           upon successful completion.
+ *    Negative number if any other error
+ */
+int nvidia_p2p_map_pages(struct device *dev,
+                struct nvidia_p2p_page_table *page_table,
+                struct nvidia_p2p_dma_mapping **map,
+                enum dma_data_direction direction);
+/*
+ * @brief
+ *   Unmap the pages previously mapped using nvidia_p2p_map_pages
+ *
+ * @param[in]   *map
+ *   A pointer to struct nvidia_p2p_dma_mapping.
+ *   The DMA mapping containing the DMA addresses to use.
+ *
+ * @return
+ *    0           upon successful completion.
+ *    Negative number if any other error
+ */
+int nvidia_p2p_unmap_pages(struct nvidia_p2p_dma_mapping *map);
+
+/*
+ * @brief
+ *   Unmap the pages previously mapped using nvidia_p2p_map_pages.
+ *  This is called  during the  execution of the free_callback().
+ *
+ * @param[in]   *map
+ *   A pointer to struct nvidia_p2p_dma_mapping.
+ *   The DMA mapping containing the DMA addresses to use.
+ *
+ * @return
+ *    0           upon successful completion.
+ *    Negative number if any other error
+ */
+int nvidia_p2p_free_dma_mapping(struct nvidia_p2p_dma_mapping *dma_mapping);
+
+#endif
diff --git a/drivers/gpu/nvgpu/os/linux/nvidia_p2p.c b/drivers/gpu/nvgpu/os/linux/nvidia_p2p.c
new file mode 100644
index 00000000..dcd37bee
--- /dev/null
+++ b/drivers/gpu/nvgpu/os/linux/nvidia_p2p.c
@@ -0,0 +1,296 @@
+/*
+ * 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.
+ */
+
+#include <linux/slab.h>
+#include <nvgpu/linux/nvidia_p2p.h>
+
+static void nvidia_p2p_mn_release(struct mmu_notifier *mn,
+        struct mm_struct *mm)
+{
+        struct nvidia_p2p_page_table *page_table = container_of(mn,
+                                                struct nvidia_p2p_page_table,
+                                                mn);
+
+        page_table->free_callback(page_table->data);
+}
+
+static void nvidia_p2p_mn_invl_range_start(struct mmu_notifier *mn,
+        struct mm_struct *mm, unsigned long start, unsigned long end)
+{
+        struct nvidia_p2p_page_table *page_table = container_of(mn,
+                                                struct nvidia_p2p_page_table,
+                                                mn);
+        u64 vaddr = 0;
+        u64 size = 0;
+
+        vaddr = page_table->vaddr;
+        size = page_table->size;
+
+        if (vaddr >= start && vaddr <= end) {
+                mmu_notifier_unregister_no_release(&page_table->mn, page_table->mm);
+                page_table->free_callback(page_table->data);
+        }
+}
+
+static struct mmu_notifier_ops nvidia_p2p_mmu_ops = {
+        .release                = nvidia_p2p_mn_release,
+        .invalidate_range_start = nvidia_p2p_mn_invl_range_start,
+};
+
+int nvidia_p2p_get_pages(u64 vaddr, u64 size,
+                struct nvidia_p2p_page_table **page_table,
+                void (*free_callback)(void *data), void *data)
+{
+        int ret = 0;
+        int user_pages = 0;
+        int locked = 0;
+        int nr_pages = size >> PAGE_SHIFT;
+        struct page **pages;
+
+        if (nr_pages <= 0) {
+                return -EINVAL;
+        }
+
+        *page_table = kzalloc(sizeof(**page_table), GFP_KERNEL);
+        if (!*page_table) {
+                return -ENOMEM;
+        }
+
+        pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
+        if (!pages) {
+                ret = -ENOMEM;
+                goto free_page_table;
+        }
+        down_read(&current->mm->mmap_sem);
+        locked = 1;
+        user_pages = get_user_pages_locked(vaddr & PAGE_MASK, nr_pages,
+                        FOLL_WRITE | FOLL_FORCE,
+                        pages, &locked);
+        up_read(&current->mm->mmap_sem);
+        if (user_pages != nr_pages) {
+                ret = user_pages < 0 ? user_pages : -ENOMEM;
+                goto free_pages;
+        }
+
+        (*page_table)->pages = pages;
+        (*page_table)->entries = user_pages;
+        (*page_table)->page_size = NVIDIA_P2P_PAGE_SIZE_4KB;
+        (*page_table)->size = size;
+
+        (*page_table)->mn.ops = &nvidia_p2p_mmu_ops;
+        (*page_table)->mm = current->mm;
+        (*page_table)->free_callback = free_callback;
+        (*page_table)->data = data;
+        (*page_table)->vaddr = vaddr;
+        nvgpu_mutex_init(&(*page_table)->lock);
+        (*page_table)->mapped = NVIDIA_P2P_PINNED;
+
+        ret = mmu_notifier_register(&(*page_table)->mn, (*page_table)->mm);
+        if (ret) {
+                goto free_pages;
+        }
+
+        return 0;
+free_pages:
+        while (--user_pages >= 0) {
+                put_page(pages[user_pages]);
+        }
+        kfree(pages);
+free_page_table:
+        kfree(*page_table);
+        *page_table = NULL;
+        return ret;
+}
+EXPORT_SYMBOL(nvidia_p2p_get_pages);
+
+int nvidia_p2p_put_pages(struct nvidia_p2p_page_table *page_table)
+{
+        if (!page_table) {
+                return -EINVAL;
+        }
+
+        mmu_notifier_unregister(&page_table->mn, page_table->mm);
+
+        return 0;
+}
+EXPORT_SYMBOL(nvidia_p2p_put_pages);
+
+int nvidia_p2p_free_page_table(struct nvidia_p2p_page_table *page_table)
+{
+        int user_pages = 0;
+        struct page **pages = NULL;
+
+        if (!page_table) {
+                return 0;
+        }
+
+        nvgpu_mutex_acquire(&page_table->lock);
+
+        if (page_table->mapped & NVIDIA_P2P_MAPPED) {
+                WARN(1, "Attempting to free unmapped pages");
+        }
+
+        if (page_table->mapped & NVIDIA_P2P_PINNED) {
+                pages = page_table->pages;
+                user_pages = page_table->entries;
+
+                while (--user_pages >= 0) {
+                        put_page(pages[user_pages]);
+                }
+
+                kfree(pages);
+                page_table->mapped &= (u32)~NVIDIA_P2P_PINNED;
+        }
+
+        nvgpu_mutex_release(&page_table->lock);
+
+        return 0;
+}
+EXPORT_SYMBOL(nvidia_p2p_free_page_table);
+
+int nvidia_p2p_map_pages(struct device *dev,
+                struct nvidia_p2p_page_table *page_table,
+                struct nvidia_p2p_dma_mapping **dma_mapping,
+                enum dma_data_direction direction)
+{
+        struct sg_table *sgt = NULL;
+        struct scatterlist *sg;
+        struct page **pages = NULL;
+        u32 nr_pages = 0;
+        int ret = 0;
+        int i, count;
+
+        if (!page_table) {
+                return -EINVAL;
+        }
+
+        nvgpu_mutex_acquire(&page_table->lock);
+
+        pages = page_table->pages;
+        nr_pages = page_table->entries;
+        if (nr_pages <= 0) {
+                nvgpu_mutex_release(&page_table->lock);
+                return -EINVAL;
+        }
+
+        *dma_mapping = kzalloc(sizeof(**dma_mapping), GFP_KERNEL);
+        if (!*dma_mapping) {
+                nvgpu_mutex_release(&page_table->lock);
+                return -ENOMEM;
+        }
+        sgt = kzalloc(sizeof(*sgt), GFP_KERNEL);
+        if (!sgt) {
+                ret = -ENOMEM;
+                goto free_dma_mapping;
+        }
+        ret = sg_alloc_table_from_pages(sgt, pages,
+                                nr_pages, 0, page_table->size, GFP_KERNEL);
+        if (ret) {
+                goto free_sgt;
+        }
+        (*dma_mapping)->sgt = sgt;
+        (*dma_mapping)->dev = dev;
+        (*dma_mapping)->direction = direction;
+        (*dma_mapping)->page_table = page_table;
+
+        count = dma_map_sg(dev, sgt->sgl, sgt->nents, direction);
+        if (count < 1) {
+                goto free_sg_table;
+        }
+
+        (*dma_mapping)->entries = count;
+
+        (*dma_mapping)->hw_address = kcalloc(count, sizeof(u64), GFP_KERNEL);
+        if (!((*dma_mapping)->hw_address)) {
+                ret = -ENOMEM;
+                goto unmap_sg;
+        }
+        (*dma_mapping)->hw_len = kcalloc(count, sizeof(u64), GFP_KERNEL);
+        if (!((*dma_mapping)->hw_len)) {
+                ret = -ENOMEM;
+                goto free_hw_address;
+        }
+
+        for_each_sg(sgt->sgl, sg, count, i) {
+                (*dma_mapping)->hw_address[i] = sg_dma_address(sg);
+                (*dma_mapping)->hw_len[i] = sg_dma_len(sg);
+        }
+        (*dma_mapping)->page_table->mapped |= NVIDIA_P2P_MAPPED;
+        nvgpu_mutex_release(&page_table->lock);
+
+        return 0;
+free_hw_address:
+        kfree((*dma_mapping)->hw_address);
+unmap_sg:
+        dma_unmap_sg(dev, sgt->sgl,
+                sgt->nents, direction);
+free_sg_table:
+        sg_free_table(sgt);
+free_sgt:
+        kfree(sgt);
+free_dma_mapping:
+        kfree(*dma_mapping);
+        *dma_mapping = NULL;
+        nvgpu_mutex_release(&page_table->lock);
+
+        return ret;
+}
+EXPORT_SYMBOL(nvidia_p2p_map_pages);
+
+int nvidia_p2p_unmap_pages(struct nvidia_p2p_dma_mapping *dma_mapping)
+{
+        struct nvidia_p2p_page_table *page_table = NULL;
+
+        if (!dma_mapping) {
+                return -EINVAL;
+        }
+
+        page_table = dma_mapping->page_table;
+        if (!page_table) {
+                return -EFAULT;
+        }
+
+        nvgpu_mutex_acquire(&page_table->lock);
+        if (page_table->mapped & NVIDIA_P2P_MAPPED) {
+                kfree(dma_mapping->hw_len);
+                kfree(dma_mapping->hw_address);
+                if (dma_mapping->entries)
+                        dma_unmap_sg(dma_mapping->dev,
+                                dma_mapping->sgt->sgl,
+                                dma_mapping->sgt->nents,
+                                dma_mapping->direction);
+                sg_free_table(dma_mapping->sgt);
+                kfree(dma_mapping->sgt);
+                kfree(dma_mapping);
+                page_table->mapped &= (u32)~NVIDIA_P2P_MAPPED;
+        }
+        nvgpu_mutex_release(&page_table->lock);
+
+        return 0;
+}
+EXPORT_SYMBOL(nvidia_p2p_unmap_pages);
+
+int nvidia_p2p_free_dma_mapping(struct nvidia_p2p_dma_mapping *dma_mapping)
+{
+        return nvidia_p2p_unmap_pages(dma_mapping);
+}
+EXPORT_SYMBOL(nvidia_p2p_free_dma_mapping);