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 <alexw@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1566628
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

4 files changed, 5 insertions, 358 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
index 57c1c0bc..700dcdf8 100644
--- a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
@@ -3044,36 +3044,6 @@ out:
         return err;
 }
 
-int gk20a_comptag_allocator_init(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 = vzalloc(BITS_TO_LONGS(size) * sizeof(long));
-        if (!allocator->bitmap)
-                return -ENOMEM;
-        allocator->size = size;
-        return 0;
-}
-
-void gk20a_comptag_allocator_destroy(struct gk20a_comptag_allocator *allocator)
-{
-        struct gr_gk20a *gr = container_of(allocator,
-                                           struct gr_gk20a, comp_tags);
-
-        /*
-         * 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(gr->g, allocator->bitmap);
-}
-
 static void gk20a_remove_gr_support(struct gr_gk20a *gr)
 {
         struct gk20a *g = gr->g;
@@ -3148,7 +3118,7 @@ static void gk20a_remove_gr_support(struct gr_gk20a *gr)
                 nvgpu_big_free(g, gr->ctx_vars.hwpm_ctxsw_buffer_offset_map);
         gr->ctx_vars.hwpm_ctxsw_buffer_offset_map = NULL;
 
-        gk20a_comptag_allocator_destroy(&gr->comp_tags);
+        gk20a_comptag_allocator_destroy(g, &gr->comp_tags);
 }
 
 static int gr_gk20a_init_gr_config(struct gk20a *g, struct gr_gk20a *gr)
diff --git a/drivers/gpu/nvgpu/gk20a/gr_gk20a.h b/drivers/gpu/nvgpu/gk20a/gr_gk20a.h
index 5fab43ca..84eb8970 100644
--- a/drivers/gpu/nvgpu/gk20a/gr_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/gr_gk20a.h
@@ -36,6 +36,8 @@
 #include "gr_ctx_gk20a.h"
 #include "mm_gk20a.h"
 
+#include <nvgpu/comptags.h>
+
 #define GR_IDLE_CHECK_DEFAULT           10 /* usec */
 #define GR_IDLE_CHECK_MAX               200 /* usec */
 #define GR_FECS_POLL_INTERVAL           5 /* usec */
@@ -342,13 +344,7 @@ struct gr_gk20a {
 
         u32 max_comptag_mem; /* max memory size (MB) for comptag */
         struct compbit_store_desc compbit_store;
-        struct gk20a_comptag_allocator {
-                struct nvgpu_mutex lock;
-                /* this bitmap starts at ctag 1. 0th cannot be taken */
-                unsigned long *bitmap;
-                /* size of bitmap, not max ctags, so one less */
-                unsigned long size;
-        } comp_tags;
+        struct gk20a_comptag_allocator comp_tags;
 
         struct gr_zcull_gk20a zcull;
 
@@ -503,10 +499,6 @@ int gk20a_init_gr_support(struct gk20a *g);
 int gk20a_enable_gr_hw(struct gk20a *g);
 int gk20a_gr_reset(struct gk20a *g);
 void gk20a_gr_wait_initialized(struct gk20a *g);
-/* real size here, but first (ctag 0) isn't used */
-int gk20a_comptag_allocator_init(struct gk20a_comptag_allocator *allocator,
-                unsigned long size);
-void gk20a_comptag_allocator_destroy(struct gk20a_comptag_allocator *allocator);
 
 int gk20a_init_gr_channel(struct channel_gk20a *ch_gk20a);
 
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
index 8936cd03..69d9e983 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
@@ -22,7 +22,6 @@
  * DEALINGS IN THE SOFTWARE.
  */
 
-#include <linux/scatterlist.h>
 #include <linux/dma-buf.h>
 #include <linux/dma-mapping.h>
 #include <linux/dma-attrs.h>
@@ -70,6 +69,7 @@
  * all the common APIs no longers have Linux stuff in them.
  */
 #include "common/linux/vm_priv.h"
+#include "common/linux/dmabuf.h"
 
 /*
  * GPU mapping life cycle
@@ -108,190 +108,6 @@ static int __must_check gk20a_init_hwpm(struct mm_gk20a *mm);
 static int __must_check gk20a_init_cde_vm(struct mm_gk20a *mm);
 static int __must_check gk20a_init_ce_vm(struct mm_gk20a *mm);
 
-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;
-};
-
-static 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;
-}
-
-static 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);
-}
-
-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);
-}
-
-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);
-}
-
-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;
-}
-
 static int gk20a_init_mm_reset_enable_hw(struct gk20a *g)
 {
         gk20a_dbg_fn("");
@@ -1037,87 +853,6 @@ int gk20a_vm_bind_channel(struct gk20a_as_share *as_share,
         return __gk20a_vm_bind_channel(as_share->vm, ch);
 }
 
-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 nvgpu_vm_map_buffer(struct vm_gk20a *vm,
                         int dmabuf_fd,
                         u64 *offset_align,
@@ -1613,34 +1348,3 @@ const struct gk20a_mmu_level *gk20a_mm_get_mmu_levels(struct gk20a *g,
         return (big_page_size == SZ_64K) ?
                  gk20a_mm_levels_64k : gk20a_mm_levels_128k;
 }
-
-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/gk20a/mm_gk20a.h b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
index 448496f5..04034d84 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
@@ -117,13 +117,6 @@ gk20a_buffer_state_from_list(struct nvgpu_list_node *node)
                 ((uintptr_t)node - offsetof(struct gk20a_buffer_state, list));
 };
 
-struct gk20a_comptags {
-        u32 offset;
-        u32 lines;
-        u32 allocated_lines;
-        bool user_mappable;
-};
-
 struct priv_cmd_queue {
         struct nvgpu_mem mem;
         u32 size;       /* num of entries in words */
@@ -369,10 +362,6 @@ void gk20a_locked_gmmu_unmap(struct vm_gk20a *vm,
                              bool sparse,
                              struct vm_gk20a_mapping_batch *batch);
 
-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);
-
 /* vm-as interface */
 struct nvgpu_as_alloc_space_args;
 struct nvgpu_as_free_space_args;
@@ -381,14 +370,6 @@ int gk20a_vm_bind_channel(struct gk20a_as_share *as_share,
                           struct channel_gk20a *ch);
 int __gk20a_vm_bind_channel(struct vm_gk20a *vm, struct channel_gk20a *ch);
 
-void gk20a_get_comptags(struct device *dev, struct dma_buf *dmabuf,
-                        struct gk20a_comptags *comptags);
-
-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);
-
 void pde_range_from_vaddr_range(struct vm_gk20a *vm,
                                               u64 addr_lo, u64 addr_hi,
                                               u32 *pde_lo, u32 *pde_hi);