gpu: nvgpu: Remove extraneous FB flush calls

gk20a_mm_fb_flush() invoked G_ELPG_FLUSH and FB_FLUSH. Remove the
invokation of G_ELPG_FLUSH. Replace calls to gk20a_mm_fb_flush() with
gk20a_mm_l2_flush() when appropriate.

Bug 1421824

Change-Id: I02af4bdc3b7bd26d0f6a8d610f70349269775a36
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/408210
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit

2 files changed, 4 insertions, 10 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
index 6a41509b..df85d852 100644
--- a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
@@ -655,8 +655,6 @@ static int gr_gk20a_commit_inst(struct channel_gk20a *c, u64 gpu_va)
 
         gk20a_dbg_fn("");
 
-        gk20a_mm_fb_flush(c->g);
-
         inst_ptr = c->inst_block.cpuva;
         if (!inst_ptr)
                 return -ENOMEM;
@@ -1573,7 +1571,7 @@ static int gr_gk20a_init_golden_ctx_image(struct gk20a *g,
         ctx_header_words =  roundup(ctx_header_bytes, sizeof(u32));
         ctx_header_words >>= 2;
 
-        gk20a_mm_fb_flush(g);
+        gk20a_mm_l2_flush(g, true);
 
         for (i = 0; i < ctx_header_words; i++) {
                 data = gk20a_mem_rd32(ctx_ptr, i);
@@ -1634,11 +1632,9 @@ int gr_gk20a_update_smpc_ctxsw_mode(struct gk20a *g,
         void *ctx_ptr = NULL;
         u32 data;
 
-        /*XXX caller responsible for making sure the channel is quiesced? */
-
         /* Channel gr_ctx buffer is gpu cacheable.
            Flush and invalidate before cpu update. */
-        gk20a_mm_fb_flush(g);
+        gk20a_mm_l2_flush(g, true);
 
         ctx_ptr = vmap(ch_ctx->gr_ctx.pages,
                         PAGE_ALIGN(ch_ctx->gr_ctx.size) >> PAGE_SHIFT,
@@ -1678,7 +1674,7 @@ static int gr_gk20a_load_golden_ctx_image(struct gk20a *g,
 
         /* Channel gr_ctx buffer is gpu cacheable.
            Flush and invalidate before cpu update. */
-        gk20a_mm_fb_flush(g);
+        gk20a_mm_l2_flush(g, true);
 
         ctx_ptr = vmap(ch_ctx->gr_ctx.pages,
                         PAGE_ALIGN(ch_ctx->gr_ctx.size) >> PAGE_SHIFT,
@@ -6686,7 +6682,7 @@ int gr_gk20a_exec_ctx_ops(struct channel_gk20a *ch,
                 goto cleanup;
         }
 
-        gk20a_mm_fb_flush(g);
+        gk20a_mm_l2_flush(g, true);
 
         /* write to appropriate place in context image,
          * first have to figure out where that really is */
diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
index dd492d92..3b0b020d 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.c
@@ -2801,8 +2801,6 @@ int gk20a_mm_fb_flush(struct gk20a *g)
 
         mutex_lock(&mm->l2_op_lock);
 
-        g->ops.ltc.elpg_flush(g);
-
         /* Make sure all previous writes are committed to the L2. There's no
            guarantee that writes are to DRAM. This will be a sysmembar internal
            to the L2. */