gpu: nvgpu: gk20a: Set lockboost size for compute

For compute channel on gk20a, set lockboost size to zero.

Bug 1573856

Change-Id: I369cebf72241e4017e7d380c82caff6014e42984
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: http://git-master/r/594843
GVS: Gerrit_Virtual_Submit

4 files changed, 39 insertions, 6 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/cde_gk20a.c b/drivers/gpu/nvgpu/gk20a/cde_gk20a.c
index 6f629e81..4acfa955 100644
--- a/drivers/gpu/nvgpu/gk20a/cde_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/cde_gk20a.c
@@ -515,7 +515,7 @@ static int gk20a_init_cde_required_class(struct gk20a_cde_ctx *cde_ctx,
         int err;
 
         alloc_obj_ctx.class_num = required_class;
-        alloc_obj_ctx.padding = 0;
+        alloc_obj_ctx.flags = 0;
 
         err = gk20a_alloc_obj_ctx(cde_ctx->ch, &alloc_obj_ctx);
         if (err) {
diff --git a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
index 37cccba3..8acc5b45 100644
--- a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
@@ -2771,7 +2771,7 @@ int gk20a_alloc_obj_ctx(struct channel_gk20a  *c,
                 if (!ch_ctx->gr_ctx) {
                         err = gr_gk20a_alloc_channel_gr_ctx(g, c,
                                                             args->class_num,
-                                                            args->padding);
+                                                            args->flags);
                         if (err) {
                                 gk20a_err(dev_from_gk20a(g),
                                         "fail to allocate gr ctx buffer");
@@ -2792,7 +2792,7 @@ int gk20a_alloc_obj_ctx(struct channel_gk20a  *c,
                         gk20a_vm_get(tsg->vm);
                         err = gr_gk20a_alloc_tsg_gr_ctx(g, tsg,
                                                         args->class_num,
-                                                        args->padding);
+                                                        args->flags);
                         if (err) {
                                 gk20a_err(dev_from_gk20a(g),
                                         "fail to allocate TSG gr ctx buffer");
@@ -2837,7 +2837,12 @@ int gk20a_alloc_obj_ctx(struct channel_gk20a  *c,
         /* tweak any perf parameters per-context here */
         if (args->class_num == KEPLER_COMPUTE_A) {
                 int begin_err;
-                u32 tex_lock_disable_mask =
+                u32 tex_lock_disable_mask;
+                u32 texlock;
+                u32 lockboost_mask;
+                u32 lockboost;
+
+                tex_lock_disable_mask =
                         gr_gpcs_tpcs_sm_sch_texlock_tex_hash_m()         |
                         gr_gpcs_tpcs_sm_sch_texlock_tex_hash_tile_m()    |
                         gr_gpcs_tpcs_sm_sch_texlock_tex_hash_phase_m()   |
@@ -2845,7 +2850,7 @@ int gk20a_alloc_obj_ctx(struct channel_gk20a  *c,
                         gr_gpcs_tpcs_sm_sch_texlock_tex_hash_timeout_m() |
                         gr_gpcs_tpcs_sm_sch_texlock_dot_t_unlock_m();
 
-                u32 texlock = gk20a_readl(g, gr_gpcs_tpcs_sm_sch_texlock_r());
+                texlock = gk20a_readl(g, gr_gpcs_tpcs_sm_sch_texlock_r());
 
                 texlock = (texlock & ~tex_lock_disable_mask) |
                 (gr_gpcs_tpcs_sm_sch_texlock_tex_hash_disable_f()         |
@@ -2855,12 +2860,24 @@ int gk20a_alloc_obj_ctx(struct channel_gk20a  *c,
                  gr_gpcs_tpcs_sm_sch_texlock_tex_hash_timeout_disable_f() |
                  gr_gpcs_tpcs_sm_sch_texlock_dot_t_unlock_disable_f());
 
+                lockboost_mask =
+                        gr_gpcs_tpcs_sm_sch_macro_sched_lockboost_size_m();
+
+                lockboost = gk20a_readl(g, gr_gpcs_tpcs_sm_sch_macro_sched_r());
+                lockboost = (lockboost & ~lockboost_mask) |
+                        gr_gpcs_tpcs_sm_sch_macro_sched_lockboost_size_f(0);
+
                 begin_err = gr_gk20a_ctx_patch_write_begin(g, ch_ctx);
 
                 if (!begin_err) {
                         err = gr_gk20a_ctx_patch_write(g, ch_ctx,
                                 gr_gpcs_tpcs_sm_sch_texlock_r(),
                                 texlock, true);
+
+                        if (!err)
+                                err = gr_gk20a_ctx_patch_write(g, ch_ctx,
+                                        gr_gpcs_tpcs_sm_sch_macro_sched_r(),
+                                        lockboost, true);
                 }
                 if ((begin_err || err)) {
                         gk20a_err(dev_from_gk20a(g),
@@ -2868,6 +2885,8 @@ int gk20a_alloc_obj_ctx(struct channel_gk20a  *c,
                 }
                 if (!begin_err)
                         gr_gk20a_ctx_patch_write_end(g, ch_ctx);
+
+                args->flags |= NVGPU_ALLOC_OBJ_FLAGS_LOCKBOOST_ZERO;
         }
 
         /* init golden image, ELPG enabled after this is done */
diff --git a/drivers/gpu/nvgpu/gk20a/hw_gr_gk20a.h b/drivers/gpu/nvgpu/gk20a/hw_gr_gk20a.h
index 4e15af5f..1de9e794 100644
--- a/drivers/gpu/nvgpu/gk20a/hw_gr_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/hw_gr_gk20a.h
@@ -3250,6 +3250,18 @@ static inline u32 gr_gpcs_tpcs_sm_sch_texlock_dot_t_unlock_disable_f(void)
 {
         return 0x0;
 }
+static inline u32 gr_gpcs_tpcs_sm_sch_macro_sched_r(void)
+{
+        return 0x00419eac;
+}
+static inline u32 gr_gpcs_tpcs_sm_sch_macro_sched_lockboost_size_f(u32 v)
+{
+        return (v & 0x1) << 2;
+}
+static inline u32 gr_gpcs_tpcs_sm_sch_macro_sched_lockboost_size_m(void)
+{
+        return 0x1 << 2;
+}
 static inline u32 gr_gpcs_tpcs_sm_dbgr_control0_r(void)
 {
         return 0x00419e10;
diff --git a/include/uapi/linux/nvgpu.h b/include/uapi/linux/nvgpu.h
index 97e791df..20acc66a 100644
--- a/include/uapi/linux/nvgpu.h
+++ b/include/uapi/linux/nvgpu.h
@@ -459,9 +459,11 @@ struct nvgpu_set_nvmap_fd_args {
         __u32 fd;
 } __packed;
 
+#define NVGPU_ALLOC_OBJ_FLAGS_LOCKBOOST_ZERO    (1 << 0)
+
 struct nvgpu_alloc_obj_ctx_args {
         __u32 class_num; /* kepler3d, 2d, compute, etc       */
-        __u32 padding;
+        __u32 flags;     /* input, output */
         __u64 obj_id;    /* output, used to free later       */
 };