gpu: nvgpu: Add gv11b_gr_clear_sm_error_state

All chips were currently using gm20b_gr_clear_sm_error_state
It was wrong for chips based on volta and later as the implementation didn't
consider non pes-aware vsms mapping

Add new HAL implementation for clear_sm_error_state for volta based and later
chips to fix this.

Bug 200448172

Change-Id: I65988c8cbb35d13089ac628e8333d9a3b58e0eb1
Signed-off-by: Anup Mahindre <amahindre@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1837188
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

3 files changed, 56 insertions, 1 deletions

diff --git a/drivers/gpu/nvgpu/gv11b/gr_gv11b.c b/drivers/gpu/nvgpu/gv11b/gr_gv11b.c
index 288bd583..bb76178e 100644
--- a/drivers/gpu/nvgpu/gv11b/gr_gv11b.c
+++ b/drivers/gpu/nvgpu/gv11b/gr_gv11b.c
@@ -5010,3 +5010,56 @@ int gr_gv11b_create_priv_addr_table(struct gk20a *g,
         *num_registers = t;
         return 0;
 }
+
+int gv11b_gr_clear_sm_error_state(struct gk20a *g,
+                struct channel_gk20a *ch, u32 sm_id)
+{
+        u32 gpc, tpc, sm, offset;
+        u32 val;
+        struct tsg_gk20a *tsg;
+
+        int err = 0;
+
+        tsg = tsg_gk20a_from_ch(ch);
+        if (tsg == NULL) {
+                return -EINVAL;
+        }
+
+        nvgpu_mutex_acquire(&g->dbg_sessions_lock);
+
+        (void)memset(&tsg->sm_error_states[sm_id], 0, sizeof(*tsg->sm_error_states));
+
+        err = gr_gk20a_disable_ctxsw(g);
+        if (err != 0) {
+                nvgpu_err(g, "unable to stop gr ctxsw");
+                goto fail;
+        }
+
+        if (gk20a_is_channel_ctx_resident(ch)) {
+                gpc = g->gr.sm_to_cluster[sm_id].gpc_index;
+                if (g->ops.gr.get_nonpes_aware_tpc != NULL) {
+                        tpc = g->ops.gr.get_nonpes_aware_tpc(g,
+                                        g->gr.sm_to_cluster[sm_id].gpc_index,
+                                        g->gr.sm_to_cluster[sm_id].tpc_index);
+                } else {
+                        tpc = g->gr.sm_to_cluster[sm_id].tpc_index;
+                }
+                sm = g->gr.sm_to_cluster[sm_id].sm_index;
+
+                offset = gk20a_gr_gpc_offset(g, gpc) +
+                                gk20a_gr_tpc_offset(g, tpc) +
+                                gv11b_gr_sm_offset(g, sm);
+
+                val = gk20a_readl(g, gr_gpc0_tpc0_sm0_hww_global_esr_r() + offset);
+                gk20a_writel(g, gr_gpc0_tpc0_sm0_hww_global_esr_r() + offset,
+                                val);
+                gk20a_writel(g, gr_gpc0_tpc0_sm0_hww_warp_esr_r() + offset,
+                                0);
+        }
+
+        err = gr_gk20a_enable_ctxsw(g);
+
+fail:
+        nvgpu_mutex_release(&g->dbg_sessions_lock);
+        return err;
+}
diff --git a/drivers/gpu/nvgpu/gv11b/gr_gv11b.h b/drivers/gpu/nvgpu/gv11b/gr_gv11b.h
index 20377acf..2f765336 100644
--- a/drivers/gpu/nvgpu/gv11b/gr_gv11b.h
+++ b/drivers/gpu/nvgpu/gv11b/gr_gv11b.h
@@ -170,6 +170,8 @@ int gv11b_gr_set_sm_debug_mode(struct gk20a *g,
         struct channel_gk20a *ch, u64 sms, bool enable);
 int gv11b_gr_record_sm_error_state(struct gk20a *g, u32 gpc, u32 tpc, u32 sm,
                 struct channel_gk20a *fault_ch);
+int gv11b_gr_clear_sm_error_state(struct gk20a *g,
+                struct channel_gk20a *ch, u32 sm_id);
 void gv11b_gr_set_hww_esr_report_mask(struct gk20a *g);
 bool gv11b_gr_sm_debugger_attached(struct gk20a *g);
 void gv11b_gr_suspend_single_sm(struct gk20a *g,
diff --git a/drivers/gpu/nvgpu/gv11b/hal_gv11b.c b/drivers/gpu/nvgpu/gv11b/hal_gv11b.c
index 6a2dae77..2548cd16 100644
--- a/drivers/gpu/nvgpu/gv11b/hal_gv11b.c
+++ b/drivers/gpu/nvgpu/gv11b/hal_gv11b.c
@@ -335,7 +335,7 @@ static const struct gpu_ops gv11b_ops = {
                 .update_hwpm_ctxsw_mode = gr_gk20a_update_hwpm_ctxsw_mode,
                 .init_hwpm_pmm_register = gr_gv100_init_hwpm_pmm_register,
                 .record_sm_error_state = gv11b_gr_record_sm_error_state,
-                .clear_sm_error_state = gm20b_gr_clear_sm_error_state,
+                .clear_sm_error_state = gv11b_gr_clear_sm_error_state,
                 .suspend_contexts = gr_gp10b_suspend_contexts,
                 .resume_contexts = gr_gk20a_resume_contexts,
                 .get_preemption_mode_flags = gr_gp10b_get_preemption_mode_flags,