gpu: nvgpu: SM/TEX exception handling support

Add TEX exception handling support. Also make SM exception handler into
a function pointer, which should allow different chips to implement
their own SM exception handling routine.

Bug 1635727
Bug 1637486

Change-Id: I429905726c1840c11e83780843d82729495dc6a5
Signed-off-by: Adeel Raza <araza@nvidia.com>
Reviewed-on: http://git-master/r/935329

6 files changed, 85 insertions, 3 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/gk20a.h b/drivers/gpu/nvgpu/gk20a/gk20a.h
index 0207588f..24c062d2 100644
--- a/drivers/gpu/nvgpu/gk20a/gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/gk20a.h
@@ -202,6 +202,10 @@ struct gpu_ops {
                                 struct channel_gk20a *fault_ch,
                                 bool *early_exit, bool *ignore_debugger);
                 u32 (*mask_hww_warp_esr)(u32 hww_warp_esr);
+                int (*handle_sm_exception)(struct gk20a *g, u32 gpc, u32 tpc,
+                        bool *post_event, struct channel_gk20a *fault_ch);
+                int (*handle_tex_exception)(struct gk20a *g, u32 gpc, u32 tpc,
+                                                bool *post_event);
         } gr;
         const char *name;
         struct {
diff --git a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
index 6e2ea548..542a6c02 100644
--- a/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
+++ b/drivers/gpu/nvgpu/gk20a/gr_gk20a.c
@@ -3996,6 +3996,7 @@ static void gk20a_gr_enable_gpc_exceptions(struct gk20a *g)
         u32 tpc_mask;
 
         gk20a_writel(g, gr_gpcs_tpcs_tpccs_tpc_exception_en_r(),
+                        gr_gpcs_tpcs_tpccs_tpc_exception_en_tex_enabled_f() |
                         gr_gpcs_tpcs_tpccs_tpc_exception_en_sm_enabled_f());
 
         tpc_mask =
@@ -5241,7 +5242,7 @@ u32 gk20a_mask_hww_warp_esr(u32 hww_warp_esr)
         return hww_warp_esr;
 }
 
-static int gk20a_gr_handle_sm_exception(struct gk20a *g, u32 gpc, u32 tpc,
+int gr_gk20a_handle_sm_exception(struct gk20a *g, u32 gpc, u32 tpc,
                 bool *post_event, struct channel_gk20a *fault_ch)
 {
         int ret = 0;
@@ -5322,6 +5323,27 @@ static int gk20a_gr_handle_sm_exception(struct gk20a *g, u32 gpc, u32 tpc,
         return ret;
 }
 
+int gr_gk20a_handle_tex_exception(struct gk20a *g, u32 gpc, u32 tpc,
+                bool *post_event)
+{
+        int ret = 0;
+        u32 offset = proj_gpc_stride_v() * gpc +
+                     proj_tpc_in_gpc_stride_v() * tpc;
+        u32 esr;
+
+        gk20a_dbg(gpu_dbg_fn | gpu_dbg_gpu_dbg, "");
+
+        esr = gk20a_readl(g,
+                         gr_gpc0_tpc0_tex_m_hww_esr_r() + offset);
+        gk20a_dbg(gpu_dbg_intr | gpu_dbg_gpu_dbg, "0x%08x", esr);
+
+        gk20a_writel(g,
+                     gr_gpc0_tpc0_tex_m_hww_esr_r() + offset,
+                     esr);
+
+        return ret;
+}
+
 static int gk20a_gr_handle_tpc_exception(struct gk20a *g, u32 gpc, u32 tpc,
                 bool *post_event, struct channel_gk20a *fault_ch)
 {
@@ -5338,8 +5360,16 @@ static int gk20a_gr_handle_tpc_exception(struct gk20a *g, u32 gpc, u32 tpc,
                         gr_gpc0_tpc0_tpccs_tpc_exception_sm_pending_v()) {
                 gk20a_dbg(gpu_dbg_intr | gpu_dbg_gpu_dbg,
                                 "GPC%d TPC%d: SM exception pending", gpc, tpc);
-                ret = gk20a_gr_handle_sm_exception(g, gpc, tpc,
-                                        post_event, fault_ch);
+                ret = g->ops.gr.handle_sm_exception(g, gpc, tpc,
+                                                        post_event, fault_ch);
+        }
+
+        /* check if a tex exeption is pending */
+        if (gr_gpc0_tpc0_tpccs_tpc_exception_tex_v(tpc_exception) ==
+                        gr_gpc0_tpc0_tpccs_tpc_exception_tex_pending_v()) {
+                gk20a_dbg(gpu_dbg_intr | gpu_dbg_gpu_dbg,
+                                "GPC%d TPC%d: TEX exception pending", gpc, tpc);
+                ret = g->ops.gr.handle_tex_exception(g, gpc, tpc, post_event);
         }
 
         return ret;
@@ -7595,4 +7625,6 @@ void gk20a_init_gr_ops(struct gpu_ops *gops)
         gops->gr.get_access_map = gr_gk20a_get_access_map;
         gops->gr.handle_fecs_error = gk20a_gr_handle_fecs_error;
         gops->gr.mask_hww_warp_esr = gk20a_mask_hww_warp_esr;
+        gops->gr.handle_sm_exception = gr_gk20a_handle_sm_exception;
+        gops->gr.handle_tex_exception = gr_gk20a_handle_tex_exception;
 }
diff --git a/drivers/gpu/nvgpu/gk20a/gr_gk20a.h b/drivers/gpu/nvgpu/gk20a/gr_gk20a.h
index ad197228..9c37fd02 100644
--- a/drivers/gpu/nvgpu/gk20a/gr_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/gr_gk20a.h
@@ -528,6 +528,10 @@ int gr_gk20a_add_zbc_depth(struct gk20a *g, struct gr_gk20a *gr,
                            struct zbc_entry *depth_val, u32 index);
 int gr_gk20a_wait_idle(struct gk20a *g, unsigned long end_jiffies,
                        u32 expect_delay);
+int gr_gk20a_handle_sm_exception(struct gk20a *g, u32 gpc, u32 tpc,
+                bool *post_event, struct channel_gk20a *fault_ch);
+int gr_gk20a_handle_tex_exception(struct gk20a *g, u32 gpc, u32 tpc,
+                                        bool *post_event);
 int gr_gk20a_init_ctx_state(struct gk20a *g);
 int gr_gk20a_submit_fecs_method_op(struct gk20a *g,
                                    struct fecs_method_op_gk20a op,
diff --git a/drivers/gpu/nvgpu/gk20a/hw_gr_gk20a.h b/drivers/gpu/nvgpu/gk20a/hw_gr_gk20a.h
index 6a155ccc..a73209f2 100644
--- a/drivers/gpu/nvgpu/gk20a/hw_gr_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/hw_gr_gk20a.h
@@ -2990,6 +2990,10 @@ static inline u32 gr_gpcs_tpcs_tpccs_tpc_exception_en_sm_enabled_f(void)
 {
         return 0x2;
 }
+static inline u32 gr_gpcs_tpcs_tpccs_tpc_exception_en_tex_enabled_f(void)
+{
+        return 0x1;
+}
 static inline u32 gr_gpc0_tpc0_tpccs_tpc_exception_en_r(void)
 {
         return 0x0050450c;
@@ -3026,6 +3030,14 @@ static inline u32 gr_gpc0_tpc0_tpccs_tpc_exception_r(void)
 {
         return 0x00504508;
 }
+static inline u32 gr_gpc0_tpc0_tpccs_tpc_exception_tex_v(u32 r)
+{
+        return (r >> 0) & 0x1;
+}
+static inline u32 gr_gpc0_tpc0_tpccs_tpc_exception_tex_pending_v(void)
+{
+        return 0x00000001;
+}
 static inline u32 gr_gpc0_tpc0_tpccs_tpc_exception_sm_v(u32 r)
 {
         return (r >> 1) & 0x1;
@@ -3170,6 +3182,14 @@ static inline u32 gr_gpc0_tpc0_sm_hww_global_esr_single_step_complete_pending_f(
 {
         return 0x40;
 }
+static inline u32 gr_gpc0_tpc0_tex_m_hww_esr_r(void)
+{
+        return 0x00504224;
+}
+static inline u32 gr_gpc0_tpc0_tex_m_hww_esr_intr_pending_f(void)
+{
+        return 0x1;
+}
 static inline u32 gr_gpc0_tpc0_sm_hww_warp_esr_r(void)
 {
         return 0x00504648;
diff --git a/drivers/gpu/nvgpu/gm20b/gr_gm20b.c b/drivers/gpu/nvgpu/gm20b/gr_gm20b.c
index a80cef8f..ab1b166d 100644
--- a/drivers/gpu/nvgpu/gm20b/gr_gm20b.c
+++ b/drivers/gpu/nvgpu/gm20b/gr_gm20b.c
@@ -1229,4 +1229,6 @@ void gm20b_init_gr(struct gpu_ops *gops)
         gops->gr.get_access_map = gr_gm20b_get_access_map;
         gops->gr.handle_fecs_error = gk20a_gr_handle_fecs_error;
         gops->gr.mask_hww_warp_esr = gk20a_mask_hww_warp_esr;
+        gops->gr.handle_sm_exception = gr_gk20a_handle_sm_exception;
+        gops->gr.handle_tex_exception = gr_gk20a_handle_tex_exception;
 }
diff --git a/drivers/gpu/nvgpu/gm20b/hw_gr_gm20b.h b/drivers/gpu/nvgpu/gm20b/hw_gr_gm20b.h
index 41a7c885..7a19e4ab 100644
--- a/drivers/gpu/nvgpu/gm20b/hw_gr_gm20b.h
+++ b/drivers/gpu/nvgpu/gm20b/hw_gr_gm20b.h
@@ -3022,6 +3022,10 @@ static inline u32 gr_gpcs_tpcs_tpccs_tpc_exception_en_sm_enabled_f(void)
 {
         return 0x2;
 }
+static inline u32 gr_gpcs_tpcs_tpccs_tpc_exception_en_tex_enabled_f(void)
+{
+        return 0x1;
+}
 static inline u32 gr_gpc0_tpc0_tpccs_tpc_exception_en_r(void)
 {
         return 0x0050450c;
@@ -3058,6 +3062,14 @@ static inline u32 gr_gpc0_tpc0_tpccs_tpc_exception_r(void)
 {
         return 0x00504508;
 }
+static inline u32 gr_gpc0_tpc0_tpccs_tpc_exception_tex_v(u32 r)
+{
+        return (r >> 0) & 0x1;
+}
+static inline u32 gr_gpc0_tpc0_tpccs_tpc_exception_tex_pending_v(void)
+{
+        return 0x00000001;
+}
 static inline u32 gr_gpc0_tpc0_tpccs_tpc_exception_sm_v(u32 r)
 {
         return (r >> 1) & 0x1;
@@ -3214,6 +3226,14 @@ static inline u32 gr_gpc0_tpc0_sm_hww_global_esr_single_step_complete_pending_f(
 {
         return 0x40;
 }
+static inline u32 gr_gpc0_tpc0_tex_m_hww_esr_r(void)
+{
+        return 0x00504224;
+}
+static inline u32 gr_gpc0_tpc0_tex_m_hww_esr_intr_pending_f(void)
+{
+        return 0x1;
+}
 static inline u32 gr_gpc0_tpc0_sm_hww_warp_esr_r(void)
 {
         return 0x00504648;