Update includes to L4T r32.7.4 and drop nvgpu/gk20a.h dependency

Also add instructions for updating `include/`. These files are now
only needed to build on Linux 4.9-based Tegra platforms.

1 files changed, 100 insertions, 8 deletions

diff --git a/include/gk20a/gr_gk20a.c b/include/gk20a/gr_gk20a.c
index 7bcf528..1eda853 100644
--- a/include/gk20a/gr_gk20a.c
+++ b/include/gk20a/gr_gk20a.c
@@ -1,7 +1,7 @@
 /*
  * GK20A Graphics
  *
- * Copyright (c) 2011-2020, NVIDIA CORPORATION.  All rights reserved.
+ * Copyright (c) 2011-2023, NVIDIA CORPORATION.  All rights reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -121,6 +121,83 @@ int gr_gk20a_get_ctx_id(struct gk20a *g,
         return 0;
 }
 
+void gk20a_gpccs_dump_falcon_stats(struct gk20a *g)
+{
+        unsigned int i;
+
+        nvgpu_err(g, "gr_gpc0_gpccs_falcon_irqstat : %d",
+                gk20a_readl(g, gr_gpc0_gpccs_falcon_irqstat_r()));
+        nvgpu_err(g, "gr_gpc0_gpccs_falcon_irqmode : %d",
+                gk20a_readl(g, gr_gpc0_gpccs_falcon_irqmode_r()));
+        nvgpu_err(g, "gr_gpc0_gpccs_falcon_irqmask : %d",
+                gk20a_readl(g, gr_gpc0_gpccs_falcon_irqmask_r()));
+        nvgpu_err(g, "gr_gpc0_gpccs_falcon_irqdest : %d",
+                gk20a_readl(g, gr_gpc0_gpccs_falcon_irqdest_r()));
+        nvgpu_err(g, "gr_gpc0_gpccs_falcon_debug1 : %d",
+                gk20a_readl(g, gr_gpc0_gpccs_falcon_debug1_r()));
+        nvgpu_err(g, "gr_gpc0_gpccs_falcon_debuginfo : %d",
+                gk20a_readl(g, gr_gpc0_gpccs_falcon_debuginfo_r()));
+        nvgpu_err(g, "gr_gpc0_gpccs_falcon_engctl : %d",
+                gk20a_readl(g, gr_gpc0_gpccs_falcon_engctl_r()));
+        nvgpu_err(g, "gr_gpc0_gpccs_falcon_curctx : %d",
+                gk20a_readl(g, gr_gpc0_gpccs_falcon_curctx_r()));
+        nvgpu_err(g, "gr_gpc0_gpccs_falcon_nxtctx : %d",
+                gk20a_readl(g, gr_gpc0_gpccs_falcon_nxtctx_r()));
+        nvgpu_err(g, "gr_gpc0_gpccs_ctxsw_status_1 : %d",
+                gk20a_readl(g, gr_gpc0_gpccs_ctxsw_status_1_r()));
+
+        for (i = 0; i < g->ops.gr.gpc0_gpccs_ctxsw_mailbox_size(); i++) {
+                nvgpu_err(g, "gr_gpc0_gpccs_ctxsw_mailbox_r(%d) : 0x%x",
+                        i, gk20a_readl(g, gr_gpc0_gpccs_ctxsw_mailbox_r(i)));
+        }
+
+
+        gk20a_writel(g, gr_gpc0_gpccs_falcon_icd_cmd_r(),
+                gr_gpc0_gpccs_falcon_icd_cmd_opc_rreg_f() |
+                gr_gpc0_gpccs_falcon_icd_cmd_idx_f(PMU_FALCON_REG_IMB));
+        nvgpu_err(g, "GPC0_GPCCS_FALCON_REG_IMB : 0x%x",
+                gk20a_readl(g, gr_gpc_gpccs_falcon_icd_rdata_r()));
+
+        gk20a_writel(g, gr_gpc0_gpccs_falcon_icd_cmd_r(),
+                gr_gpc0_gpccs_falcon_icd_cmd_opc_rreg_f() |
+                gr_gpc0_gpccs_falcon_icd_cmd_idx_f(PMU_FALCON_REG_DMB));
+        nvgpu_err(g, "GPC0_GPCCS_FALCON_REG_DMB : 0x%x",
+                gk20a_readl(g, gr_gpc_gpccs_falcon_icd_rdata_r()));
+
+        gk20a_writel(g, gr_gpc0_gpccs_falcon_icd_cmd_r(),
+                gr_gpc0_gpccs_falcon_icd_cmd_opc_rreg_f() |
+                gr_gpc0_gpccs_falcon_icd_cmd_idx_f(PMU_FALCON_REG_CSW));
+        nvgpu_err(g, "GPC0_GPCCS_FALCON_REG_CSW : 0x%x",
+                gk20a_readl(g, gr_gpc_gpccs_falcon_icd_rdata_r()));
+
+        gk20a_writel(g, gr_gpc0_gpccs_falcon_icd_cmd_r(),
+                gr_gpc0_gpccs_falcon_icd_cmd_opc_rreg_f() |
+                gr_gpc0_gpccs_falcon_icd_cmd_idx_f(PMU_FALCON_REG_CTX));
+        nvgpu_err(g, "GPC0_GPCCS_FALCON_REG_CTX : 0x%x",
+                gk20a_readl(g, gr_gpc_gpccs_falcon_icd_rdata_r()));
+
+        gk20a_writel(g, gr_gpc0_gpccs_falcon_icd_cmd_r(),
+                gr_gpc0_gpccs_falcon_icd_cmd_opc_rreg_f() |
+                gr_gpc0_gpccs_falcon_icd_cmd_idx_f(PMU_FALCON_REG_EXCI));
+        nvgpu_err(g, "GPC0_GPCCS_FALCON_REG_EXCI : 0x%x",
+                gk20a_readl(g, gr_gpc_gpccs_falcon_icd_rdata_r()));
+
+
+        for (i = 0; i < 4U; i++) {
+                gk20a_writel(g, gr_gpc0_gpccs_falcon_icd_cmd_r(),
+                        gr_gpc0_gpccs_falcon_icd_cmd_opc_rreg_f() |
+                        gr_gpc0_gpccs_falcon_icd_cmd_idx_f(PMU_FALCON_REG_PC));
+                nvgpu_err(g, "GPC0_GPCCS_FALCON_REG_PC : 0x%x",
+                        gk20a_readl(g, gr_gpc_gpccs_falcon_icd_rdata_r()));
+
+                gk20a_writel(g, gr_gpc0_gpccs_falcon_icd_cmd_r(),
+                        gr_gpc0_gpccs_falcon_icd_cmd_opc_rreg_f() |
+                        gr_gpc0_gpccs_falcon_icd_cmd_idx_f(PMU_FALCON_REG_SP));
+                nvgpu_err(g, "GPC0_GPCCS_FALCON_REG_SP : 0x%x",
+                        gk20a_readl(g, gr_gpc_gpccs_falcon_icd_rdata_r()));
+        }
+}
+
 void gk20a_fecs_dump_falcon_stats(struct gk20a *g)
 {
         unsigned int i;
@@ -527,6 +604,7 @@ int gr_gk20a_ctx_wait_ucode(struct gk20a *g, u32 mailbox_id,
                            "timeout waiting on mailbox=%d value=0x%08x",
                            mailbox_id, reg);
                 gk20a_fecs_dump_falcon_stats(g);
+                gk20a_gpccs_dump_falcon_stats(g);
                 gk20a_gr_debug_dump(g);
                 return -1;
         } else if (check == WAIT_UCODE_ERROR) {
@@ -534,6 +612,7 @@ int gr_gk20a_ctx_wait_ucode(struct gk20a *g, u32 mailbox_id,
                            "ucode method failed on mailbox=%d value=0x%08x",
                            mailbox_id, reg);
                 gk20a_fecs_dump_falcon_stats(g);
+                gk20a_gpccs_dump_falcon_stats(g);
                 return -1;
         }
 
@@ -1583,6 +1662,10 @@ static int gr_gk20a_init_golden_ctx_image(struct gk20a *g,
                              sw_ctx_load->l[i].value);
         }
 
+        if (g->ops.gr.disable_rd_coalesce) {
+                g->ops.gr.disable_rd_coalesce(g);
+        }
+
         if (g->ops.gr.init_preemption_state) {
                 g->ops.gr.init_preemption_state(g);
         }
@@ -4625,16 +4708,16 @@ static int gk20a_init_gr_setup_hw(struct gk20a *g)
                 g->ops.fb.init_cbc(g, gr);
         }
 
-        if (g->ops.gr.disable_rd_coalesce) {
-                g->ops.gr.disable_rd_coalesce(g);
-        }
-
         /* load ctx init */
         for (i = 0; i < sw_ctx_load->count; i++) {
                 gk20a_writel(g, sw_ctx_load->l[i].addr,
                              sw_ctx_load->l[i].value);
         }
 
+        if (g->ops.gr.disable_rd_coalesce) {
+                g->ops.gr.disable_rd_coalesce(g);
+        }
+
         err = gr_gk20a_wait_idle(g, gk20a_get_gr_idle_timeout(g),
                                  GR_IDLE_CHECK_DEFAULT);
         if (err != 0U) {
@@ -5297,9 +5380,10 @@ int gk20a_gr_handle_fecs_error(struct gk20a *g, struct channel_gk20a *ch,
         } else if ((gr_fecs_intr &
                         gr_fecs_host_int_status_watchdog_active_f()) != 0U) {
                 /* currently, recovery is not initiated */
-                nvgpu_err(g, "fecs watchdog triggered for channel %u, "
-                                "cannot ctxsw anymore !!", chid);
+                nvgpu_err(g, "fecs watchdog triggered for channel %u", chid);
                 gk20a_fecs_dump_falcon_stats(g);
+                gk20a_gpccs_dump_falcon_stats(g);
+                gk20a_gr_debug_dump(g);
         } else if ((gr_fecs_intr &
                 gr_fecs_host_int_status_ctxsw_intr_f(CTXSW_INTR0)) != 0U) {
                 u32 mailbox_value = gk20a_readl(g, gr_fecs_ctxsw_mailbox_r(6));
@@ -5321,8 +5405,9 @@ int gk20a_gr_handle_fecs_error(struct gk20a *g, struct channel_gk20a *ch,
         } else {
                 nvgpu_err(g,
                         "unhandled fecs error interrupt 0x%08x for channel %u",
-                        gr_fecs_intr, ch->chid);
+                        gr_fecs_intr, chid);
                 gk20a_fecs_dump_falcon_stats(g);
+                gk20a_gpccs_dump_falcon_stats(g);
         }
 
         gk20a_writel(g, gr_fecs_host_int_clear_r(), gr_fecs_intr);
@@ -8851,6 +8936,10 @@ int gr_gk20a_wait_for_pause(struct gk20a *g, struct nvgpu_warpstate *w_state)
         u32 gpc, tpc, sm, sm_id;
         u32 global_mask;
 
+        if (!g->ops.gr.get_sm_no_lock_down_hww_global_esr_mask ||
+            !g->ops.gr.lock_down_sm || !g->ops.gr.bpt_reg_info)
+                return -EINVAL;
+
         /* Wait for the SMs to reach full stop. This condition is:
          * 1) All SMs with valid warps must be in the trap handler (SM_IN_TRAP_MODE)
          * 2) All SMs in the trap handler must have equivalent VALID and PAUSED warp
@@ -8908,6 +8997,9 @@ int gr_gk20a_clear_sm_errors(struct gk20a *g)
         u32 global_esr;
         u32 sm_per_tpc = nvgpu_get_litter_value(g, GPU_LIT_NUM_SM_PER_TPC);