gpu: nvgpu: Add NVGPU_IOCTL_CHANNEL_SET_BOOSTED_CTX

This ioctl can be used on gp10b to set a flag in the context header
indicating this context should be run at elevated clock
frequency. FECS ctxsw ucode will read this flag as part of the context
switch and will request higher GPU clock frequencies from BPMP for the
duration of the context execution.

Bug 1819874

Change-Id: I84bf580923d95585095716d49cea24e58c9440ed
Signed-off-by: Peter Boonstoppel <pboonstoppel@nvidia.com>
Reviewed-on: http://git-master/r/1292746
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/gk20a/channel_gk20a.c

@@ -3874,6 +3874,24 @@ long gk20a_channel_ioctl(struct file *filp,
 			err = -EINVAL;
 		}
 		break;
+	case NVGPU_IOCTL_CHANNEL_SET_BOOSTED_CTX:
+		if (ch->g->ops.gr.set_boosted_ctx) {
+			bool boost =
+				((struct nvgpu_boosted_ctx_args *)buf)->boost;
+
+			err = gk20a_busy(dev);
+			if (err) {
+				dev_err(dev,
+					"%s: failed to host gk20a for ioctl cmd: 0x%x",
+					__func__, cmd);
+				break;
+			}
+			err = ch->g->ops.gr.set_boosted_ctx(ch, boost);
+			gk20a_idle(dev);
+		} else {
+			err = -EINVAL;
+		}
+		break;
 	default:
 		dev_dbg(dev, "unrecognized ioctl cmd: 0x%x", cmd);
 		err = -ENOTTY;

drivers/gpu/nvgpu/gk20a/gk20a.h

@@ -319,6 +319,10 @@ struct gpu_ops {
 				struct vm_gk20a *vm, u32 class,
 				u32 graphics_preempt_mode,
 				u32 compute_preempt_mode);
+		int (*set_boosted_ctx)(struct channel_gk20a *ch, bool boost);
+		void (*update_boosted_ctx)(struct gk20a *g,
+					   struct mem_desc *mem,
+					   struct gr_ctx_desc *gr_ctx);
 		int (*fuse_override)(struct gk20a *g);
 		void (*init_sm_id_table)(struct gk20a *g);
 		int (*load_smid_config)(struct gk20a *g);

drivers/gpu/nvgpu/gk20a/gr_gk20a.c

@@ -2109,6 +2109,9 @@ int gr_gk20a_load_golden_ctx_image(struct gk20a *g,
 	if (g->ops.gr.update_ctxsw_preemption_mode)
 		g->ops.gr.update_ctxsw_preemption_mode(g, ch_ctx, mem);
 
+	if (g->ops.gr.update_boosted_ctx)
+		g->ops.gr.update_boosted_ctx(g, mem, ch_ctx->gr_ctx);
+
 	virt_addr_lo = u64_lo32(ch_ctx->patch_ctx.mem.gpu_va);
 	virt_addr_hi = u64_hi32(ch_ctx->patch_ctx.mem.gpu_va);
 

drivers/gpu/nvgpu/gk20a/gr_gk20a.h

@@ -391,6 +391,7 @@ struct gr_ctx_desc {
 
 	u32 graphics_preempt_mode;
 	u32 compute_preempt_mode;
+	bool boosted_ctx;
 #ifdef CONFIG_ARCH_TEGRA_18x_SOC
 	struct gr_ctx_desc_t18x t18x;
 #endif

drivers/gpu/nvgpu/gp106/gr_gp106.c

@@ -243,5 +243,6 @@ void gp106_init_gr(struct gpu_ops *gops)
 	gops->gr.init_preemption_state = NULL;
 	gops->gr.set_ctxsw_preemption_mode = gr_gp106_set_ctxsw_preemption_mode;
 	gops->gr.create_gr_sysfs = NULL;
-
+	gops->gr.set_boosted_ctx = NULL;
+	gops->gr.update_boosted_ctx = NULL;
 }

drivers/gpu/nvgpu/gp10b/gr_gp10b.c

@@ -2060,6 +2060,49 @@ clean_up:
 	return err;
 }
 
+static int gr_gp10b_set_boosted_ctx(struct channel_gk20a *ch,
+				    bool boost)
+{
+	struct gr_ctx_desc *gr_ctx = ch->ch_ctx.gr_ctx;
+	struct gk20a *g = ch->g;
+	struct mem_desc *mem = &gr_ctx->mem;
+	int err = 0;
+
+	gr_ctx->boosted_ctx = boost;
+
+	if (gk20a_mem_begin(g, mem))
+		return -ENOMEM;
+
+	err = gk20a_disable_channel_tsg(g, ch);
+	if (err)
+		goto unmap_ctx;
+
+	err = gk20a_fifo_preempt(g, ch);
+	if (err)
+		goto enable_ch;
+
+	if (g->ops.gr.update_boosted_ctx)
+		g->ops.gr.update_boosted_ctx(g, mem, gr_ctx);
+	else
+		err = -ENOSYS;
+
+enable_ch:
+	gk20a_enable_channel_tsg(g, ch);
+unmap_ctx:
+	gk20a_mem_end(g, mem);
+
+	return err;
+}
+
+static void gr_gp10b_update_boosted_ctx(struct gk20a *g, struct mem_desc *mem,
+				       struct gr_ctx_desc *gr_ctx) {
+	u32 v;
+
+	v = ctxsw_prog_main_image_pmu_options_boost_clock_frequencies_f(
+		gr_ctx->boosted_ctx);
+	gk20a_mem_wr(g, mem, ctxsw_prog_main_image_pmu_options_o(), v);
+}
+
 static int gr_gp10b_set_preemption_mode(struct channel_gk20a *ch,
 					u32 graphics_preempt_mode,
 					u32 compute_preempt_mode)
@@ -2271,4 +2314,6 @@ void gp10b_init_gr(struct gpu_ops *gops)
 	gops->gr.get_preemption_mode_flags = gr_gp10b_get_preemption_mode_flags;
 	gops->gr.fuse_override = gp10b_gr_fuse_override;
 	gops->gr.load_smid_config = gr_gp10b_load_smid_config;
+	gops->gr.set_boosted_ctx = gr_gp10b_set_boosted_ctx;
+	gops->gr.update_boosted_ctx = gr_gp10b_update_boosted_ctx;
 }

drivers/gpu/nvgpu/vgpu/gr_vgpu.c

@@ -1134,4 +1134,6 @@ void vgpu_init_gr_ops(struct gpu_ops *gops)
 	gops->gr.update_hwpm_ctxsw_mode = vgpu_gr_update_hwpm_ctxsw_mode;
 	gops->gr.clear_sm_error_state = vgpu_gr_clear_sm_error_state;
 	gops->gr.dump_gr_regs = NULL;
+	gops->gr.set_boosted_ctx = NULL;
+	gops->gr.update_boosted_ctx = NULL;
 }

include/uapi/linux/nvgpu.h

@@ -1537,6 +1537,13 @@ struct nvgpu_preemption_mode_args {
 	__u32 compute_preempt_mode; /* in */
 };
 
+struct nvgpu_boosted_ctx_args {
+#define NVGPU_BOOSTED_CTX_MODE_NORMAL			(0U)
+#define NVGPU_BOOSTED_CTX_MODE_BOOSTED_EXECUTION	(1U)
+	__u32 boost;
+	__u32 padding;
+};
+
 #define NVGPU_IOCTL_CHANNEL_SET_NVMAP_FD	\
 	_IOW(NVGPU_IOCTL_MAGIC, 5, struct nvgpu_set_nvmap_fd_args)
 #define NVGPU_IOCTL_CHANNEL_SET_TIMEOUT	\
@@ -1585,9 +1592,11 @@ struct nvgpu_preemption_mode_args {
 	_IOW(NVGPU_IOCTL_MAGIC, 122, struct nvgpu_preemption_mode_args)
 #define NVGPU_IOCTL_CHANNEL_ALLOC_GPFIFO_EX	\
 	_IOW(NVGPU_IOCTL_MAGIC, 123, struct nvgpu_alloc_gpfifo_ex_args)
+#define NVGPU_IOCTL_CHANNEL_SET_BOOSTED_CTX	\
+	_IOW(NVGPU_IOCTL_MAGIC, 124, struct nvgpu_boosted_ctx_args)
 
 #define NVGPU_IOCTL_CHANNEL_LAST	\
-	_IOC_NR(NVGPU_IOCTL_CHANNEL_ALLOC_GPFIFO_EX)
+	_IOC_NR(NVGPU_IOCTL_CHANNEL_SET_BOOSTED_CTX)
 #define NVGPU_IOCTL_CHANNEL_MAX_ARG_SIZE sizeof(struct nvgpu_alloc_gpfifo_ex_args)
 
 /*