gpu: gp10b: add gfxp_wfi_timeout sysfs node

Add a sysfs node to allow root user to set PRI_FE_GFXP_WFI_TIMEOUT, for gp10b
only, in units of sysclk cycles. Store the set value in a variable, and write
the set value to register after GPU is un-railgated.

NV_PGRAPH_PRI_FE_GFXP_WFI_TIMEOUT is engine_reset after Bug 1623341.

Change default value to be specified in cycles, rather than time.  This value
is almost the current value in cycles calculated each boot.

Bug 1932782

Change-Id: I0a4207e637cd1413a1be95abe2bcce3adccf76fa
Signed-off-by: Jonathan McCaffrey <jmccaffrey@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1540939
Reviewed-by: Automatic_Commit_Validation_User
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>

drivers/gpu/nvgpu/gk20a/gr_gk20a.c

@@ -75,6 +75,7 @@
 #define CTXSW_MEM_SCRUBBING_TIMEOUT_DEFAULT 10
 #define FECS_ARB_CMD_TIMEOUT_MAX 40
 #define FECS_ARB_CMD_TIMEOUT_DEFAULT 2
+#define GFXP_WFI_TIMEOUT_COUNT_DEFAULT 100000
 
 static int gk20a_init_gr_bind_fecs_elpg(struct gk20a *g);
 
@@ -5174,6 +5175,8 @@ static int gk20a_init_gr_setup_sw(struct gk20a *g)
 
 	gr_gk20a_load_zbc_default_table(g, gr);
 
+	gr->gfxp_wfi_timeout_count = GFXP_WFI_TIMEOUT_COUNT_DEFAULT;
+
 	nvgpu_mutex_init(&gr->ctx_mutex);
 	nvgpu_spinlock_init(&gr->ch_tlb_lock);
 

drivers/gpu/nvgpu/gk20a/gr_gk20a.h

@@ -322,6 +322,7 @@ struct gr_gk20a {
 	u32 timeslice_mode;
 	u32 czf_bypass;
 	u32 pd_max_batches;
+	u32 gfxp_wfi_timeout_count;
 
 	struct gr_ctx_buffer_desc global_ctx_buffer[NR_GLOBAL_CTX_BUF];
 

drivers/gpu/nvgpu/gp10b/gp10b_sysfs.c

@@ -106,6 +106,56 @@ static ssize_t pd_max_batches_read(struct device *dev,
 
 static DEVICE_ATTR(pd_max_batches, ROOTRW, pd_max_batches_read, pd_max_batches_store);
 
+static int write_gfxp_wfi_timeout_count(struct gk20a *g, u32 val)
+{
+	gk20a_writel(g, gr_fe_gfxp_wfi_timeout_r(),
+		gr_fe_gfxp_wfi_timeout_count_f(val));
+
+	return 0;
+}
+
+static ssize_t gfxp_wfi_timeout_count_store(struct device *dev,
+		struct device_attribute *attr, const char *buf, size_t count)
+{
+	struct gk20a *g = get_gk20a(dev);
+	struct gr_gk20a *gr = &g->gr;
+	unsigned long val = 0;
+	int err = -1;
+
+	if (kstrtoul(buf, 10, &val) < 0)
+		return -EINVAL;
+
+	if (val >= 100*1000*1000) /* 100ms @ 1Ghz */
+		return -EINVAL;
+
+	gr->gfxp_wfi_timeout_count = val;
+
+	if (!g->power_on)
+		return count;
+
+	err = gk20a_busy(g);
+	if (err)
+		return err;
+	gr_gk20a_elpg_protected_call(g,
+		write_gfxp_wfi_timeout_count(g, val));
+	gk20a_idle(g);
+
+	return count;
+}
+
+static ssize_t gfxp_wfi_timeout_count_read(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	struct gk20a *g = get_gk20a(dev);
+	struct gr_gk20a *gr = &g->gr;
+	u32 val = gr->gfxp_wfi_timeout_count;
+
+	return snprintf(buf, PAGE_SIZE, "%d\n", val);
+}
+
+static DEVICE_ATTR(gfxp_wfi_timeout_count, ROOTRW,
+		gfxp_wfi_timeout_count_read, gfxp_wfi_timeout_count_store);
+
 void gp10b_create_sysfs(struct device *dev)
 {
 	struct gk20a *g = get_gk20a(dev);
@@ -116,6 +166,7 @@ void gp10b_create_sysfs(struct device *dev)
 	error |= device_create_file(dev, &dev_attr_ecc_enable);
 	error |= device_create_file(dev, &dev_attr_czf_bypass);
 	error |= device_create_file(dev, &dev_attr_pd_max_batches);
+	error |= device_create_file(dev, &dev_attr_gfxp_wfi_timeout_count);
 	if (error)
 		dev_err(dev, "Failed to create sysfs attributes!\n");
 }
@@ -125,4 +176,5 @@ void gp10b_remove_sysfs(struct device *dev)
 	device_remove_file(dev, &dev_attr_ecc_enable);
 	device_remove_file(dev, &dev_attr_czf_bypass);
 	device_remove_file(dev, &dev_attr_pd_max_batches);
+	device_remove_file(dev, &dev_attr_gfxp_wfi_timeout_count);
 }

drivers/gpu/nvgpu/gp10b/gr_gp10b.c

@@ -38,8 +38,6 @@
 #include <nvgpu/hw/gp10b/hw_ctxsw_prog_gp10b.h>
 #include <nvgpu/hw/gp10b/hw_mc_gp10b.h>
 
-#define NVGPU_GFXP_WFI_TIMEOUT_US	100LL
-
 static bool gr_gp10b_is_valid_class(struct gk20a *g, u32 class_num)
 {
 	bool valid = false;
@@ -2295,13 +2293,9 @@ static int gp10b_gr_fuse_override(struct gk20a *g)
 
 static int gr_gp10b_init_preemption_state(struct gk20a *g)
 {
-	struct gk20a_platform *platform = gk20a_get_platform(g->dev);
 	u32 debug_2;
-	u64 sysclk_rate;
+	struct gr_gk20a *gr = &g->gr;
-	u32 sysclk_cycles;
+	u32 sysclk_cycles = gr->gfxp_wfi_timeout_count;
-
-	sysclk_rate = platform->clk_get_rate(g->dev);
-	sysclk_cycles = (u32)((sysclk_rate * NVGPU_GFXP_WFI_TIMEOUT_US) / 1000000ULL);
 	gk20a_writel(g, gr_fe_gfxp_wfi_timeout_r(),
 			gr_fe_gfxp_wfi_timeout_count_f(sysclk_cycles));