gpu: nvgpu: Move rest of CDE structures to Linux

Move rest of CDE structures to common/linux. This includes moving
the per-chip firmware file interpretation functions, and removing CDE
ops from HAL and adding it to nvgpu_os_linux.

JIRA NVGPU-259

Change-Id: I59d8f44bddadecef81ad3c455b363a14034c5e13
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1570403
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>

drivers/gpu/nvgpu/Makefile.nvgpu

@@ -48,6 +48,8 @@ nvgpu-y := \
 	common/linux/cde.o \
 	common/linux/io.o \
 	common/linux/rwsem.o \
+	common/linux/cde_gm20b.o \
+	common/linux/cde_gp10b.o \
 	common/mm/nvgpu_allocator.o \
 	common/mm/bitmap_allocator.o \
 	common/mm/buddy_allocator.o \
@@ -115,7 +117,6 @@ nvgpu-y := \
 	gm20b/pmu_gm20b.o \
 	gm20b/mm_gm20b.o \
 	gm20b/regops_gm20b.o \
-	gm20b/cde_gm20b.o \
 	gm20b/therm_gm20b.o \
 	boardobj/boardobj.o \
 	boardobj/boardobjgrp.o \
@@ -192,7 +193,6 @@ nvgpu-y += \
 	gp10b/rpfb_gp10b.o \
 	gp10b/gp10b_gating_reglist.o \
 	gp10b/regops_gp10b.o \
-	gp10b/cde_gp10b.o \
 	gp10b/therm_gp10b.o \
 	gp10b/fecs_trace_gp10b.o \
 	gp10b/priv_ring_gp10b.o \

drivers/gpu/nvgpu/common/linux/cde.c

@@ -1068,8 +1068,8 @@ __releases(&l->cde_app->mutex)
 	}
 
 	if (scatterbuffer_byte_offset &&
-	    g->ops.cde.need_scatter_buffer &&
-	    g->ops.cde.need_scatter_buffer(g)) {
+	    l->ops.cde.need_scatter_buffer &&
+	    l->ops.cde.need_scatter_buffer(g)) {
 		struct sg_table *sgt;
 		void *scatter_buffer;
 
@@ -1092,7 +1092,7 @@ __releases(&l->cde_app->mutex)
 			err = -EINVAL;
 			goto exit_unmap_surface;
 		} else {
-			err = g->ops.cde.populate_scatter_buffer(g, sgt,
+			err = l->ops.cde.populate_scatter_buffer(g, sgt,
 					compbits_byte_offset, scatter_buffer,
 					scatterbuffer_size);
 			WARN_ON(err);
@@ -1463,8 +1463,8 @@ static int gk20a_buffer_convert_gpu_to_cde_v1(
 	int hprog = -1;
 	int vprog = -1;
 
-	if (g->ops.cde.get_program_numbers)
-		g->ops.cde.get_program_numbers(g, block_height_log2,
+	if (l->ops.cde.get_program_numbers)
+		l->ops.cde.get_program_numbers(g, block_height_log2,
 					       l->cde_app.shader_parameter,
 					       &hprog, &vprog);
 	else {

drivers/gpu/nvgpu/common/linux/cde_gm20b.c

@@ -35,7 +35,7 @@ enum programs {
 	PROG_PASSTHROUGH        = 6,
 };
 
-void gm20b_cde_get_program_numbers(struct gk20a *g,
+static void gm20b_cde_get_program_numbers(struct gk20a *g,
 					  u32 block_height_log2,
 					  u32 shader_parameter,
 					  int *hprog_out, int *vprog_out)
@@ -56,3 +56,9 @@ void gm20b_cde_get_program_numbers(struct gk20a *g,
 	*hprog_out = hprog;
 	*vprog_out = vprog;
 }
+
+struct nvgpu_os_linux_ops gm20b_cde_ops = {
+	.cde = {
+		.get_program_numbers = gm20b_cde_get_program_numbers,
+	},
+};

drivers/gpu/nvgpu/common/linux/cde_gm20b.h

@@ -25,11 +25,8 @@
 #ifndef _NVHOST_GM20B_CDE
 #define _NVHOST_GM20B_CDE
 
-struct gk20a;
+#include "os_linux.h"
 
-void gm20b_cde_get_program_numbers(struct gk20a *g,
-					  u32 block_height_log2,
-					  u32 shader_parameter,
-					  int *hprog_out, int *vprog_out);
+extern struct nvgpu_os_linux_ops gm20b_cde_ops;
 
 #endif

drivers/gpu/nvgpu/common/linux/cde_gp10b.c

@@ -150,3 +150,11 @@ int gp10b_populate_scatter_buffer(struct gk20a *g,
 
 	return 0;
 }
+
+struct nvgpu_os_linux_ops gp10b_cde_ops = {
+	.cde = {
+		.get_program_numbers = gp10b_cde_get_program_numbers,
+		.need_scatter_buffer = gp10b_need_scatter_buffer,
+		.populate_scatter_buffer = gp10b_populate_scatter_buffer,
+	},
+};

drivers/gpu/nvgpu/common/linux/cde_gp10b.h

@@ -25,18 +25,8 @@
 #ifndef _NVHOST_GP10B_CDE
 #define _NVHOST_GP10B_CDE
 
-struct gk20a;
-struct sg_table;
+#include "os_linux.h"
 
-void gp10b_cde_get_program_numbers(struct gk20a *g,
-					  u32 block_height_log2,
-					  u32 shader_parameter,
-					  int *hprog_out, int *vprog_out);
-bool gp10b_need_scatter_buffer(struct gk20a *g);
-int gp10b_populate_scatter_buffer(struct gk20a *g,
-					 struct sg_table *sgt,
-					 size_t surface_size,
-					 void *scatter_buffer_ptr,
-					 size_t scatter_buffer_size);
+extern struct nvgpu_os_linux_ops gp10b_cde_ops;
 
 #endif

drivers/gpu/nvgpu/common/linux/module.c

@@ -49,6 +49,8 @@
 #endif
 #endif
 #include "os_linux.h"
+#include "cde_gm20b.h"
+#include "cde_gp10b.h"
 
 #define CLASS_NAME "nvidia-gpu"
 /* TODO: Change to e.g. "nvidia-gpu%s" once we have symlinks in place. */
@@ -154,6 +156,26 @@ static int gk20a_restore_registers(struct gk20a *g)
 	return 0;
 }
 
+static int nvgpu_init_os_linux_ops(struct nvgpu_os_linux *l) {
+	struct gk20a *g = &l->g;
+	u32 ver = g->gpu_characteristics.arch + g->gpu_characteristics.impl;
+
+	switch (ver) {
+	case GK20A_GPUID_GM20B:
+	case GK20A_GPUID_GM20B_B:
+		l->ops.cde = gm20b_cde_ops.cde;
+		break;
+	case NVGPU_GPUID_GP10B:
+		l->ops.cde = gp10b_cde_ops.cde;
+		break;
+	default:
+		/* CDE is optional, so today ignoring unknown chip is fine */
+		break;
+	}
+
+	return 0;
+}
+
 int gk20a_pm_finalize_poweron(struct device *dev)
 {
 	struct gk20a *g = get_gk20a(dev);
@@ -198,6 +220,10 @@ int gk20a_pm_finalize_poweron(struct device *dev)
 
 	trace_gk20a_finalize_poweron_done(dev_name(dev));
 
+	err = nvgpu_init_os_linux_ops(l);
+	if (err)
+		goto done;
+
 	enable_irq(g->irq_stall);
 	if (g->irq_stall != g->irq_nonstall)
 		enable_irq(g->irq_nonstall);

drivers/gpu/nvgpu/common/linux/os_linux.h

@@ -24,6 +24,21 @@
 #include "gk20a/gk20a.h"
 #include "cde.h"
 
+struct nvgpu_os_linux_ops {
+	struct {
+		void (*get_program_numbers)(struct gk20a *g,
+					    u32 block_height_log2,
+					    u32 shader_parameter,
+					    int *hprog, int *vprog);
+		bool (*need_scatter_buffer)(struct gk20a *g);
+		int (*populate_scatter_buffer)(struct gk20a *g,
+					       struct sg_table *sgt,
+					       size_t surface_size,
+					       void *scatter_buffer_ptr,
+					       size_t scatter_buffer_size);
+	} cde;
+};
+
 struct nvgpu_os_linux {
 	struct gk20a g;
 	struct device *dev;
@@ -99,6 +114,9 @@ struct nvgpu_os_linux {
 #ifdef CONFIG_TEGRA_19x_GPU
 	struct nvgpu_os_linux_t19x t19x;
 #endif
+
+	struct nvgpu_os_linux_ops ops;
+
 #ifdef CONFIG_DEBUG_FS
 	struct dentry *debugfs;
 	struct dentry *debugfs_alias;

drivers/gpu/nvgpu/gk20a/gk20a.h

@@ -943,18 +943,6 @@ struct gpu_ops {
 		int (*perfbuffer_enable)(struct gk20a *g, u64 offset, u32 size);
 		int (*perfbuffer_disable)(struct gk20a *g);
 	} dbg_session_ops;
-	struct {
-		void (*get_program_numbers)(struct gk20a *g,
-					    u32 block_height_log2,
-					    u32 shader_parameter,
-					    int *hprog, int *vprog);
-		bool (*need_scatter_buffer)(struct gk20a *g);
-		int (*populate_scatter_buffer)(struct gk20a *g,
-					       struct sg_table *sgt,
-					       size_t surface_size,
-					       void *scatter_buffer_ptr,
-					       size_t scatter_buffer_size);
-	} cde;
 
 	int (*get_litter_value)(struct gk20a *g, int value);
 	int (*chip_init_gpu_characteristics)(struct gk20a *g);

drivers/gpu/nvgpu/gm20b/hal_gm20b.c

@@ -50,7 +50,6 @@
 #include "pmu_gm20b.h"
 #include "clk_gm20b.h"
 #include "regops_gm20b.h"
-#include "cde_gm20b.h"
 #include "therm_gm20b.h"
 #include "bus_gm20b.h"
 #include "hal_gm20b.h"
@@ -539,9 +538,6 @@ static const struct gpu_ops gm20b_ops = {
 		.perfbuffer_enable = gk20a_perfbuf_enable_locked,
 		.perfbuffer_disable = gk20a_perfbuf_disable_locked,
 	},
-	.cde = {
-		.get_program_numbers = gm20b_cde_get_program_numbers,
-	},
 	.bus = {
 		.init_hw = gk20a_bus_init_hw,
 		.isr = gk20a_bus_isr,
@@ -601,7 +597,6 @@ int gm20b_init_hal(struct gk20a *g)
 	gops->mc = gm20b_ops.mc;
 	gops->dbg_session_ops = gm20b_ops.dbg_session_ops;
 	gops->debug = gm20b_ops.debug;
-	gops->cde = gm20b_ops.cde;
 	gops->bus = gm20b_ops.bus;
 #if defined(CONFIG_GK20A_CYCLE_STATS)
 	gops->css = gm20b_ops.css;

drivers/gpu/nvgpu/gp106/hal_gp106.c

@@ -45,7 +45,6 @@
 #include "gp10b/mm_gp10b.h"
 #include "gp10b/ce_gp10b.h"
 #include "gp10b/regops_gp10b.h"
-#include "gp10b/cde_gp10b.h"
 #include "gp10b/priv_ring_gp10b.h"
 #include "gp10b/fifo_gp10b.h"
 #include "gp10b/fb_gp10b.h"
@@ -654,11 +653,6 @@ static const struct gpu_ops gp106_ops = {
 		.perfbuffer_enable = gk20a_perfbuf_enable_locked,
 		.perfbuffer_disable = gk20a_perfbuf_disable_locked,
 	},
-	.cde = {
-		.get_program_numbers = gp10b_cde_get_program_numbers,
-		.need_scatter_buffer = gp10b_need_scatter_buffer,
-		.populate_scatter_buffer = gp10b_populate_scatter_buffer,
-	},
 	.bus = {
 		.init_hw = gk20a_bus_init_hw,
 		.isr = gk20a_bus_isr,
@@ -738,7 +732,6 @@ int gp106_init_hal(struct gk20a *g)
 	gops->mc = gp106_ops.mc;
 	gops->debug = gp106_ops.debug;
 	gops->dbg_session_ops = gp106_ops.dbg_session_ops;
-	gops->cde = gp106_ops.cde;
 	gops->bus = gp106_ops.bus;
 #if defined(CONFIG_GK20A_CYCLE_STATS)
 	gops->css = gp106_ops.css;

drivers/gpu/nvgpu/gp10b/hal_gp10b.c

@@ -51,7 +51,6 @@
 #include "gp10b/fifo_gp10b.h"
 #include "gp10b/gp10b_gating_reglist.h"
 #include "gp10b/regops_gp10b.h"
-#include "gp10b/cde_gp10b.h"
 #include "gp10b/therm_gp10b.h"
 #include "gp10b/priv_ring_gp10b.h"
 
@@ -574,11 +573,6 @@ static const struct gpu_ops gp10b_ops = {
 		.perfbuffer_enable = gk20a_perfbuf_enable_locked,
 		.perfbuffer_disable = gk20a_perfbuf_disable_locked,
 	},
-	.cde = {
-		.get_program_numbers = gp10b_cde_get_program_numbers,
-		.need_scatter_buffer = gp10b_need_scatter_buffer,
-		.populate_scatter_buffer = gp10b_populate_scatter_buffer,
-	},
 	.bus = {
 		.init_hw = gk20a_bus_init_hw,
 		.isr = gk20a_bus_isr,
@@ -628,7 +622,6 @@ int gp10b_init_hal(struct gk20a *g)
 	gops->mc = gp10b_ops.mc;
 	gops->debug = gp10b_ops.debug;
 	gops->dbg_session_ops = gp10b_ops.dbg_session_ops;
-	gops->cde = gp10b_ops.cde;
 	gops->bus = gp10b_ops.bus;
 #if defined(CONFIG_GK20A_CYCLE_STATS)
 	gops->css = gp10b_ops.css;

drivers/gpu/nvgpu/vgpu/gm20b/vgpu_hal_gm20b.c

@@ -42,7 +42,6 @@
 #include "gm20b/pmu_gm20b.h"
 #include "gm20b/fb_gm20b.h"
 #include "gm20b/bus_gm20b.h"
-#include "gm20b/cde_gm20b.h"
 #include "gm20b/regops_gm20b.h"
 #include "gm20b/clk_gm20b.h"
 #include "gm20b/therm_gm20b.h"
@@ -438,9 +437,6 @@ static const struct gpu_ops vgpu_gm20b_ops = {
 		.perfbuffer_enable = vgpu_perfbuffer_enable,
 		.perfbuffer_disable = vgpu_perfbuffer_disable,
 	},
-	.cde = {
-		.get_program_numbers = gm20b_cde_get_program_numbers,
-	},
 	.bus = {
 		.init_hw = gk20a_bus_init_hw,
 		.isr = gk20a_bus_isr,
@@ -501,7 +497,6 @@ int vgpu_gm20b_init_hal(struct gk20a *g)
 	gops->mc = vgpu_gm20b_ops.mc;
 	gops->dbg_session_ops = vgpu_gm20b_ops.dbg_session_ops;
 	gops->debug = vgpu_gm20b_ops.debug;
-	gops->cde = vgpu_gm20b_ops.cde;
 	gops->bus = vgpu_gm20b_ops.bus;
 #if defined(CONFIG_GK20A_CYCLE_STATS)
 	gops->css = vgpu_gm20b_ops.css;

drivers/gpu/nvgpu/vgpu/gp10b/vgpu_hal_gp10b.c

@@ -50,7 +50,6 @@
 #include "gp10b/fifo_gp10b.h"
 #include "gp10b/gp10b_gating_reglist.h"
 #include "gp10b/regops_gp10b.h"
-#include "gp10b/cde_gp10b.h"
 #include "gp10b/therm_gp10b.h"
 #include "gp10b/priv_ring_gp10b.h"
 
@@ -478,11 +477,6 @@ static const struct gpu_ops vgpu_gp10b_ops = {
 		.perfbuffer_enable = vgpu_perfbuffer_enable,
 		.perfbuffer_disable = vgpu_perfbuffer_disable,
 	},
-	.cde = {
-		.get_program_numbers = gp10b_cde_get_program_numbers,
-		.need_scatter_buffer = gp10b_need_scatter_buffer,
-		.populate_scatter_buffer = gp10b_populate_scatter_buffer,
-	},
 	.bus = {
 		.init_hw = gk20a_bus_init_hw,
 		.isr = gk20a_bus_isr,
@@ -533,7 +527,6 @@ int vgpu_gp10b_init_hal(struct gk20a *g)
 	gops->mc = vgpu_gp10b_ops.mc;
 	gops->debug = vgpu_gp10b_ops.debug;
 	gops->dbg_session_ops = vgpu_gp10b_ops.dbg_session_ops;
-	gops->cde = vgpu_gp10b_ops.cde;
 	gops->bus = vgpu_gp10b_ops.bus;
 #if defined(CONFIG_GK20A_CYCLE_STATS)
 	gops->css = vgpu_gp10b_ops.css;