nv-tegra-mirror/linux-nvgpu
1
0
Fork 0
mirror of git://nv-tegra.nvidia.com/linux-nvgpu.git synced 2025-12-23 18:16:01 +03:00
Code Issues Packages Projects Releases Wiki Activity

gpu: nvgpu: vgpu: manage gr_ctx as independent resource

Browse Source 
gr_ctx will managed as independent resource in RM server
and vgpu can get a gr_ctx handle.

Bug 1702773

Change-Id: I87251af61711f0d7997ce90df8a3de196a9b481a
Signed-off-by: Richard Zhao <rizhao@nvidia.com>
Reviewed-on: http://git-master/r/1144931
(cherry picked from commit 2efbd143adaf60570121f1c212dc6b6f3d5a1661)
Reviewed-on: http://git-master/r/1150704
GVS: Gerrit_Virtual_Submit
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
This commit is contained in:
Richard Zhao
2016-05-09 15:53:39 -07:00
committed by Terje Bergstrom
parent 9c4f3799d1
commit a71ce831fb
2 changed files with 80 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

94
drivers/gpu/nvgpu/vgpu/gr_vgpu.c
View File

@@ -21,7 +21,7 @@ static int vgpu_gr_commit_inst(struct channel_gk20a *c, u64 gpu_va)
{ {
struct gk20a_platform *platform = gk20a_get_platform(c->g->dev); struct gk20a_platform *platform = gk20a_get_platform(c->g->dev);
struct tegra_vgpu_cmd_msg msg; struct tegra_vgpu_cmd_msg msg;
struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx; struct tegra_vgpu_ch_ctx_params *p = &msg.params.ch_ctx;
int err; int err;
gk20a_dbg_fn(""); gk20a_dbg_fn("");
@@ -39,7 +39,7 @@ static int vgpu_gr_commit_global_ctx_buffers(struct gk20a *g,
{ {
struct gk20a_platform *platform = gk20a_get_platform(g->dev); struct gk20a_platform *platform = gk20a_get_platform(g->dev);
struct tegra_vgpu_cmd_msg msg; struct tegra_vgpu_cmd_msg msg;
struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx; struct tegra_vgpu_ch_ctx_params *p = &msg.params.ch_ctx;
int err; int err;
gk20a_dbg_fn(""); gk20a_dbg_fn("");
@@ -58,7 +58,7 @@ static int vgpu_gr_load_golden_ctx_image(struct gk20a *g,
{ {
struct gk20a_platform *platform = gk20a_get_platform(g->dev); struct gk20a_platform *platform = gk20a_get_platform(g->dev);
struct tegra_vgpu_cmd_msg msg; struct tegra_vgpu_cmd_msg msg;
struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx; struct tegra_vgpu_ch_ctx_params *p = &msg.params.ch_ctx;
int err; int err;
gk20a_dbg_fn(""); gk20a_dbg_fn("");
@@ -130,7 +130,7 @@ static int vgpu_gr_map_global_ctx_buffers(struct gk20a *g,
{ {
struct gk20a_platform *platform = gk20a_get_platform(g->dev); struct gk20a_platform *platform = gk20a_get_platform(g->dev);
struct tegra_vgpu_cmd_msg msg; struct tegra_vgpu_cmd_msg msg;
struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx; struct tegra_vgpu_ch_ctx_params *p = &msg.params.ch_ctx;
struct vm_gk20a *ch_vm = c->vm; struct vm_gk20a *ch_vm = c->vm;
u64 *g_bfr_va = c->ch_ctx.global_ctx_buffer_va; u64 *g_bfr_va = c->ch_ctx.global_ctx_buffer_va;
u64 *g_bfr_size = c->ch_ctx.global_ctx_buffer_size; u64 *g_bfr_size = c->ch_ctx.global_ctx_buffer_size;
@@ -219,7 +219,7 @@ static void vgpu_gr_unmap_global_ctx_buffers(struct channel_gk20a *c)
if (c->ch_ctx.global_ctx_buffer_mapped) { if (c->ch_ctx.global_ctx_buffer_mapped) {
struct tegra_vgpu_cmd_msg msg; struct tegra_vgpu_cmd_msg msg;
struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx; struct tegra_vgpu_ch_ctx_params *p = &msg.params.ch_ctx;
int err; int err;
msg.cmd = TEGRA_VGPU_CMD_CHANNEL_UNMAP_GR_GLOBAL_CTX; msg.cmd = TEGRA_VGPU_CMD_CHANNEL_UNMAP_GR_GLOBAL_CTX;
@@ -246,10 +246,10 @@ int vgpu_gr_alloc_gr_ctx(struct gk20a *g,
u32 flags) u32 flags)
{ {
struct gk20a_platform *platform = gk20a_get_platform(g->dev); struct gk20a_platform *platform = gk20a_get_platform(g->dev);
struct tegra_vgpu_cmd_msg msg; struct tegra_vgpu_cmd_msg msg = {0};
struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx; struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx;
struct gr_gk20a *gr = &g->gr; struct gr_gk20a *gr = &g->gr;
struct gr_ctx_desc *gr_ctx = *__gr_ctx; struct gr_ctx_desc *gr_ctx;
int err; int err;
gk20a_dbg_fn(""); gk20a_dbg_fn("");
@@ -261,6 +261,10 @@ int vgpu_gr_alloc_gr_ctx(struct gk20a *g,
gr->ctx_vars.buffer_size = gr->ctx_vars.golden_image_size; gr->ctx_vars.buffer_size = gr->ctx_vars.golden_image_size;
gr->ctx_vars.buffer_total_size = gr->ctx_vars.golden_image_size; gr->ctx_vars.buffer_total_size = gr->ctx_vars.golden_image_size;
gr_ctx = kzalloc(sizeof(*gr_ctx), GFP_KERNEL);
if (!gr_ctx)
return -ENOMEM;
gr_ctx->mem.size = gr->ctx_vars.buffer_total_size; gr_ctx->mem.size = gr->ctx_vars.buffer_total_size;
gr_ctx->mem.gpu_va = gk20a_vm_alloc_va(vm, gr_ctx->mem.gpu_va = gk20a_vm_alloc_va(vm,
gr_ctx->mem.size, gr_ctx->mem.size,
@@ -271,49 +275,27 @@ int vgpu_gr_alloc_gr_ctx(struct gk20a *g,
return -ENOMEM; return -ENOMEM;
} }
msg.cmd = TEGRA_VGPU_CMD_CHANNEL_ALLOC_GR_CTX; msg.cmd = TEGRA_VGPU_CMD_GR_CTX_ALLOC;
msg.handle = platform->virt_handle; msg.handle = platform->virt_handle;
p->handle = gr_ctx->virt_ctx; p->as_handle = vm->handle;
p->gr_ctx_va = gr_ctx->mem.gpu_va; p->gr_ctx_va = gr_ctx->mem.gpu_va;
p->class_num = class; p->class_num = class;
err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg));
err = err ? err : msg.ret;
if (err || msg.ret) { if (unlikely(err)) {
gk20a_err(dev_from_gk20a(g), "fail to alloc gr_ctx");
gk20a_vm_free_va(vm, gr_ctx->mem.gpu_va, gk20a_vm_free_va(vm, gr_ctx->mem.gpu_va,
gr_ctx->mem.size, 0); gr_ctx->mem.size, 0);
kfree(gr_ctx); kfree(gr_ctx);
err = -ENOMEM; } else {
gr_ctx->virt_ctx = p->gr_ctx_handle;
*__gr_ctx = gr_ctx;
} }
return err; return err;
} }
static int vgpu_gr_alloc_channel_gr_ctx(struct gk20a *g,
struct channel_gk20a *c,
u32 class,
u32 flags)
{
struct gr_ctx_desc **gr_ctx = &c->ch_ctx.gr_ctx;
struct gr_ctx_desc *__gr_ctx = kzalloc(sizeof(*__gr_ctx), GFP_KERNEL);
int err;
gk20a_dbg_fn("");
if (!__gr_ctx)
return -ENOMEM;
__gr_ctx->virt_ctx = c->virt_ctx;
*gr_ctx = __gr_ctx;
err = g->ops.gr.alloc_gr_ctx(g, gr_ctx, c->vm, class, flags);
if (err) {
kfree(__gr_ctx);
return err;
}
c->ch_ctx.gr_ctx = __gr_ctx;
return 0;
}
void vgpu_gr_free_gr_ctx(struct gk20a *g, struct vm_gk20a *vm, void vgpu_gr_free_gr_ctx(struct gk20a *g, struct vm_gk20a *vm,
struct gr_ctx_desc *gr_ctx) struct gr_ctx_desc *gr_ctx)
{ {
@@ -325,9 +307,9 @@ void vgpu_gr_free_gr_ctx(struct gk20a *g, struct vm_gk20a *vm,
struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx; struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx;
int err; int err;
msg.cmd = TEGRA_VGPU_CMD_CHANNEL_FREE_GR_CTX; msg.cmd = TEGRA_VGPU_CMD_GR_CTX_FREE;
msg.handle = platform->virt_handle; msg.handle = platform->virt_handle;
p->handle = gr_ctx->virt_ctx; p->gr_ctx_handle = gr_ctx->virt_ctx;
err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg)); err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg));
WARN_ON(err || msg.ret); WARN_ON(err || msg.ret);
@@ -351,7 +333,7 @@ static int vgpu_gr_alloc_channel_patch_ctx(struct gk20a *g,
struct patch_desc *patch_ctx = &c->ch_ctx.patch_ctx; struct patch_desc *patch_ctx = &c->ch_ctx.patch_ctx;
struct vm_gk20a *ch_vm = c->vm; struct vm_gk20a *ch_vm = c->vm;
struct tegra_vgpu_cmd_msg msg; struct tegra_vgpu_cmd_msg msg;
struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx; struct tegra_vgpu_ch_ctx_params *p = &msg.params.ch_ctx;
int err; int err;
gk20a_dbg_fn(""); gk20a_dbg_fn("");
@@ -387,7 +369,7 @@ static void vgpu_gr_free_channel_patch_ctx(struct channel_gk20a *c)
if (patch_ctx->mem.gpu_va) { if (patch_ctx->mem.gpu_va) {
struct tegra_vgpu_cmd_msg msg; struct tegra_vgpu_cmd_msg msg;
struct tegra_vgpu_gr_ctx_params *p = &msg.params.gr_ctx; struct tegra_vgpu_ch_ctx_params *p = &msg.params.ch_ctx;
int err; int err;
msg.cmd = TEGRA_VGPU_CMD_CHANNEL_FREE_GR_PATCH_CTX; msg.cmd = TEGRA_VGPU_CMD_CHANNEL_FREE_GR_PATCH_CTX;
@@ -443,6 +425,26 @@ static void vgpu_gr_free_channel_ctx(struct channel_gk20a *c)
c->first_init = false; c->first_init = false;
} }
static int vgpu_gr_ch_bind_gr_ctx(struct channel_gk20a *c)
{
struct gk20a_platform *platform = gk20a_get_platform(c->g->dev);
struct gr_ctx_desc *gr_ctx = c->ch_ctx.gr_ctx;
struct tegra_vgpu_cmd_msg msg = {0};
struct tegra_vgpu_channel_bind_gr_ctx_params *p =
&msg.params.ch_bind_gr_ctx;
int err;
msg.cmd = TEGRA_VGPU_CMD_CHANNEL_BIND_GR_CTX;
msg.handle = platform->virt_handle;
p->ch_handle = c->virt_ctx;
p->gr_ctx_handle = gr_ctx->virt_ctx;
err = vgpu_comm_sendrecv(&msg, sizeof(msg), sizeof(msg));
err = err ? err : msg.ret;
WARN_ON(err);
return err;
}
static int vgpu_gr_alloc_obj_ctx(struct channel_gk20a *c, static int vgpu_gr_alloc_obj_ctx(struct channel_gk20a *c,
struct nvgpu_alloc_obj_ctx_args *args) struct nvgpu_alloc_obj_ctx_args *args)
{ {
@@ -476,9 +478,13 @@ static int vgpu_gr_alloc_obj_ctx(struct channel_gk20a *c,
/* allocate gr ctx buffer */ /* allocate gr ctx buffer */
if (!ch_ctx->gr_ctx) { if (!ch_ctx->gr_ctx) {
err = vgpu_gr_alloc_channel_gr_ctx(g, c, err = g->ops.gr.alloc_gr_ctx(g, &c->ch_ctx.gr_ctx,
args->class_num, c->vm,
args->flags); args->class_num,
args->flags);
if (!err)
err = vgpu_gr_ch_bind_gr_ctx(c);
if (err) { if (err) {
gk20a_err(dev_from_gk20a(g), gk20a_err(dev_from_gk20a(g),
"fail to allocate gr ctx buffer"); "fail to allocate gr ctx buffer");

35
include/linux/tegra_vgpu.h
View File

@@ -54,8 +54,8 @@ enum {
TEGRA_VGPU_CMD_CHANNEL_DISABLE, TEGRA_VGPU_CMD_CHANNEL_DISABLE,
TEGRA_VGPU_CMD_CHANNEL_PREEMPT, TEGRA_VGPU_CMD_CHANNEL_PREEMPT,
TEGRA_VGPU_CMD_CHANNEL_SETUP_RAMFC, TEGRA_VGPU_CMD_CHANNEL_SETUP_RAMFC,
TEGRA_VGPU_CMD_CHANNEL_ALLOC_GR_CTX, TEGRA_VGPU_CMD_CHANNEL_ALLOC_GR_CTX, /* deprecated */
TEGRA_VGPU_CMD_CHANNEL_FREE_GR_CTX, TEGRA_VGPU_CMD_CHANNEL_FREE_GR_CTX, /* deprecated */
TEGRA_VGPU_CMD_CHANNEL_COMMIT_GR_CTX, TEGRA_VGPU_CMD_CHANNEL_COMMIT_GR_CTX,
TEGRA_VGPU_CMD_CHANNEL_ALLOC_GR_PATCH_CTX, TEGRA_VGPU_CMD_CHANNEL_ALLOC_GR_PATCH_CTX,
TEGRA_VGPU_CMD_CHANNEL_FREE_GR_PATCH_CTX, TEGRA_VGPU_CMD_CHANNEL_FREE_GR_PATCH_CTX,
@@ -84,6 +84,10 @@ enum {
TEGRA_VGPU_CMD_CHANNEL_SET_SMPC_CTXSW_MODE, TEGRA_VGPU_CMD_CHANNEL_SET_SMPC_CTXSW_MODE,
TEGRA_VGPU_CMD_CHANNEL_SET_HWPM_CTXSW_MODE, TEGRA_VGPU_CMD_CHANNEL_SET_HWPM_CTXSW_MODE,
TEGRA_VGPU_CMD_CHANNEL_FREE_HWPM_CTX, TEGRA_VGPU_CMD_CHANNEL_FREE_HWPM_CTX,
TEGRA_VGPU_CMD_GR_CTX_ALLOC,
TEGRA_VGPU_CMD_GR_CTX_FREE,
TEGRA_VGPU_CMD_CHANNEL_BIND_GR_CTX,
TEGRA_VGPU_CMD_TSG_BIND_GR_CTX,
}; };
struct tegra_vgpu_connect_params { struct tegra_vgpu_connect_params {
@@ -192,7 +196,7 @@ struct tegra_vgpu_ramfc_params {
u8 iova; u8 iova;
}; };
struct tegra_vgpu_gr_ctx_params { struct tegra_vgpu_ch_ctx_params {
u64 handle; u64 handle;
u64 gr_ctx_va; u64 gr_ctx_va;
u64 patch_ctx_va; u64 patch_ctx_va;
@@ -268,10 +272,11 @@ struct tegra_vgpu_zbc_query_table_params {
#define TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_MAX 4 #define TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_MAX 4
struct tegra_vgpu_gr_bind_ctxsw_buffers_params { struct tegra_vgpu_gr_bind_ctxsw_buffers_params {
u64 handle; u64 handle; /* deprecated */
u64 gpu_va[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_MAX]; u64 gpu_va[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_MAX];
u64 size[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_MAX]; u64 size[TEGRA_VGPU_GR_BIND_CTXSW_BUFFER_MAX];
u32 mode; u32 mode;
u64 gr_ctx_handle;
}; };
struct tegra_vgpu_mmu_debug_mode { struct tegra_vgpu_mmu_debug_mode {
@@ -339,6 +344,23 @@ struct tegra_vgpu_channel_free_hwpm_ctx {
u64 handle; u64 handle;
}; };
struct tegra_vgpu_gr_ctx_params {
u64 gr_ctx_handle;
u64 as_handle;
u64 gr_ctx_va;
u32 class_num;
};
struct tegra_vgpu_channel_bind_gr_ctx_params {
u64 ch_handle;
u64 gr_ctx_handle;
};
struct tegra_vgpu_tsg_bind_gr_ctx_params {
u32 tsg_id;
u64 gr_ctx_handle;
};
struct tegra_vgpu_cmd_msg { struct tegra_vgpu_cmd_msg {
u32 cmd; u32 cmd;
int ret; int ret;
@@ -354,7 +376,7 @@ struct tegra_vgpu_cmd_msg {
struct tegra_vgpu_as_invalidate_params as_invalidate; struct tegra_vgpu_as_invalidate_params as_invalidate;
struct tegra_vgpu_channel_config_params channel_config; struct tegra_vgpu_channel_config_params channel_config;
struct tegra_vgpu_ramfc_params ramfc; struct tegra_vgpu_ramfc_params ramfc;
struct tegra_vgpu_gr_ctx_params gr_ctx; struct tegra_vgpu_ch_ctx_params ch_ctx;
struct tegra_vgpu_zcull_bind_params zcull_bind; struct tegra_vgpu_zcull_bind_params zcull_bind;
struct tegra_vgpu_cache_maint_params cache_maint; struct tegra_vgpu_cache_maint_params cache_maint;
struct tegra_vgpu_runlist_params runlist; struct tegra_vgpu_runlist_params runlist;
@@ -372,6 +394,9 @@ struct tegra_vgpu_cmd_msg {
struct tegra_vgpu_fecs_trace_filter fecs_trace_filter; struct tegra_vgpu_fecs_trace_filter fecs_trace_filter;
struct tegra_vgpu_channel_set_ctxsw_mode set_ctxsw_mode; struct tegra_vgpu_channel_set_ctxsw_mode set_ctxsw_mode;
struct tegra_vgpu_channel_free_hwpm_ctx free_hwpm_ctx; struct tegra_vgpu_channel_free_hwpm_ctx free_hwpm_ctx;
struct tegra_vgpu_gr_ctx_params gr_ctx;
struct tegra_vgpu_channel_bind_gr_ctx_params ch_bind_gr_ctx;
struct tegra_vgpu_tsg_bind_gr_ctx_params tsg_bind_gr_ctx;
char padding[192]; char padding[192];
} params; } params;
}; };