mika/libnx
1
0
Fork 0
mirror of https://github.com/switchbrew/libnx.git synced 2025-06-21 12:32:40 +02:00
Code Issues Packages Projects Releases Wiki Activity

Added better code for gpfifo handling which is disabled, currently broken with the data written into the gpfifo-data mem. Updated nvgfxEventWait(). Load the classnum for nvioctlChannel_AllocObjCtx() from gpu_characteristics. Added disabled code for gpfifo in gfx.c + added a comment.

Browse Source
This commit is contained in:
yellows8 2017-12-17 20:05:38 -05:00
parent 2a26b50459
commit 824dadd291
3 changed files with 55 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
nx/include/switch/gfx/nvgfx.h
View File

@ -1,4 +1,5 @@
Result nvgfxInitialize(void);
void nvgfxExit(void);
Result nvgfxEventWait(u32 syncpt_id, u32 threshold, s32 timeout);
Result nvgfxSubmitGpfifo(void);
Result nvgfxGetFramebuffer(u8 **buffer, size_t *size);

8
nx/source/gfx/gfx.c
View File

@ -27,6 +27,7 @@ extern u32 __nx_applet_type;
extern u32 g_nvgfx_totalframebufs;
extern size_t g_nvgfx_singleframebuf_size;
extern nvioctl_fence g_nvgfx_nvhostgpu_gpfifo_fence;
//static Result _gfxGetDisplayResolution(u64 *width, u64 *height);
@ -87,6 +88,8 @@ static Result _gfxDequeueBuffer(bool handle_fence) {
if (R_SUCCEEDED(rc)) g_gfxCurrentBuffer = (g_gfxCurrentBuffer + 1) & (g_nvgfx_totalframebufs-1);
//if (R_SUCCEEDED(rc)) rc = nvgfxSubmitGpfifo();
return rc;
}
@ -96,6 +99,9 @@ static Result _gfxQueueBuffer(s32 buf) {
if (buf == -1) return 0;
g_gfxQueueBufferData.timestamp = svcGetSystemTick();//This is probably not the proper value for the timestamp, but shouldn't(?) matter.
//if (g_nvgfx_nvhostgpu_gpfifo_fence.id) memcpy(&g_gfxQueueBufferData.fence.nv_fences[0], &g_nvgfx_nvhostgpu_gpfifo_fence, sizeof(nvioctl_fence));
//if (g_nvgfx_nvhostgpu_gpfifo_fence.id) rc = nvgfxEventWait(g_nvgfx_nvhostgpu_gpfifo_fence.id, g_nvgfx_nvhostgpu_gpfifo_fence.value, -1);
if (R_FAILED(rc)) return rc;
rc = bufferProducerQueueBuffer(buf, &g_gfxQueueBufferData, &g_gfx_QueueBuffer_QueueBufferOutput);
if (R_FAILED(rc)) return rc;
@ -150,7 +156,7 @@ static Result _gfxInit(viServiceType servicetype, const char *DisplayName, u32 L
if (R_SUCCEEDED(rc)) rc = nvgfxGetFramebuffer(&g_gfxFramebuf, &g_gfxFramebufSize);
if (R_SUCCEEDED(rc)) {
if (R_SUCCEEDED(rc)) { //TODO: Merge this into gfxSwapBuffers()?
for(i=0; i<2; i++) {
rc = _gfxDequeueBuffer(0);
if (R_FAILED(rc)) break;

57
nx/source/gfx/nvgfx.c
View File

@ -2,6 +2,8 @@
#include <malloc.h>
#include <switch.h>
//#include "nvgfx_gpu_gpfifo_data0_bin.h"
typedef struct {
bool initialized;
u32 handle;
@ -28,7 +30,7 @@ static u32 g_nvgfx_zcullinfo[40>>2];
static nvioctl_va_region g_nvgfx_nvhostasgpu_varegions[2];
static nvioctl_l2_state g_nvgfx_l2state;
static nvioctl_fence g_nvgfx_nvhost_fence;
static nvioctl_fence g_nvgfx_nvhostgpu_gpfifo_fence;
nvioctl_fence g_nvgfx_nvhostgpu_gpfifo_fence;
static u8 *g_nvgfx_nvhost_userdata;
static size_t g_nvgfx_nvhost_userdata_size;
static u32 g_nvgfx_nvhostctrl_eventres;
@ -38,9 +40,12 @@ size_t g_nvgfx_singleframebuf_size = /*0x3c0000*/ 1280*768*4;
static nvmapobj nvmap_objs[18];
static u64 nvmap_obj3_mapbuffer_x0_offset;
static u64 nvmap_obj4_mapbuffer_x0_offset;
static u64 nvmap_obj6_mapbuffer_xdb_offset;
static u64 g_nvgfx_gpfifo_pos = 0;
//Some of this struct is based on tegra_dc_ext_flip_windowattr.
static u32 g_gfxprod_BufferInitData[0x178>>2] = {
0x1, 0x16c, 0x0,
@ -132,8 +137,6 @@ Result nvgfxInitialize(void) {
u32 framebuf_nvmap_handle = 0;//Special handle ID for framebuf/windowbuf.
//nvioctl_gpfifo_entry gpfifo_entries[2] = {{0x00030000, 0x00177a05}, {0x00031778, 0x80002e05}};
g_nvgfx_fd_nvhostctrlgpu = 0;
g_nvgfx_fd_nvhostasgpu = 0;
g_nvgfx_fd_nvmap = 0;
@ -155,10 +158,13 @@ Result nvgfxInitialize(void) {
memset(&g_nvgfx_nvhostgpu_gpfifo_fence, 0, sizeof(g_nvgfx_nvhostgpu_gpfifo_fence));
g_nvgfx_nvhostasgpu_allocspace_offset = 0;
g_nvgfx_zcullctxsize = 0;
nvmap_obj3_mapbuffer_x0_offset = 0;
nvmap_obj4_mapbuffer_x0_offset = 0;
nvmap_obj6_mapbuffer_xdb_offset = 0;
g_nvgfx_nvhostctrl_eventres = 0;
g_nvgfx_gpfifo_pos = 0;
//All of the below sizes for nvmapobjInitialize are from certain official sw.
if (R_SUCCEEDED(rc)) rc = nvmapobjInitialize(&nvmap_objs[0], 0x1000);
if (R_SUCCEEDED(rc)) rc = nvmapobjInitialize(&nvmap_objs[1], 0x10000);
@ -230,7 +236,7 @@ Result nvgfxInitialize(void) {
if (R_SUCCEEDED(rc)) rc = nvioctlChannel_AllocGpfifoEx2(g_nvgfx_fd_nvhostgpu, 0x800, 0x1, 0, 0, 0, 0, &g_nvgfx_nvhost_fence);
if (R_SUCCEEDED(rc)) rc = nvioctlChannel_AllocObjCtx(g_nvgfx_fd_nvhostgpu, NVIOCTL_CHANNEL_OBJ_CLASSNUM_3d, 0);
if (R_SUCCEEDED(rc)) rc = nvioctlChannel_AllocObjCtx(g_nvgfx_fd_nvhostgpu, g_nvgfx_gpu_characteristics.threed_class, 0);
if (R_SUCCEEDED(rc)) rc = nvQueryEvent(g_nvgfx_fd_nvhostgpu, 3, &g_nvgfx_nvhostgpu_event3);
@ -242,7 +248,7 @@ Result nvgfxInitialize(void) {
if (R_SUCCEEDED(rc)) rc = nvmapobjSetup(&nvmap_objs[3], 0, 0, 0x20000, 0);
if (R_SUCCEEDED(rc)) rc = nvioctlNvhostAsGpu_MapBufferEx(g_nvgfx_fd_nvhostasgpu, 0, 0, nvmap_objs[3].handle, 0x10000, 0, 0, 0, NULL);
if (R_SUCCEEDED(rc)) rc = nvioctlNvhostAsGpu_MapBufferEx(g_nvgfx_fd_nvhostasgpu, 0, 0, nvmap_objs[3].handle, 0x10000, 0, 0, 0, &nvmap_obj3_mapbuffer_x0_offset);
if (R_SUCCEEDED(rc)) rc = nvioctlNvhostAsGpu_MapBufferEx(g_nvgfx_fd_nvhostasgpu, 0, 0xfe, nvmap_objs[3].handle, 0x10000, 0, 0, 0, NULL);
if (R_SUCCEEDED(rc)) rc = nvmapobjSetup(&nvmap_objs[4], 0, 0x1, 0x20000, 0);
@ -252,10 +258,10 @@ Result nvgfxInitialize(void) {
if (R_SUCCEEDED(rc)) rc = nvioctlChannel_ZCullBind(g_nvgfx_fd_nvhostgpu, nvmap_obj4_mapbuffer_x0_offset+0x8000, 0x2);
//Officially, ipcQueryPointerBufferSize and NVGPU_IOCTL_CHANNEL_SUBMIT_GPFIFO(nvioctlChannel_SubmitGPFIFO) are used here with the duplicate service session setup during nv serv init.
//TODO: This is probably used for GPU rendering? Is this really needed when not doing actual GPU rendering?
//Skip this since this causes a white-screen hang.
//if (R_SUCCEEDED(rc)) rc = nvioctlChannel_SubmitGPFIFO(g_nvgfx_fd_nvhostgpu, gpfifo_entries, 2, 0x104, &g_nvgfx_nvhostgpu_gpfifo_fence);
//Officially, ipcQueryPointerBufferSize and NVGPU_IOCTL_CHANNEL_SUBMIT_GPFIFO(nvioctlChannel_SubmitGpfifo) are used here with the duplicate service session setup during nv serv init.
//This is used for GPU rendering.
//TODO: Is this really needed when not doing actual GPU rendering?
//if (R_SUCCEEDED(rc)) rc = nvgfxSubmitGpfifo();
if (R_SUCCEEDED(rc)) rc = nvmapobjSetup(&nvmap_objs[5], 0, 0, 0x20000, 0);
@ -448,10 +454,11 @@ Result nvgfxEventWait(u32 syncpt_id, u32 threshold, s32 timeout) {
if (R_SUCCEEDED(rc)) {
do {
rc = nvioctlNvhostCtrl_EventWait(g_nvgfx_fd_nvhostctrl, /*0x42, 0x1ca7*/syncpt_id, threshold, /*0x64*/timeout, 0, &g_nvgfx_nvhostctrl_eventres);
rc = nvioctlNvhostCtrl_EventWait(g_nvgfx_fd_nvhostctrl, syncpt_id, threshold, timeout, 0, &g_nvgfx_nvhostctrl_eventres);
} while(rc==5);//timeout error
}
//Official sw only uses the below block when event-waiting timeout occurs.
//Currently broken.
//if (R_SUCCEEDED(rc)) rc = nvQueryEvent(g_nvgfx_fd_nvhostctrl, g_nvgfx_nvhostctrl_eventres, &g_nvgfx_nvhostctrl_eventhandle);
@ -465,6 +472,36 @@ Result nvgfxEventWait(u32 syncpt_id, u32 threshold, s32 timeout) {
return rc;
}
Result nvgfxSubmitGpfifo(void) {
//Extracted from memory of certain official sw.
u8 gpfifo_data[] = {0x00, 0x00, 0x00, 0x00, 0x51, 0x04, 0x00, 0x80, 0xB2, 0x00, 0x01, 0x20, 0x42, 0x00, 0x10, 0x00, 0x51, 0x04, 0x00, 0x80, 0xC0, 0x06, 0x04, 0x20, 0x05, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x10, 0xF0, 0x00, 0x10, 0xE0, 0x03, 0x00, 0x80};
nvioctl_gpfifo_entry gpfifo_entries[2];
u64 total_size = 0x1778+sizeof(gpfifo_data);
u64 tmpva;
if (g_nvgfx_gpfifo_pos >= nvmap_objs[3].mem_size || g_nvgfx_gpfifo_pos+total_size >= nvmap_objs[3].mem_size) g_nvgfx_gpfifo_pos = 0;
//memcpy(&nvmap_objs[3].mem[g_nvgfx_gpfifo_pos], nvgfx_gpu_gpfifo_data0_bin, nvgfx_gpu_gpfifo_data0_bin_size);
memcpy(&nvmap_objs[3].mem[g_nvgfx_gpfifo_pos+0x1778], gpfifo_data, sizeof(gpfifo_data));
armDCacheFlush(nvmap_objs[3].mem, nvmap_objs[3].mem_size);
tmpva = nvmap_obj3_mapbuffer_x0_offset+g_nvgfx_gpfifo_pos;
gpfifo_entries[0].entry0 = (u32)tmpva | (0x00177aULL<<40);
gpfifo_entries[0].entry1 = ((u32)(tmpva>>32)) | (0x00177a<<8);
tmpva = nvmap_obj3_mapbuffer_x0_offset+g_nvgfx_gpfifo_pos+0x1778;
gpfifo_entries[1].entry0 = (u32)tmpva;
gpfifo_entries[1].entry1 = ((u32)(tmpva>>32)) | (0x80002e<<8);
g_nvgfx_gpfifo_pos+= total_size;
Result rc = nvioctlChannel_SubmitGpfifo(g_nvgfx_fd_nvhostgpu, gpfifo_entries, 2, 0x104, &g_nvgfx_nvhostgpu_gpfifo_fence);
return rc;
}
Result nvgfxGetFramebuffer(u8 **buffer, size_t *size) {
if(!g_nvgfxInitialized)return MAKERESULT(MODULE_LIBNX, LIBNX_NOTINITIALIZED);