path: root/src/lnc_cmdbuf.c

/*
 * INTEL CONFIDENTIAL
 * Copyright 2007 Intel Corporation. All Rights Reserved.
 *
 * The source code contained or described herein and all documents related to
 * the source code ("Material") are owned by Intel Corporation or its suppliers
 * or licensors. Title to the Material remains with Intel Corporation or its
 * suppliers and licensors. The Material may contain trade secrets and
 * proprietary and confidential information of Intel Corporation and its
 * suppliers and licensors, and is protected by worldwide copyright and trade
 * secret laws and treaty provisions. No part of the Material may be used,
 * copied, reproduced, modified, published, uploaded, posted, transmitted,
 * distributed, or disclosed in any way without Intel's prior express written
 * permission. 
 * 
 * No license under any patent, copyright, trade secret or other intellectual
 * property right is granted to or conferred upon you by disclosure or delivery
 * of the Materials, either expressly, by implication, inducement, estoppel or
 * otherwise. Any license under such intellectual property rights must be
 * express and approved by Intel in writing.
 */


#include "lnc_cmdbuf.h"

#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <errno.h>
#include <string.h>
#include <wsbm/wsbm_manager.h>

#include "psb_def.h"
#include "lnc_hostcode.h"
#include "psb_ws_driver.h"
#include "psb_drm.h"

/*
 * Buffer layout:
 *         cmd_base <= cmd_idx < CMD_END() == reloc_base
 *         reloc_base <= reloc_idx < RELOC_END() == (reloc_size)
 */

#define RELOC_END(cmdbuf)     (cmdbuf->cmd_base + cmdbuf->size)

#define CMD_END(cmdbuf)       (cmdbuf->reloc_base)

#define RELOC_SIZE            (0x3000)

#define CMD_SIZE              (0x3000)

#define RELOC_MARGIN          (0x0800)

#define CMD_MARGIN            (0x0400)


#define MAX_CMD_COUNT         12

#define MTX_SEG_SIZE          (0x0800)


/*
 * Create command buffer
 */
VAStatus lnc_cmdbuf_create(
    object_context_p obj_context,    
    psb_driver_data_p driver_data,
    lnc_cmdbuf_p cmdbuf)
{
    context_ENC_p ctx = (context_ENC_p) obj_context->format_data;
    VAStatus vaStatus = VA_STATUS_SUCCESS;
    unsigned int size = CMD_SIZE + RELOC_SIZE;
    
    cmdbuf->size = 0;
    cmdbuf->cmd_base = NULL;
    cmdbuf->cmd_idx = NULL;
    cmdbuf->reloc_base = NULL;
    cmdbuf->reloc_idx = NULL;
    cmdbuf->buffer_refs_count = 0;
    cmdbuf->buffer_refs_allocated = 10;
    cmdbuf->buffer_refs = (psb_buffer_p *) calloc(1, sizeof(psb_buffer_p) * cmdbuf->buffer_refs_allocated);
    if (NULL == cmdbuf->buffer_refs)
    {
        cmdbuf->buffer_refs_allocated = 0;
        vaStatus = VA_STATUS_ERROR_ALLOCATION_FAILED;
    }
    if (VA_STATUS_SUCCESS == vaStatus)
    {
        vaStatus = psb_buffer_create( driver_data, size, psb_bt_cpu_only, &cmdbuf->buf );
        cmdbuf->size = size;
    }

    if (VA_STATUS_SUCCESS != vaStatus) {
        free(cmdbuf->buffer_refs);
        cmdbuf->buffer_refs = NULL;
        cmdbuf->buffer_refs_allocated = 0;
        return vaStatus;
    }
    
    /* create topaz parameter buffer */
    vaStatus = psb_buffer_create(driver_data, ctx->pic_params_size, psb_bt_cpu_vpu, &cmdbuf->pic_params);
    if(VA_STATUS_SUCCESS != vaStatus)
        goto error_out3;

    /* create header buffer */
    vaStatus = psb_buffer_create(driver_data, ctx->header_buffer_size, psb_bt_cpu_vpu, &cmdbuf->header_mem);
    if(VA_STATUS_SUCCESS != vaStatus) 
        goto error_out2;

    /* create slice parameter buffer */
    vaStatus = psb_buffer_create(driver_data, ctx->sliceparam_buffer_size, psb_bt_cpu_vpu, &cmdbuf->slice_params);
    if(VA_STATUS_SUCCESS != vaStatus)
        goto error_out1;

    /* all cmdbuf share one MTX_CURRENT_IN_PARAMS since every MB has a MTX_CURRENT_IN_PARAMS structure
     * and filling this structure for all MB is very time-consuming
     */
    cmdbuf->topaz_in_params_I = &ctx->topaz_in_params_I;
    cmdbuf->topaz_in_params_P = &ctx->topaz_in_params_P;
    cmdbuf->topaz_above_bellow_params = &ctx->topaz_above_bellow_params;
    
    return vaStatus;
    
  error_out1:
    psb_buffer_destroy(&cmdbuf->header_mem);
  error_out2:
    psb_buffer_destroy(&cmdbuf->pic_params);
  error_out3:
    lnc_cmdbuf_destroy(cmdbuf);
    
    return vaStatus;
}

/*
 * Destroy buffer
 */   
void lnc_cmdbuf_destroy( lnc_cmdbuf_p cmdbuf )
{
    if (cmdbuf->size)
    {
        psb_buffer_destroy( &cmdbuf->buf );
        cmdbuf->size = 0;
    }
    if (cmdbuf->buffer_refs_allocated)
    {
        free( cmdbuf->buffer_refs );
        cmdbuf->buffer_refs = NULL;
        cmdbuf->buffer_refs_allocated = 0;
    }

    psb_buffer_destroy( &cmdbuf->pic_params );
    psb_buffer_destroy( &cmdbuf->header_mem );
    psb_buffer_destroy( &cmdbuf->slice_params );
    
}

/*
 * Reset buffer & map
 *
 * Returns 0 on success
 */
int lnc_cmdbuf_reset( lnc_cmdbuf_p cmdbuf )
{
    int ret;

    cmdbuf->cmd_base = NULL;
    cmdbuf->cmd_idx = NULL;
    cmdbuf->reloc_base = NULL;
    cmdbuf->reloc_idx = NULL;

    cmdbuf->buffer_refs_count = 0;
    cmdbuf->cmd_count = 0;
   
    ret = psb_buffer_map( &cmdbuf->buf, &cmdbuf->cmd_base );
    if (ret)
    {
        return ret;
    }

    cmdbuf->cmd_start = cmdbuf->cmd_base;
    cmdbuf->cmd_idx = (uint32_t *) cmdbuf->cmd_base;

    cmdbuf->reloc_base = cmdbuf->cmd_base + CMD_SIZE;
    cmdbuf->reloc_idx = (struct drm_psb_reloc *) cmdbuf->reloc_base;

    /* Add ourselves to the buffer list */
    lnc_cmdbuf_buffer_ref(cmdbuf, &cmdbuf->buf); /* cmd buf == 0 */
    return ret;
}

/*
 * Unmap buffer
 *
 * Returns 0 on success
 */
int lnc_cmdbuf_unmap( lnc_cmdbuf_p cmdbuf )
{
    cmdbuf->cmd_base = NULL;
    cmdbuf->cmd_start = NULL;
    cmdbuf->cmd_idx = NULL;
    cmdbuf->reloc_base = NULL;
    cmdbuf->reloc_idx = NULL;
    cmdbuf->cmd_count = 0;
    psb_buffer_unmap( &cmdbuf->buf );
    return 0;
}


/*
 * Reference an addtional buffer "buf" in the command stream
 * Returns a reference index that can be used to refer to "buf" in
 * relocation records, -1 on error
 */
int lnc_cmdbuf_buffer_ref( lnc_cmdbuf_p cmdbuf, psb_buffer_p buf )
{
    int item_loc = 0;

    while( (item_loc < cmdbuf->buffer_refs_count) && (cmdbuf->buffer_refs[item_loc] != buf) )
    {
        item_loc++;
    }
    if (item_loc == cmdbuf->buffer_refs_count)
    {
        /* Add new entry */
        if (item_loc >= cmdbuf->buffer_refs_allocated)
        {
            /* Allocate more entries */
            int new_size = cmdbuf->buffer_refs_allocated+10;
            psb_buffer_p *new_array;
            new_array = (psb_buffer_p *) calloc(1, sizeof(psb_buffer_p) * new_size);
            if (NULL == new_array)
            {
                return -1; /* Allocation failure */
            }
            memcpy(new_array, cmdbuf->buffer_refs, sizeof(psb_buffer_p) * cmdbuf->buffer_refs_allocated);
            free(cmdbuf->buffer_refs);
            cmdbuf->buffer_refs_allocated = new_size;
            cmdbuf->buffer_refs = new_array;
        }
        cmdbuf->buffer_refs[item_loc] = buf;
        cmdbuf->buffer_refs_count++;
        buf->status = psb_bs_queued;
    }
    return item_loc;
}

/* Creates a relocation record for a DWORD in the mapped "cmdbuf" at address
 * "addr_in_cmdbuf"
 * The relocation is based on the device virtual address of "ref_buffer"
 * "buf_offset" is be added to the device virtual address, and the sum is then
 * right shifted with "align_shift".
 * "mask" determines which bits of the target DWORD will be updated with the so
 * constructed address. The remaining bits will be filled with bits from "background".
 */
void lnc_cmdbuf_add_relocation( lnc_cmdbuf_p cmdbuf,
                                uint32_t *addr_in_dst_buffer,/*addr of dst_buffer for the DWORD*/
                                psb_buffer_p ref_buffer,	
                                uint32_t buf_offset,
                                uint32_t mask,
                                uint32_t background,
                                uint32_t align_shift,
                                uint32_t dst_buffer,
                                uint32_t *start_of_dst_buffer) /*Index of the list refered by cmdbuf->buffer_refs */
{
    struct drm_psb_reloc *reloc = cmdbuf->reloc_idx;
    uint64_t presumed_offset = wsbmBOOffsetHint(ref_buffer->drm_buf);

    reloc->where = addr_in_dst_buffer - start_of_dst_buffer; /* Offset in DWORDs */

    reloc->buffer = lnc_cmdbuf_buffer_ref( cmdbuf, ref_buffer );
    ASSERT( reloc->buffer != -1 );
    
    reloc->reloc_op = PSB_RELOC_OP_OFFSET;
#ifndef VA_EMULATOR
    if (presumed_offset)
    {
        uint32_t new_val =  presumed_offset + buf_offset;

        new_val = ((new_val >> align_shift) << (align_shift << PSB_RELOC_ALSHIFT_SHIFT));
        new_val = (background & ~mask) | (new_val & mask);
        *addr_in_dst_buffer = new_val;
    } 
    else
    {
        *addr_in_dst_buffer = PSB_RELOC_MAGIC;
    }
#else
    /* indicate subscript of relocation buffer */
    *addr_in_dst_buffer = reloc - (struct drm_psb_reloc *)cmdbuf->reloc_base;
#endif    
    reloc->mask = mask;
    reloc->shift = align_shift << PSB_RELOC_ALSHIFT_SHIFT;
    reloc->pre_add = buf_offset;
    reloc->background = background;
    reloc->dst_buffer = dst_buffer;
    cmdbuf->reloc_idx++;

    ASSERT( ((void *) (cmdbuf->reloc_idx)) < RELOC_END(cmdbuf) );
}

/*
 * Advances "obj_context" to the next cmdbuf
 *
 * Returns 0 on success
 */
int lnc_context_get_next_cmdbuf( object_context_p obj_context )
{
    lnc_cmdbuf_p cmdbuf;
    int ret;
    
    if (obj_context->lnc_cmdbuf)
    {
        return 0;
    }

    obj_context->cmdbuf_current++;
    if (obj_context->cmdbuf_current >= LNC_MAX_CMDBUFS_ENCODE)
    {
        obj_context->cmdbuf_current = 0;
    }
    
    cmdbuf = obj_context->lnc_cmdbuf_list[obj_context->cmdbuf_current];
    ret = lnc_cmdbuf_reset( cmdbuf );
    if (!ret)
    {
        /* Success */
        obj_context->lnc_cmdbuf = cmdbuf;
    }

    /* added pic_params/slice_params into ref, so the index is 1/2 */
    lnc_cmdbuf_buffer_ref(cmdbuf,&cmdbuf->pic_params);
    lnc_cmdbuf_buffer_ref(cmdbuf,&cmdbuf->slice_params);
    
    return ret;
}

/*
 * This is the user-space do-it-all interface to the drm cmdbuf ioctl.
 * It allows different buffers as command- and reloc buffer. A list of
 * cliprects to apply and whether to copy the clipRect content to all
 * scanout buffers (damage = 1).
 */
/*
 * Don't add debug statements in this function, it gets called with the
 * DRM lock held and output to an X terminal can cause X to deadlock
 */
static int
lncDRMCmdBuf(int fd, int ioctl_offset, psb_buffer_p *buffer_list, int buffer_count, unsigned cmdBufHandle,
             unsigned cmdBufOffset, unsigned cmdBufSize,
             unsigned relocBufHandle, unsigned relocBufOffset,
             unsigned numRelocs, drm_clip_rect_t * clipRects, int damage,
             unsigned engine, unsigned fence_flags, struct psb_ttm_fence_rep *fence_rep)
{
    drm_psb_cmdbuf_arg_t ca;
    struct psb_validate_arg *arg_list;
    int i;
    int ret;
    uint64_t mask = PSB_GPU_ACCESS_MASK;

    arg_list = (struct psb_validate_arg *) calloc(1, sizeof(struct psb_validate_arg)*buffer_count);
    if (arg_list == NULL) {
        psb__error_message("Allocate memory failed\n");
        return -ENOMEM;
    }

    for( i = 0; i < buffer_count; i++)
    {
        struct psb_validate_arg *arg = &(arg_list[i]);
        struct psb_validate_req *req = &arg->d.req;

        memset(arg, 0, sizeof(*arg));
        req->next = (unsigned long) &(arg_list[i+1]);
            
        req->buffer_handle = wsbmKBufHandle(wsbmKBuf(buffer_list[i]->drm_buf));
        req->group = 0;
        req->set_flags = (PSB_GPU_ACCESS_READ | PSB_GPU_ACCESS_WRITE) & mask;
        req->clear_flags = (~(PSB_GPU_ACCESS_READ | PSB_GPU_ACCESS_WRITE)) & mask;
#if 1
        req->presumed_gpu_offset = (uint64_t)wsbmBOOffsetHint(buffer_list[i]->drm_buf);
        req->presumed_flags = PSB_USE_PRESUMED;
	if ((req->presumed_gpu_offset >> 28) & 0x1) {
		psb__error_message("buffer is at the address topaz can not access\n");
		ret = -1;
		goto out;
	}
#else
	req->presumed_flags = 0;
#endif
    }
    arg_list[buffer_count-1].d.req.next = 0;
	
    memset(&ca,0,sizeof(ca));

    ca.buffer_list = (uint64_t)((unsigned long)arg_list);
    ca.clip_rects = (uint64_t)((unsigned long)clipRects);
    ca.cmdbuf_handle = cmdBufHandle;
    ca.cmdbuf_offset = cmdBufOffset;
    ca.cmdbuf_size = cmdBufSize;
    ca.reloc_handle = relocBufHandle;
    ca.reloc_offset = relocBufOffset;
    ca.num_relocs = numRelocs;
    ca.engine = engine;
    ca.fence_flags = fence_flags;
    ca.fence_arg = (uint64_t) ((unsigned long)fence_rep);
    ca.damage = damage;


    do {
        ret = drmCommandWrite(fd, ioctl_offset, &ca, sizeof(ca));
    } while (ret == EAGAIN);

    if (ret)
        goto out;

    for( i = 0; i < buffer_count; i++)
    {
        struct psb_validate_arg *arg = &(arg_list[i]);
        struct psb_validate_rep *rep = &arg->d.rep;

        if (!arg->handled) {
            ret = -EFAULT;
            goto out;
        }
        if (arg->ret != 0)
        {
            ret = arg->ret;
            goto out;
        }
        wsbmUpdateKBuf(wsbmKBuf(buffer_list[i]->drm_buf),
                       rep->gpu_offset, rep->placement, rep->fence_type_mask);
    }
  out:
    free(arg_list);
    for( i = 0; i < buffer_count; i++)
    {
        /*
         * Buffer no longer queued in userspace
         */
        switch (buffer_list[i]->status)
        {
        case psb_bs_queued:
            buffer_list[i]->status = psb_bs_ready;
            break;

        case psb_bs_abandoned:
            psb_buffer_destroy(buffer_list[i]);
            free(buffer_list[i]);
            break;

        default:
            /* Not supposed to happen */
            ASSERT(0);
        }
    }
	
    return ret;
}

#if 0
static struct _WsbmFenceObject *
lnc_fence_wait(psb_driver_data_p driver_data,
               struct psb_ttm_fence_rep *fence_rep, int *status)

{
    struct _WsbmFenceObject *fence = NULL;
    int ret = -1;
    
    /* copy fence information */
    if (fence_rep->error != 0) {
        psb__error_message("drm failed to create a fence"
                           " and has idled the HW\n");
        DEBUG_FAILURE_RET;
        return NULL;
    }

    fence = wsbmFenceCreate(driver_data->fence_mgr, fence_rep->fence_class,
                            fence_rep->fence_type,
                            (void *)fence_rep->handle,
                            0);
    if (fence) 
        *status = wsbmFenceFinish(fence,fence_rep->fence_type,0);
    
    return fence;
}
#endif

/*
 * Submits the current cmdbuf
 *
 * Returns 0 on success
 */
int lnc_context_submit_cmdbuf( object_context_p obj_context )
{
    
    return 0;
}

/*
 * FrameSkip is only meaningful for RC enabled mode
 * Topaz raises this flag after surface N encoding is finished (vaSyncSurface gets back)
 * then for the next encode surface N+1 (ctx->src_surface) frameskip flag is cleared in vaBeginPicuture
 * and is always set in vaEndPicture:lnc_PatchRCMode
 * vaQuerySurfaceStatus is supposed only to be called after vaEndPicture/vaSyncSurface,
 * The caller should ensure the surface pertains to an encode context
 */
int lnc_surface_get_frameskip(object_context_p obj_context, psb_surface_p surface, int *frame_skip)
{
    struct drm_lnc_video_getparam_arg arg;
    unsigned long temp;
    int ret = 0;
    psb_driver_data_p driver_data = obj_context->driver_data;
    context_ENC_p ctx = (context_ENC_p) obj_context->format_data;

    if (ctx->sRCParams.RCEnable == 0)
	return 0;

    /* bit31 indicate if frameskip is already settled, it is used to record the frame skip flag for old surfaces
     * because current FRAMESKIP in hardware can't be applied to the old surfaces
     * bit31 is cleared when the surface is used as encode render target or reference/reconstrucure target
     */
    if (GET_SURFACE_INFO_skipped_flag(surface) & SURFACE_INFO_SKIP_FLAG_SETTLED) {
        *frame_skip = GET_SURFACE_INFO_skipped_flag(surface) & 1;
        return 0;
    }

    /* not settled, we get it from current HW FRAMESKIP flag */ 
    arg.key = LNC_VIDEO_FRAME_SKIP;
    arg.value = (uint64_t)((unsigned long) &temp);
    ret = drmCommandWriteRead(driver_data->drm_fd, driver_data->getParamIoctlOffset,
                              &arg, sizeof(arg));
    if (ret==0) {
        SET_SURFACE_INFO_skipped_flag(surface, temp);
        *frame_skip = temp;
        if (temp == 1)
            psb__information_message("Detected a skipped frame for encode\n");
    }
    
    return ret;
}


/*
 * Flushes all cmdbufs
 */
int lnc_context_flush_cmdbuf( object_context_p obj_context )
{
    lnc_cmdbuf_p cmdbuf = obj_context->lnc_cmdbuf;
    psb_driver_data_p driver_data = obj_context->driver_data;
    unsigned int fence_flags;
    struct psb_ttm_fence_rep fence_rep;
    unsigned int reloc_offset;
    unsigned int num_relocs;
    int ret;
    unsigned int cmdbuffer_size = (void *) cmdbuf->cmd_idx - cmdbuf->cmd_start; /* In bytes */

    ASSERT(cmdbuffer_size < CMD_SIZE);
    ASSERT((void *) cmdbuf->cmd_idx < CMD_END(cmdbuf));
    /* LOCK */
    ret = LOCK_HARDWARE(driver_data);
    if (ret)
    {
        UNLOCK_HARDWARE(driver_data);
        DEBUG_FAILURE_RET;
        return ret;
    }

    /* Now calculate the total number of relocations */
    reloc_offset = cmdbuf->reloc_base - cmdbuf->cmd_base;
    num_relocs = (((void *) cmdbuf->reloc_idx) - cmdbuf->reloc_base) / sizeof(struct drm_psb_reloc);
    
    lnc_cmdbuf_unmap( cmdbuf );
    
    ASSERT( NULL == cmdbuf->reloc_base );

#ifdef DEBUG_TRACE
    fence_flags = 0;
#else
    fence_flags = DRM_PSB_FENCE_NO_USER;
#endif    
    
#ifndef LNC_ENGINE_ENCODE
#define LNC_ENGINE_ENCODE  5
#endif

    wsbmWriteLockKernelBO();
    ret = lncDRMCmdBuf(driver_data->drm_fd, driver_data->execIoctlOffset,
                       cmdbuf->buffer_refs, cmdbuf->buffer_refs_count,wsbmKBufHandle(wsbmKBuf(cmdbuf->buf.drm_buf)),
                       0, cmdbuffer_size,/*unsigned cmdBufSize*/
                       wsbmKBufHandle(wsbmKBuf(cmdbuf->buf.drm_buf)),reloc_offset, num_relocs,
                       0 /* clipRects */, 0, LNC_ENGINE_ENCODE, fence_flags, &fence_rep);
    wsbmWriteUnlockKernelBO();    
    UNLOCK_HARDWARE(driver_data);
        
    if (ret)
    {
        obj_context->lnc_cmdbuf = NULL;

        DEBUG_FAILURE_RET;
        return ret;
    }

#if 0 /*DEBUG_TRACE*/
    int status = -1;
    struct _WsbmFenceObject *fence=NULL;

    fence = lnc_fence_wait(driver_data, &fence_rep, &status);
    psb__information_message("psb_fence_wait returns: %d (fence=0x%08x)\n",status, fence);
    
    if (fence)
	wsbmFenceUnreference(fence);

#endif

    obj_context->lnc_cmdbuf = NULL;
    
    return 0;
}