linux-zen-server/drivers/media/platform/ti/omap/omap_vout_vrfb.c

420 lines
11 KiB
C
Raw Permalink Normal View History

2023-08-30 17:53:23 +02:00
/*
* omap_vout_vrfb.c
*
* Copyright (C) 2010 Texas Instruments.
*
* This file is licensed under the terms of the GNU General Public License
* version 2. This program is licensed "as is" without any warranty of any
* kind, whether express or implied.
*
*/
#include <linux/sched.h>
#include <linux/platform_device.h>
#include <linux/videodev2.h>
#include <linux/slab.h>
#include <media/v4l2-device.h>
#include <video/omapvrfb.h>
#include "omap_voutdef.h"
#include "omap_voutlib.h"
#include "omap_vout_vrfb.h"
#define OMAP_DMA_NO_DEVICE 0
/*
* Function for allocating video buffers
*/
static int omap_vout_allocate_vrfb_buffers(struct omap_vout_device *vout,
unsigned int *count, int startindex)
{
int i, j;
for (i = 0; i < *count; i++) {
if (!vout->smsshado_virt_addr[i]) {
vout->smsshado_virt_addr[i] =
omap_vout_alloc_buffer(vout->smsshado_size,
&vout->smsshado_phy_addr[i]);
}
if (!vout->smsshado_virt_addr[i] && startindex != -1) {
if (vout->vq.memory == V4L2_MEMORY_MMAP && i >= startindex)
break;
}
if (!vout->smsshado_virt_addr[i]) {
for (j = 0; j < i; j++) {
omap_vout_free_buffer(
vout->smsshado_virt_addr[j],
vout->smsshado_size);
vout->smsshado_virt_addr[j] = 0;
vout->smsshado_phy_addr[j] = 0;
}
*count = 0;
return -ENOMEM;
}
memset((void *)(long)vout->smsshado_virt_addr[i], 0,
vout->smsshado_size);
}
return 0;
}
/*
* Wakes up the application once the DMA transfer to VRFB space is completed.
*/
static void omap_vout_vrfb_dma_tx_callback(void *data)
{
struct vid_vrfb_dma *t = (struct vid_vrfb_dma *) data;
t->tx_status = 1;
wake_up_interruptible(&t->wait);
}
/*
* Free VRFB buffers
*/
void omap_vout_free_vrfb_buffers(struct omap_vout_device *vout)
{
int j;
for (j = 0; j < VRFB_NUM_BUFS; j++) {
if (vout->smsshado_virt_addr[j]) {
omap_vout_free_buffer(vout->smsshado_virt_addr[j],
vout->smsshado_size);
vout->smsshado_virt_addr[j] = 0;
vout->smsshado_phy_addr[j] = 0;
}
}
}
int omap_vout_setup_vrfb_bufs(struct platform_device *pdev, int vid_num,
bool static_vrfb_allocation)
{
int ret = 0, i, j;
struct omap_vout_device *vout;
struct video_device *vfd;
dma_cap_mask_t mask;
int image_width, image_height;
int vrfb_num_bufs = VRFB_NUM_BUFS;
struct v4l2_device *v4l2_dev = platform_get_drvdata(pdev);
struct omap2video_device *vid_dev =
container_of(v4l2_dev, struct omap2video_device, v4l2_dev);
vout = vid_dev->vouts[vid_num];
vfd = vout->vfd;
for (i = 0; i < VRFB_NUM_BUFS; i++) {
if (omap_vrfb_request_ctx(&vout->vrfb_context[i])) {
dev_info(&pdev->dev, ": VRFB allocation failed\n");
for (j = 0; j < i; j++)
omap_vrfb_release_ctx(&vout->vrfb_context[j]);
return -ENOMEM;
}
}
/* Calculate VRFB memory size */
/* allocate for worst case size */
image_width = VID_MAX_WIDTH / TILE_SIZE;
if (VID_MAX_WIDTH % TILE_SIZE)
image_width++;
image_width = image_width * TILE_SIZE;
image_height = VID_MAX_HEIGHT / TILE_SIZE;
if (VID_MAX_HEIGHT % TILE_SIZE)
image_height++;
image_height = image_height * TILE_SIZE;
vout->smsshado_size = PAGE_ALIGN(image_width * image_height * 2 * 2);
/*
* Request and Initialize DMA, for DMA based VRFB transfer
*/
dma_cap_zero(mask);
dma_cap_set(DMA_INTERLEAVE, mask);
vout->vrfb_dma_tx.chan = dma_request_chan_by_mask(&mask);
if (IS_ERR(vout->vrfb_dma_tx.chan)) {
vout->vrfb_dma_tx.req_status = DMA_CHAN_NOT_ALLOTED;
} else {
size_t xt_size = sizeof(struct dma_interleaved_template) +
sizeof(struct data_chunk);
vout->vrfb_dma_tx.xt = kzalloc(xt_size, GFP_KERNEL);
if (!vout->vrfb_dma_tx.xt) {
dma_release_channel(vout->vrfb_dma_tx.chan);
vout->vrfb_dma_tx.req_status = DMA_CHAN_NOT_ALLOTED;
}
}
if (vout->vrfb_dma_tx.req_status == DMA_CHAN_NOT_ALLOTED)
dev_info(&pdev->dev,
": failed to allocate DMA Channel for video%d\n",
vfd->minor);
init_waitqueue_head(&vout->vrfb_dma_tx.wait);
/*
* statically allocated the VRFB buffer is done through
* command line arguments
*/
if (static_vrfb_allocation) {
if (omap_vout_allocate_vrfb_buffers(vout, &vrfb_num_bufs, -1)) {
ret = -ENOMEM;
goto release_vrfb_ctx;
}
vout->vrfb_static_allocation = true;
}
return 0;
release_vrfb_ctx:
for (j = 0; j < VRFB_NUM_BUFS; j++)
omap_vrfb_release_ctx(&vout->vrfb_context[j]);
return ret;
}
/*
* Release the VRFB context once the module exits
*/
void omap_vout_release_vrfb(struct omap_vout_device *vout)
{
int i;
for (i = 0; i < VRFB_NUM_BUFS; i++)
omap_vrfb_release_ctx(&vout->vrfb_context[i]);
if (vout->vrfb_dma_tx.req_status == DMA_CHAN_ALLOTED) {
vout->vrfb_dma_tx.req_status = DMA_CHAN_NOT_ALLOTED;
kfree(vout->vrfb_dma_tx.xt);
dmaengine_terminate_sync(vout->vrfb_dma_tx.chan);
dma_release_channel(vout->vrfb_dma_tx.chan);
}
}
/*
* Allocate the buffers for the VRFB space. Data is copied from V4L2
* buffers to the VRFB buffers using the DMA engine.
*/
int omap_vout_vrfb_buffer_setup(struct omap_vout_device *vout,
unsigned int *count, unsigned int startindex)
{
int i;
bool yuv_mode;
if (!is_rotation_enabled(vout))
return 0;
/* If rotation is enabled, allocate memory for VRFB space also */
*count = *count > VRFB_NUM_BUFS ? VRFB_NUM_BUFS : *count;
/* Allocate the VRFB buffers only if the buffers are not
* allocated during init time.
*/
if (!vout->vrfb_static_allocation)
if (omap_vout_allocate_vrfb_buffers(vout, count, startindex))
return -ENOMEM;
if (vout->dss_mode == OMAP_DSS_COLOR_YUV2 ||
vout->dss_mode == OMAP_DSS_COLOR_UYVY)
yuv_mode = true;
else
yuv_mode = false;
for (i = 0; i < *count; i++)
omap_vrfb_setup(&vout->vrfb_context[i],
vout->smsshado_phy_addr[i], vout->pix.width,
vout->pix.height, vout->bpp, yuv_mode);
return 0;
}
int omap_vout_prepare_vrfb(struct omap_vout_device *vout,
struct vb2_buffer *vb)
{
struct dma_async_tx_descriptor *tx;
enum dma_ctrl_flags flags = DMA_PREP_INTERRUPT | DMA_CTRL_ACK;
struct dma_chan *chan = vout->vrfb_dma_tx.chan;
struct dma_interleaved_template *xt = vout->vrfb_dma_tx.xt;
dma_cookie_t cookie;
dma_addr_t buf_phy_addr = vb2_dma_contig_plane_dma_addr(vb, 0);
enum dma_status status;
enum dss_rotation rotation;
size_t dst_icg;
u32 pixsize;
if (!is_rotation_enabled(vout))
return 0;
/* If rotation is enabled, copy input buffer into VRFB
* memory space using DMA. We are copying input buffer
* into VRFB memory space of desired angle and DSS will
* read image VRFB memory for 0 degree angle
*/
pixsize = vout->bpp * vout->vrfb_bpp;
dst_icg = MAX_PIXELS_PER_LINE * pixsize - vout->pix.width * vout->bpp;
xt->src_start = buf_phy_addr;
xt->dst_start = vout->vrfb_context[vb->index].paddr[0];
xt->numf = vout->pix.height;
xt->frame_size = 1;
xt->sgl[0].size = vout->pix.width * vout->bpp;
xt->sgl[0].icg = dst_icg;
xt->dir = DMA_MEM_TO_MEM;
xt->src_sgl = false;
xt->src_inc = true;
xt->dst_sgl = true;
xt->dst_inc = true;
tx = dmaengine_prep_interleaved_dma(chan, xt, flags);
if (tx == NULL) {
pr_err("%s: DMA interleaved prep error\n", __func__);
return -EINVAL;
}
tx->callback = omap_vout_vrfb_dma_tx_callback;
tx->callback_param = &vout->vrfb_dma_tx;
cookie = dmaengine_submit(tx);
if (dma_submit_error(cookie)) {
pr_err("%s: dmaengine_submit failed (%d)\n", __func__, cookie);
return -EINVAL;
}
vout->vrfb_dma_tx.tx_status = 0;
dma_async_issue_pending(chan);
wait_event_interruptible_timeout(vout->vrfb_dma_tx.wait,
vout->vrfb_dma_tx.tx_status == 1,
VRFB_TX_TIMEOUT);
status = dma_async_is_tx_complete(chan, cookie, NULL, NULL);
if (vout->vrfb_dma_tx.tx_status == 0) {
pr_err("%s: Timeout while waiting for DMA\n", __func__);
dmaengine_terminate_sync(chan);
return -EINVAL;
} else if (status != DMA_COMPLETE) {
pr_err("%s: DMA completion %s status\n", __func__,
status == DMA_ERROR ? "error" : "busy");
dmaengine_terminate_sync(chan);
return -EINVAL;
}
/* Store buffers physical address into an array. Addresses
* from this array will be used to configure DSS */
rotation = calc_rotation(vout);
vout->queued_buf_addr[vb->index] =
vout->vrfb_context[vb->index].paddr[rotation];
return 0;
}
/*
* Calculate the buffer offsets from which the streaming should
* start. This offset calculation is mainly required because of
* the VRFB 32 pixels alignment with rotation.
*/
void omap_vout_calculate_vrfb_offset(struct omap_vout_device *vout)
{
enum dss_rotation rotation;
bool mirroring = vout->mirror;
struct v4l2_rect *crop = &vout->crop;
struct v4l2_pix_format *pix = &vout->pix;
int *cropped_offset = &vout->cropped_offset;
int vr_ps = 1, ps = 2, temp_ps = 2;
int offset = 0, ctop = 0, cleft = 0, line_length = 0;
rotation = calc_rotation(vout);
if (V4L2_PIX_FMT_YUYV == pix->pixelformat ||
V4L2_PIX_FMT_UYVY == pix->pixelformat) {
if (is_rotation_enabled(vout)) {
/*
* ps - Actual pixel size for YUYV/UYVY for
* VRFB/Mirroring is 4 bytes
* vr_ps - Virtually pixel size for YUYV/UYVY is
* 2 bytes
*/
ps = 4;
vr_ps = 2;
} else {
ps = 2; /* otherwise the pixel size is 2 byte */
}
} else if (V4L2_PIX_FMT_RGB32 == pix->pixelformat) {
ps = 4;
} else if (V4L2_PIX_FMT_RGB24 == pix->pixelformat) {
ps = 3;
}
vout->ps = ps;
vout->vr_ps = vr_ps;
if (is_rotation_enabled(vout)) {
line_length = MAX_PIXELS_PER_LINE;
ctop = (pix->height - crop->height) - crop->top;
cleft = (pix->width - crop->width) - crop->left;
} else {
line_length = pix->width;
}
vout->line_length = line_length;
switch (rotation) {
case dss_rotation_90_degree:
offset = vout->vrfb_context[0].yoffset *
vout->vrfb_context[0].bytespp;
temp_ps = ps / vr_ps;
if (!mirroring) {
*cropped_offset = offset + line_length *
temp_ps * cleft + crop->top * temp_ps;
} else {
*cropped_offset = offset + line_length * temp_ps *
cleft + crop->top * temp_ps + (line_length *
((crop->width / (vr_ps)) - 1) * ps);
}
break;
case dss_rotation_180_degree:
offset = ((MAX_PIXELS_PER_LINE * vout->vrfb_context[0].yoffset *
vout->vrfb_context[0].bytespp) +
(vout->vrfb_context[0].xoffset *
vout->vrfb_context[0].bytespp));
if (!mirroring) {
*cropped_offset = offset + (line_length * ps * ctop) +
(cleft / vr_ps) * ps;
} else {
*cropped_offset = offset + (line_length * ps * ctop) +
(cleft / vr_ps) * ps + (line_length *
(crop->height - 1) * ps);
}
break;
case dss_rotation_270_degree:
offset = MAX_PIXELS_PER_LINE * vout->vrfb_context[0].xoffset *
vout->vrfb_context[0].bytespp;
temp_ps = ps / vr_ps;
if (!mirroring) {
*cropped_offset = offset + line_length *
temp_ps * crop->left + ctop * ps;
} else {
*cropped_offset = offset + line_length *
temp_ps * crop->left + ctop * ps +
(line_length * ((crop->width / vr_ps) - 1) *
ps);
}
break;
case dss_rotation_0_degree:
if (!mirroring) {
*cropped_offset = (line_length * ps) *
crop->top + (crop->left / vr_ps) * ps;
} else {
*cropped_offset = (line_length * ps) *
crop->top + (crop->left / vr_ps) * ps +
(line_length * (crop->height - 1) * ps);
}
break;
default:
*cropped_offset = (line_length * ps * crop->top) /
vr_ps + (crop->left * ps) / vr_ps +
((crop->width / vr_ps) - 1) * ps;
break;
}
}