[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <7967c1df-ec41-e5ee-021d-d671021bc779@collabora.com>
Date: Mon, 18 Jul 2022 11:51:19 +0200
From: AngeloGioacchino Del Regno
<angelogioacchino.delregno@...labora.com>
To: Irui Wang <irui.wang@...iatek.com>,
Hans Verkuil <hverkuil-cisco@...all.nl>,
Mauro Carvalho Chehab <mchehab@...nel.org>,
Rob Herring <robh+dt@...nel.org>,
Matthias Brugger <matthias.bgg@...il.com>,
Tzung-Bi Shih <tzungbi@...omium.org>,
Tiffany Lin <tiffany.lin@...iatek.com>,
Andrew-CT Chen <andrew-ct.chen@...iatek.com>
Cc: Yong Wu <yong.wu@...iatek.com>,
Maoguang Meng <maoguang.meng@...iatek.com>,
Longfei Wang <longfei.wang@...iatek.com>,
Yunfei Dong <yunfei.dong@...iatek.com>,
linux-media@...r.kernel.org, devicetree@...r.kernel.org,
linux-kernel@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
srv_heupstream@...iatek.com, linux-mediatek@...ts.infradead.org,
Project_Global_Chrome_Upstream_Group@...iatek.com
Subject: Re: [PATCH 1/5] media: mediatek: vcodec: Add encoder driver support
for 34-bit iova
Il 16/07/22 11:38, Irui Wang ha scritto:
> Encoder driver got iova from IOMMU is 34-bit, for example:
>
> Here is the sample code:
> encoder input frame buffer dma address is:
> frm_buf =
> vb2_dma_contig_plane_dma_addr(&vb2_v4l2_buffer->vb2_buf, 0);
> the value of frm_buf is 0x1_ff30_0000.
>
> encoder driver got the frm_buf and send the iova to SCP firmware
> through SCP IPI message, then write to encoder hardware in SCP.
> The iova is stored in IPI message as uint32_t data type, so the
> value will be truncated from *0x1_ff30_0000* to *0xff30_0000*,
> and then *0xff30_0000* will be written to encoder hardware, but
> IOMMU will help to add the high *0x1_* bit back, so IOMMU can
> translate the iova to PA correctly, encoder hardware can access
> the correct memory for encoding.
> Another reason to do this is the encoder hardware can't access
> the 34-bit iova, IOMMU will help to add the remaining high bits
> of iova. But for mt8188, encoder hardware can access 34-bit iova
> directly, and encoder driver need write all 34-bit iova because
> IOMMU can't help driver do this if the hardware support access
> 34-bit iova.
> For the reasons above, this patch is added to support transfer
> 34-bit iova between kernel and SCP encoder driver. Use uint64_t
> data type to store the iova, for compatibility with old chipsets,
> add some new struct definitions for 34-bit.
>
> Signed-off-by: Irui Wang <irui.wang@...iatek.com>
> ---
> .../platform/mediatek/vcodec/mtk_vcodec_drv.h | 3 +
> .../mediatek/vcodec/venc/venc_h264_if.c | 200 +++++++++++++++---
> .../platform/mediatek/vcodec/venc_ipi_msg.h | 24 +++
> .../platform/mediatek/vcodec/venc_vpu_if.c | 34 ++-
> 4 files changed, 227 insertions(+), 34 deletions(-)
>
> diff --git a/drivers/media/platform/mediatek/vcodec/mtk_vcodec_drv.h b/drivers/media/platform/mediatek/vcodec/mtk_vcodec_drv.h
> index ef4584a46417..ab80e1b1979e 100644
> --- a/drivers/media/platform/mediatek/vcodec/mtk_vcodec_drv.h
> +++ b/drivers/media/platform/mediatek/vcodec/mtk_vcodec_drv.h
> @@ -401,6 +401,7 @@ struct mtk_vcodec_dec_pdata {
> * @output_formats: array of supported output formats
> * @num_output_formats: number of entries in output_formats
> * @core_id: stand for h264 or vp8 encode index
> + * @is_34bit: whether the encoder uses 34bit iova
> */
> struct mtk_vcodec_enc_pdata {
> bool uses_ext;
> @@ -411,9 +412,11 @@ struct mtk_vcodec_enc_pdata {
> const struct mtk_video_fmt *output_formats;
> size_t num_output_formats;
> int core_id;
> + bool is_34bit;
> };
>
> #define MTK_ENC_CTX_IS_EXT(ctx) ((ctx)->dev->venc_pdata->uses_ext)
> +#define MTK_ENC_IOVA_IS_34BIT(ctx) ((ctx)->dev->venc_pdata->is_34bit)
>
> /**
> * struct mtk_vcodec_dev - driver data
> diff --git a/drivers/media/platform/mediatek/vcodec/venc/venc_h264_if.c b/drivers/media/platform/mediatek/vcodec/venc/venc_h264_if.c
> index 4d9b8798dffe..3a5af6cca040 100644
> --- a/drivers/media/platform/mediatek/vcodec/venc/venc_h264_if.c
> +++ b/drivers/media/platform/mediatek/vcodec/venc/venc_h264_if.c
> @@ -127,6 +127,72 @@ struct venc_h264_vsi {
> struct venc_h264_vpu_buf work_bufs[VENC_H264_VPU_WORK_BUF_MAX];
> };
>
> +/**
> + * struct venc_h264_vpu_config_ext - Structure for h264 encoder configuration
> + * AP-W/R : AP is writer/reader on this item
> + * VPU-W/R: VPU is write/reader on this item
> + * @input_fourcc: input fourcc
> + * @bitrate: target bitrate (in bps)
> + * @pic_w: picture width. Picture size is visible stream resolution, in pixels,
> + * to be used for display purposes; must be smaller or equal to buffer
> + * size.
> + * @pic_h: picture height
> + * @buf_w: buffer width. Buffer size is stream resolution in pixels aligned to
> + * hardware requirements.
> + * @buf_h: buffer height
> + * @gop_size: group of picture size (idr frame)
> + * @intra_period: intra frame period
> + * @framerate: frame rate in fps
> + * @profile: as specified in standard
> + * @level: as specified in standard
> + * @wfd: WFD mode 1:on, 0:off
> + * @max_qp: max quant parameter
> + * @min_qp: min quant parameter
> + * @reserved: reserved configs
> + */
> +struct venc_h264_vpu_config_ext {
> + u32 input_fourcc;
> + u32 bitrate;
> + u32 pic_w;
> + u32 pic_h;
> + u32 buf_w;
> + u32 buf_h;
> + u32 gop_size;
> + u32 intra_period;
> + u32 framerate;
> + u32 profile;
> + u32 level;
> + u32 wfd;
> + u32 max_qp;
> + u32 min_qp;
> + u32 reserved[8];
> +};
> +
> +/**
> + * struct venc_h264_vpu_buf_34 - Structure for 34 bit buffer information
> + * AP-W/R : AP is writer/reader on this item
> + * VPU-W/R: VPU is write/reader on this item
> + * @iova: 34 bit IO virtual address
> + * @vpua: VPU side memory addr which is used by RC_CODE
> + * @size: buffer size (in bytes)
> + */
> +struct venc_h264_vpu_buf_34 {
> + u64 iova;
> + u32 vpua;
> + u32 size;
> +};
> +
> +/**
> + * struct venc_h264_vsi_64 - Structure for VPU driver control and info share
Typo here -------------- ^^^^
> + * Used for 64 bit iova sharing
> + * @config: h264 encoder configuration
> + * @work_bufs: working buffer information in VPU side
> + */
> +struct venc_h264_vsi_34 {
> + struct venc_h264_vpu_config_ext config;
> + struct venc_h264_vpu_buf_34 work_bufs[VENC_H264_VPU_WORK_BUF_MAX];
> +};
> +
> /*
> * struct venc_h264_inst - h264 encoder AP driver instance
> * @hw_base: h264 encoder hardware register base
> @@ -140,6 +206,8 @@ struct venc_h264_vsi {
..snip..
> diff --git a/drivers/media/platform/mediatek/vcodec/venc_vpu_if.c b/drivers/media/platform/mediatek/vcodec/venc_vpu_if.c
> index d3570c4c177d..25c1b13559c9 100644
> --- a/drivers/media/platform/mediatek/vcodec/venc_vpu_if.c
> +++ b/drivers/media/platform/mediatek/vcodec/venc_vpu_if.c
> @@ -228,17 +228,28 @@ int vpu_enc_encode(struct venc_vpu_inst *vpu, unsigned int bs_mode,
> struct venc_frame_info *frame_info)
> {
That's practically 75% flow differences (or more, actually)... and there's going
to always be a useless memzero, because a platform will always use either 34, or 32
bits code.
At this point I think that for both performance and readability purposes, you
should simply create another function.
Perhaps something like
static int vpu_enc_encode_32bits(struct venc_vpu_inst *vpu, unsigned int bs_mode,
struct venc_frm_buf *frm_buf,
struct mtk_vcodec_mem *bs_buf,
struct venc_frame_info *frame_info)
{
..... function .....
}
static int vpu_enc_encode_34bits(struct venc_vpu_inst *vpu, unsigned int bs_mode,
struct venc_frm_buf *frm_buf,
struct mtk_vcodec_mem *bs_buf,
struct venc_frame_info *frame_info)
{
...... function ......
}
int vpu_enc_encode(struct venc_vpu_inst *vpu, unsigned int bs_mode,
struct venc_frm_buf *frm_buf,
struct mtk_vcodec_mem *bs_buf,
struct venc_frame_info *frame_info)
{
int ret;
if (MTK_ENC_IOVA_IS_34BIT(vpu->ctx))
ret = vpu_enc_encode_34bits(......);
else
ret = vpu_enc_encode_32bits(....);
if (ret)
return ret;
mtk_vcodec_debug(vpu, "bs_mode %d state %d size %d key_frm %d <-",
bs_mode, vpu->state, vpu->bs_size, vpu->is_key_frm);
return 0;
}
> const bool is_ext = MTK_ENC_CTX_IS_EXT(vpu->ctx);
> + const bool is_34bit = MTK_ENC_IOVA_IS_34BIT(vpu->ctx);
> +
> size_t msg_size = is_ext ?
> sizeof(struct venc_ap_ipi_msg_enc_ext) :
> sizeof(struct venc_ap_ipi_msg_enc);
> + int status;
> struct venc_ap_ipi_msg_enc_ext out;
> + struct venc_ap_ipi_msg_enc_ext_34 out_34;
>
> mtk_vcodec_debug(vpu, "bs_mode %d ->", bs_mode);
>
> memset(&out, 0, sizeof(out));
> + memset(&out_34, 0, sizeof(out_34));
> +
> out.base.msg_id = AP_IPIMSG_ENC_ENCODE;
> out.base.vpu_inst_addr = vpu->inst_addr;
> out.base.bs_mode = bs_mode;
> +
> + out_34.msg_id = AP_IPIMSG_ENC_ENCODE;
> + out_34.vpu_inst_addr = vpu->inst_addr;
> + out_34.bs_mode = bs_mode;
> +
> if (frm_buf) {
> if ((frm_buf->fb_addr[0].dma_addr % 16 == 0) &&
> (frm_buf->fb_addr[1].dma_addr % 16 == 0) &&
> @@ -246,6 +257,10 @@ int vpu_enc_encode(struct venc_vpu_inst *vpu, unsigned int bs_mode,
> out.base.input_addr[0] = frm_buf->fb_addr[0].dma_addr;
> out.base.input_addr[1] = frm_buf->fb_addr[1].dma_addr;
> out.base.input_addr[2] = frm_buf->fb_addr[2].dma_addr;
> +
> + out_34.input_addr[0] = frm_buf->fb_addr[0].dma_addr;
> + out_34.input_addr[1] = frm_buf->fb_addr[1].dma_addr;
> + out_34.input_addr[2] = frm_buf->fb_addr[2].dma_addr;
> } else {
> mtk_vcodec_err(vpu, "dma_addr not align to 16");
> return -EINVAL;
> @@ -254,14 +269,31 @@ int vpu_enc_encode(struct venc_vpu_inst *vpu, unsigned int bs_mode,
> if (bs_buf) {
> out.base.bs_addr = bs_buf->dma_addr;
> out.base.bs_size = bs_buf->size;
> +
> + out_34.bs_addr = bs_buf->dma_addr;
> + out_34.bs_size = bs_buf->size;
> }
> +
> if (is_ext && frame_info) {
> out.data_item = 3;
> out.data[0] = frame_info->frm_count;
> out.data[1] = frame_info->skip_frm_count;
> out.data[2] = frame_info->frm_type;
> +
> + out_34.data_item = 3;
> + out_34.data[0] = frame_info->frm_count;
> + out_34.data[1] = frame_info->skip_frm_count;
> + out_34.data[2] = frame_info->frm_type;
> }
> - if (vpu_enc_send_msg(vpu, &out, msg_size)) {
> +
> + if (is_34bit) {
> + msg_size = sizeof(struct venc_ap_ipi_msg_enc_ext_34);
> + status = vpu_enc_send_msg(vpu, &out_34, msg_size);
> + } else {
> + status = vpu_enc_send_msg(vpu, &out, msg_size);
> + }
> +
> + if (status) {
> mtk_vcodec_err(vpu, "AP_IPIMSG_ENC_ENCODE %d fail",
> bs_mode);
> return -EINVAL;
Powered by blists - more mailing lists