Re: [PATCH v3] media: mtk-vpu: Ensure alignment of 8 for DTCM buffer

[Dafna Hirschfeld]

On 01.09.21 10:50, Hans Verkuil wrote:
> On 25/08/2021 12:17, Dafna Hirschfeld wrote:
> > From: Alexandre Courbot <acourbot@xxxxxxxxxxxx>
> >
> > When running memcpy_toio:
> > memcpy_toio(send_obj->share_buf, buf, len);
> > it was found that errors appear if len is not a multiple of 8:
> >
> > [58.350841] mtk-mdp 14001000.rdma: processing failed: -22
> >
> > This patch ensures the copy of a multile of 8 size by calling
>
> multile -> multiple
>
> > round_up(len, 8) when copying
> >
> > Fixes: e6599adfad30 ("media: mtk-vpu: avoid unaligned access to DTCM buffer.")
> > Signed-off-by: Alexandre Courbot <acourbot@xxxxxxxxxxxx>
> > Signed-off-by: Enric Balletbo i Serra <enric.balletbo@xxxxxxxxxxxxx>
> > Signed-off-by: Dafna Hirschfeld <dafna.hirschfeld@xxxxxxxxxxxxx>
> > ---
> > changes since v2:
> > 1. do the extra copy only if len is not multiple of 8
> >
> > changes since v1:
> > 1. change sign-off-by tags
> > 2. change values to memset
> >
> >   drivers/media/platform/mtk-vpu/mtk_vpu.c | 11 ++++++++++-
> >   1 file changed, 10 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/media/platform/mtk-vpu/mtk_vpu.c b/drivers/media/platform/mtk-vpu/mtk_vpu.c
> > index ec290dde59cf..658161ee3e4e 100644
> > --- a/drivers/media/platform/mtk-vpu/mtk_vpu.c
> > +++ b/drivers/media/platform/mtk-vpu/mtk_vpu.c
> > @@ -349,7 +349,16 @@ int vpu_ipi_send(struct platform_device *pdev,
> >   		}
> >   	} while (vpu_cfg_readl(vpu, HOST_TO_VPU));
> >   
> > -	memcpy_toio(send_obj->share_buf, buf, len);
> > +	if (len % 8 != 0) {
>
> You need to add a comment here explaining why this is done (basically what
> you also say in the commit log).
>
> Otherwise people would have to dig into the git log to figure out why this
> code is the way it is.
>
> Is len often not a multiple of 8? If that's the case, then it might be easier
> to just always do the memset/memcpy.

I wrote a program that prints the sizes of all possible messages: http://ix.io/3zsr
compiling it to arm64 gives:

sizeof(mdp_ipi_comm) is 24
sizeof(mdp_ipi_init) is 16

sizeof(venc_ap_ipi_msg_deinit) is 8
sizeof(venc_ap_ipi_msg_enc_ext) is 164
sizeof(venc_ap_ipi_msg_enc) is 32
sizeof(venc_ap_ipi_msg_set_param_ext) is 144
sizeof(venc_ap_ipi_msg_set_param) is 48
sizeof(venc_ap_ipi_msg_init) is 16

sizeof(vdec_ap_ipi_cmd) is 8
sizeof(vdec_ap_ipi_init) is 16
sizeof(vdec_ap_ipi_dec_start) is 24

BUT, when printing the size in the kernel I got "sizeof(mdp_ipi_comm) = 20" and found out
that this is due to the macro #pragma pack(push, 4) in mtk_mdp_ipi.h

so for mdp it makes sens to always do the memset/memcpy but for mtk-vcodec it is not necessary.
In addition, if the message is one of 'venc_ap_ipi_msg_enc_ext' or 'venc_ap_ipi_msg_set_param_ext' then vpu_ipi_send
will fail in the beginning of the function since in those cases we have 'len > sizeof(send_obj->share_buf)'.
so I should send a patch that set SHARE_BUF_SIZE to e.g. 200

I also found out that adding #pragma pack(push, 4) to the definitions of the mtk-vcodec messages has no influence but
maybe this should be added since we copy those messages to hw , so might avoid future bugs.

thanks,
Dafna

> Regards,
>
> 	Hans
>
> > +		unsigned char data[SHARE_BUF_SIZE];
> > +
> > +		memset(data + len, 0, sizeof(data) - len);
> > +		memcpy(data, buf, len);
> > +		memcpy_toio(send_obj->share_buf, data, round_up(len, 8));
> > +	} else {
> > +		memcpy_toio(send_obj->share_buf, buf, len);
> > +	}
> > +
> >   	writel(len, &send_obj->len);
> >   	writel(id, &send_obj->id);
> >