Hi Kieran,
Thanks for the feedback.
> Subject: Re: [PATCH v6 3/3] media: renesas: vsp1: Add support for RZ/G2L
> VSPD
>
> Quoting Biju Das (2022-03-16 11:55:51)
> > The RZ/G2L VSPD provides a single VSPD instance. It has the following
> > sub modules MAU, CTU, RPF, DPR, LUT, BRS, WPF and LIF.
> >
> > The VSPD block on RZ/G2L does not have a version register, so added a
> > new compatible string "renesas,rzg2l-vsp2" with a data pointer
> > containing the info structure. Also the reset line is shared with the DU
> module.
> >
> > Signed-off-by: Biju Das <biju.das.jz@xxxxxxxxxxxxxx>
> > Reviewed-by: Lad Prabhakar <prabhakar.mahadev-lad.rj@xxxxxxxxxxxxxx>
> > Reviewed-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> > ---
> > v5->v6:
> > * Rebased to media_staging and updated commit header
> > * Removed the extra tab from rzg2l_vsp2_device_info
> > * Changed the function vsp1_lookup->vsp1_lookup_info and
> > all info match related code moved here.
> > * Add VI6_IP_VERSION_VSP and VI6_IP_VERSION_VSP_SW macros to
> > distinguish HW & SW IP_VSP_Version.
> > * Used 0x80 for RZG2L VSPD model and SoC identification
> > * Updated Switch() for LIF0 buffer attribute handling.
> > v4->v5:
> > * Fixed typo VI6_IP_VERSION_MODEL_MASK->VI6_IP_VERSION_MASK
> > * To be consistent with other SoC's, introduced VI6_IP_VERSION_SOC_G2L
> > for RZ/G2L SoC's.
> > v3->v4:
> > * Added Rb tag from Geert
> > * Add switch() for LIF0 buffer attribute handling for RZ/G2L and V3M
> > v2->v3:
> > * Fixed version comparison in vsp1_lookup()
> > v1->v2:
> > * Changed the compatible from vsp2-rzg2l->rzg2l-vsp2
> > * Added standalone device info for rzg2l-vsp2.
> > * Added vsp1_lookup helper function.
> > * Updated comments for LIF0 buffer attribute register
> > * Used last ID for rzg2l-vsp2.
> > RFC->v1:
> > * Used data pointer containing info structure to retrieve version
> > information
> > RFC:
> > ---
> > .../media/platform/renesas/vsp1/vsp1_drv.c | 56 ++++++++++++++-----
> > .../media/platform/renesas/vsp1/vsp1_lif.c | 18 ++++--
> > .../media/platform/renesas/vsp1/vsp1_regs.h | 8 +++
> > 3 files changed, 62 insertions(+), 20 deletions(-)
> >
> > diff --git a/drivers/media/platform/renesas/vsp1/vsp1_drv.c
> > b/drivers/media/platform/renesas/vsp1/vsp1_drv.c
> > index 159b68fa0829..f1f52c0c1c59 100644
> > --- a/drivers/media/platform/renesas/vsp1/vsp1_drv.c
> > +++ b/drivers/media/platform/renesas/vsp1/vsp1_drv.c
> > @@ -812,11 +812,47 @@ static const struct vsp1_device_info
> vsp1_device_infos[] = {
> > },
> > };
> >
> > +static const struct vsp1_device_info rzg2l_vsp2_device_info = {
> > + .version = VI6_IP_VERSION_MODEL_VSPD_RZG2L,
> > + .model = "VSP2-D",
> > + .gen = 3,
> > + .features = VSP1_HAS_BRS | VSP1_HAS_WPF_VFLIP | VSP1_HAS_EXT_DL,
> > + .lif_count = 1,
> > + .rpf_count = 2,
> > + .wpf_count = 1,
> > +};
> > +
> > +static const struct vsp1_device_info *vsp1_lookup_info(struct
> > +vsp1_device *vsp1) {
> > + const struct vsp1_device_info *info;
> > + unsigned int i;
> > +
> > + /*
> > + * Try the info stored in match data first for devices that
> don't have
> > + * a version register.
> > + */
> > + info = of_device_get_match_data(vsp1->dev);
> > + if (info)
>
> Presumably - as this will not call vsp1_read(vsp1, VI6_IP_VERSION), we
> could/should always set vsp1->version here, or'ing in the _SW flag with
> the derived version and SoC identifiers from the info structure.
OK, I have prototyped as per your suggestion, and it looks good.
Here it is
if (info) {
vsp1->quirks = LIF_BUF_ATTR_QUIRKS;
vsp1->version = VI6_IP_VERSION_VSP_SW | info->version |
VI6_IP_VERSION_SOC_RZG2L;
>
> > + return info;
> > +
> > + vsp1->version = vsp1_read(vsp1, VI6_IP_VERSION);
> > +
> > + for (i = 0; i < ARRAY_SIZE(vsp1_device_infos); ++i) {
> > + info = &vsp1_device_infos[i];
> > +
> > + if ((vsp1->version & VI6_IP_VERSION_MODEL_MASK) == info-
> >version)
Here it is
if ((vsp1->version & VI6_IP_VERSION_MODEL_MASK) == info->version) {
if ((vsp1->version & VI6_IP_VERSION_MASK) ==
(VI6_IP_VERSION_MODEL_VSPD_V3 | VI6_IP_VERSION_SOC_V3M))
vsp1->quirks = LIF_BUF_ATTR_QUIRKS;
> > + return info;
> > + }
> > +
> > + dev_err(vsp1->dev, "unsupported IP version 0x%08x\n",
> > + vsp1->version);
> > +
> > + return NULL;
> > +}
> > +
> > static int vsp1_probe(struct platform_device *pdev) {
> > struct vsp1_device *vsp1;
> > struct device_node *fcp_node;
> > - unsigned int i;
> > int ret;
> > int irq;
> >
> > @@ -872,25 +908,16 @@ static int vsp1_probe(struct platform_device
> *pdev)
> > if (ret < 0)
> > goto done;
> >
> > - vsp1->version = vsp1_read(vsp1, VI6_IP_VERSION);
> > -
> > - for (i = 0; i < ARRAY_SIZE(vsp1_device_infos); ++i) {
> > - if ((vsp1->version & VI6_IP_VERSION_MODEL_MASK) ==
> > - vsp1_device_infos[i].version) {
> > - vsp1->info = &vsp1_device_infos[i];
> > - break;
> > - }
> > - }
> > -
> > + vsp1->info = vsp1_lookup_info(vsp1);
> > if (!vsp1->info) {
> > - dev_err(&pdev->dev, "unsupported IP version 0x%08x\n",
> > - vsp1->version);
> > vsp1_device_put(vsp1);
> > ret = -ENXIO;
> > goto done;
> > }
> >
> > - dev_dbg(&pdev->dev, "IP version 0x%08x\n", vsp1->version);
> > + if ((vsp1->version & VI6_IP_VERSION_VSP_MASK) !=
> VI6_IP_VERSION_VSP)
> > + vsp1->version = VI6_IP_VERSION_VSP_SW | vsp1->info-
> >version |
> > + VI6_IP_VERSION_SOC_RZG2L;
>
> It seems odd to have this specific version assignment here. Shouldn't that
> be set during vsp1_lookup_info() in the case that there is a match from
> of_device_get_match_data()? That way it would be extendable by adding just
> a new vsp1_device_info structure for the next platform that has this
> issue. This implies that they will 'always' be RZG2L but that information
> should live in the vsp1_device_info structure I think.
>
We can remove this assignment from here and move vsp1_lookup_info.
> Could be handled when/if we get a new device added I guess, but I think
> that VI6_IP_VERSION_SOC_RZG2L should be something that is retrieved from
> the vsp1_device_info structure.
>
> Re-reading the vsp1_lookup_info() function - it does seem like something
> suited to there, as the vsp1->version is never read from hardware in the
> new case.
Ok, Agreed.
>
> >
> > /*
> > * Previous use of the hardware (e.g. by the bootloader) could
> > leave @@ -941,6 +968,7 @@ static int vsp1_remove(struct
> > platform_device *pdev) static const struct of_device_id vsp1_of_match[]
> = {
> > { .compatible = "renesas,vsp1" },
> > { .compatible = "renesas,vsp2" },
> > + { .compatible = "renesas,rzg2l-vsp2", .data =
> > + &rzg2l_vsp2_device_info },
> > { },
> > };
> > MODULE_DEVICE_TABLE(of, vsp1_of_match); diff --git
> > a/drivers/media/platform/renesas/vsp1/vsp1_lif.c
> > b/drivers/media/platform/renesas/vsp1/vsp1_lif.c
> > index 6a6857ac9327..e36ed2d2b22b 100644
> > --- a/drivers/media/platform/renesas/vsp1/vsp1_lif.c
> > +++ b/drivers/media/platform/renesas/vsp1/vsp1_lif.c
> > @@ -107,6 +107,7 @@ static void lif_configure_stream(struct
> > vsp1_entity *entity,
> >
> > case VI6_IP_VERSION_MODEL_VSPDL_GEN3:
> > case VI6_IP_VERSION_MODEL_VSPD_V3:
> > + case VI6_IP_VERSION_MODEL_VSPD_RZG2L:
> > hbth = 0;
> > obth = 1500;
> > lbth = 0;
> > @@ -130,16 +131,21 @@ static void lif_configure_stream(struct
> vsp1_entity *entity,
> > VI6_LIF_CTRL_REQSEL | VI6_LIF_CTRL_LIF_EN);
> >
> > /*
> > - * On R-Car V3M the LIF0 buffer attribute register has to be set
> to a
> > - * non-default value to guarantee proper operation (otherwise
> artifacts
> > - * may appear on the output). The value required by the manual
> is not
> > - * explained but is likely a buffer size or threshold.
> > + * On R-Car V3M and RZ/G2L the LIF0 buffer attribute register
> has to be
> > + * set to a non-default value to guarantee proper operation
> (otherwise
> > + * artifacts may appear on the output). The value required by
> the
> > + * manual is not explained but is likely a buffer size or
> threshold.
> > */
> > - if ((entity->vsp1->version & VI6_IP_VERSION_MASK) ==
> > - (VI6_IP_VERSION_MODEL_VSPD_V3 | VI6_IP_VERSION_SOC_V3M))
> > + switch (entity->vsp1->version) {
> > + case (VI6_IP_VERSION_VSP | VI6_IP_VERSION_MODEL_VSPD_V3 |
> > + VI6_IP_VERSION_SOC_V3M):
> > + case (VI6_IP_VERSION_VSP_SW | VI6_IP_VERSION_MODEL_VSPD_RZG2L |
> > + VI6_IP_VERSION_SOC_RZG2L):
>
> If this is going to grow - I would think it would be better served with a
> feature flag - although this isn't so much of a feature, and more of a
> quirk, so I wonder if that would push us closer to getting a quirks flag.
>
> I'm weary that this may not scale otherwise, but ... for now this works,
> but I think it means we have multiple ways of handling platform specific
> code already.
Here it is
if (lif->quirks)
vsp1_lif_write(lif, dlb, VI6_LIF_LBA,
VI6_LIF_LBA_LBA0 |
(1536 << VI6_LIF_LBA_LBA1_SHIFT));
And the below change in vsp1_lif_create
lif->quirks = vsp1->quirks & LIF_BUF_ATTR_QUIRKS;
>
>
> > vsp1_lif_write(lif, dlb, VI6_LIF_LBA,
> > VI6_LIF_LBA_LBA0 |
> > (1536 << VI6_LIF_LBA_LBA1_SHIFT));
> > + break;
> > + }
> > }
> >
> > static const struct vsp1_entity_operations lif_entity_ops = { diff
> > --git a/drivers/media/platform/renesas/vsp1/vsp1_regs.h
> > b/drivers/media/platform/renesas/vsp1/vsp1_regs.h
> > index fae7286eb01e..e66553c42e50 100644
> > --- a/drivers/media/platform/renesas/vsp1/vsp1_regs.h
> > +++ b/drivers/media/platform/renesas/vsp1/vsp1_regs.h
> > @@ -767,6 +767,8 @@
> > #define VI6_IP_VERSION_MODEL_VSPDL_GEN3 (0x19 << 8)
> > #define VI6_IP_VERSION_MODEL_VSPBS_GEN3 (0x1a << 8)
> > #define VI6_IP_VERSION_MODEL_VSPD_V3U (0x1c << 8)
> > +/* RZ/G2L SoC's have no version register, So use 0x80 as the model
> version */
> > +#define VI6_IP_VERSION_MODEL_VSPD_RZG2L (0x80 << 8)
> >
> > #define VI6_IP_VERSION_SOC_MASK (0xff << 0)
> > #define VI6_IP_VERSION_SOC_H2 (0x01 << 0)
> > @@ -780,6 +782,12 @@
> > #define VI6_IP_VERSION_SOC_M3N (0x04 << 0)
> > #define VI6_IP_VERSION_SOC_E3 (0x04 << 0)
> > #define VI6_IP_VERSION_SOC_V3U (0x05 << 0)
> > +/* RZ/G2L SoC's have no version register, So use 0x80 for SoC
> Identification */
> > +#define VI6_IP_VERSION_SOC_RZG2L (0x80 << 0)
> > +
> > +#define VI6_IP_VERSION_VSP_MASK (0xffff << 16)
> > +#define VI6_IP_VERSION_VSP (0x0101 << 16) /* HW VSP version
> */
>
> Is this constant on all supported platforms? both Gen2 and Gen3? (Is there
> a gen1?). Does it need to be specified to the generation?
I have checked Gen1 and Gen2 HW manual I don't find this info. So I would like
to remove this macro as it is unused after quirk changes.
I am planning to send V7 with these changes, please let me know if you have any feedback.
Cheers,
Biju
>
> There's nothing specifically complex there or blocking I don't think - so
> with comments considered as required:
>
> Reviewed-by: Kieran Bingham <kieran.bingham+renesas@xxxxxxxxxxxxxxxx>
>
> > +#define VI6_IP_VERSION_VSP_SW (0xfffe << 16) /* SW VSP version
> */
> >
> > /* --------------------------------------------------------------------
> ---------
> > * RPF CLUT Registers
> > --
> > 2.17.1
> >
