[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20210216144426.h45b42g6w4hswjuy@uno.localdomain>
Date: Tue, 16 Feb 2021 15:44:26 +0100
From: Jacopo Mondi <jacopo@...ndi.org>
To: Andrey Konovalov <andrey.konovalov@...aro.org>
Cc: junak.pub@...il.com, robert.foss@...aro.org,
sakari.ailus@...ux.intel.com, todor.too@...il.com,
agross@...nel.org, bjorn.andersson@...aro.org, mchehab@...nel.org,
laurent.pinchart@...asonboard.com, linux-media@...r.kernel.org,
linux-arm-msm@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH 1/2] media: camss: use v4l2_get_link_freq() to calculate
the relevant clocks
Hi Andrey,
On Mon, Feb 15, 2021 at 11:58:27PM +0300, Andrey Konovalov wrote:
> Hi Jacopo,
>
> Thank you for the detailed review!
>
> On 15.02.2021 15:00, Jacopo Mondi wrote:
> > Hi Andrey,
> > nice to see progress in this direction
> >
> > On Mon, Feb 15, 2021 at 12:34:03AM +0300, Andrey Konovalov wrote:
> > > There are places in the camss driver where camss_get_pixel_clock() is
> > > called to get the pixel rate (using V4L2_CID_PIXEL_RATE control) and to
> > > calculate the link frequency from it. There is a case when this would
> > > not work: when V4L2_CID_PIXEL_RATE gets the rate at which the pixels are
> > > read (sampled) from the sensor's pixel array, and this rate is different
> > > from the pixel transmission rate over the CSI link, the link frequency
> > > value can't be calculated from the pixel rate. One needs to use
> > > V4L2_CID_LINK_FREQ to get the link frequency in this case.
> > >
> > > Replace such calls to camss_get_pixel_clock() with calls to a wrapper
> > > around v4l2_get_link_freq(). v4l2_get_link_freq() tries V4L2_CID_LINK_FREQ
> > > first, and if it is not implemented by the camera sensor driver, falls
> > > back to V4L2_CID_PIXEL_RATE to calculate the link frequency value from.
> >
> > Is it worth warning in the core function that the subdevice should
> > support LINK_FREQ and if we fallback to use PIXEL_RATE the calculation
> > result might not be accurate ?
>
> Yes, this might be useful. I'll add a separate patch for that in v2.
>
> > > Calls to camss_get_pixel_clock() from vfe_[check,set]_clock_rates()
> > > are left intact as it looks like this VFE clock does depend on the
> > > rate the pixel samples comes out of the camera sensor, not on the
> > > frequency at which the link between the sensor and the CSI receiver
> > > operates.
> > >
> > > Signed-off-by: Andrey Konovalov <andrey.konovalov@...aro.org>
> > > ---
> > > .../media/platform/qcom/camss/camss-csid.c | 22 ++++++------
> > > .../qcom/camss/camss-csiphy-2ph-1-0.c | 22 ++++++------
> > > .../qcom/camss/camss-csiphy-3ph-1-0.c | 22 ++++++------
> > > .../media/platform/qcom/camss/camss-csiphy.c | 36 +++++++++----------
> > > .../media/platform/qcom/camss/camss-csiphy.h | 2 +-
> > > drivers/media/platform/qcom/camss/camss.c | 23 ++++++++++++
> > > drivers/media/platform/qcom/camss/camss.h | 2 ++
> > > 7 files changed, 73 insertions(+), 56 deletions(-)
> > >
> > > diff --git a/drivers/media/platform/qcom/camss/camss-csid.c b/drivers/media/platform/qcom/camss/camss-csid.c
> > > index be3fe76f3dc3..b2cbf4b65949 100644
> > > --- a/drivers/media/platform/qcom/camss/camss-csid.c
> > > +++ b/drivers/media/platform/qcom/camss/camss-csid.c
> > > @@ -462,13 +462,19 @@ static irqreturn_t csid_isr(int irq, void *dev)
> > > static int csid_set_clock_rates(struct csid_device *csid)
> > > {
> > > struct device *dev = csid->camss->dev;
> > > - u32 pixel_clock;
> > > + s64 link_freq;
> > > int i, j;
> > > int ret;
> > >
> > > - ret = camss_get_pixel_clock(&csid->subdev.entity, &pixel_clock);
> > > - if (ret)
> > > - pixel_clock = 0;
> > > + const struct csid_format *f = csid_get_fmt_entry(
> > > + csid->formats,
> > > + csid->nformats,
> > > + csid->fmt[MSM_CSIPHY_PAD_SINK].code);
> >
> > Weird indent :/
>
> This part was in this driver before - I've just moved it a few lines up.
> But it doesn't look very nice, agreed.
>
> > I would either keep the arguments on one line or align after the open
> > ( if it doesn't go past 80-cols
>
> - as far as I can see, the "csid->fmt[MSM_CSIPHY_PAD_SINK].code);" wouldn't
> fit into 80-cols either way.
>
> I'll try to think something out in v2.
>
> > > + u8 num_lanes = csid->phy.lane_cnt;
> > > + link_freq = camss_get_link_freq(&csid->subdev.entity, f->bpp,
> >
> > Empy line maybe ?
>
> Ack. checkpatch suggests the same.
>
> > > + 2 * num_lanes);
> >
> > I see you pass in 2 * num_lanes and I assume it's for CSI-2 DDR.
>
> Right.
>
> > Can't this be handled in camss_get_link_freq() so that you here only
> > pass in the actual number of lanes ?
>
> OK, this should look more clear. Will do that in v2.
>
> > > + if (link_freq < 0)
> > > + link_freq = 0;
> >
> > I don't know this driver, but I wonder if it wouldn't be better to
> > fail instead of defaulting to 0, which might be dangerous if used as a
> > divider.
>
> This is in accordance with the logic implemented in the current driver:
>
> -----8<-----
> u64 min_rate = link_freq / 4;
> <snip>
> camss_add_clock_margin(&min_rate);
> <if min_rate is 0, camss_add_clock_margin() leaves min_rate as 0>
> <snip>
> /* if sensor pixel clock is not available */
> /* set highest possible CSID clock rate */
Sorry, we clarified on irc and I never replied to this email.
As I've said I'm not familiar with the driver, so if it's best to
return 0 and use the maximum available CSID rate, then please do so.
> if (min_rate == 0)
> j = clock->nfreqs - 1;
> -----8<-----
>
> > >
> > > for (i = 0; i < csid->nclocks; i++) {
> > > struct camss_clock *clock = &csid->clock[i];
> > > @@ -477,13 +483,7 @@ static int csid_set_clock_rates(struct csid_device *csid)
> > > !strcmp(clock->name, "csi1") ||
> > > !strcmp(clock->name, "csi2") ||
> > > !strcmp(clock->name, "csi3")) {
> > > - const struct csid_format *f = csid_get_fmt_entry(
> > > - csid->formats,
> > > - csid->nformats,
> > > - csid->fmt[MSM_CSIPHY_PAD_SINK].code);
> > > - u8 num_lanes = csid->phy.lane_cnt;
> > > - u64 min_rate = pixel_clock * f->bpp /
> > > - (2 * num_lanes * 4);
> > > + u64 min_rate = link_freq / 4;
> >
> > Why 4 ? :)
>
> min_rate = (pixel_clock * f->bpp / (2 * num_lanes)) / 4;
> But (pixel_clock * f->bpp / (2 * num_lanes)) equals link_freq.
> Actually the "APQ8016E Technical Reference Manual", LM80-P0436-100 Rev. D [1] at page 960
> says exactly that:
> "F(GCC_CAMSS_CSIx_CLK) >= F(DDR_CLK)/4"
>
> [1] https://developer.qualcomm.com/download/sd410/snapdragon-410e-technical-reference-manual.pdf
>
Ah ok, it's an HW requirement then. I felt I was missing something
obvious...
> > > long rate;
> > >
> > > camss_add_clock_margin(&min_rate);
> > > diff --git a/drivers/media/platform/qcom/camss/camss-csiphy-2ph-1-0.c b/drivers/media/platform/qcom/camss/camss-csiphy-2ph-1-0.c
> > > index 12bce391d71f..30b454c369ab 100644
> > > --- a/drivers/media/platform/qcom/camss/camss-csiphy-2ph-1-0.c
> > > +++ b/drivers/media/platform/qcom/camss/camss-csiphy-2ph-1-0.c
> > > @@ -51,16 +51,13 @@ static void csiphy_reset(struct csiphy_device *csiphy)
> > > *
> > > * Helper function to calculate settle count value. This is
> > > * based on the CSI2 T_hs_settle parameter which in turn
> > > - * is calculated based on the CSI2 transmitter pixel clock
> > > - * frequency.
> > > + * is calculated based on the CSI2 transmitter link frequency.
> > > *
> > > - * Return settle count value or 0 if the CSI2 pixel clock
> > > - * frequency is not available
> > > + * Return settle count value or 0 if the CSI2 link frequency
> > > + * is not available
> > > */
> > > -static u8 csiphy_settle_cnt_calc(u32 pixel_clock, u8 bpp, u8 num_lanes,
> > > - u32 timer_clk_rate)
> > > +static u8 csiphy_settle_cnt_calc(s64 link_freq, u32 timer_clk_rate)
> > > {
> > > - u32 mipi_clock; /* Hz */
> > > u32 ui; /* ps */
> > > u32 timer_period; /* ps */
> > > u32 t_hs_prepare_max; /* ps */
> > > @@ -68,8 +65,10 @@ static u8 csiphy_settle_cnt_calc(u32 pixel_clock, u8 bpp, u8 num_lanes,
> > > u32 t_hs_settle; /* ps */
> > > u8 settle_cnt;
> > >
> > > - mipi_clock = pixel_clock * bpp / (2 * num_lanes);
> > > - ui = div_u64(1000000000000LL, mipi_clock);
> > > + if (link_freq <= 0)
> > > + return 0;
> >
> > If you error out if the link frequency cannot be calculated, can this
> > be skipped ?
>
> Do you mean removing the "if (link_freq <= 0) return 0;" from csiphy_settle_cnt_calc(), and
> making the caller of csiphy_settle_cnt_calc() to do this check?
>
> > > +
> > > + ui = div_u64(1000000000000LL, link_freq);
> > > ui /= 2;
> > > t_hs_prepare_max = 85000 + 6 * ui;
> > > t_hs_prepare_zero_min = 145000 + 10 * ui;
> > > @@ -83,15 +82,14 @@ static u8 csiphy_settle_cnt_calc(u32 pixel_clock, u8 bpp, u8 num_lanes,
> > >
> > > static void csiphy_lanes_enable(struct csiphy_device *csiphy,
> > > struct csiphy_config *cfg,
> > > - u32 pixel_clock, u8 bpp, u8 lane_mask)
> > > + s64 link_freq, u8 lane_mask)
> > > {
> > > struct csiphy_lanes_cfg *c = &cfg->csi2->lane_cfg;
> > > u8 settle_cnt;
> > > u8 val, l = 0;
> > > int i = 0;
> > >
> > > - settle_cnt = csiphy_settle_cnt_calc(pixel_clock, bpp, c->num_data,
> > > - csiphy->timer_clk_rate);
> > > + settle_cnt = csiphy_settle_cnt_calc(link_freq, csiphy->timer_clk_rate);
> > >
> > > writel_relaxed(0x1, csiphy->base +
> > > CAMSS_CSI_PHY_GLBL_T_INIT_CFG0);
> > > diff --git a/drivers/media/platform/qcom/camss/camss-csiphy-3ph-1-0.c b/drivers/media/platform/qcom/camss/camss-csiphy-3ph-1-0.c
> > > index 97cb9de85031..da7c3d3f9a10 100644
> > > --- a/drivers/media/platform/qcom/camss/camss-csiphy-3ph-1-0.c
> > > +++ b/drivers/media/platform/qcom/camss/camss-csiphy-3ph-1-0.c
> > > @@ -107,24 +107,23 @@ static irqreturn_t csiphy_isr(int irq, void *dev)
> > > *
> > > * Helper function to calculate settle count value. This is
> > > * based on the CSI2 T_hs_settle parameter which in turn
> > > - * is calculated based on the CSI2 transmitter pixel clock
> > > - * frequency.
> > > + * is calculated based on the CSI2 transmitter link frequency.
> > > *
> > > - * Return settle count value or 0 if the CSI2 pixel clock
> > > - * frequency is not available
> > > + * Return settle count value or 0 if the CSI2 link frequency
> > > + * is not available
> > > */
> > > -static u8 csiphy_settle_cnt_calc(u32 pixel_clock, u8 bpp, u8 num_lanes,
> > > - u32 timer_clk_rate)
> > > +static u8 csiphy_settle_cnt_calc(s64 link_freq, u32 timer_clk_rate)
> > > {
> > > - u32 mipi_clock; /* Hz */
> > > u32 ui; /* ps */
> > > u32 timer_period; /* ps */
> > > u32 t_hs_prepare_max; /* ps */
> > > u32 t_hs_settle; /* ps */
> > > u8 settle_cnt;
> > >
> > > - mipi_clock = pixel_clock * bpp / (2 * num_lanes);
> > > - ui = div_u64(1000000000000LL, mipi_clock);
> > > + if (link_freq <= 0)
> > > + return 0;
> > > +
> > > + ui = div_u64(1000000000000LL, link_freq);
> > > ui /= 2;
> > > t_hs_prepare_max = 85000 + 6 * ui;
> > > t_hs_settle = t_hs_prepare_max;
> > > @@ -137,15 +136,14 @@ static u8 csiphy_settle_cnt_calc(u32 pixel_clock, u8 bpp, u8 num_lanes,
> > >
> > > static void csiphy_lanes_enable(struct csiphy_device *csiphy,
> > > struct csiphy_config *cfg,
> > > - u32 pixel_clock, u8 bpp, u8 lane_mask)
> > > + s64 link_freq, u8 lane_mask)
> > > {
> > > struct csiphy_lanes_cfg *c = &cfg->csi2->lane_cfg;
> > > u8 settle_cnt;
> > > u8 val, l = 0;
> > > int i;
> > >
> > > - settle_cnt = csiphy_settle_cnt_calc(pixel_clock, bpp, c->num_data,
> > > - csiphy->timer_clk_rate);
> > > + settle_cnt = csiphy_settle_cnt_calc(link_freq, csiphy->timer_clk_rate);
> > >
> > > val = BIT(c->clk.pos);
> > > for (i = 0; i < c->num_data; i++)
> > > diff --git a/drivers/media/platform/qcom/camss/camss-csiphy.c b/drivers/media/platform/qcom/camss/camss-csiphy.c
> > > index 509c9a59c09c..9b5fe6fc7664 100644
> > > --- a/drivers/media/platform/qcom/camss/camss-csiphy.c
> > > +++ b/drivers/media/platform/qcom/camss/camss-csiphy.c
> > > @@ -102,23 +102,23 @@ static u8 csiphy_get_bpp(const struct csiphy_format *formats,
> > > static int csiphy_set_clock_rates(struct csiphy_device *csiphy)
> > > {
> > > struct device *dev = csiphy->camss->dev;
> > > - u32 pixel_clock;
> > > + s64 link_freq;
> > > int i, j;
> > > int ret;
> > >
> > > - ret = camss_get_pixel_clock(&csiphy->subdev.entity, &pixel_clock);
> > > - if (ret)
> > > - pixel_clock = 0;
> > > + u8 bpp = csiphy_get_bpp(csiphy->formats, csiphy->nformats,
> > > + csiphy->fmt[MSM_CSIPHY_PAD_SINK].code);
> > > + u8 num_lanes = csiphy->cfg.csi2->lane_cfg.num_data;
> > > + link_freq = camss_get_link_freq(&csiphy->subdev.entity, bpp,
> > > + 2 * num_lanes);
> > > + if (link_freq < 0)
> > > + link_freq = 0;
> > >
> > > for (i = 0; i < csiphy->nclocks; i++) {
> > > struct camss_clock *clock = &csiphy->clock[i];
> > >
> > > if (csiphy->rate_set[i]) {
> > > - u8 bpp = csiphy_get_bpp(csiphy->formats,
> > > - csiphy->nformats,
> > > - csiphy->fmt[MSM_CSIPHY_PAD_SINK].code);
> > > - u8 num_lanes = csiphy->cfg.csi2->lane_cfg.num_data;
> > > - u64 min_rate = pixel_clock * bpp / (2 * num_lanes * 4);
> > > + u64 min_rate = link_freq / 4;
> > > long round_rate;
> > >
> > > camss_add_clock_margin(&min_rate);
> > > @@ -238,22 +238,18 @@ static u8 csiphy_get_lane_mask(struct csiphy_lanes_cfg *lane_cfg)
> > > static int csiphy_stream_on(struct csiphy_device *csiphy)
> > > {
> > > struct csiphy_config *cfg = &csiphy->cfg;
> > > - u32 pixel_clock;
> > > + s64 link_freq;
> > > u8 lane_mask = csiphy_get_lane_mask(&cfg->csi2->lane_cfg);
> > > u8 bpp = csiphy_get_bpp(csiphy->formats, csiphy->nformats,
> > > csiphy->fmt[MSM_CSIPHY_PAD_SINK].code);
> > > + u8 num_lanes = csiphy->cfg.csi2->lane_cfg.num_data;
> > > u8 val;
> > > - int ret;
> > >
> > > - ret = camss_get_pixel_clock(&csiphy->subdev.entity, &pixel_clock);
> > > - if (ret) {
> > > - dev_err(csiphy->camss->dev,
> > > - "Cannot get CSI2 transmitter's pixel clock\n");
> > > - return -EINVAL;
> > > - }
> > > - if (!pixel_clock) {
> > > + link_freq = camss_get_link_freq(&csiphy->subdev.entity, bpp,
> > > + 2 * num_lanes);
> > > + if (link_freq < 0) {
> > > dev_err(csiphy->camss->dev,
> > > - "Got pixel clock == 0, cannot continue\n");
> > > + "Cannot get CSI2 transmitter's link frequency\n");
> > > return -EINVAL;
> > > }
> > >
> > > @@ -268,7 +264,7 @@ static int csiphy_stream_on(struct csiphy_device *csiphy)
> > > writel_relaxed(val, csiphy->base_clk_mux);
> > > wmb();
> > >
> > > - csiphy->ops->lanes_enable(csiphy, cfg, pixel_clock, bpp, lane_mask);
> > > + csiphy->ops->lanes_enable(csiphy, cfg, link_freq, lane_mask);
> > >
> > > return 0;
> > > }
> > > diff --git a/drivers/media/platform/qcom/camss/camss-csiphy.h b/drivers/media/platform/qcom/camss/camss-csiphy.h
> > > index f7967ef836dc..d71b8bc6ec00 100644
> > > --- a/drivers/media/platform/qcom/camss/camss-csiphy.h
> > > +++ b/drivers/media/platform/qcom/camss/camss-csiphy.h
> > > @@ -50,7 +50,7 @@ struct csiphy_hw_ops {
> > > void (*reset)(struct csiphy_device *csiphy);
> > > void (*lanes_enable)(struct csiphy_device *csiphy,
> > > struct csiphy_config *cfg,
> > > - u32 pixel_clock, u8 bpp, u8 lane_mask);
> > > + s64 link_freq, u8 lane_mask);
> > > void (*lanes_disable)(struct csiphy_device *csiphy,
> > > struct csiphy_config *cfg);
> > > irqreturn_t (*isr)(int irq, void *dev);
> > > diff --git a/drivers/media/platform/qcom/camss/camss.c b/drivers/media/platform/qcom/camss/camss.c
> > > index 7c0f669f8aa6..2888c7ef2303 100644
> > > --- a/drivers/media/platform/qcom/camss/camss.c
> > > +++ b/drivers/media/platform/qcom/camss/camss.c
> > > @@ -548,6 +548,29 @@ struct media_entity *camss_find_sensor(struct media_entity *entity)
> > > }
> > > }
> > >
> > > +/**
> > > + * camss_get_link_freq - Get link frequency from sensor
> > > + * @entity: Media entity in the current pipeline
> > > + * @bpp: Number of bits per pixel for the current format
> > > + * @lanes: Number of lanes in the link to the sensor
> > > + *
> > > + * Return link frequency on success or a negative error code otherwise
> > > + */
> > > +s64 camss_get_link_freq(struct media_entity *entity, unsigned int bpp,
> > > + unsigned int lanes)
> > > +{
> > > + struct media_entity *sensor;
> > > + struct v4l2_subdev *subdev;
> > > +
> > > + sensor = camss_find_sensor(entity);
> > > + if (!sensor)
> > > + return -ENODEV;
> >
> > Can this happen ?
>
> In general case yes, it could.
> This would definitely happen if we used camss_get_link_freq() with the vfe
> entities when there is no sensor accessible from the given vfe entity via
> the enabled links.
> But as for the vfe entities we call camss_get_pixel_clock() instead,
> it looks like camss_get_link_freq() should not fail to find the sensor.
> Does it hurts, and isn't this safer to check the return value of
> camss_find_sensor() in camss_get_link_freq() as well (just in case)?
I don't mind. If the fact that there might be no sensor on the other
end of the link is accepted by the driver, please keep this check
here.
Thanks
j
>
> Thanks,
> Andrey
>
> > Thanks
> > j
> >
> > > +
> > > + subdev = media_entity_to_v4l2_subdev(sensor);
> > > +
> > > + return v4l2_get_link_freq(subdev->ctrl_handler, bpp, lanes);
> > > +}
> > > +
> > > /*
> > > * camss_get_pixel_clock - Get pixel clock rate from sensor
> > > * @entity: Media entity in the current pipeline
> > > diff --git a/drivers/media/platform/qcom/camss/camss.h b/drivers/media/platform/qcom/camss/camss.h
> > > index 3a0484683cd6..86cdc25189eb 100644
> > > --- a/drivers/media/platform/qcom/camss/camss.h
> > > +++ b/drivers/media/platform/qcom/camss/camss.h
> > > @@ -108,6 +108,8 @@ int camss_enable_clocks(int nclocks, struct camss_clock *clock,
> > > struct device *dev);
> > > void camss_disable_clocks(int nclocks, struct camss_clock *clock);
> > > struct media_entity *camss_find_sensor(struct media_entity *entity);
> > > +s64 camss_get_link_freq(struct media_entity *entity, unsigned int bpp,
> > > + unsigned int lanes);
> > > int camss_get_pixel_clock(struct media_entity *entity, u32 *pixel_clock);
> > > int camss_pm_domain_on(struct camss *camss, int id);
> > > void camss_pm_domain_off(struct camss *camss, int id);
> > > --
> > > 2.17.1
> > >
Powered by blists - more mailing lists