| Message ID | 1490706496-4959-6-git-send-email-tomi.valkeinen@ti.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Hi Tomi, Thank you for the patch. On Tuesday 28 Mar 2017 16:07:51 Tomi Valkeinen wrote: > The clock source selection for the LCD outputs is too hardcoded at the > moment. For example, LCD3 is set to use PLL2_1, and PLL2 doesn't exist > on DRA72x SoCs. > > There are quite many ways to configure the clocks, even using HDMI PLL > for LCD outputs, but enabling full configuration of the clocks is rather > tricky. > > This patch improves the situation a bit by checking if the PLL about to > be used exists, and if not, tries another one. > > Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com> > --- > drivers/gpu/drm/omapdrm/dss/dpi.c | 47 +++++++++++++++++++++++++++--------- > 1 file changed, 37 insertions(+), 10 deletions(-) > > diff --git a/drivers/gpu/drm/omapdrm/dss/dpi.c > b/drivers/gpu/drm/omapdrm/dss/dpi.c index e0b0c5c24c55..0f32d5d078c6 100644 > --- a/drivers/gpu/drm/omapdrm/dss/dpi.c > +++ b/drivers/gpu/drm/omapdrm/dss/dpi.c > @@ -67,6 +67,42 @@ static struct dpi_data *dpi_get_data_from_pdev(struct > platform_device *pdev) return dev_get_drvdata(&pdev->dev); > } > > +static enum dss_clk_source dpi_get_clk_src_dra7xx(enum omap_channel > channel) > +{ > + /* > + * Possible clock sources: > + * LCD1: FCK/PLL1_1/HDMI_PLL > + * LCD2: FCK/PLL1_3/HDMI_PLL (DRA74x: PLL2_3) > + * LCD3: FCK/PLL1_3/HDMI_PLL (DRA74x: PLL2_1) > + */ > + > + switch (channel) { > + case OMAP_DSS_CHANNEL_LCD: > + { > + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL1_1)) > + return DSS_CLK_SRC_PLL1_1; > + } Aren't you missing break statements ? > + case OMAP_DSS_CHANNEL_LCD2: > + { > + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL1_3)) > + return DSS_CLK_SRC_PLL1_3; > + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL2_3)) > + return DSS_CLK_SRC_PLL2_3; > + } > + case OMAP_DSS_CHANNEL_LCD3: > + { > + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL2_1)) > + return DSS_CLK_SRC_PLL2_1; > + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL1_3)) > + return DSS_CLK_SRC_PLL1_3; What happens if LCD2 is already using PLL1_3 ? > + } > + default: > + break; > + } > + > + return DSS_CLK_SRC_FCK; > +} > + > static enum dss_clk_source dpi_get_clk_src(enum omap_channel channel) > { > /* > @@ -107,16 +143,7 @@ static enum dss_clk_source dpi_get_clk_src(enum > omap_channel channel) } > > case OMAPDSS_VER_DRA7xx: > - switch (channel) { > - case OMAP_DSS_CHANNEL_LCD: > - return DSS_CLK_SRC_PLL1_1; > - case OMAP_DSS_CHANNEL_LCD2: > - return DSS_CLK_SRC_PLL1_3; > - case OMAP_DSS_CHANNEL_LCD3: > - return DSS_CLK_SRC_PLL2_1; > - default: > - return DSS_CLK_SRC_FCK; > - } > + return dpi_get_clk_src_dra7xx(channel); > > default: > return DSS_CLK_SRC_FCK;
On 29/03/17 11:19, Laurent Pinchart wrote: > Hi Tomi, > > Thank you for the patch. > > On Tuesday 28 Mar 2017 16:07:51 Tomi Valkeinen wrote: >> The clock source selection for the LCD outputs is too hardcoded at the >> moment. For example, LCD3 is set to use PLL2_1, and PLL2 doesn't exist >> on DRA72x SoCs. >> >> There are quite many ways to configure the clocks, even using HDMI PLL >> for LCD outputs, but enabling full configuration of the clocks is rather >> tricky. >> >> This patch improves the situation a bit by checking if the PLL about to >> be used exists, and if not, tries another one. >> >> Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com> >> --- >> drivers/gpu/drm/omapdrm/dss/dpi.c | 47 +++++++++++++++++++++++++++--------- >> 1 file changed, 37 insertions(+), 10 deletions(-) >> >> diff --git a/drivers/gpu/drm/omapdrm/dss/dpi.c >> b/drivers/gpu/drm/omapdrm/dss/dpi.c index e0b0c5c24c55..0f32d5d078c6 100644 >> --- a/drivers/gpu/drm/omapdrm/dss/dpi.c >> +++ b/drivers/gpu/drm/omapdrm/dss/dpi.c >> @@ -67,6 +67,42 @@ static struct dpi_data *dpi_get_data_from_pdev(struct >> platform_device *pdev) return dev_get_drvdata(&pdev->dev); >> } >> >> +static enum dss_clk_source dpi_get_clk_src_dra7xx(enum omap_channel >> channel) >> +{ >> + /* >> + * Possible clock sources: >> + * LCD1: FCK/PLL1_1/HDMI_PLL >> + * LCD2: FCK/PLL1_3/HDMI_PLL (DRA74x: PLL2_3) >> + * LCD3: FCK/PLL1_3/HDMI_PLL (DRA74x: PLL2_1) >> + */ >> + >> + switch (channel) { >> + case OMAP_DSS_CHANNEL_LCD: >> + { >> + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL1_1)) >> + return DSS_CLK_SRC_PLL1_1; >> + } > > Aren't you missing break statements ? Huh, indeed I am. Thanks! Interesting that we've never hit that bug (this has been in TI kernel for some time). > >> + case OMAP_DSS_CHANNEL_LCD2: >> + { >> + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL1_3)) >> + return DSS_CLK_SRC_PLL1_3; >> + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL2_3)) >> + return DSS_CLK_SRC_PLL2_3; >> + } >> + case OMAP_DSS_CHANNEL_LCD3: >> + { >> + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL2_1)) >> + return DSS_CLK_SRC_PLL2_1; >> + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL1_3)) >> + return DSS_CLK_SRC_PLL1_3; > > What happens if LCD2 is already using PLL1_3 ? Bad things, I guess? =) It is a limitation with the current implementation that some combinations are not really possible. To support all the cases, we should somehow share the clocks, and manage the needs of the displays using the same clocks... We have nothing for that. So if you design a board that uses LCD2 and LCD3, and you don't have PLL2, then the driver won't work. Sharing the clock with two displays of exact same pixel clock possibly works even now, but I have not tested it. This patch improves the situation a bit, by allowing all cases where we don't share the clocks, but it doesn't try to fix the sharing problem. Tomi
diff --git a/drivers/gpu/drm/omapdrm/dss/dpi.c b/drivers/gpu/drm/omapdrm/dss/dpi.c index e0b0c5c24c55..0f32d5d078c6 100644 --- a/drivers/gpu/drm/omapdrm/dss/dpi.c +++ b/drivers/gpu/drm/omapdrm/dss/dpi.c @@ -67,6 +67,42 @@ static struct dpi_data *dpi_get_data_from_pdev(struct platform_device *pdev) return dev_get_drvdata(&pdev->dev); } +static enum dss_clk_source dpi_get_clk_src_dra7xx(enum omap_channel channel) +{ + /* + * Possible clock sources: + * LCD1: FCK/PLL1_1/HDMI_PLL + * LCD2: FCK/PLL1_3/HDMI_PLL (DRA74x: PLL2_3) + * LCD3: FCK/PLL1_3/HDMI_PLL (DRA74x: PLL2_1) + */ + + switch (channel) { + case OMAP_DSS_CHANNEL_LCD: + { + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL1_1)) + return DSS_CLK_SRC_PLL1_1; + } + case OMAP_DSS_CHANNEL_LCD2: + { + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL1_3)) + return DSS_CLK_SRC_PLL1_3; + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL2_3)) + return DSS_CLK_SRC_PLL2_3; + } + case OMAP_DSS_CHANNEL_LCD3: + { + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL2_1)) + return DSS_CLK_SRC_PLL2_1; + if (dss_pll_find_by_src(DSS_CLK_SRC_PLL1_3)) + return DSS_CLK_SRC_PLL1_3; + } + default: + break; + } + + return DSS_CLK_SRC_FCK; +} + static enum dss_clk_source dpi_get_clk_src(enum omap_channel channel) { /* @@ -107,16 +143,7 @@ static enum dss_clk_source dpi_get_clk_src(enum omap_channel channel) } case OMAPDSS_VER_DRA7xx: - switch (channel) { - case OMAP_DSS_CHANNEL_LCD: - return DSS_CLK_SRC_PLL1_1; - case OMAP_DSS_CHANNEL_LCD2: - return DSS_CLK_SRC_PLL1_3; - case OMAP_DSS_CHANNEL_LCD3: - return DSS_CLK_SRC_PLL2_1; - default: - return DSS_CLK_SRC_FCK; - } + return dpi_get_clk_src_dra7xx(channel); default: return DSS_CLK_SRC_FCK;
The clock source selection for the LCD outputs is too hardcoded at the moment. For example, LCD3 is set to use PLL2_1, and PLL2 doesn't exist on DRA72x SoCs. There are quite many ways to configure the clocks, even using HDMI PLL for LCD outputs, but enabling full configuration of the clocks is rather tricky. This patch improves the situation a bit by checking if the PLL about to be used exists, and if not, tries another one. Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com> --- drivers/gpu/drm/omapdrm/dss/dpi.c | 47 ++++++++++++++++++++++++++++++--------- 1 file changed, 37 insertions(+), 10 deletions(-)