| Message ID | 1551950425-7346-1-git-send-email-abel.vesa@nxp.com (mailing list archive) |
|---|---|
| State | RFC |
| Headers | show |
| Series | [RFC] clk: imx: Allow re-parenting by default on set rate | expand |
Hi Abel, On Thu, Mar 07, 2019 at 09:20:37AM +0000, Abel Vesa wrote: > By default, the muxes should re-parent on set_rate. > This would allow the drivers to control only the leaf clock node, > leaving the rest to the clock driver, that way simplifying the > clock control. I am afraid of this change. Besides the rate there might be other reasons to choose one mux input over another, consider for example low power audio playback where we need one specific mux setting because it provides a clock which runs at low power mode. On the IPU on i.MX5/6 there are clocks being used as pixel clocks derived from different muxes. I don't think you want to pick an input clock just because it happens to deliver the best clock rate at that point in time, but really is shared with some other clock that changes its rate in the next moment. I have no concrete examples for things that break with this change, but I would be more confident if we change the behaviour explicitly only for the muxes that we have reviewed to cope with this change. Sascha
On 19-03-11 11:28:25, Sascha Hauer wrote: > Hi Abel, > > On Thu, Mar 07, 2019 at 09:20:37AM +0000, Abel Vesa wrote: > > By default, the muxes should re-parent on set_rate. > > This would allow the drivers to control only the leaf clock node, > > leaving the rest to the clock driver, that way simplifying the > > clock control. > > I am afraid of this change. Besides the rate there might be other > reasons to choose one mux input over another, consider for example low > power audio playback where we need one specific mux setting because it > provides a clock which runs at low power mode. > On the IPU on i.MX5/6 there are clocks being used as pixel clocks > derived from different muxes. I don't think you want to pick an input > clock just because it happens to deliver the best clock rate at that > point in time, but really is shared with some other clock that changes > its rate in the next moment. > > I have no concrete examples for things that break with this change, but > I would be more confident if we change the behaviour explicitly only for > the muxes that we have reviewed to cope with this change. > Fair enough. We could replace all the imx_clk_mux with imx_clk_mux_noreparent and after that we can independently switch the clocks that are safe (to switch) to imx_clk_mux (which would not have the noreparent flag set). The main idea is to simplify the clock control from drivers point of view. The end goal here would be to only have noreparent flag set for the clocks that need specific mux control. > Sascha > > -- > Pengutronix e.K. | | > Industrial Linux Solutions | https://emea01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.pengutronix.de%2F&data=02%7C01%7Cabel.vesa%40nxp.com%7Cde43ba7dcd774a3709df08d6a60c4d90%7C686ea1d3bc2b4c6fa92cd99c5c301635%7C0%7C0%7C636878969100157679&sdata=4eU6YmuXDhHJZUKVKJhiRwlzq7b9XIrUfLSl2JFlRd8%3D&reserved=0 | > Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0 | > Amtsgericht Hildesheim, HRA 2686 | Fax: +49-5121-206917-5555 |
On Mon, Mar 11, 2019 at 10:41:40AM +0000, Abel Vesa wrote: > On 19-03-11 11:28:25, Sascha Hauer wrote: > > Hi Abel, > > > > On Thu, Mar 07, 2019 at 09:20:37AM +0000, Abel Vesa wrote: > > > By default, the muxes should re-parent on set_rate. > > > This would allow the drivers to control only the leaf clock node, > > > leaving the rest to the clock driver, that way simplifying the > > > clock control. > > > > I am afraid of this change. Besides the rate there might be other > > reasons to choose one mux input over another, consider for example low > > power audio playback where we need one specific mux setting because it > > provides a clock which runs at low power mode. > > On the IPU on i.MX5/6 there are clocks being used as pixel clocks > > derived from different muxes. I don't think you want to pick an input > > clock just because it happens to deliver the best clock rate at that > > point in time, but really is shared with some other clock that changes > > its rate in the next moment. > > > > > I have no concrete examples for things that break with this change, but > > I would be more confident if we change the behaviour explicitly only for > > the muxes that we have reviewed to cope with this change. > > > > Fair enough. We could replace all the imx_clk_mux with imx_clk_mux_noreparent > and after that we can independently switch the clocks that are safe (to switch) > to imx_clk_mux (which would not have the noreparent flag set). Ok with me. > > The main idea is to simplify the clock control from drivers point of view. Which drivers do you have in mind? I hardly ever missed reparenting on rate changes, so where is this feature useful? Sascha
On Wed, Mar 13, 2019 at 12:42 PM Sascha Hauer <s.hauer@pengutronix.de> wrote: > > On Mon, Mar 11, 2019 at 10:41:40AM +0000, Abel Vesa wrote: > > On 19-03-11 11:28:25, Sascha Hauer wrote: > > > Hi Abel, > > > > > > On Thu, Mar 07, 2019 at 09:20:37AM +0000, Abel Vesa wrote: > > > > By default, the muxes should re-parent on set_rate. > > > > This would allow the drivers to control only the leaf clock node, > > > > leaving the rest to the clock driver, that way simplifying the > > > > clock control. > > > > > > I am afraid of this change. Besides the rate there might be other > > > reasons to choose one mux input over another, consider for example low > > > power audio playback where we need one specific mux setting because it > > > provides a clock which runs at low power mode. > > > On the IPU on i.MX5/6 there are clocks being used as pixel clocks > > > derived from different muxes. I don't think you want to pick an input > > > clock just because it happens to deliver the best clock rate at that > > > point in time, but really is shared with some other clock that changes > > > its rate in the next moment. > > > > > > > > I have no concrete examples for things that break with this change, but > > > I would be more confident if we change the behaviour explicitly only for > > > the muxes that we have reviewed to cope with this change. > > > > > > > Fair enough. We could replace all the imx_clk_mux with imx_clk_mux_noreparent > > and after that we can independently switch the clocks that are safe (to switch) > > to imx_clk_mux (which would not have the noreparent flag set). > > Ok with me. > > > > > The main idea is to simplify the clock control from drivers point of view. > > Which drivers do you have in mind? I hardly ever missed reparenting on > rate changes, so where is this feature useful? Hi Sascha, Reparenting is very useful for Audio IPs (SAI, SPDIF, PDM, FSL DSP). We often need clock rates that are multiples of 8000 or 11025 which can be obtain from different PLLs. But in order for this to work we need to have the correct parent. In our current tree we do the re parenting manually inside drivers. For example: https://source.codeaurora.org/external/imx/linux-imx/tree/sound/soc/fsl/fsl_sai.c?h=imx_4.14.78_1.0.0_ga#n244 This is far from optimal. Abel's patch will be of a great help for Audio. thanks, Daniel.
diff --git a/drivers/clk/imx/clk.h b/drivers/clk/imx/clk.h index 5748ec8..055b602 100644 --- a/drivers/clk/imx/clk.h +++ b/drivers/clk/imx/clk.h @@ -148,7 +148,7 @@ static inline struct clk *imx_clk_mux_ldb(const char *name, void __iomem *reg, int num_parents) { return clk_register_mux(NULL, name, parents, num_parents, - CLK_SET_RATE_NO_REPARENT | CLK_SET_RATE_PARENT, reg, + CLK_SET_RATE_PARENT, reg, shift, width, CLK_MUX_READ_ONLY, &imx_ccm_lock); } @@ -321,7 +321,7 @@ static inline struct clk *imx_clk_mux(const char *name, void __iomem *reg, int num_parents) { return clk_register_mux(NULL, name, parents, num_parents, - CLK_SET_RATE_NO_REPARENT, reg, shift, + 0, reg, shift, width, 0, &imx_ccm_lock); } @@ -330,7 +330,7 @@ static inline struct clk *imx_clk_mux2(const char *name, void __iomem *reg, int num_parents) { return clk_register_mux(NULL, name, parents, num_parents, - CLK_SET_RATE_NO_REPARENT | CLK_OPS_PARENT_ENABLE, + CLK_OPS_PARENT_ENABLE, reg, shift, width, 0, &imx_ccm_lock); } @@ -340,7 +340,6 @@ static inline struct clk_hw *imx_clk_hw_mux2(const char *name, void __iomem *reg int num_parents) { return clk_hw_register_mux(NULL, name, parents, num_parents, - CLK_SET_RATE_NO_REPARENT | CLK_OPS_PARENT_ENABLE, reg, shift, width, 0, &imx_ccm_lock); } @@ -351,7 +350,7 @@ static inline struct clk *imx_clk_mux_flags(const char *name, unsigned long flags) { return clk_register_mux(NULL, name, parents, num_parents, - flags | CLK_SET_RATE_NO_REPARENT, reg, shift, width, 0, + flags, reg, shift, width, 0, &imx_ccm_lock); } @@ -361,7 +360,7 @@ static inline struct clk *imx_clk_mux2_flags(const char *name, int num_parents, unsigned long flags) { return clk_register_mux(NULL, name, parents, num_parents, - flags | CLK_SET_RATE_NO_REPARENT | CLK_OPS_PARENT_ENABLE, + flags | CLK_OPS_PARENT_ENABLE, reg, shift, width, 0, &imx_ccm_lock); } @@ -373,7 +372,7 @@ static inline struct clk_hw *imx_clk_hw_mux_flags(const char *name, unsigned long flags) { return clk_hw_register_mux(NULL, name, parents, num_parents, - flags | CLK_SET_RATE_NO_REPARENT, + flags, reg, shift, width, 0, &imx_ccm_lock); } @@ -389,7 +388,7 @@ struct clk *imx8m_clk_composite_flags(const char *name, #define __imx8m_clk_composite(name, parent_names, reg, flags) \ imx8m_clk_composite_flags(name, parent_names, \ ARRAY_SIZE(parent_names), reg, \ - flags | CLK_SET_RATE_NO_REPARENT | CLK_OPS_PARENT_ENABLE) + flags | CLK_OPS_PARENT_ENABLE) #define imx8m_clk_composite(name, parent_names, reg) \ __imx8m_clk_composite(name, parent_names, reg, 0)
By default, the muxes should re-parent on set_rate. This would allow the drivers to control only the leaf clock node, leaving the rest to the clock driver, that way simplifying the clock control. Signed-off-by: Abel Vesa <abel.vesa@nxp.com> --- drivers/clk/imx/clk.h | 15 +++++++-------- 1 file changed, 7 insertions(+), 8 deletions(-)