| Message ID | 1389236699-10387-1-git-send-email-Guangyu.Chen@freescale.com |
|---|---|
| State | New |
| Headers | show |
On Thu, Jan 09, 2014 at 11:58:12AM +0800, Shawn Guo wrote: > > static struct clk *clk[clk_max]; > > @@ -355,6 +355,7 @@ static void __init imx6q_clocks_init(struct device_node *ccm_node) > > clk[ecspi5] = imx_clk_gate2("ecspi5", "ecspi_root", base + 0x6c, 8); > > clk[enet] = imx_clk_gate2("enet", "ipg", base + 0x6c, 10); > > clk[esai] = imx_clk_gate2("esai", "esai_podf", base + 0x6c, 16); > > + clk[esai_ahb] = imx_clk_gate2("esai_ahb", "ahb", base + 0x6c, 16); > > Hmm, having two clocks operating on the same gate bit will get us > problem in clock disabling. Clock enabling is fine, since either > one who calls clk_enable() first will just set the gate bit. But in > case that clk_enable() is called on both clocks, and then when either > clock calls clk_disable(), the gate bit will be cleared and thus breaks > the other one that might still be in use. Understood. But how could we handle this situation? The only way I can figure out is to make sure the driver open/close them at the same time, it's not a perfect way though. Nicolin
On Thu, Jan 09, 2014 at 11:04:59AM +0800, Nicolin Chen wrote: > esai_ahb clock is derived from ahb and used to provide ESAI the capability of > register accessing and FSYS clock source for I2S clocks dividing. Although the > gate of this esai_ahb is duplicated with esai clock -- the baud clock, yet > considering about the differences of their clock rates, it's quite essential > to patch this missing clock. > > Signed-off-by: Nicolin Chen <Guangyu.Chen@freescale.com> > --- > Documentation/devicetree/bindings/clock/imx6q-clock.txt | 1 + > arch/arm/mach-imx/clk-imx6q.c | 3 ++- > 2 files changed, 3 insertions(+), 1 deletion(-) > > diff --git a/Documentation/devicetree/bindings/clock/imx6q-clock.txt b/Documentation/devicetree/bindings/clock/imx6q-clock.txt > index 6aab72b..90ec91f 100644 > --- a/Documentation/devicetree/bindings/clock/imx6q-clock.txt > +++ b/Documentation/devicetree/bindings/clock/imx6q-clock.txt > @@ -220,6 +220,7 @@ clocks and IDs. > lvds2_sel 205 > lvds1_gate 206 > lvds2_gate 207 > + esai_ahb 208 > > Examples: > > diff --git a/arch/arm/mach-imx/clk-imx6q.c b/arch/arm/mach-imx/clk-imx6q.c > index af2e582..20215b9 100644 > --- a/arch/arm/mach-imx/clk-imx6q.c > +++ b/arch/arm/mach-imx/clk-imx6q.c > @@ -107,7 +107,7 @@ enum mx6q_clks { > sata_ref, sata_ref_100m, pcie_ref, pcie_ref_125m, enet_ref, usbphy1_gate, > usbphy2_gate, pll4_post_div, pll5_post_div, pll5_video_div, eim_slow, > spdif, cko2_sel, cko2_podf, cko2, cko, vdoa, pll4_audio_div, > - lvds1_sel, lvds2_sel, lvds1_gate, lvds2_gate, clk_max > + lvds1_sel, lvds2_sel, lvds1_gate, lvds2_gate, esai_ahb, clk_max > }; > > static struct clk *clk[clk_max]; > @@ -355,6 +355,7 @@ static void __init imx6q_clocks_init(struct device_node *ccm_node) > clk[ecspi5] = imx_clk_gate2("ecspi5", "ecspi_root", base + 0x6c, 8); > clk[enet] = imx_clk_gate2("enet", "ipg", base + 0x6c, 10); > clk[esai] = imx_clk_gate2("esai", "esai_podf", base + 0x6c, 16); > + clk[esai_ahb] = imx_clk_gate2("esai_ahb", "ahb", base + 0x6c, 16); Hmm, having two clocks operating on the same gate bit will get us problem in clock disabling. Clock enabling is fine, since either one who calls clk_enable() first will just set the gate bit. But in case that clk_enable() is called on both clocks, and then when either clock calls clk_disable(), the gate bit will be cleared and thus breaks the other one that might still be in use. Shawn > clk[gpt_ipg] = imx_clk_gate2("gpt_ipg", "ipg", base + 0x6c, 20); > clk[gpt_ipg_per] = imx_clk_gate2("gpt_ipg_per", "ipg_per", base + 0x6c, 22); > if (cpu_is_imx6dl()) > -- > 1.8.4 > >
On Thu, Jan 09, 2014 at 11:49:41AM +0800, Nicolin Chen wrote: > On Thu, Jan 09, 2014 at 11:58:12AM +0800, Shawn Guo wrote: > > > static struct clk *clk[clk_max]; > > > @@ -355,6 +355,7 @@ static void __init imx6q_clocks_init(struct device_node *ccm_node) > > > clk[ecspi5] = imx_clk_gate2("ecspi5", "ecspi_root", base + 0x6c, 8); > > > clk[enet] = imx_clk_gate2("enet", "ipg", base + 0x6c, 10); > > > clk[esai] = imx_clk_gate2("esai", "esai_podf", base + 0x6c, 16); > > > + clk[esai_ahb] = imx_clk_gate2("esai_ahb", "ahb", base + 0x6c, 16); > > > > Hmm, having two clocks operating on the same gate bit will get us > > problem in clock disabling. Clock enabling is fine, since either > > one who calls clk_enable() first will just set the gate bit. But in > > case that clk_enable() is called on both clocks, and then when either > > clock calls clk_disable(), the gate bit will be cleared and thus breaks > > the other one that might still be in use. > > Understood. But how could we handle this situation? The only way I can figure > out is to make sure the driver open/close them at the same time, it's not a > perfect way though. Hmm, we generally leave the gate bit to the clock used to access register, because usually it's the first one to be on and the last one to be off. Shawn
On Thu, Jan 09, 2014 at 02:57:42PM +0800, Shawn Guo wrote: > On Thu, Jan 09, 2014 at 11:49:41AM +0800, Nicolin Chen wrote: > > On Thu, Jan 09, 2014 at 11:58:12AM +0800, Shawn Guo wrote: > > > > static struct clk *clk[clk_max]; > > > > @@ -355,6 +355,7 @@ static void __init imx6q_clocks_init(struct device_node *ccm_node) > > > > clk[ecspi5] = imx_clk_gate2("ecspi5", "ecspi_root", base + 0x6c, 8); > > > > clk[enet] = imx_clk_gate2("enet", "ipg", base + 0x6c, 10); > > > > clk[esai] = imx_clk_gate2("esai", "esai_podf", base + 0x6c, 16); > > > > + clk[esai_ahb] = imx_clk_gate2("esai_ahb", "ahb", base + 0x6c, 16); > > > > > > Hmm, having two clocks operating on the same gate bit will get us > > > problem in clock disabling. Clock enabling is fine, since either > > > one who calls clk_enable() first will just set the gate bit. But in > > > case that clk_enable() is called on both clocks, and then when either > > > clock calls clk_disable(), the gate bit will be cleared and thus breaks > > > the other one that might still be in use. > > > > Understood. But how could we handle this situation? The only way I can figure > > out is to make sure the driver open/close them at the same time, it's not a > > perfect way though. > > Hmm, we generally leave the gate bit to the clock used to access > register, because usually it's the first one to be on and the last one > to be off. Then we should attach CLK_IGNORE_UNUSED to clk[esai] since clk[esai_ahb] is the clock used to access memory, shouldn't we? Thank you. Nicolin
On Thu, Jan 09, 2014 at 08:58:28AM +0100, Sascha Hauer wrote: > On Thu, Jan 09, 2014 at 03:41:38PM +0800, Nicolin Chen wrote: > > On Thu, Jan 09, 2014 at 02:57:42PM +0800, Shawn Guo wrote: > > > On Thu, Jan 09, 2014 at 11:49:41AM +0800, Nicolin Chen wrote: > > > > On Thu, Jan 09, 2014 at 11:58:12AM +0800, Shawn Guo wrote: > > > > > > static struct clk *clk[clk_max]; > > > > > > @@ -355,6 +355,7 @@ static void __init imx6q_clocks_init(struct device_node *ccm_node) > > > > > > clk[ecspi5] = imx_clk_gate2("ecspi5", "ecspi_root", base + 0x6c, 8); > > > > > > clk[enet] = imx_clk_gate2("enet", "ipg", base + 0x6c, 10); > > > > > > clk[esai] = imx_clk_gate2("esai", "esai_podf", base + 0x6c, 16); > > > > > > + clk[esai_ahb] = imx_clk_gate2("esai_ahb", "ahb", base + 0x6c, 16); > > > > > > > > > > Hmm, having two clocks operating on the same gate bit will get us > > > > > problem in clock disabling. Clock enabling is fine, since either > > > > > one who calls clk_enable() first will just set the gate bit. But in > > > > > case that clk_enable() is called on both clocks, and then when either > > > > > clock calls clk_disable(), the gate bit will be cleared and thus breaks > > > > > the other one that might still be in use. > > > > > > > > Understood. But how could we handle this situation? The only way I can figure > > > > out is to make sure the driver open/close them at the same time, it's not a > > > > perfect way though. > > > > > > Hmm, we generally leave the gate bit to the clock used to access > > > register, because usually it's the first one to be on and the last one > > > to be off. > > > > Then we should attach CLK_IGNORE_UNUSED to clk[esai] since clk[esai_ahb] is > > the clock used to access memory, shouldn't we? > > Please wait for Mikes input or let's look how a proper solution can look > like. I've already seen the case that a single bit controls multiple > clocks. Hacking around this issue each time is not a solution. Okay. Thank you, Sascha.
[Added Mike Turquette to Cc] On Thu, Jan 09, 2014 at 11:04:59AM +0800, Nicolin Chen wrote: > esai_ahb clock is derived from ahb and used to provide ESAI the capability of > register accessing and FSYS clock source for I2S clocks dividing. Although the > gate of this esai_ahb is duplicated with esai clock -- the baud clock, yet > considering about the differences of their clock rates, it's quite essential > to patch this missing clock. > [...] > static struct clk *clk[clk_max]; > @@ -355,6 +355,7 @@ static void __init imx6q_clocks_init(struct device_node *ccm_node) > clk[ecspi5] = imx_clk_gate2("ecspi5", "ecspi_root", base + 0x6c, 8); > clk[enet] = imx_clk_gate2("enet", "ipg", base + 0x6c, 10); > clk[esai] = imx_clk_gate2("esai", "esai_podf", base + 0x6c, 16); > + clk[esai_ahb] = imx_clk_gate2("esai_ahb", "ahb", base + 0x6c, 16); We have the situation here that a single bit controls two clocks. As Shawn mentioned just using two gates on the same bit doesn't work properly. Do we need a new basic clock type or expand the common gate code somehow? This situation happens from time to time and I haven't seen a solution for this. Sascha
On Thu, Jan 09, 2014 at 03:41:38PM +0800, Nicolin Chen wrote: > On Thu, Jan 09, 2014 at 02:57:42PM +0800, Shawn Guo wrote: > > On Thu, Jan 09, 2014 at 11:49:41AM +0800, Nicolin Chen wrote: > > > On Thu, Jan 09, 2014 at 11:58:12AM +0800, Shawn Guo wrote: > > > > > static struct clk *clk[clk_max]; > > > > > @@ -355,6 +355,7 @@ static void __init imx6q_clocks_init(struct device_node *ccm_node) > > > > > clk[ecspi5] = imx_clk_gate2("ecspi5", "ecspi_root", base + 0x6c, 8); > > > > > clk[enet] = imx_clk_gate2("enet", "ipg", base + 0x6c, 10); > > > > > clk[esai] = imx_clk_gate2("esai", "esai_podf", base + 0x6c, 16); > > > > > + clk[esai_ahb] = imx_clk_gate2("esai_ahb", "ahb", base + 0x6c, 16); > > > > > > > > Hmm, having two clocks operating on the same gate bit will get us > > > > problem in clock disabling. Clock enabling is fine, since either > > > > one who calls clk_enable() first will just set the gate bit. But in > > > > case that clk_enable() is called on both clocks, and then when either > > > > clock calls clk_disable(), the gate bit will be cleared and thus breaks > > > > the other one that might still be in use. > > > > > > Understood. But how could we handle this situation? The only way I can figure > > > out is to make sure the driver open/close them at the same time, it's not a > > > perfect way though. > > > > Hmm, we generally leave the gate bit to the clock used to access > > register, because usually it's the first one to be on and the last one > > to be off. > > Then we should attach CLK_IGNORE_UNUSED to clk[esai] since clk[esai_ahb] is > the clock used to access memory, shouldn't we? Please wait for Mikes input or let's look how a proper solution can look like. I've already seen the case that a single bit controls multiple clocks. Hacking around this issue each time is not a solution. Sascha
On Thu, Jan 09, 2014 at 08:55 +0100, Sascha Hauer wrote: > > [ ... ] > > We have the situation here that a single bit controls two clocks. As > Shawn mentioned just using two gates on the same bit doesn't work > properly. Do we need a new basic clock type or expand the common gate > code somehow? > This situation happens from time to time and I haven't seen a solution > for this. You may want to lookup the following message: Date: Tue, 23 Jul 2013 15:14:06 +0200 From: Gerhard Sittig <gsi@denx.de> To: linuxppc-dev@lists.ozlabs.org, Anatolij Gustschin <agust@denx.de>, Mike Turquette <mturquette@linaro.org>, linux-arm-kernel@lists.infradead.org, devicetree@vger.kernel.org Cc: [ ... ] Subject: Re: [PATCH v3 17/31] clk: mpc512x: introduce COMMON_CLK for MPC512x http://lists.infradead.org/pipermail/linux-arm-kernel/2013-July/185687.html The specific situation was for MS-CAN on PowerPC, but it inspired my outlining an approach to "shared clock gates". See an example implementation towards the end of the message with both the clk-gate.c extension, as well as rather generic example use. My approach turned out to not be needed, but it might serve as a starting point for you. You'd have to add support for static declaration though, but this should be straight forward. virtually yours Gerhard Sittig
Sascha, your messages have the Mail-Followup-To: header...
Mail-Followup-To: Nicolin Chen <Guangyu.Chen@freescale.com>,
shawn.guo@linaro.org, kernel@pengutronix.de, linux@arm.linux.org.uk,
linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org,
rob.herring@calxeda.com, pawel.moll@arm.com, mark.rutland@arm.com,
ijc+devicetree@hellion.org.uk, galak@codeaurora.org,
devicetree@vger.kernel.org, rob@landley.net, mturquette@linaro.org
resulting in me (and others) getting mails marked as To: me where they're
not supposed to be To: me in the first place.
Russell, On Thu, Jan 09, 2014 at 03:01:41PM +0000, Russell King - ARM Linux wrote: > Sascha, your messages have the Mail-Followup-To: header... > > Mail-Followup-To: Nicolin Chen <Guangyu.Chen@freescale.com>, > shawn.guo@linaro.org, kernel@pengutronix.de, linux@arm.linux.org.uk, > linux-arm-kernel@lists.infradead.org, linux-kernel@vger.kernel.org, > rob.herring@calxeda.com, pawel.moll@arm.com, mark.rutland@arm.com, > ijc+devicetree@hellion.org.uk, galak@codeaurora.org, > devicetree@vger.kernel.org, rob@landley.net, mturquette@linaro.org > > resulting in me (and others) getting mails marked as To: me where they're > not supposed to be To: me in the first place. This should be fixed with this mail. Sascha
diff --git a/Documentation/devicetree/bindings/clock/imx6q-clock.txt b/Documentation/devicetree/bindings/clock/imx6q-clock.txt index 6aab72b..90ec91f 100644 --- a/Documentation/devicetree/bindings/clock/imx6q-clock.txt +++ b/Documentation/devicetree/bindings/clock/imx6q-clock.txt @@ -220,6 +220,7 @@ clocks and IDs. lvds2_sel 205 lvds1_gate 206 lvds2_gate 207 + esai_ahb 208 Examples: diff --git a/arch/arm/mach-imx/clk-imx6q.c b/arch/arm/mach-imx/clk-imx6q.c index af2e582..20215b9 100644 --- a/arch/arm/mach-imx/clk-imx6q.c +++ b/arch/arm/mach-imx/clk-imx6q.c @@ -107,7 +107,7 @@ enum mx6q_clks { sata_ref, sata_ref_100m, pcie_ref, pcie_ref_125m, enet_ref, usbphy1_gate, usbphy2_gate, pll4_post_div, pll5_post_div, pll5_video_div, eim_slow, spdif, cko2_sel, cko2_podf, cko2, cko, vdoa, pll4_audio_div, - lvds1_sel, lvds2_sel, lvds1_gate, lvds2_gate, clk_max + lvds1_sel, lvds2_sel, lvds1_gate, lvds2_gate, esai_ahb, clk_max }; static struct clk *clk[clk_max]; @@ -355,6 +355,7 @@ static void __init imx6q_clocks_init(struct device_node *ccm_node) clk[ecspi5] = imx_clk_gate2("ecspi5", "ecspi_root", base + 0x6c, 8); clk[enet] = imx_clk_gate2("enet", "ipg", base + 0x6c, 10); clk[esai] = imx_clk_gate2("esai", "esai_podf", base + 0x6c, 16); + clk[esai_ahb] = imx_clk_gate2("esai_ahb", "ahb", base + 0x6c, 16); clk[gpt_ipg] = imx_clk_gate2("gpt_ipg", "ipg", base + 0x6c, 20); clk[gpt_ipg_per] = imx_clk_gate2("gpt_ipg_per", "ipg_per", base + 0x6c, 22); if (cpu_is_imx6dl())
esai_ahb clock is derived from ahb and used to provide ESAI the capability of register accessing and FSYS clock source for I2S clocks dividing. Although the gate of this esai_ahb is duplicated with esai clock -- the baud clock, yet considering about the differences of their clock rates, it's quite essential to patch this missing clock. Signed-off-by: Nicolin Chen <Guangyu.Chen@freescale.com> --- Documentation/devicetree/bindings/clock/imx6q-clock.txt | 1 + arch/arm/mach-imx/clk-imx6q.c | 3 ++- 2 files changed, 3 insertions(+), 1 deletion(-)