| Message ID | 1404701631-13870-1-git-send-email-shawn.guo@freescale.com |
|---|---|
| State | New |
| Headers | show |
On Sun, Jul 6, 2014 at 11:53 PM, Shawn Guo <shawn.guo@freescale.com> wrote: > Let's say clock A and B are two gate clocks that share the same register > bit in hardware. Therefore they are registered as shared gate clocks > with imx_clk_gate2_shared(). > > In a scenario that only clock A is enabled by clk_enable(A) while B is > not used, the shared gate will be unexpectedly disabled in hardware. > It happens because clk_enable(A) increments the share_count from 0 to 1, > while clock B is unused to clock core, and therefore the core function > will just disable B by calling clk->ops->disable() directly. The > consequence of that call is share_count is decremented to 0 and the gate > is disabled in hardware, even though clock A is still in use. > > The patch fixes the issue by initializing the share_count per hardware > state and returns enable state per share_count from .is_enabled() hook, > in case it's a shared gate. > > While at it, add a check in clk_gate2_disable() to ensure it's never > called with a zero share_count. > > Reported-by: Fabio Estevam <fabio.estevam@freescale.com> > Fixes: f9f28cdf2167 ("ARM: imx: add shared gate clock support") > Signed-off-by: Shawn Guo <shawn.guo@freescale.com> > --- > Fabio, > > Per my testing, it fixes the shared gate clock issue you reported. > But I'd like to get your confirmation before I ask arm-soc folks to > apply it for 3.16-rc. I can get audio working now and I have also inspected CCGR5 register and it looks correct now: Tested-by: Fabio Estevam <fabio.estevam@freescale.com>
On Mon, Jul 07, 2014 at 09:49:00AM -0300, Fabio Estevam wrote: > On Sun, Jul 6, 2014 at 11:53 PM, Shawn Guo <shawn.guo@freescale.com> wrote: > > Let's say clock A and B are two gate clocks that share the same register > > bit in hardware. Therefore they are registered as shared gate clocks > > with imx_clk_gate2_shared(). > > > > In a scenario that only clock A is enabled by clk_enable(A) while B is > > not used, the shared gate will be unexpectedly disabled in hardware. > > It happens because clk_enable(A) increments the share_count from 0 to 1, > > while clock B is unused to clock core, and therefore the core function > > will just disable B by calling clk->ops->disable() directly. The > > consequence of that call is share_count is decremented to 0 and the gate > > is disabled in hardware, even though clock A is still in use. > > > > The patch fixes the issue by initializing the share_count per hardware > > state and returns enable state per share_count from .is_enabled() hook, > > in case it's a shared gate. > > > > While at it, add a check in clk_gate2_disable() to ensure it's never > > called with a zero share_count. > > > > Reported-by: Fabio Estevam <fabio.estevam@freescale.com> > > Fixes: f9f28cdf2167 ("ARM: imx: add shared gate clock support") > > Signed-off-by: Shawn Guo <shawn.guo@freescale.com> > > --- > > Fabio, > > > > Per my testing, it fixes the shared gate clock issue you reported. > > But I'd like to get your confirmation before I ask arm-soc folks to > > apply it for 3.16-rc. > > I can get audio working now and I have also inspected CCGR5 register > and it looks correct now: > > Tested-by: Fabio Estevam <fabio.estevam@freescale.com> Thanks, Fabio. Arnd, Olof, Can you please apply this patch for 3.16? Shawn
On Mon, Jul 07, 2014 at 09:56:37PM +0800, Shawn Guo wrote: > On Mon, Jul 07, 2014 at 09:49:00AM -0300, Fabio Estevam wrote: > > On Sun, Jul 6, 2014 at 11:53 PM, Shawn Guo <shawn.guo@freescale.com> wrote: > > > Let's say clock A and B are two gate clocks that share the same register > > > bit in hardware. Therefore they are registered as shared gate clocks > > > with imx_clk_gate2_shared(). > > > > > > In a scenario that only clock A is enabled by clk_enable(A) while B is > > > not used, the shared gate will be unexpectedly disabled in hardware. > > > It happens because clk_enable(A) increments the share_count from 0 to 1, > > > while clock B is unused to clock core, and therefore the core function > > > will just disable B by calling clk->ops->disable() directly. The > > > consequence of that call is share_count is decremented to 0 and the gate > > > is disabled in hardware, even though clock A is still in use. > > > > > > The patch fixes the issue by initializing the share_count per hardware > > > state and returns enable state per share_count from .is_enabled() hook, > > > in case it's a shared gate. > > > > > > While at it, add a check in clk_gate2_disable() to ensure it's never > > > called with a zero share_count. > > > > > > Reported-by: Fabio Estevam <fabio.estevam@freescale.com> > > > Fixes: f9f28cdf2167 ("ARM: imx: add shared gate clock support") > > > Signed-off-by: Shawn Guo <shawn.guo@freescale.com> > > > --- > > > Fabio, > > > > > > Per my testing, it fixes the shared gate clock issue you reported. > > > But I'd like to get your confirmation before I ask arm-soc folks to > > > apply it for 3.16-rc. > > > > I can get audio working now and I have also inspected CCGR5 register > > and it looks correct now: > > > > Tested-by: Fabio Estevam <fabio.estevam@freescale.com> > > Thanks, Fabio. > > Arnd, Olof, > > Can you please apply this patch for 3.16? Done, thanks. -Olof
diff --git a/arch/arm/mach-imx/clk-gate2.c b/arch/arm/mach-imx/clk-gate2.c index 4ba587da89d2..84acdfd1d715 100644 --- a/arch/arm/mach-imx/clk-gate2.c +++ b/arch/arm/mach-imx/clk-gate2.c @@ -67,8 +67,12 @@ static void clk_gate2_disable(struct clk_hw *hw) spin_lock_irqsave(gate->lock, flags); - if (gate->share_count && --(*gate->share_count) > 0) - goto out; + if (gate->share_count) { + if (WARN_ON(*gate->share_count == 0)) + goto out; + else if (--(*gate->share_count) > 0) + goto out; + } reg = readl(gate->reg); reg &= ~(3 << gate->bit_idx); @@ -78,19 +82,26 @@ out: spin_unlock_irqrestore(gate->lock, flags); } -static int clk_gate2_is_enabled(struct clk_hw *hw) +static int clk_gate2_reg_is_enabled(void __iomem *reg, u8 bit_idx) { - u32 reg; - struct clk_gate2 *gate = to_clk_gate2(hw); + u32 val = readl(reg); - reg = readl(gate->reg); - - if (((reg >> gate->bit_idx) & 1) == 1) + if (((val >> bit_idx) & 1) == 1) return 1; return 0; } +static int clk_gate2_is_enabled(struct clk_hw *hw) +{ + struct clk_gate2 *gate = to_clk_gate2(hw); + + if (gate->share_count) + return !!(*gate->share_count); + else + return clk_gate2_reg_is_enabled(gate->reg, gate->bit_idx); +} + static struct clk_ops clk_gate2_ops = { .enable = clk_gate2_enable, .disable = clk_gate2_disable, @@ -116,6 +127,10 @@ struct clk *clk_register_gate2(struct device *dev, const char *name, gate->bit_idx = bit_idx; gate->flags = clk_gate2_flags; gate->lock = lock; + + /* Initialize share_count per hardware state */ + if (share_count) + *share_count = clk_gate2_reg_is_enabled(reg, bit_idx) ? 1 : 0; gate->share_count = share_count; init.name = name;
Let's say clock A and B are two gate clocks that share the same register bit in hardware. Therefore they are registered as shared gate clocks with imx_clk_gate2_shared(). In a scenario that only clock A is enabled by clk_enable(A) while B is not used, the shared gate will be unexpectedly disabled in hardware. It happens because clk_enable(A) increments the share_count from 0 to 1, while clock B is unused to clock core, and therefore the core function will just disable B by calling clk->ops->disable() directly. The consequence of that call is share_count is decremented to 0 and the gate is disabled in hardware, even though clock A is still in use. The patch fixes the issue by initializing the share_count per hardware state and returns enable state per share_count from .is_enabled() hook, in case it's a shared gate. While at it, add a check in clk_gate2_disable() to ensure it's never called with a zero share_count. Reported-by: Fabio Estevam <fabio.estevam@freescale.com> Fixes: f9f28cdf2167 ("ARM: imx: add shared gate clock support") Signed-off-by: Shawn Guo <shawn.guo@freescale.com> --- Fabio, Per my testing, it fixes the shared gate clock issue you reported. But I'd like to get your confirmation before I ask arm-soc folks to apply it for 3.16-rc. Shawn arch/arm/mach-imx/clk-gate2.c | 31 +++++++++++++++++++++++-------- 1 file changed, 23 insertions(+), 8 deletions(-)