There are two failures which DT can return when looking up a phandle:
-ENOENT - when DT finds that the desired property/index is not present
-EINVAL - when DT finds that the desired property/index is present, but
there is a problem parsing it.
We should only fall through to clk_get_sys() (the table driven clock
lookup) when DT indicates that there was no entry in the OF tables
for the clock. Doing otherwise causes the clk API to always indicate
that there is no entry for this clock, which is not correct behaviour.
Cc: <[email protected]>
Signed-off-by: Russell King <[email protected]>
---
Grant, Rob,
Please let me know ASAP if this gives you any cause for concern. This
looks to me like a long standing bug which really needs fixing. Part
of the motivation here is similar to Jean-Francois' patch to clkdev
which allows a DT specified clock which isn't get present to have
clk_get() and friends return -EPROBE_DEFER - again, something that I
think really should happen. Jean-Francois hasn't been able to get
much traction for his patch, so let's start with fixing this bug first.
drivers/clk/clkdev.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/clk/clkdev.c b/drivers/clk/clkdev.c
index 48f67218247c..4e65d0e10b05 100644
--- a/drivers/clk/clkdev.c
+++ b/drivers/clk/clkdev.c
@@ -165,7 +165,7 @@ struct clk *clk_get(struct device *dev, const char *con_id)
if (dev) {
clk = of_clk_get_by_name(dev->of_node, con_id);
- if (!IS_ERR(clk))
+ if (!IS_ERR(clk) || PTR_ERR(clk) != -ENOENT)
return clk;
}
--
1.8.3.1
On 04/26/2014 05:49 PM, Russell King wrote:
> There are two failures which DT can return when looking up a phandle:
> -ENOENT - when DT finds that the desired property/index is not present
> -EINVAL - when DT finds that the desired property/index is present, but
> there is a problem parsing it.
>
> We should only fall through to clk_get_sys() (the table driven clock
> lookup) when DT indicates that there was no entry in the OF tables
> for the clock. Doing otherwise causes the clk API to always indicate
> that there is no entry for this clock, which is not correct behaviour.
>
> Cc: <[email protected]>
> Signed-off-by: Russell King <[email protected]>
> ---
> Grant, Rob,
>
> Please let me know ASAP if this gives you any cause for concern. This
> looks to me like a long standing bug which really needs fixing. Part
> of the motivation here is similar to Jean-Francois' patch to clkdev
> which allows a DT specified clock which isn't get present to have
> clk_get() and friends return -EPROBE_DEFER - again, something that I
> think really should happen. Jean-Francois hasn't been able to get
> much traction for his patch, so let's start with fixing this bug first.
How about upstream commit a34cd4666f3da84228a82f70c94b8d9b692034ea,
I thought it already solves the issue you're trying to address in
this patch ?
> drivers/clk/clkdev.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/clk/clkdev.c b/drivers/clk/clkdev.c
> index 48f67218247c..4e65d0e10b05 100644
> --- a/drivers/clk/clkdev.c
> +++ b/drivers/clk/clkdev.c
> @@ -165,7 +165,7 @@ struct clk *clk_get(struct device *dev, const char *con_id)
>
> if (dev) {
> clk = of_clk_get_by_name(dev->of_node, con_id);
> - if (!IS_ERR(clk))
> + if (!IS_ERR(clk) || PTR_ERR(clk) != -ENOENT)
> return clk;
> }
--
Regards,
Sylwester
On Sun, Apr 27, 2014 at 09:08:54PM +0200, Sylwester Nawrocki wrote:
> On 04/26/2014 05:49 PM, Russell King wrote:
>> There are two failures which DT can return when looking up a phandle:
>> -ENOENT - when DT finds that the desired property/index is not present
>> -EINVAL - when DT finds that the desired property/index is present, but
>> there is a problem parsing it.
>>
>> We should only fall through to clk_get_sys() (the table driven clock
>> lookup) when DT indicates that there was no entry in the OF tables
>> for the clock. Doing otherwise causes the clk API to always indicate
>> that there is no entry for this clock, which is not correct behaviour.
>>
>> Cc: <[email protected]>
>> Signed-off-by: Russell King <[email protected]>
>> ---
>> Grant, Rob,
>>
>> Please let me know ASAP if this gives you any cause for concern. This
>> looks to me like a long standing bug which really needs fixing. Part
>> of the motivation here is similar to Jean-Francois' patch to clkdev
>> which allows a DT specified clock which isn't get present to have
>> clk_get() and friends return -EPROBE_DEFER - again, something that I
>> think really should happen. Jean-Francois hasn't been able to get
>> much traction for his patch, so let's start with fixing this bug first.
>
> How about upstream commit a34cd4666f3da84228a82f70c94b8d9b692034ea,
> I thought it already solves the issue you're trying to address in
> this patch ?
FFS. You know, I thought _I_ was the clkdev maintainer, not Mike.
That commit is wrong for the reasons I've stated above. Why should we
fall through to clk_get_sys() if the DT stuff returns -EINVAL?
Okay, I'll re-spin against that patch and fix this crap.
--
FTTC broadband for 0.8mile line: now at 9.7Mbps down 460kbps up... slowly
improving, and getting towards what was expected from it.
On Sat, Apr 26, 2014 at 10:49 AM, Russell King
<[email protected]> wrote:
> There are two failures which DT can return when looking up a phandle:
> -ENOENT - when DT finds that the desired property/index is not present
> -EINVAL - when DT finds that the desired property/index is present, but
> there is a problem parsing it.
>
> We should only fall through to clk_get_sys() (the table driven clock
> lookup) when DT indicates that there was no entry in the OF tables
> for the clock. Doing otherwise causes the clk API to always indicate
> that there is no entry for this clock, which is not correct behaviour.
>
> Cc: <[email protected]>
> Signed-off-by: Russell King <[email protected]>
> ---
> Grant, Rob,
>
> Please let me know ASAP if this gives you any cause for concern. This
> looks to me like a long standing bug which really needs fixing. Part
> of the motivation here is similar to Jean-Francois' patch to clkdev
> which allows a DT specified clock which isn't get present to have
> clk_get() and friends return -EPROBE_DEFER - again, something that I
> think really should happen. Jean-Francois hasn't been able to get
> much traction for his patch, so let's start with fixing this bug first.
I have no issues. A platform would have to be pretty broken to be
relying on the current behavior.
Rob
>
> drivers/clk/clkdev.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/clk/clkdev.c b/drivers/clk/clkdev.c
> index 48f67218247c..4e65d0e10b05 100644
> --- a/drivers/clk/clkdev.c
> +++ b/drivers/clk/clkdev.c
> @@ -165,7 +165,7 @@ struct clk *clk_get(struct device *dev, const char *con_id)
>
> if (dev) {
> clk = of_clk_get_by_name(dev->of_node, con_id);
> - if (!IS_ERR(clk))
> + if (!IS_ERR(clk) || PTR_ERR(clk) != -ENOENT)
> return clk;
> }
>
> --
> 1.8.3.1
>