| Message ID | 1486937212-10093-1-git-send-email-pierre-louis.bossart@linux.intel.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
On 02/12, Pierre-Louis Bossart wrote: > Due to timing requirements, TI and Conexant manage the audio > reference clock from their ASoC codec drivers using the "mclk" > string. This patch adds a default alias to "pmc_plt_clk_3" to > avoid Intel-specific tests in those codec drivers and use code > as-is. > "pmc_plt_clk_3" is used exclusively for audio on all known > Baytrail/CherryTrail designs and is e.g. routed on the MCLK > (pin 26) of the MinnowBoardMAX Turbot LSE connector. > > Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> > --- > > Notes: > This patch only applies on top of linux-next or clk-next, the > previous clk-related patches are not yet merged in the audio > trees > > drivers/clk/x86/clk-pmc-atom.c | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/drivers/clk/x86/clk-pmc-atom.c b/drivers/clk/x86/clk-pmc-atom.c > index 2b60577..ba00d27 100644 > --- a/drivers/clk/x86/clk-pmc-atom.c > +++ b/drivers/clk/x86/clk-pmc-atom.c > @@ -340,6 +340,8 @@ static int plt_clk_probe(struct platform_device *pdev) > > plt_clk_free_parent_names_loop(parent_names, data->nparents); > > + clk_add_alias("mclk", dev_name(&pdev->dev), "pmc_plt_clk_3", NULL); > + This leaks an alias when the driver is removed. Honestly, clk_add_alias() doesn't work well because of that problem. Can you just add another lookup with the pointer you already have instead of passing a NULL device to do a global lookup?
On 2/14/17 1:04 PM, Stephen Boyd wrote: > On 02/12, Pierre-Louis Bossart wrote: >> Due to timing requirements, TI and Conexant manage the audio >> reference clock from their ASoC codec drivers using the "mclk" >> string. This patch adds a default alias to "pmc_plt_clk_3" to >> avoid Intel-specific tests in those codec drivers and use code >> as-is. >> "pmc_plt_clk_3" is used exclusively for audio on all known >> Baytrail/CherryTrail designs and is e.g. routed on the MCLK >> (pin 26) of the MinnowBoardMAX Turbot LSE connector. >> >> Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> >> --- >> >> Notes: >> This patch only applies on top of linux-next or clk-next, the >> previous clk-related patches are not yet merged in the audio >> trees >> >> drivers/clk/x86/clk-pmc-atom.c | 2 ++ >> 1 file changed, 2 insertions(+) >> >> diff --git a/drivers/clk/x86/clk-pmc-atom.c b/drivers/clk/x86/clk-pmc-atom.c >> index 2b60577..ba00d27 100644 >> --- a/drivers/clk/x86/clk-pmc-atom.c >> +++ b/drivers/clk/x86/clk-pmc-atom.c >> @@ -340,6 +340,8 @@ static int plt_clk_probe(struct platform_device *pdev) >> >> plt_clk_free_parent_names_loop(parent_names, data->nparents); >> >> + clk_add_alias("mclk", dev_name(&pdev->dev), "pmc_plt_clk_3", NULL); >> + > > This leaks an alias when the driver is removed. Honestly, > clk_add_alias() doesn't work well because of that problem. Can > you just add another lookup with the pointer you already have > instead of passing a NULL device to do a global lookup? This is a builtin driver that cannot be configured as a module, is the leaked alias problematic? I don't mind trying something different but I am not familiar enough with the framework to understand what you are hinting at. Are you suggesting a change on the last parameter such as: clk_add_alias("mclk", dev_name(&pdev->dev), "pmc_plt_clk_3", &pdev->dev) Thanks! -Pierre -- To unsubscribe from this list: send the line "unsubscribe linux-clk" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On 02/14, Pierre-Louis Bossart wrote: > On 2/14/17 1:04 PM, Stephen Boyd wrote: > > > >This leaks an alias when the driver is removed. Honestly, > >clk_add_alias() doesn't work well because of that problem. Can > >you just add another lookup with the pointer you already have > >instead of passing a NULL device to do a global lookup? > > This is a builtin driver that cannot be configured as a module, is > the leaked alias problematic? Well we don't suppress driver unbinding via sysfs here, unless I missed something, so the leak could be triggered that way. > I don't mind trying something different but I am not familiar enough > with the framework to understand what you are hinting at. Are you > suggesting a change on the last parameter such as: > > clk_add_alias("mclk", dev_name(&pdev->dev), "pmc_plt_clk_3", &pdev->dev) > I mean: lookup = clkdev_hw_create(data->clks[3].hw, "mclk", NULL); and then freeing that lookup with clkdev_drop in the remove of the driver. "mclk" is really generic for a connection name without an associated device id, so you may want to pass some device as the last argument here, but dev_name(&pdev->dev) seems odd because that's the clock controller device, not whatever device would be calling clk_get() with this created lookup in mind.
On Thu, Feb 16, 2017 at 2:56 AM, Stephen Boyd <sboyd@codeaurora.org> wrote: > On 02/14, Pierre-Louis Bossart wrote: >> On 2/14/17 1:04 PM, Stephen Boyd wrote: >> This is a builtin driver that cannot be configured as a module, is >> the leaked alias problematic? > > Well we don't suppress driver unbinding via sysfs here, unless I > missed something, so the leak could be triggered that way. It has ->remove() hook, so, definitely it's possible to unbind. > lookup = clkdev_hw_create(data->clks[3].hw, "mclk", NULL); > > and then freeing that lookup with clkdev_drop in the remove of > the driver. "mclk" is really generic for a connection name > without an associated device id, so you may want to pass some > device as the last argument here, but dev_name(&pdev->dev) seems > odd because that's the clock controller device, not whatever > device would be calling clk_get() with this created lookup in > mind. Here is the problem with CLK framework since it's (was?) too oriented to ARM and clock trees. The best option for x86 is to place it in some platform code, which is bad idea per se. For example, GPIO, PWM, etc have static look up tables and glue drivers (like PCI part of the driver, or I2C, or similar) can fill that table based on ID. My personal opinion that the best place is not the clock driver for that, rather converting existing drivers to take clock name based on some (platform related) ID.
On Thu, Feb 16, 2017 at 04:05:19AM +0200, Andy Shevchenko wrote: > On Thu, Feb 16, 2017 at 2:56 AM, Stephen Boyd <sboyd@codeaurora.org> wrote: > > and then freeing that lookup with clkdev_drop in the remove of > > the driver. "mclk" is really generic for a connection name > > without an associated device id, so you may want to pass some > > device as the last argument here, but dev_name(&pdev->dev) seems > > odd because that's the clock controller device, not whatever > > device would be calling clk_get() with this created lookup in > > mind. > Here is the problem with CLK framework since it's (was?) too oriented > to ARM and clock trees. That's not really much more of a problem for x86 than it is for ARM, things got easier for ARM with the introduction of device tree but this stuff all predates device tree. > The best option for x86 is to place it in some platform code, which is > bad idea per se. > For example, GPIO, PWM, etc have static look up tables and glue > drivers (like PCI part of the driver, or I2C, or similar) can fill > that table based on ID. Platform code seems fine? > My personal opinion that the best place is not the clock driver for > that, rather converting existing drivers to take clock name based on > some (platform related) ID. That doesn't scale so well, you end up needing the platform code anyway and it's one more thing to go wrong when configuring the platform code so you may as well just do something idiomatic with the platform code. Some systems did historically do things like this, that's where the use of the device pointer for namespacing came from.
diff --git a/drivers/clk/x86/clk-pmc-atom.c b/drivers/clk/x86/clk-pmc-atom.c index 2b60577..ba00d27 100644 --- a/drivers/clk/x86/clk-pmc-atom.c +++ b/drivers/clk/x86/clk-pmc-atom.c @@ -340,6 +340,8 @@ static int plt_clk_probe(struct platform_device *pdev) plt_clk_free_parent_names_loop(parent_names, data->nparents); + clk_add_alias("mclk", dev_name(&pdev->dev), "pmc_plt_clk_3", NULL); + platform_set_drvdata(pdev, data); return 0;
Due to timing requirements, TI and Conexant manage the audio reference clock from their ASoC codec drivers using the "mclk" string. This patch adds a default alias to "pmc_plt_clk_3" to avoid Intel-specific tests in those codec drivers and use code as-is. "pmc_plt_clk_3" is used exclusively for audio on all known Baytrail/CherryTrail designs and is e.g. routed on the MCLK (pin 26) of the MinnowBoardMAX Turbot LSE connector. Signed-off-by: Pierre-Louis Bossart <pierre-louis.bossart@linux.intel.com> --- Notes: This patch only applies on top of linux-next or clk-next, the previous clk-related patches are not yet merged in the audio trees drivers/clk/x86/clk-pmc-atom.c | 2 ++ 1 file changed, 2 insertions(+)