| Message ID | 55e59c4792d64ff6336fcaa85ec15590553e9d63.1659903516.git.christophe.jaillet@wanadoo.fr (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | ASoC: Intel: kbl_rt5663_max98927: Simplify clk_get() usage | expand |
On 8/7/22 22:18, Christophe JAILLET wrote: > If clk_get() returns -ENOENT, there is no need to defer the driver, -ENOENT > will be returned the same for each retries. > So, return the error code directly instead of -EPROBE_DEFER. > > Remove this special case and use dev_err_probe() to simplify code. It will > also be less verbose if the clk is really deferred. > > Fixes: f7f61e08fe58 ("ASoC: Intel: kbl: Enable mclk and ssp sclk early") > Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> > --- > This is based on my understanding of clk_get(). > Review with care. > > Not sure the Fixes tag is needed. The patch does not fix anything. > If devm_clk_get() returns -ENOENT, it will just loop several time until > the framework gives up. > If it returns -EPROBE_DEFER, this case is already handled by the > "return ret;" > > So this patch should be a no-op, just a clean-up. I can't pretend understanding the clk framework in depth, but the only case where -ENOENT is returned seems to be this block in clk_hw_create_clk() if (!try_module_get(core->owner)) { free_clk(clk); return ERR_PTR(-ENOENT); } I have no idea why this would be converted to a -EPROBE_DEFER. May to account for module loading? > --- > sound/soc/intel/boards/kbl_rt5663_max98927.c | 31 ++++---------------- > 1 file changed, 6 insertions(+), 25 deletions(-) > > diff --git a/sound/soc/intel/boards/kbl_rt5663_max98927.c b/sound/soc/intel/boards/kbl_rt5663_max98927.c > index 2d4224c5b152..07b00af2fa3c 100644 > --- a/sound/soc/intel/boards/kbl_rt5663_max98927.c > +++ b/sound/soc/intel/boards/kbl_rt5663_max98927.c > @@ -989,7 +989,6 @@ static int kabylake_audio_probe(struct platform_device *pdev) > { > struct kbl_rt5663_private *ctx; > struct snd_soc_acpi_mach *mach; > - int ret; > > ctx = devm_kzalloc(&pdev->dev, sizeof(*ctx), GFP_KERNEL); > if (!ctx) > @@ -1009,32 +1008,14 @@ static int kabylake_audio_probe(struct platform_device *pdev) > &constraints_dmic_2ch : &constraints_dmic_channels; > > ctx->mclk = devm_clk_get(&pdev->dev, "ssp1_mclk"); > - if (IS_ERR(ctx->mclk)) { > - ret = PTR_ERR(ctx->mclk); > - if (ret == -ENOENT) { > - dev_info(&pdev->dev, > - "Failed to get ssp1_sclk, defer probe\n"); > - return -EPROBE_DEFER; > - } > - > - dev_err(&pdev->dev, "Failed to get ssp1_mclk with err:%d\n", > - ret); > - return ret; > - } > + if (IS_ERR(ctx->mclk)) > + return dev_err_probe(&pdev->dev, PTR_ERR(ctx->mclk), > + "Failed to get ssp1_mclk\n"); > > ctx->sclk = devm_clk_get(&pdev->dev, "ssp1_sclk"); > - if (IS_ERR(ctx->sclk)) { > - ret = PTR_ERR(ctx->sclk); > - if (ret == -ENOENT) { > - dev_info(&pdev->dev, > - "Failed to get ssp1_sclk, defer probe\n"); > - return -EPROBE_DEFER; > - } > - > - dev_err(&pdev->dev, "Failed to get ssp1_sclk with err:%d\n", > - ret); > - return ret; > - } > + if (IS_ERR(ctx->sclk)) > + return dev_err_probe(&pdev->dev, PTR_ERR(ctx->sclk), > + "Failed to get ssp1_sclk\n"); > > return devm_snd_soc_register_card(&pdev->dev, kabylake_audio_card); > }
On Sun, Aug 07, 2022 at 10:18:54PM +0200, Christophe JAILLET wrote:
> Not sure the Fixes tag is needed. The patch does not fix anything.
You just answered your own question there...
On Sun, Aug 07, 2022 at 10:18:54PM +0200, Christophe JAILLET wrote: > If clk_get() returns -ENOENT, there is no need to defer the driver, -ENOENT > will be returned the same for each retries. > So, return the error code directly instead of -EPROBE_DEFER. Are you *sure* that this is the case on Intel platforms where we don't have a full firmware description for clocks and therefore can't identify cases where we expect a clock to at some point to become available.
Le 10/08/2022 à 15:50, Mark Brown a écrit : > On Sun, Aug 07, 2022 at 10:18:54PM +0200, Christophe JAILLET wrote: >> If clk_get() returns -ENOENT, there is no need to defer the driver, -ENOENT >> will be returned the same for each retries. >> So, return the error code directly instead of -EPROBE_DEFER. > > Are you *sure* that this is the case on Intel platforms where we don't > have a full firmware description for clocks and therefore can't identify > cases where we expect a clock to at some point to become available. No, I'm *not* sure. This looked odd enough to deserve a patch proposal, that's all. (based on my grep and coccinelle scripts, this is the only place in the kernel where the result of a clk_get() is handled that way) There are many intel.com in cc:. Would be nice if s.o. could confirm if the patch is valid or not. CJ
diff --git a/sound/soc/intel/boards/kbl_rt5663_max98927.c b/sound/soc/intel/boards/kbl_rt5663_max98927.c index 2d4224c5b152..07b00af2fa3c 100644 --- a/sound/soc/intel/boards/kbl_rt5663_max98927.c +++ b/sound/soc/intel/boards/kbl_rt5663_max98927.c @@ -989,7 +989,6 @@ static int kabylake_audio_probe(struct platform_device *pdev) { struct kbl_rt5663_private *ctx; struct snd_soc_acpi_mach *mach; - int ret; ctx = devm_kzalloc(&pdev->dev, sizeof(*ctx), GFP_KERNEL); if (!ctx) @@ -1009,32 +1008,14 @@ static int kabylake_audio_probe(struct platform_device *pdev) &constraints_dmic_2ch : &constraints_dmic_channels; ctx->mclk = devm_clk_get(&pdev->dev, "ssp1_mclk"); - if (IS_ERR(ctx->mclk)) { - ret = PTR_ERR(ctx->mclk); - if (ret == -ENOENT) { - dev_info(&pdev->dev, - "Failed to get ssp1_sclk, defer probe\n"); - return -EPROBE_DEFER; - } - - dev_err(&pdev->dev, "Failed to get ssp1_mclk with err:%d\n", - ret); - return ret; - } + if (IS_ERR(ctx->mclk)) + return dev_err_probe(&pdev->dev, PTR_ERR(ctx->mclk), + "Failed to get ssp1_mclk\n"); ctx->sclk = devm_clk_get(&pdev->dev, "ssp1_sclk"); - if (IS_ERR(ctx->sclk)) { - ret = PTR_ERR(ctx->sclk); - if (ret == -ENOENT) { - dev_info(&pdev->dev, - "Failed to get ssp1_sclk, defer probe\n"); - return -EPROBE_DEFER; - } - - dev_err(&pdev->dev, "Failed to get ssp1_sclk with err:%d\n", - ret); - return ret; - } + if (IS_ERR(ctx->sclk)) + return dev_err_probe(&pdev->dev, PTR_ERR(ctx->sclk), + "Failed to get ssp1_sclk\n"); return devm_snd_soc_register_card(&pdev->dev, kabylake_audio_card); }
If clk_get() returns -ENOENT, there is no need to defer the driver, -ENOENT will be returned the same for each retries. So, return the error code directly instead of -EPROBE_DEFER. Remove this special case and use dev_err_probe() to simplify code. It will also be less verbose if the clk is really deferred. Fixes: f7f61e08fe58 ("ASoC: Intel: kbl: Enable mclk and ssp sclk early") Signed-off-by: Christophe JAILLET <christophe.jaillet@wanadoo.fr> --- This is based on my understanding of clk_get(). Review with care. Not sure the Fixes tag is needed. The patch does not fix anything. If devm_clk_get() returns -ENOENT, it will just loop several time until the framework gives up. If it returns -EPROBE_DEFER, this case is already handled by the "return ret;" So this patch should be a no-op, just a clean-up. --- sound/soc/intel/boards/kbl_rt5663_max98927.c | 31 ++++---------------- 1 file changed, 6 insertions(+), 25 deletions(-)