On 02/03/2015 06:26 PM, Lars-Peter Clausen wrote:
On 02/03/2015 06:17 PM, Russell King - ARM Linux wrote:
On Tue, Feb 03, 2015 at 05:53:48PM +0100, Lars-Peter Clausen wrote:
On 02/03/2015 01:44 PM, Mark Brown wrote:
On Tue, Feb 03, 2015 at 08:54:57AM +0100, Manuel Lauss wrote:
- wm8731->mclk = devm_clk_get(&spi->dev, "mclk");
- if (IS_ERR(wm8731->mclk)) {
wm8731->mclk = NULL;dev_warn(&spi->dev, "assuming static MCLK\n");- }
This is broken for both deferred probe and in the case where the clock API genuinely returns a NULL clock. Other than that it's the kind of thing that we've done for some other drivers, though it's not good to have to do this. Check them for correct behaviour.
Ideally we'd introduce a {devm_}clk_get_optional(), with the same semantics as gpiod_get_optional(), which handles the finer details of differentiating between clock specified, but not yet probed, clock specified, but incorrectly and no clock specified, so this doesn't have to be done over and over by each driver.
No, we don't need to. It clk_get() already knows this distinction, and it appropriately returns -ENOENT vs -EPROBE_DEFER according to whether there's a clock specified in DT or not.
I know, but it returns a error when no clock is specified (-ENOENT), whereas gpiod_get_optional()-like semantics mean, it would return no error.
What I wanted to say is that pretty much every user of clk_get() that wants a optional clock gets the handling wrong. E.g. they check for PTR_ERR(clk) == -EPROBE_DEFER rather than checking for PTR_ERR(clk) != -ENOENT. Which causes errors when the clock is specified, but incorrectly specified (e.g. invalid phandle or specifier) to be silently ignored.
My hope is that having a explicit API for requesting a optional clock might make it easier for users to gets things right.
If you have coccinelle you can use the following script to find good and bad users:
@@ expression clk; @@ clk = ( devm_clk_get | clk_get ) (...); <+... ( *PTR_ERR(clk) == -EPROBE_DEFER | *PTR_ERR(clk) != -ENOENT ) ...+>