On Wed, 2023-03-29 at 17:11 +0300, Andy Shevchenko wrote: > On Wed, Mar 29, 2023 at 05:08:44PM +0300, Andy Shevchenko wrote: > > On Tue, Mar 28, 2023 at 04:01:21PM +0000, Sahin, Okan wrote: > > > > On Tue, Mar 28, 2023 at 04:51:18PM +0200, Nuno Sá wrote: > > > > > On Tue, 2023-03-28 at 17:35 +0300, Andy Shevchenko wrote: > > > > > > On Tue, Mar 28, 2023 at 04:18:30PM +0200, Nuno Sá wrote: > > > > > > > On Tue, 2023-03-28 at 14:46 +0100, Mark Brown wrote: > > > > > > > > On Tue, Mar 28, 2023 at 03:26:44PM +0200, Nuno Sá > > > > > > > > wrote: > > > > > > > > > > > > > > > > > IIRC, regmap_read() is not really reentrant and it is > > > > > > > > > used in > > > > > > > > > the IIO driver on the sysfs interface. So, yeah, I > > > > > > > > > think you > > > > > > > > > need the regmap lock and better just leave the config > > > > > > > > > as is. > > > > > > > > > Yes, the lock is > > > > > > > > > opt- > > > > > > > > > out > > > > > > > > > so let's not disable it :) > > > > > > > > > > > > > > > > All the regmap operations are fully thread safe. > > > > > > > > > > > > > > Even if 'config->disable_locking' is set? I think that is > > > > > > > what's > > > > > > > being discussed in here... > > > > > > > > > > > > In case the driver has its own lock to serialize IO how on > > > > > > earth the > > > > > > regmap lock is needed. That's what I asked the author of > > > > > > the driver. > > > > > > He told the code > > > > > > > > > > Well, if the driver has it's own locking, then sure we do not > > > > > need > > > > > regmap's lock... > > > > > > > > > > > doesn't require the regmap lock, and I tend to believe the > > > > > > author. > > > > > > So, why to > > > > > > keep it? > > > > > > > > > > However, if you look at the adc driver, I can see plain > > > > > regmap_read() > > > > > calls without any "outside" locking. > > > > > > > > Then author of the code should know what they are doing. Right? > > > > > Actually, I do not want to disable regmap lock that's why I did > > > not update it. > > > > If you have something like > > > > func1() > > regmap_read(reg1) > > regmap_read/write(reg2) > > I would also add that as soon as you have more than one access to regmap (or spi, or i2c...) in one func and that function is accessible through sysfs (and likely any other userspace interface), then you should already take care. So, it might be even the case that func1() and func2() don't have to run in parallel. - Nuno Sá >
