On Mon, Jan 12, 2015 at 02:37:21PM +0200, Roger Quadros wrote: > On 12/01/15 14:05, Mark Brown wrote: > > The usual thing to do here is an explicit write clearing the latch, > > either immediately after setting it or immediately before setting it. > > If the register is marked as volatile and the hardware doesn't read back > > the latched state that also does the trick. > How does this work if driver has access to only 1 bit that can only be > written with 1 to clear a condition? Writing a 0 is no-op. The expectation is that for hardware like that writes of zero will be ignored so you can just do one (generally this is the point of such things - you write to explicitly clear the bits you want to clear so by extension other bits are ignored. > It can read back 0 or 1 depending on the condition. If the hardware readback has nothing to do with what's read back then by definition this isn't something you should be updating with update_bits() - it's doing a read/modify/write cycle so if the read part doesn't correspond to the write part things will go wrong. > I didn't understand the volatile trick :P. If the hardware always reads back zero then if the register isn't cached (which it sounds like this one shouldn't be) the comparison done by update_bits() will always show that the latch bit needs writing.
Attachment:
signature.asc
Description: Digital signature
