On Thu, Jan 24, 2013 at 01:02:38PM +0100, Stijn Devriendt wrote: > As a fictive example, consider the i2c-bitbang driver, which you could > optimize > by using block-gpio with sda/scl in a single block. By offering the > block-gpio API > even when you cannot set all bits at once, you could cause timing issues. > You might be toggling the clock line before pushing out data, for example.
> The same holds below, for a driver that has separate hi/lo bits. If there's a strict ordering requirement on updates then I would expect a user to explictly code that in hardware otherwise there may be hardware level issues with unpredictable results; besides in general it seems silly to force users to open code both versions if they don't want to rely on this API.
signature.asc
Description: Digital signature
