On Tue, 2012-05-08 at 16:52 +0530, Archit Taneja wrote: > On Tuesday 08 May 2012 04:20 PM, Tomi Valkeinen wrote: > > Checking the validity of all the settings is a bit tricky, but currently > > I think, as a rule of thumb, we should accept any settings when things > > are disabled. So, until the interface driver sets the timings before > > enabling the output, all ovl/mgr settings should be allowed. And we > > We have 2 ways to go about this, one is to have an initial set of > 'always valid' values like I have done, the other option is to ignore > manager timing related checks if the manager is disabled, i.e all > configs are okay. To implement the second option, I think our function > dss_check_settings_low() would get more complicated. We would now have > to pass mp->enabled outside of apply, and pass it to dss_mgr_check() > which would further need to pass this to dss_ovl_check(). Hence I used > the first approach. I'm not sure about that. We already do it for overlay. In ovl.c we have dss_ovl_simple_check() and dss_ovl_check(). The simple check sees if the settings pass basic sanity check. The check sees if all the ovl & mgr settings are compatible with each other. Simple check is done when setting the config, called from dss_ovl_set_info(). The proper check is done later when the actual config is about to be taken into use. If mp->enabled == false, can't we just skip dss_check_settings_low() totally for that manager? We skip the check for ovls the same way. > > shouldn't even write any shadow registers until the moment the output is > > enabled. > > That's being done correctly even now I guess, with the checks for > mp->enabled in wrtie_regs() and set_go_bits(). Yes, for timings. I was thinking more about the other settings done in dpi.c currently, like dispc_mgr_set_pol_freq(). That writes directly to registers, so we need runtime_get for that. Tomi
Attachment:
signature.asc
Description: This is a digitally signed message part
