[PATCH 04/14] ARM: OMAP2+: Add function for configuring GPMC settings

Philip, Avinash avinashphilip at ti.com
Mon Mar 4 21:05:41 EST 2013


On Fri, Mar 01, 2013 at 21:13:37, Hunter, Jon wrote:
> 
> On 02/28/2013 11:33 PM, Philip, Avinash wrote:
> > On Thu, Feb 28, 2013 at 22:42:37, Hunter, Jon wrote:
> >>
> >> On 02/28/2013 09:52 AM, Jon Hunter wrote:
> >>>
> >>> On 02/28/2013 12:05 AM, Philip, Avinash wrote:
> >>>> On Tue, Feb 26, 2013 at 23:00:31, Hunter, Jon wrote:
> >>>>> The GPMC has various different configuration options such as bus-width,
> >>>>> synchronous or asychronous mode selection, burst mode options etc.
> >>>>> Currently, there is no common function for configuring these options and
> >>>>> various devices set these options by either programming the GPMC CONFIG1
> >>>>> register directly or by calling gpmc_cs_configure() to set some of the
> >>>>> options.
> >>>>>
> >>>>> Add a new function for configuring all of the GPMC options. Having a common
> >>>>> function for configuring this options will simplify code and ease the
> >>>>> migration to device-tree.
> >>>>>
> >>>>> Signed-off-by: Jon Hunter <jon-hunter at ti.com>
> >>>>> ---
> >>>>>  arch/arm/mach-omap2/gpmc.c |   65 ++++++++++++++++++++++++++++++++++++++++++++
> >>>>>  arch/arm/mach-omap2/gpmc.h |    6 ++++
> >>>>>  2 files changed, 71 insertions(+)
> >>>>>
> >>>>> diff --git a/arch/arm/mach-omap2/gpmc.c b/arch/arm/mach-omap2/gpmc.c
> >>>>> index 465cac9..fb8dfd2 100644
> >>>>> --- a/arch/arm/mach-omap2/gpmc.c
> >>>>> +++ b/arch/arm/mach-omap2/gpmc.c
> >>>>> @@ -1137,6 +1137,71 @@ int gpmc_calc_timings(struct gpmc_timings *gpmc_t,
> >>>>>  	return 0;
> >>>>>  }
> >>>>>  
> >>>>> +int gpmc_cs_program_settings(int cs, struct gpmc_settings *p)
> >>>>> +{
> >>>>> +	u32 config1;
> >>>>> +
> >>>>> +	if ((!p->device_width) || (p->device_width > GPMC_DEVWIDTH_16BIT)) {
> >>>>> +		pr_err("%s: invalid width %d!", __func__, p->device_width);
> >>>>> +		return -EINVAL;
> >>>>> +	}
> >>>>> +
> >>>>> +	/* Address-data multiplexing not supported for NAND devices */
> >>>>> +	if (p->device_nand && p->mux_add_data) {
> >>>>> +		pr_err("%s: invalid configuration!\n", __func__);
> >>>>> +		return -EINVAL;
> >>>>> +	}
> >>>>> +
> >>>>> +	/* Page/burst mode supports lengths of 4, 8 and 16 bytes */
> >>>>> +	if (p->burst_read || p->burst_write) {
> >>>>> +		switch (p->burst_len) {
> >>>>> +		case GPMC_BURST_4:
> >>>>> +		case GPMC_BURST_8:
> >>>>> +		case GPMC_BURST_16:
> >>>>> +			break;
> >>>>> +		default:
> >>>>> +			pr_err("%s: invalid page/burst-length (%d)\n",
> >>>>> +			       __func__, p->burst_len);
> >>>>> +			return -EINVAL;
> >>>>> +		}
> >>>>> +	}
> >>>>> +
> >>>>> +	if ((p->wait_on_read || p->wait_on_write) &&
> >>>>> +	    (p->wait_pin > gpmc_nr_waitpins)) {
> >>>>> +		pr_err("%s: invalid wait-pin (%d)\n", __func__, p->wait_pin);
> >>>>> +		return -EINVAL;
> >>>>> +	}
> >>>>> +
> >>>>> +	config1 = GPMC_CONFIG1_DEVICESIZE((p->device_width - 1));
> >>>>
> >>>> Can you consider read_modify approach?
> >>>
> >>> I was purposely trying to avoid that. The intent here is to program all
> >>> the settings and get away from any boot-loader dependencies. If we use a
> >>> read-modify approach then it will never be clear in the kernel what
> >>> should actually be programmed.
> >>
> >> By the way, it would be good to know what your motivation for this is.
> >>
> >> My goal is for all gpmc settings to eventually live in the DT blob and
> >> there be no dependency on the bootloader whatsoever. That may mean that
> >> settings are re-programmed again by the kernel but IMO that is best.
> > 
> > Yes I understood now and the right thing to do.
> > Suggested read modify approach because of some confusion as GPMC_CONFIG1 is already
> > modified in gpmc_cs_set_timings() and is called from omap2_nand_gpmc_retime().
> > So the data modified in gpmc_cs_set_timings() will lost (not significant)
> > May be better and cleaner solution is to remove GPMC_CONFIG1 modification in
> > gpmc_cs_set_timings() also.
> 
> Actually that is a good point. There are some timings in CONFIG1 that
> are not being configured in gpmc_cs_program_settings() but
> gpmc_cs_set_timings(). I had left it this way intentionally to keep
> timings together, which makes sense for where gpmc clock could change at
> run time. However, this means that gpmc_cs_program_settings() needs to
> be called before gpmc_cs_set_timings(). Really this is no different to
> how the code was before, because the code was writing directly to
> CONFIG1 and now we have a function.
> 
> So the intent is for gpmc_cs_program_settings() to only be called once
> for a given device and then gpmc_cs_set_timings() be called whenever the
> timings need to be updated. May be I can updated the kernel-doc for
> gpmc_cs_program_settings() to reflect this. Let me know if you have any
> other concerns with this.

Fine enough.

Thanks
Avinash
> 
> Cheers
> Jon
> 



More information about the devicetree-discuss mailing list