[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: [PATCH v2] [arm64] Add more options



On Tue, 2017-01-24 at 11:35 +0200, Riku Voipio wrote:
> On 24 January 2017 at 04:21, Ben Hutchings <ben@decadent.org.uk>
> wrote:
> > On Fri, 2017-01-13 at 16:41 +0200, riku.voipio@linaro.org wrote:
> > > From: Riku Voipio <riku.voipio@linaro.org>
> > > 
> > > * Enable Cavium ThunderX server architecutre
> > > * Enable server related options
> > > * Enable CMA for DB410c DRM
> > 
> > CMA and NUMA both require an ABI bump (which we'll do at some point
> > before release, but probably not the next upload).
> > 
> > > * Some missing gpio driver enabled in upstream defconfig
> > > * DT cpufreq, spidev and missing DB410c audio setting
> > > ---
> > 
> > [...]
> > > +##
> > > +## file: drivers/base/Kconfig
> > > +##
> > > +CONFIG_DMA_CMA=y
> > > +CONFIG_CMA_SIZE_MBYTES=64
> > 
> > I notice this is higher than the default (16).  Is it a sensible
> > default for all/most platforms that will need CMA?  What about
> > platforms that don't use CMA at all, where the CMA zone will be
> > wasted?
> 
> CMA memory isn't wasted. It is populated with movable pages until a
> CMA user needs a continuous block from it. Userspace (unless mlocked)
> and cache/buffers are all movable, so I don't think CMA would be
> underutilized in memory pressure scenarios.

Oh, yes I see.

> 16M was not enough for getting DRM display on dragonboard 410c. I can
> check what is the minimum needed.

Given a single full-HD frame needs 8 MiB, I can see 16 MiB total would
be too small.  But then, multiply by 4 for 4K frames, and 64 MiB would
also be constraining.  (I don't know whether we currently support any
hardware with 4K output.)

Perhaps it would make sense to set a percentage instead (e.g.
CONFIG_CMA_SIZE_MBYTES=0, CONFIG_CMA_SIZE_PERCENTAGE=10)?

> > [...]
> > >  @@ -682,6 +734,18 @@ CONFIG_USB_DWC3_DUAL_ROLE=y
> > >  ## file: drivers/usb/gadget/Kconfig
> > >  ##
> > >  CONFIG_USB_GADGET=m
> > > +CONFIG_USB_CONFIGFS=m
> > 
> > I don't think that does anything useful unless we also enable some more
> > CONFIG_USB_CONFIGFS_* options.
> 
> Gah, the set of configfs, functionfs and gadgetfs is so confusing
> here..
[...]

Yes it is!

Ben.

-- 
Ben Hutchings
Never put off till tomorrow what you can avoid all together.

Attachment: signature.asc
Description: This is a digitally signed message part


Reply to: