On Thu, May 15, 2014 at 07:45:28PM +0530, Balaji T K wrote: > On Monday 12 May 2014 08:20 PM, Tom Rini wrote: > >On Mon, May 12, 2014 at 07:12:44PM +0530, Balaji T K wrote: > >>On Monday 12 May 2014 06:58 PM, Tom Rini wrote: > >>>On Fri, May 02, 2014 at 07:25:20PM +0530, Balaji T K wrote: > >>> > >>>>MMC instance 1 and 2 is capable of ADMA in omap4, omap5. > >>>>Add support for ADMA and enable ADMA for read/write to > >>>>improve mmc throughput. > >>>[snip] > >>>>@@ -44,12 +45,30 @@ > >>>> #undef OMAP_HSMMC_USE_GPIO > >>>> #endif > >>>> > >>>>+#ifdef CONFIG_SPL_BUILD > >>>>+#undef CONFIG_OMAP_MMC_ADMA > >>>>+#endif > >>> > >>>Why? Especially since a number of the folks interested in this for > >>>performance want it for SPL OS mode. Thanks! > >> > >>Because in SoCs like OMAP4/5 mmc1/2 adma doesn't have access to sram. > >>So can't have descriptor or src / destination buffers (allocated on stack) > >>given by mmc core on sram. > > > >And we can't malloc them? > > > > For descriptor yes, but for src / dest, I doubt since > it would be difficult to force the upper (fs, mmc..) layers for controller > limitation > and some ARCH might require the buffers to be cache aligned > > Does malloc return buffers aligned to cache boundary ?
Yes, with memalign which upper layers should also be using for such cases. FWIW, the ADMA patches I see in the omapzoom tree I'm fairly certain do SPL (since that's a focus of theirs) but incorrectly don't bother ensuring alignment with memalign. -- Tom
signature.asc
Description: Digital signature
_______________________________________________ U-Boot mailing list U-Boot@lists.denx.de http://lists.denx.de/mailman/listinfo/u-boot