On 07/23/2012 06:30 PM, Benjamin Herrenschmidt wrote: > On Tue, 2012-07-24 at 09:29 +1000, Benjamin Herrenschmidt wrote: > >> The layers in between, not the well behaved drivers. Again, we have >> ZONE_DMA32 specifically for the purpose, why use something else ? >> >> In any case, make the whole thing at the very least a config option, I >> don't want sane HW to have to deal with split zones. > > Or if possible a flag set by machine probe()
I suggested making the threshold configurable by platform code. Sane hardware would leave it at its default of infinity (~0ULL), and you wouldn't have a split zone. Our hardware would set it at 31-bit. It looks like this is already sort-of done for ISA_DMA_THRESHOLD for 32-bit non-coherent-DMA platforms (amigaone sets a 24-bit threshold), though I don't see where ZONE_DMA is limited accordingly. We'd add an additional threshold for ZONE_DMA32, make it actually affect the zone definition, and make the standard alloc_coherent() honor it. -Scott _______________________________________________ Linuxppc-dev mailing list Linuxppc-dev@lists.ozlabs.org https://lists.ozlabs.org/listinfo/linuxppc-dev
