On Tue, Dec 08, 2015 at 02:19:39PM -0800, Andrew Morton wrote: > On Tue, 8 Dec 2015 10:18:50 +0000 Qais Yousef <qais.yousef@xxxxxxxxxx> wrote: > > > Recent changes to how GFP_ATOMIC is defined seems to have broken the condition > > to use mips_alloc_from_contiguous() in mips_dma_alloc_coherent(). > > > > I couldn't bottom out the exact change but I think it's this one > > > > d0164adc89f6 (mm, page_alloc: distinguish between being unable to sleep, > > unwilling to sleep and avoiding waking kswapd) > > > > >From what I see GFP_ATOMIC has multiple bits set and the check for !(gfp > > & GFP_ATOMIC) isn't enough. To verify if the flag is atomic we need to make > > sure that (gfp & GFP_ATOMIC) == GFP_ATOMIC to verify that all bits rquired to > > satisfy GFP_ATOMIC condition are set. > > > > ... > > > > --- a/arch/mips/mm/dma-default.c > > +++ b/arch/mips/mm/dma-default.c > > @@ -145,7 +145,7 @@ static void *mips_dma_alloc_coherent(struct device *dev, size_t size, > > > > gfp = massage_gfp_flags(dev, gfp); > > > > - if (IS_ENABLED(CONFIG_DMA_CMA) && !(gfp & GFP_ATOMIC)) > > + if (IS_ENABLED(CONFIG_DMA_CMA) && ((gfp & GFP_ATOMIC) != GFP_ATOMIC)) > > page = dma_alloc_from_contiguous(dev, > > count, get_order(size)); > > if (!page) > > hm. It seems that the code is asking "can I do a potentially-sleeping > memory allocation"? > > The way to do that under the new regime is > > if (IS_ENABLED(CONFIG_DMA_CMA) && gfpflags_allow_blocking(gfp)) > > Mel, can you please confirm? Yes, this is the correct way it should be checked. The full flags cover watermark and kswapd treatment which potentially could be altered by the caller. -- Mel Gorman SUSE Labs
