On Mon, Jul 29, 2024 at 01:51:26PM +0300, Baruch Siach wrote: > diff --git a/kernel/dma/pool.c b/kernel/dma/pool.c > index 410a7b40e496..ded3d841c88c 100644 > --- a/kernel/dma/pool.c > +++ b/kernel/dma/pool.c > @@ -12,6 +12,7 @@ > #include <linux/set_memory.h> > #include <linux/slab.h> > #include <linux/workqueue.h> > +#include <linux/memblock.h> > > static struct gen_pool *atomic_pool_dma __ro_after_init; > static unsigned long pool_size_dma; > @@ -70,7 +71,7 @@ static bool cma_in_zone(gfp_t gfp) > /* CMA can't cross zone boundaries, see cma_activate_area() */ > end = cma_get_base(cma) + size - 1; > if (IS_ENABLED(CONFIG_ZONE_DMA) && (gfp & GFP_DMA)) > - return end <= zone_dma_limit; > + return end <= memblock_start_of_DRAM() + zone_dma_limit; I think this patch is entirely wrong. After the previous patch, zone_dma_limit is already a physical/CPU address, not some offset or range - of_dma_get_max_cpu_address() returns the absolute physical address. Adding memblock_start_of_DRAM() to it does not make any sense. It made sense when we had zone_dma_bits but since we are trying to move away from bitmasks to absolute CPU addresses, zone_dma_limit already includes the start of DRAM. What problems do you see without this patch? Is it because DMA_BIT_MASK(32) can be lower than zone_dma_limit as I mentioned on the previous patch? -- Catalin
