arm64: mm: keep low RAM dma zone

Commit ba0fb44 ("dma-mapping: replace zone_dma_bits by
zone_dma_limit") optimistically assumed that device-tree dma-ranges
property describes the system DMA limits. That assumption ignores DMA
limits of individual devices that are not encoded in device tree.

Commit 833bd28 ("arm64: mm: fix DMA zone when dma-ranges is
missing") fixed part of the problem for platforms that do not provide
dma-ranges at all. However platforms like SM8550-HDK provide DMA bus
limit, but have devices with stronger DMA limits.
of_dma_get_max_cpu_address() does not take device limitations into
account.

These platforms implicitly rely on DMA zone in low 32-bit RAM area.
Until we find a better way to figure out the optimal DMA zone range,
restore the low RAM DMA zone we had before commit ba0fb44.

Fixes: ba0fb44 ("dma-mapping: replace zone_dma_bits by zone_dma_limit")
Closes: https://lore.kernel.org/r/[email protected]
Reported-by: Neil Armstrong <[email protected]>
Reviewed-by: Catalin Marinas <[email protected]>
Suggested-by: Robin Murphy <[email protected]>
Signed-off-by: Baruch Siach <[email protected]>
Tested-by: Neil Armstrong <[email protected]> # on SM8550-HDK
Tested-by: Marek Szyprowski <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>

Author: baruchsiach

arch/arm64/mm/init.c

@@ -116,8 +116,14 @@ static void __init arch_reserve_crashkernel(void)
 
 static phys_addr_t __init max_zone_phys(phys_addr_t zone_limit)
 {
-	if (zone_limit == PHYS_ADDR_MAX)
-		zone_limit = U32_MAX;
+	/**
+	 * Information we get from firmware (e.g. DT dma-ranges) describe DMA
+	 * bus constraints. Devices using DMA might have their own limitations.
+	 * Some of them rely on DMA zone in low 32-bit memory. Keep low RAM
+	 * DMA zone on platforms that have RAM there.
+	 */
+	if (memblock_start_of_DRAM() < U32_MAX)
+		zone_limit = min(zone_limit, U32_MAX);
 
 	return min(zone_limit, memblock_end_of_DRAM() - 1) + 1;
 }