[PATCH] riscv: mm: Move dma_contiguous_reserve() after NUMA initialization

[PATCH] riscv: mm: Move dma_contiguous_reserve() after NUMA initialization

[Eder Zulian]

Move the dma_contiguous_reserve() call from setup_bootmem() to
misc_mem_init(), placing it after arch_numa_init(). This ensures that
NUMA topology is initialized when reserving contiguous memory for DMA.

Tested with CMA_SIZE_PERNUMA enabled and two NUMA nodes on QEMU:

qemu-system-riscv64 \
	-machine virt \
	-nographic \
	-smp 2 -m 512M \
	-numa node,nodeid=0,cpus=0,memdev=m0 \
	-numa node,nodeid=1,cpus=1,memdev=m1 \
	-object memory-backend-ram,id=m0,size=256M \
	-object memory-backend-ram,id=m1,size=256M \
	-kernel arch/riscv/boot/Image \
	-append "console=ttyS0 earlycon loglevel=7 cma=16M"

Unpatched kernel's log (one global CMA pool and no per-nodepools, 16 MiB
cma-reserved):

	[    0.000000] cma: Reserved 16 MiB at 0x000000009ee00000
	...
	[    0.000000] Initmem setup node 0 [mem 0x0000000080000000-0x000000008fffffff]
	[    0.000000] Initmem setup node 1 [mem 0x0000000090000000-0x000000009fffffff]
	...
	[    0.056397] smp: Brought up 2 nodes, 2 CPUs
	[    0.066135] Memory: 449288K/524288K available (12279K kernel code, 5980K rwdata, 6144K rodata, 2467K init, 482K bss, 54504K reserved, 16384K cma-reserved)

Patched kernel's log (three CMA pools are created, 48 MiB total),

	[    0.000000] cma: Reserved 16 MiB at 0x000000009ee00000
	[    0.000000] cma: Reserved 16 MiB at 0x000000008ee00000
	[    0.000000] cma: Reserved 16 MiB at 0x000000009de00000
	...
	[    0.054251] smp: Brought up 2 nodes, 2 CPUs
	[    0.064082] Memory: 416520K/524288K available (12279K kernel code, 5980K rwdata, 6144K rodata, 2467K init, 482K bss, 54504K reserved, 49152K cma-reserved)

Signed-off-by: Eder Zulian <ezulian@redhat.com>
---
 arch/riscv/mm/init.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
index 85bdcce63a3e..79a658ebe74d 100644
--- a/arch/riscv/mm/init.c
+++ b/arch/riscv/mm/init.c
@@ -295,8 +295,6 @@ static void __init setup_bootmem(void)
 	 */
 	if (!IS_ENABLED(CONFIG_BUILTIN_DTB))
 		memblock_reserve(dtb_early_pa, fdt_totalsize(dtb_early_va));
-
-	dma_contiguous_reserve(dma32_phys_limit);
 }
 
 #ifdef CONFIG_RELOCATABLE
@@ -1349,6 +1347,7 @@ void __init misc_mem_init(void)
 {
 	early_memtest(min_low_pfn << PAGE_SHIFT, max_low_pfn << PAGE_SHIFT);
 	arch_numa_init();
+	dma_contiguous_reserve(dma32_phys_limit);
 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 	/* The entire VMEMMAP region has been populated. Flush TLB for this region */
 	local_flush_tlb_kernel_range(VMEMMAP_START, VMEMMAP_END);
-- 
2.55.0

Re: [PATCH] riscv: mm: Move dma_contiguous_reserve() after NUMA initialization

[Paul Walmsley]

Hi,

On Tue, 14 Jul 2026, Eder Zulian wrote:

> Move the dma_contiguous_reserve() call from setup_bootmem() to
> misc_mem_init(), placing it after arch_numa_init(). This ensures that
> NUMA topology is initialized when reserving contiguous memory for DMA.
> 
> Tested with CMA_SIZE_PERNUMA enabled and two NUMA nodes on QEMU:
> 
> qemu-system-riscv64 \
> 	-machine virt \
> 	-nographic \
> 	-smp 2 -m 512M \
> 	-numa node,nodeid=0,cpus=0,memdev=m0 \
> 	-numa node,nodeid=1,cpus=1,memdev=m1 \
> 	-object memory-backend-ram,id=m0,size=256M \
> 	-object memory-backend-ram,id=m1,size=256M \
> 	-kernel arch/riscv/boot/Image \
> 	-append "console=ttyS0 earlycon loglevel=7 cma=16M"
> 
> Unpatched kernel's log (one global CMA pool and no per-nodepools, 16 MiB
> cma-reserved):
> 
> 	[    0.000000] cma: Reserved 16 MiB at 0x000000009ee00000
> 	...
> 	[    0.000000] Initmem setup node 0 [mem 0x0000000080000000-0x000000008fffffff]
> 	[    0.000000] Initmem setup node 1 [mem 0x0000000090000000-0x000000009fffffff]
> 	...
> 	[    0.056397] smp: Brought up 2 nodes, 2 CPUs
> 	[    0.066135] Memory: 449288K/524288K available (12279K kernel code, 5980K rwdata, 6144K rodata, 2467K init, 482K bss, 54504K reserved, 16384K cma-reserved)
> 
> Patched kernel's log (three CMA pools are created, 48 MiB total),
> 
> 	[    0.000000] cma: Reserved 16 MiB at 0x000000009ee00000
> 	[    0.000000] cma: Reserved 16 MiB at 0x000000008ee00000
> 	[    0.000000] cma: Reserved 16 MiB at 0x000000009de00000
> 	...
> 	[    0.054251] smp: Brought up 2 nodes, 2 CPUs
> 	[    0.064082] Memory: 416520K/524288K available (12279K kernel code, 5980K rwdata, 6144K rodata, 2467K init, 482K bss, 54504K reserved, 49152K cma-reserved)
> 
> Signed-off-by: Eder Zulian <ezulian@redhat.com>

Thanks, queued for v7.3.


- Paul

Re: [PATCH] riscv: mm: Move dma_contiguous_reserve() after NUMA initialization

[patchwork-bot+linux-riscv]

Hello:

This patch was applied to riscv/linux.git (for-next)
by Paul Walmsley <pjw@kernel.org>:

On Tue, 14 Jul 2026 20:56:48 +0200 you wrote:
> Move the dma_contiguous_reserve() call from setup_bootmem() to
> misc_mem_init(), placing it after arch_numa_init(). This ensures that
> NUMA topology is initialized when reserving contiguous memory for DMA.
> 
> Tested with CMA_SIZE_PERNUMA enabled and two NUMA nodes on QEMU:
> 
> qemu-system-riscv64 \
> 	-machine virt \
> 	-nographic \
> 	-smp 2 -m 512M \
> 	-numa node,nodeid=0,cpus=0,memdev=m0 \
> 	-numa node,nodeid=1,cpus=1,memdev=m1 \
> 	-object memory-backend-ram,id=m0,size=256M \
> 	-object memory-backend-ram,id=m1,size=256M \
> 	-kernel arch/riscv/boot/Image \
> 	-append "console=ttyS0 earlycon loglevel=7 cma=16M"
> 
> [...]

Here is the summary with links:
  - riscv: mm: Move dma_contiguous_reserve() after NUMA initialization
    https://git.kernel.org/riscv/c/76e885f2b89d

You are awesome, thank you!
-- 
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html