[PATCH v5 3/6] dma: swiotlb: Centralize minimal pool sizing

[Robin Murphy]

On 21/09/2026 7:36 am, Aneesh Kumar K.V (Arm) wrote:
> A default SWIOTLB pool used only for unaligned kmalloc bouncing can be
> smaller than one required for limited DMA addressing. Arm64 and RISC-V
> currently implement this sizing independently, using 1 MiB of SWIOTLB
> space per GiB of RAM.
> 
> The SWIOTLB core now identifies this case with SWIOTLB_POOL_MINIMAL.
> Size that policy in swiotlb_adjust_pool_size() and remove the
> architecture-specific adjustments.
> 
> Explicit swiotlb= sizing remains unchanged because swiotlb_adjust_size()
> preserves a user-configured size.
> 
> Signed-off-by: Aneesh Kumar K.V (Arm) <aneesh.kumar at kernel.org>
> ---
>   arch/arm64/mm/init.c | 11 -----------
>   arch/riscv/mm/init.c |  4 ----
>   kernel/dma/swiotlb.c | 11 ++++++++++-
>   3 files changed, 10 insertions(+), 16 deletions(-)
> 
> diff --git a/arch/arm64/mm/init.c b/arch/arm64/mm/init.c
> index 7d51fb5f5a71..95075b5c8207 100644
> --- a/arch/arm64/mm/init.c
> +++ b/arch/arm64/mm/init.c
> @@ -340,17 +340,6 @@ void __init arch_mm_preinit(void)
>   {
>   	unsigned int flags = SWIOTLB_VERBOSE;
>   
> -	if (max_pfn <= PFN_DOWN(arm64_dma_phys_limit)) {
> -		/*
> -		 * If no bouncing needed for ZONE_DMA, reduce the swiotlb
> -		 * buffer for kmalloc() bouncing to 1MB per 1GB of RAM.
> -		 */
> -		unsigned long size =
> -			DIV_ROUND_UP(memblock_phys_mem_size(), 1024);
> -
> -		swiotlb_adjust_size(min(swiotlb_size_or_default(), size));
> -	}
> -
>   	if (max_pfn > PFN_DOWN(arm64_dma_phys_limit))
>   		flags |= SWIOTLB_INIT_ADDRESSING_LIMIT;
>   
> diff --git a/arch/riscv/mm/init.c b/arch/riscv/mm/init.c
> index e0e4c103d603..96856114b387 100644
> --- a/arch/riscv/mm/init.c
> +++ b/arch/riscv/mm/init.c
> @@ -183,10 +183,6 @@ void __init arch_mm_preinit(void)
>   		 * memory when DRAM starts above dma32_phys_limit.  Allocate
>   		 * ~1 MB per 1 GB of RAM.
>   		 */
> -		unsigned long size =
> -			DIV_ROUND_UP(memblock_phys_mem_size(), 1024);
> -		swiotlb_adjust_size(min(swiotlb_size_or_default(), size));
> -		swiotlb = true;
>   		swiotlb_flags |= SWIOTLB_ANY;
>   	}
>   
> diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
> index 8f86deb25be2..f368a73f4ed0 100644
> --- a/kernel/dma/swiotlb.c
> +++ b/kernel/dma/swiotlb.c
> @@ -483,9 +483,18 @@ static bool __init swiotlb_kmalloc_needs_bounce(void)
>   static void __init
>   swiotlb_adjust_pool_size(enum swiotlb_pool_policy policy)
>   {
> +	if (swiotlb_default_size_changed())
> +		return;
> +

This appears to be entirely redundant, as ultimately the point of this 
function is to call swiotlb_adjust_size() (if it does anything at all), 
and the first thing that does is this same exact check. We hardly need 
to micro-optimise short-circuiting a handful of arithmetic in a one-off 
setup path, and it's convoluted enough as it is, so please try to avoid 
redundancy that makes it even more confusing to follow.

>   	switch (policy) {
> -	case SWIOTLB_POOL_MINIMAL:
> +	case SWIOTLB_POOL_MINIMAL: {
> +		unsigned long size;
> +
> +		/* Use 1MB per 1GB of RAM for kmalloc() bouncing. */
> +		size = DIV_ROUND_UP(memblock_phys_mem_size(), 1024);
> +		swiotlb_adjust_size(min(swiotlb_size_or_default(), size));
>   		break;

Similarly I think it would be clearer if we had a common 
swiotlb_adjust_size() call at the end of the function, and then either 
calculate a size or return early in each switch case as appropriate.

Furthermore, swiotlb_size_or_default() is awful IMO - and in fact after 
this series we could perhaps clean it up entirely by making the size 
implicit in swiotlb_init_late() - not to mention misleadingly redundant. 
I'd say just open-code "default_nslabs << IO_TLB_SHIFT" like elsewhere 
in the file, but in fact it may as well just be IO_TLB_DEFAULT_SIZE 
(think about it...)

Thanks,
Robin.

> +	}
>   	case SWIOTLB_POOL_CC_GUEST:
>   		break;
>   	case SWIOTLB_POOL_NONE: