2022-12-14 18:05:43

by Luben Tuikov

Subject: [PATCH] dma-direct: Optimize get_required_mask

Optimize dma_direct_get_required_mask(), in that we don't need to multiply by
two if we don't subtract 1 from the exponent. That is,

(1 << (n - 1)) * 2 - 1 <==>
2^(n-1) * 2^1 - 1 = (by rule of exponents)
2^n - 1 <==>
(1 << n) - 1.

Cc: Christoph Hellwig <[email protected]>
Cc: Robin Murphy <[email protected]>
Cc: Alex Deucher <[email protected]>
Cc: Christian König <[email protected]>
Cc: [email protected]
Cc: Linux Kernel Mailing List <[email protected]>
Signed-off-by: Luben Tuikov <[email protected]>
---
kernel/dma/direct.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c
index 63859a101ed831..bb416a3949dac0 100644
--- a/kernel/dma/direct.c
+++ b/kernel/dma/direct.c
@@ -41,7 +41,7 @@ u64 dma_direct_get_required_mask(struct device *dev)
phys_addr_t phys = (phys_addr_t)(max_pfn - 1) << PAGE_SHIFT;
u64 max_dma = phys_to_dma_direct(dev, phys);

- return (1ULL << (fls64(max_dma) - 1)) * 2 - 1;
+ return (1ULL << fls64(max_dma)) - 1;
}

static gfp_t dma_direct_optimal_gfp_mask(struct device *dev, u64 dma_mask,

base-commit: e2ca6ba6ba0152361aa4fcbf6067db71b2c7a770
--
2.39.0

2022-12-14 21:27:15

by Robin Murphy

Subject: Re: [PATCH] dma-direct: Optimize get_required_mask

On 2022-12-14 18:00, Luben Tuikov wrote:
> Optimize dma_direct_get_required_mask(), in that we don't need to multiply by
> two if we don't subtract 1 from the exponent. That is,
>
> (1 << (n - 1)) * 2 - 1 <==>
> 2^(n-1) * 2^1 - 1 = (by rule of exponents)
> 2^n - 1 <==>
> (1 << n) - 1.

...except when n==64 (for the actual code below), in which case the
result of the shift becomes undefined.

Robin.

> Cc: Christoph Hellwig <[email protected]>
> Cc: Robin Murphy <[email protected]>
> Cc: Alex Deucher <[email protected]>
> Cc: Christian König <[email protected]>
> Cc: [email protected]
> Cc: Linux Kernel Mailing List <[email protected]>
> Signed-off-by: Luben Tuikov <[email protected]>
> ---
> kernel/dma/direct.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c
> index 63859a101ed831..bb416a3949dac0 100644
> --- a/kernel/dma/direct.c
> +++ b/kernel/dma/direct.c
> @@ -41,7 +41,7 @@ u64 dma_direct_get_required_mask(struct device *dev)
> phys_addr_t phys = (phys_addr_t)(max_pfn - 1) << PAGE_SHIFT;
> u64 max_dma = phys_to_dma_direct(dev, phys);
>
> - return (1ULL << (fls64(max_dma) - 1)) * 2 - 1;
> + return (1ULL << fls64(max_dma)) - 1;
> }
>
> static gfp_t dma_direct_optimal_gfp_mask(struct device *dev, u64 dma_mask,
>
> base-commit: e2ca6ba6ba0152361aa4fcbf6067db71b2c7a770

2022-12-15 05:09:21

by Luben Tuikov

Subject: Re: [PATCH] dma-direct: Optimize get_required_mask

On 2022-12-14 15:57, Robin Murphy wrote:
> On 2022-12-14 18:00, Luben Tuikov wrote:
>> Optimize dma_direct_get_required_mask(), in that we don't need to multiply by
>> two if we don't subtract 1 from the exponent. That is,
>>
>> (1 << (n - 1)) * 2 - 1 <==>
>> 2^(n-1) * 2^1 - 1 = (by rule of exponents)
>> 2^n - 1 <==>
>> (1 << n) - 1.
>
> ...except when n==64 (for the actual code below), in which case the
> result of the shift becomes undefined.

Oh, right, for bit 63 being set. Forgot about that one. Good call.

Thanks,
Luben