[v4,0/1] riscv: improving uaccess with logs from network bench

Message

Akira Tsukamoto July 19, 2021, 12:51 p.m. UTC

Hi Guenter, Geert and Qiu,

I fixed the bug which was overrunning the copy when the size was in the
between 8*SZREG to 9*SZREG. The SZREG holds the bytes per register size
which is 4 for RV32 and 8 for RV64.

Do you mind trying this patch? It works OK at my place.

Since I had to respin the patch I added word copy without unrolling when
the size is in the between 2*SZREG to 9*SZREG to reduce the number of byte
copies which has heavy overhead as Palmer has mentioned when he included
this patch to riscv/for-next.


I rewrote the functions but heavily influenced by Garry's memcpy
function [1]. It must be written in assembler to handle page faults
manually inside the function unlike other memcpy functions.

This patch will reduce cpu usage dramatically in kernel space especially
for applications which use sys-call with large buffer size, such as network
applications. The main reason behind this is that every unaligned memory
access will raise exceptions and switch between s-mode and m-mode causing
large overhead.

---
v3 -> v4:
- Fixed overrun copy
- Added word copy without unrolling to reduce byte copy for left over

v2 -> v3:
- Merged all patches

v1 -> v2:
- Added shift copy
- Separated patches for readability of changes in assembler
- Using perf results

[1] https://lkml.org/lkml/2021/2/16/778

Akira Tsukamoto (1):
  riscv: __asm_copy_to-from_user: Optimize unaligned memory access and
    pipeline stall

 arch/riscv/lib/uaccess.S | 218 ++++++++++++++++++++++++++++++++-------
 1 file changed, 183 insertions(+), 35 deletions(-)

Comments

Akira Tsukamoto July 19, 2021, 2:55 p.m. UTC | #1

Hi Palmer,

Please do not bather with this patch.
It still have bug for rv32 reported by Guenter and I will regenerate the 
patch against v5.14-rc* since I made the patch against v5.13.x for this
patch.

Akira

On 7/19/2021 9:51 PM, Akira Tsukamoto wrote:
> Hi Guenter, Geert and Qiu,
> 
> I fixed the bug which was overrunning the copy when the size was in the
> between 8*SZREG to 9*SZREG. The SZREG holds the bytes per register size
> which is 4 for RV32 and 8 for RV64.
> 
> Do you mind trying this patch? It works OK at my place.
> 
> Since I had to respin the patch I added word copy without unrolling when
> the size is in the between 2*SZREG to 9*SZREG to reduce the number of byte
> copies which has heavy overhead as Palmer has mentioned when he included
> this patch to riscv/for-next.
> 
> 
> I rewrote the functions but heavily influenced by Garry's memcpy
> function [1]. It must be written in assembler to handle page faults
> manually inside the function unlike other memcpy functions.
> 
> This patch will reduce cpu usage dramatically in kernel space especially
> for applications which use sys-call with large buffer size, such as network
> applications. The main reason behind this is that every unaligned memory
> access will raise exceptions and switch between s-mode and m-mode causing
> large overhead.
> 
> ---
> v3 -> v4:
> - Fixed overrun copy
> - Added word copy without unrolling to reduce byte copy for left over
> 
> v2 -> v3:
> - Merged all patches
> 
> v1 -> v2:
> - Added shift copy
> - Separated patches for readability of changes in assembler
> - Using perf results
> 
> [1] https://lkml.org/lkml/2021/2/16/778
> 
> Akira Tsukamoto (1):
>   riscv: __asm_copy_to-from_user: Optimize unaligned memory access and
>     pipeline stall
> 
>  arch/riscv/lib/uaccess.S | 218 ++++++++++++++++++++++++++++++++-------
>  1 file changed, 183 insertions(+), 35 deletions(-)
>