Re: [RFC PATCH] drivers: ata: ahci_sunxi: Increased SATA/AHCI DMA TX/RX FIFOs

From: U.Mutlu
Date: Sat May 11 2019 - 14:13:44 EST


Stefan Monnier wrote on 05/11/2019 03:37 PM:
Increasing the SATA/AHCI DMA TX/RX FIFOs (P0DMACR.TXTS and .RXTS) from
default 0x0 each to 0x3 each gives a write performance boost of 120MB/s
from lame 36MB/s to 45MB/s previously. Read performance is about 200MB/s
[tested on SSD using dd bs=4K count=512K].

Such a simple patch to fix such a long-standing performance problem that
everyone [ well, apparently not quite everyone ] assumed was a hardware
limitation...

And yet, April 1st is long gone.

Is it really for real?

Yes, it's indeed real, Stefan; really no April 1st joke. :-)

As you indicated, this problem of slow SATA write-speed
with these small devices lasts now for more than 5 years.
This patch finally solves the problem.

On my test device (BPI-R1) the optimum blocksize seems to be 12K
as it then gives even 129 MB/s write speed.

Here are some test results with different blocksizes, all giving
a write speed of 125 to 129 MB/s:

time sh -c "dd if=/dev/zero of=test.tmp bs=$bs count=$count conv=fdatasync"


------------ bs=8K / count=256K / 1 ------------------
262144+0 records in
262144+0 records out
2147483648 bytes (2.1 GB) copied, 16.9237 s, 127 MB/s

real 0m16.935s
user 0m0.388s
sys 0m15.777s

------------ bs=8K / count=256K / 2 ------------------
262144+0 records in
262144+0 records out
2147483648 bytes (2.1 GB) copied, 16.9916 s, 126 MB/s

real 0m17.973s
user 0m0.326s
sys 0m16.806s

------------ bs=8K / count=256K / 3 ------------------
262144+0 records in
262144+0 records out
2147483648 bytes (2.1 GB) copied, 17.0085 s, 126 MB/s

real 0m17.993s
user 0m0.442s
sys 0m16.588s

------------ bs=12K / count=171K / 1 ------------------
175104+0 records in
175104+0 records out
2151677952 bytes (2.2 GB) copied, 16.8474 s, 128 MB/s

real 0m16.860s
user 0m0.205s
sys 0m15.705s

------------ bs=12K / count=171K / 2 ------------------
175104+0 records in
175104+0 records out
2151677952 bytes (2.2 GB) copied, 16.6934 s, 129 MB/s

real 0m17.669s
user 0m0.227s
sys 0m16.355s

------------ bs=12K / count=171K / 3 ------------------
175104+0 records in
175104+0 records out
2151677952 bytes (2.2 GB) copied, 16.6684 s, 129 MB/s

real 0m17.654s
user 0m0.388s
sys 0m16.118s

------------ bs=16K / count=128K / 1 ------------------
131072+0 records in
131072+0 records out
2147483648 bytes (2.1 GB) copied, 17.1845 s, 125 MB/s

real 0m17.200s
user 0m0.251s
sys 0m16.060s

------------ bs=16K / count=128K / 2 ------------------
131072+0 records in
131072+0 records out
2147483648 bytes (2.1 GB) copied, 16.9221 s, 127 MB/s

real 0m17.902s
user 0m0.170s
sys 0m16.763s

------------ bs=16K / count=128K / 3 ------------------
131072+0 records in
131072+0 records out
2147483648 bytes (2.1 GB) copied, 16.8845 s, 127 MB/s

real 0m17.868s
user 0m0.167s
sys 0m16.736s