Hi Ulf,
On 2017/8/22 17:05, Ulf Hansson wrote:
On 8 August 2017 at 13:44, Shawn Lin <shawn.lin@xxxxxxxxxxxxxx> wrote:
Per SD specification version 4.10, section 4.3.4, it says "pre-erased
(ACMD23) will make a following Multiple Block Write operation faster
compared to the same operation without preceding ACMD23". ACMD23 is
mandatory for all sd cards and the spec also says "Command STOP_TRAN
(CMD12) shall be used to stop the transmission in Write Multiple Block
whether or not the preerase (ACMD23) feature is used."
However current code only check to see if CMD23 is available. If not,
it would fallback to open-ending mode. So this catch-all RFC patch
is trying to activate ACMD23 support for SD cards when doing multiple
write.
I have been testing it for while with all different kinds of SD cards
. But what makes me hesitate to submit a formal patch is that
(1)
If the SD card does support CMD23 as well as the host, then we don't
use ACMD23 as I don't see *obvious* write performance boost for
ACMD23 vs CMD23 when doing multiple write for most of the cards,
although some of them do improve a little. So this makes me though that
the behaviour of dealing with ACMD23 or CMD23 depends on the firmware
designed.
Just so I get this right...
If the card supports CMD23, then you mean we are fine with the current
code as is?
Yep.
However, if the card don't support CMD23, but ACMD23, $subject patch
*may* offer an improvement. Right?
Right, at least the spec explicitly say that.
That leads to yet another question. Are there many SD cards that don't
support CMD23 but ACMD23?
ACMD23 is mandatory but CMD23 is not. I also have to say, very few of
SD cards at my hand(I have dozens of card :)) actually support CMD23.
(2)
If enabling ACMD23 for SD cards, I did see a interesting thing that
some of the cards fail to return to tran state even when the host
fires CMD12 which meets the requirement of spec. When debugging, I saw
these buggy cards was still in recv state and never changed. So I did
some hack to trace the card status and fire CMD12 again if possible.
(3)
I test all of my SD cards at hand, and try to figure out if we benefit
from ACMD23 vs open-ending multiple write. When using ACMD23, we fire
extra ACMD23 which would lower the write performance theoretically,
although I also didn't see any *obviuos* thoughput drop. The only
visible benefit is that pre-erased multiple write is more 'stable'
than open-ending multiple write, just as the same reason for why we
prefer to use CMD23 instead of open-ending mode.
Overall, I wouldn't have any issues of enabling ACMD23 for SD cards -
if only it could be justified as providing an improved behavior.
I guess we will need more peoples opinion and help in test to understand better.
Agree! Maybe we could still wait some more time until after 4.14 merge
window, and activate that in linux-next to help test this?
[...]
Kind regards
Uffe
--
To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
