Re: [RFC PATCH 0/7] Add MMC packed function

[Baolin Wang]

On Mon, 12 Aug 2019 at 18:52, Adrian Hunter <adrian.hunter@xxxxxxxxx> wrote:
>
> On 12/08/19 12:44 PM, Baolin Wang wrote:
> > Hi Adrian,
> >
> > On Mon, 12 Aug 2019 at 16:59, Adrian Hunter <adrian.hunter@xxxxxxxxx> wrote:
> >>
> >> On 12/08/19 8:20 AM, Baolin Wang wrote:
> >>> Hi,
> >>>
> >>> On Mon, 22 Jul 2019 at 21:10, Baolin Wang <baolin.wang@xxxxxxxxxx> wrote:
> >>>>
> >>>> Hi All,
> >>>>
> >>>> Now some SD/MMC controllers can support packed command or packed request,
> >>>> that means it can package multiple requests to host controller to be handled
> >>>> at one time, which can improve the I/O performence. Thus this patchset is
> >>>> used to add the MMC packed function to support packed request or packed
> >>>> command.
> >>>>
> >>>> In this patch set, I implemented the SD host ADMA3 transfer mode to support
> >>>> packed request. The ADMA3 transfer mode can process a multi-block data transfer
> >>>> by using a pair of command descriptor and ADMA2 descriptor. In future we can
> >>>> easily expand the MMC packed function to support packed command.
> >>>>
> >>>> Below are some comparison data between packed request and non-packed request
> >>>> with fio tool. The fio command I used is like below with changing the
> >>>> '--rw' parameter and enabling the direct IO flag to measure the actual hardware
> >>>> transfer speed.
> >>>>
> >>>> ./fio --filename=/dev/mmcblk0p30 --direct=1 --iodepth=20 --rw=read --bs=4K --size=512M --group_reporting --numjobs=20 --name=test_read
> >>>>
> >>>> My eMMC card working at HS400 Enhanced strobe mode:
> >>>> [    2.229856] mmc0: new HS400 Enhanced strobe MMC card at address 0001
> >>>> [    2.237566] mmcblk0: mmc0:0001 HBG4a2 29.1 GiB
> >>>> [    2.242621] mmcblk0boot0: mmc0:0001 HBG4a2 partition 1 4.00 MiB
> >>>> [    2.249110] mmcblk0boot1: mmc0:0001 HBG4a2 partition 2 4.00 MiB
> >>>> [    2.255307] mmcblk0rpmb: mmc0:0001 HBG4a2 partition 3 4.00 MiB, chardev (248:0)
> >>>>
> >>>> 1. Non-packed request
> >>>> I tested 3 times for each case and output a average speed.
> >>>>
> >>>> 1) Sequential read:
> >>>> Speed: 28.9MiB/s, 26.4MiB/s, 30.9MiB/s
> >>>> Average speed: 28.7MiB/s
> >>
> >> This seems surprising low for a HS400ES card.  Do you know why that is?
> >
> > I've set the clock to 400M, but it seems the hardware did not output
> > the corresponding clock. I will check my hardware.
> >
> >>>>
> >>>> 2) Random read:
> >>>> Speed: 18.2MiB/s, 8.9MiB/s, 15.8MiB/s
> >>>> Average speed: 14.3MiB/s
> >>>>
> >>>> 3) Sequential write:
> >>>> Speed: 21.1MiB/s, 27.9MiB/s, 25MiB/s
> >>>> Average speed: 24.7MiB/s
> >>>>
> >>>> 4) Random write:
> >>>> Speed: 21.5MiB/s, 18.1MiB/s, 18.1MiB/s
> >>>> Average speed: 19.2MiB/s
> >>>>
> >>>> 2. Packed request
> >>>> In packed request mode, I set the host controller can package maximum 10
> >>>> requests at one time (Actually I can increase the package number), and I
> >>>> enabled read/write packed request mode. Also I tested 3 times for each
> >>>> case and output a average speed.
> >>>>
> >>>> 1) Sequential read:
> >>>> Speed: 165MiB/s, 167MiB/s, 164MiB/s
> >>>> Average speed: 165.3MiB/s
> >>>>
> >>>> 2) Random read:
> >>>> Speed: 147MiB/s, 141MiB/s, 144MiB/s
> >>>> Average speed: 144MiB/s
> >>>>
> >>>> 3) Sequential write:
> >>>> Speed: 87.8MiB/s, 89.1MiB/s, 90.0MiB/s
> >>>> Average speed: 89MiB/s
> >>>>
> >>>> 4) Random write:
> >>>> Speed: 90.9MiB/s, 89.8MiB/s, 90.4MiB/s
> >>>> Average speed: 90.4MiB/s
> >>>>
> >>>> Form above data, we can see the packed request can improve the performance greatly.
> >>>> Any comments are welcome. Thanks a lot.
> >>>
> >>> Any comments for this patch set? Thanks.
> >>
> >> Did you consider adapting the CQE interface?
> >
> > I am not very familiar with CQE, since my controller did not support
> > it. But the MMC packed function had introduced some callbacks to help
> > for different controllers to do packed request, so I think it is easy
> > to adapt the CQE interface.
> >
>
> I meant did you consider using the CQE interface instead of creating another
> one?

Sorry for misunderstanding. I think the core/core.c modification can
use the CQE interface, but there are some difference in core/block.c,
and I think they are different mechanisms, also I want to keep avoid
affecting CQE and normal transfer, so I think adding MMC packed
related interfaces will be easy to read and maintain.

-- 
Baolin Wang
Best Regards