On 2018/5/28 20:09, Martin Hicks wrote:
On Mon, May 28, 2018 at 01:28:14PM +0200, Ulf Hansson wrote:
On 25 May 2018 at 09:21, Shawn Lin <shawn.lin@xxxxxxxxxxxxxx> wrote:
Hi Martin,
For long running discard operarations, like a full-device secure discard,
this change drops the interrupt rates on my single-core NXP I.MX6UL from
45000/s to about 20/s, and greatly improves system responsiveness.
Impressive. I don't have a single core platform at hand now, but manual
unplug other cpus on my high-end platform with 32GB eMMC, shows your
patch works fine wrt. the interrupt and the perf tool also shows it
relinquish CPU sanely after doing erase.
Btw, I was CCing you to my patchset and could you kindly help test it? I
guess we might rebase your excellent work based on my trival cleanup
patchset.
Martin, Shawn,
Actually, I don't think the "cleanup" pointed out by Shawn above is
trivial, rather also very nice work which improves polling for other
situations as well. Honestly, I think we are going to iterate that
series a few times before we apply it.
That said, I think Martins changes here are ready and should improve
polling for erase. Thus I decided to apply his latest version for
next. Shawn, I hope that shouldn't cause you that much trouble to
re-base on top, right!?
In response to Shawn's request for me to test his patch series. I gave it
a quick go on Friday with 4.17-rc, but my board just barely works with
Top-of-Tree (I have to ramboot, SPI flash doesn't work, nor does ethernet).
My production kernel tree is the nxp 4.9 tree, which has thousands of
patches... I sent out the erase patch because it still applied to 4.17-rc,
but much of Shawn's work touches things that have changed between 4.9 and 4.17.
I was able to boot 4.17-rc and run bonnie++ and performance seems unchanged
I think my series doesn't improve your case but it will help if you add
your sleep into the new mmc_poll_for_busy(). Anyway, thanks for
testing.
with Shawn's patch series. (Although it's much worse than the 4.9 nxp tree,
which probably means there's MMC driver changes that make things work a lot
better for my board)
Thanks,
mh
--
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
