Debian on Synology RS816 (Armada 385)

Richard,

> Bottom line is it appears rootwait wont work for
> us on raid although I think it still makes sense
> for USB or SD on other boxes which have that.

rootwait is not needed on HDD, SSD, or USB rootfs. Only needed on boxes with an SD card (MMC) rootfs.

-bodhi
===========================
Forum Wiki
bodhi's corner

@bodhi

> rootwait is not needed on HDD, SSD, or USB rootfs. Only needed on boxes with an SD card (MMC) rootfs.

Back when we first started working on the RS816 I originally put rootdelay=10 into the bootargs because the USB occasionally did not come up in time. Its up to you whether we leave rootdelay=10 / 20 or use rootwait but I suggest you DO need one of them on USB.

The main point of my 2am testing was to check if rootwait would avoid mdadm arrays from starting in degraded mode which it does not and my post was principally reporting on that.

It was your suggestion to use rootwait for RAID not mine and it would have been polite to acknowledge the outcome of the test.

Richard

Hey Richard,

Actually I have not read all your recent posts closely (too busy to do that). But I will do so.

Thanks for all your contribution to this forum.

-bodhi
===========================
Forum Wiki
bodhi's corner

@bohdi

>Thanks for all your contribution to this forum.

Zero sleep and no prospect of any for several hours has probably led me to be more 'sensitive' than events would dictate appropriate.

Thank you for the comment.

I think we both need to clear our desks a bit before we do much more, I suggest we park this project till Monday next week or longer if you are snowed under ?

Richard

> 7) Once boot gets to rootfs seen the drives which
> have not spun up seem to be just marked as
> 'missing' and need to be manually added once boot
> is complete. (I would need to dig into Debian
> config files to confirm this but it does appear to
> be the case from these trials and previous
> observations with 8TB 7200rpm drives)

Bummer! then rootwait can not be used in your RAID configuration.

> Bottom line is it appears rootwait wont work for
> us on raid although I think it still makes sense
> for USB or SD on other boxes which have that.

As I mentioned, rootwait is only needed for slow spin-up drives. USBs and HDDs don't have that problem. MMC dirves do need it sometime.

> On a plus note the apparent issue with mdadm not
> being in rootfs has been confirmed as not valid.

Good to hear!

-bodhi
===========================
Forum Wiki
bodhi's corner

Final edit for Section I:

https://forum.doozan.com/read.php?2,76314,91094#msg-91094

-bodhi
===========================
Forum Wiki
bodhi's corner

I've updated released thread Linux Kernel 5.2.9 MVEBU package and Debian armhf rootfs to include Section I.

Quote

Zyxel NAS326 Installation, see this thread.
WD My Cloud EX2100 Installation: see this thread.
GlobalScale Technologies Mirabox Installation, see this thread.
Thecus N2350 Installation: see TBD.
Synolgy RS816 Installation: see Installation with USB rootfs (Section I). There will be 3 more sections for other installation methods.

-bodhi
===========================
Forum Wiki
bodhi's corner

> I think we both need to clear our desks a bit
> before we do much more, I suggest we park this
> project till Monday next week or longer if you are
> snowed under ?

Please take your leisure time to answer posts. After all, this is a hobbyist forum! things are taking pace slowly or fast, depending on how interesting the subject is.

-bodhi
===========================
Forum Wiki
bodhi's corner

@bodhi

>Final edit for Section I:

All looks good to me :-)

My only minor comment is regarding the rootdelay=20 could be changed to rootdelay=10 or even 5 probably but to be honest it's not material.

Just recapping the history of the rootdelay= addition.
I originally added rootdelay=10 because very occasional the USB rootfs boot would drop to intramfs with root not found. It is possible the issue was my USB stick getting a bit long in the tooth rather than it being an issue with USB generally.

Once I started to use RAID on <4TB 5400rpm drives I needed to bump to rootdelay=20 which I then used everywhere including USB rootfs. With >4TB 7200 rpm drives we needed to bump again this time to rootdelay=40 which should be safe for all drives, I suggest we use for the RAID tutorial.

The final USB rootdelay I leave to your judgement / experience but as I say its not greatly material and If it were me I would probably not bother changing it.

Richard

I successfully installed on Synology DS116. I am looking forward to approved installation instructions on HDD. Thanks for amazing work!

bonk1990 Wrote:
-------------------------------------------------------
> I successfully installed on Synology DS116. I am
> looking forward to approved installation
> instructions on HDD. Thanks for amazing work!

Cool! I'll take a look at that again when I have some free time, probably in a week or so.

-bodhi
===========================
Forum Wiki
bodhi's corner

could you please paste the url of the tarball?

youxiaojie Wrote:
-------------------------------------------------------
> could you please paste the url of the tarball?


Which tarball?

-bodhi
===========================
Forum Wiki
bodhi's corner

bodhi Wrote:
-------------------------------------------------------
> Richard,
>
> Don't need to find it. Let me see if I can upload
> my Marvell xxx 38x u-boot GPL tarball that I got
> from the manufacturer for you to try to compile
> later. I'm also swamped right now with works so
> cannot spend much time to do this.


this link you said 1.4g and you have a slow speed of internet.

youxiaojie,

> this link you said 1.4g and you have a slow speed
> of internet.

I forgot all about this! will have to look if I had uploaded this tarball :)

-bodhi
===========================
Forum Wiki
bodhi's corner

@bodhi

Just a heads up that linux-5.5.3-mvebu-tld-1 may not work correctly on the RS816 if my experience with my own kernel is indicative.

On my own kernel I was ok up to linux-5.4.6 but sata broke with linux-5.4.7. What appeared to be happening is the drives were brought up by our DTS (I could hear them start) but then set back down again by the kernel, the first two drives, attached to controller 0, would then come back up but not the second controller 1 pair. lsblk etc showed no trace of them and my thought was the power had been switched back off for that controller.

At the time I was very swamped with work and didn't have time to investigate instead choosing to stick with linux-5.4.6 I did, however, try to compile later kernels in the hope the issue got fixed, I tried up to linux-5.5.1 before I gave up. Around this time, as a company, we decided to switch from offering our customers access to their projects on our servers instead switching to externally hosted servers (cost reasons mainly) from that point the RS816 has not been used and I've not tried to look into the issue since.

On a personal level I still have an interest in this stuff and if I get a chance I will try to USB boot linux-5.5.3-mvebu-tld-1 with your rootfs and see if it exhibits the same issue as I saw on my kernel. I can't promise that I will be able to test in the very near future and hence the heads up.

Richard

Richard,

> On my own kernel I was ok up to linux-5.4.6 but
> sata broke with linux-5.4.7. What appeared to be
> happening is the drives were brought up by our DTS
> (I could hear them start) but then set back down
> again by the kernel, the first two drives,
> attached to controller 0, would then come back up
> but not the second controller 1 pair. lsblk etc
> showed no trace of them and my thought was the
> power had been switched back off for that
> controller.

That might be related to a regulator core problem. I'll look at this box DTS to see if anything needs to be update.

-bodhi
===========================
Forum Wiki
bodhi's corner

Yes. Just take a peek.

I need to post a new DTB for this box for you and others to try when convenient.

-bodhi
===========================
Forum Wiki
bodhi's corner

Here is the new DTB. Please try and observe the HDDs spin up behavior.

-bodhi
===========================
Forum Wiki
bodhi's corner

ugh, did the binding change? Guess I have some device trees to update.

@1000001101000,

> ugh, did the binding change? Guess I have some
> device trees to update.

The binding implementation must haved been changed in mainline (probably unintentional), or just a new bug in core.

There were some changes in regulator code recently that might have affected the regulator-boot-on. It used to work just by iself, but not anymore in my testing the Armada boxes (all have old Marvell u-boot). We now need regulator-always-on to tell the kernel not to bring it down after the first activation.

There was a discussion a few years ago regarding boot-on behavior. It was to save some seconds in boot time. I've suspected that implementation was somehow broken by the code change in the regulator core area.

-bodhi
===========================
Forum Wiki
bodhi's corner

I guess it's time to start my testing/Bullseye branch. I suppose it was around this time on the cycle i started working on Buster.

@bodhi

Revised DTS / DTB works for me with both your rootfs and kernel and my own.

I herby award you a large shiraz :-)

Richard

> Revised DTS / DTB works for me with both your
> rootfs and kernel and my own.

Good to hear!

-bodhi
===========================
Forum Wiki
bodhi's corner