About how usable/stable is ARM64 now, in particular Rockchip, what about PCIe and XHCI?

classic Classic list List threaded Threaded
6 messages Options
Reply | Threaded
Open this post in threaded view
|

About how usable/stable is ARM64 now, in particular Rockchip, what about PCIe and XHCI?

tinkr
Hi!

About how usable and stable is ARM64 now, in particular Rockchip?

Are we more at "happy if it boots & doesn't crash in 5 minutes and PCI
express is experimental", or might it survive with some load and for
long, a good home router?

What about PCIe and USB/XHCI?

Thanks!
Tinker

Reply | Threaded
Open this post in threaded view
|

Re: About how usable/stable is ARM64 now, in particular Rockchip, what about PCIe and XHCI?

Stuart Henderson
On 2017/06/15 03:59, Tinker wrote:

> Hi!
>
> About how usable and stable is ARM64 now, in particular Rockchip?
>
> Are we more at "happy if it boots & doesn't crash in 5 minutes and PCI
> express is experimental", or might it survive with some load and for long, a
> good home router?
>
> What about PCIe and USB/XHCI?
>
> Thanks!
> Tinker
>

I don't know about Rockchip in particular. With the Overdrive 1000 there are
some segfaults in sh processes, which I hit in the configure script for
ruby/python (need to re-attempt builds many times to get past that stage).
And there are deadlocks relating to the vp pool when the machine is working
hard. When I was doing package builds I was seeing it roughly every 12-48
hours (I have stopped these builds for now because with those problems
progress is just too slow to be useful).

Reply | Threaded
Open this post in threaded view
|

Re: About how usable/stable is ARM64 now, in particular Rockchip, what about PCIe and XHCI?

tinkr
On 2017-06-15 08:50, Stuart Henderson wrote:

> On 2017/06/15 03:59, Tinker wrote:
>> Hi!
>>
>> About how usable and stable is ARM64 now, in particular Rockchip?
>>
>> Are we more at "happy if it boots & doesn't crash in 5 minutes and PCI
>> express is experimental", or might it survive with some load and for
>> long, a
>> good home router?
>>
>> What about PCIe and USB/XHCI?
>>
>> Thanks!
>> Tinker
>>
>
> I don't know about Rockchip in particular. With the Overdrive 1000
> there are
> some segfaults in sh processes, which I hit in the configure script for
> ruby/python (need to re-attempt builds many times to get past that
> stage).
> And there are deadlocks relating to the vp pool when the machine is
> working
> hard. When I was doing package builds I was seeing it roughly every
> 12-48
> hours (I have stopped these builds for now because with those problems
> progress is just too slow to be useful).

Aha. Thanks for letting me know!

Three brief followup Q:s if relevant,

  * Regarding stability, are kernel trace bug reports of crash occasions
possible & useful, or might it be so early that it's not in that game
yet?

  * Any idea if there's PCI express or XHCI yet?

  * The vp pool and other crashes, may those relate to the ARM platform's
weaker memory consistency model so memory read/write sync/barriers in
the kernel need to be tuned for that? (ARM32 is only singlecore while
ARM64 is SMP already.)


(Not sure how to read
https://cvsweb.openbsd.org/cgi-bin/cvsweb/src/sys/arch/arm64/ to extract
these answers.)

Tinker

Reply | Threaded
Open this post in threaded view
|

Re: About how usable/stable is ARM64 now, in particular Rockchip, what about PCIe and XHCI?

Stuart Henderson
On 2017/06/15 10:21, Tinker wrote:

> On 2017-06-15 08:50, Stuart Henderson wrote:
> > On 2017/06/15 03:59, Tinker wrote:
> > > Hi!
> > >
> > > About how usable and stable is ARM64 now, in particular Rockchip?
> > >
> > > Are we more at "happy if it boots & doesn't crash in 5 minutes and PCI
> > > express is experimental", or might it survive with some load and for
> > > long, a
> > > good home router?
> > >
> > > What about PCIe and USB/XHCI?
> > >
> > > Thanks!
> > > Tinker
> > >
> >
> > I don't know about Rockchip in particular. With the Overdrive 1000 there
> > are
> > some segfaults in sh processes, which I hit in the configure script for
> > ruby/python (need to re-attempt builds many times to get past that
> > stage).
> > And there are deadlocks relating to the vp pool when the machine is
> > working
> > hard. When I was doing package builds I was seeing it roughly every
> > 12-48
> > hours (I have stopped these builds for now because with those problems
> > progress is just too slow to be useful).
>
> Aha. Thanks for letting me know!
>
> Three brief followup Q:s if relevant,
>
>  * Regarding stability, are kernel trace bug reports of crash occasions
> possible & useful, or might it be so early that it's not in that game yet?

In my case I get deadlocks rather than kernel crashes.

>  * Any idea if there's PCI express or XHCI yet?

On the Overdrive 1000 the lan-on-motherboard is connected by PCIe, so
that works. No card connector though.

>  * The vp pool and other crashes, may those relate to the ARM platform's
> weaker memory consistency model so memory read/write sync/barriers in the
> kernel need to be tuned for that? (ARM32 is only singlecore while ARM64 is
> SMP already.)

There's some explanation from Mark's investigation into the problems
with the vp pool at http://marc.info/?l=openbsd-tech&m=149207874802781&w=2

I don't recall reading of any ideas about the bin/sh segfaults yet.

Reply | Threaded
Open this post in threaded view
|

Re: About how usable/stable is ARM64 now, in particular Rockchip, what about PCIe and XHCI?

tinkr
On 2017-06-15 14:04, Stuart Henderson wrote:
..

Aha. Have there been any new commits so that users(me etc.) should go
test, and report kernel debugger output in case of any instability?

Reply | Threaded
Open this post in threaded view
|

Re: About how usable/stable is ARM64 now, in particular Rockchip, what about PCIe and XHCI?

Mark Kettenis
> Date: Mon, 19 Jun 2017 16:44:35 +0000
> From: Tinker <[hidden email]>
>
> On 2017-06-15 14:04, Stuart Henderson wrote:
> ..
>
> Aha. Have there been any new commits so that users(me etc.) should go
> test, and report kernel debugger output in case of any instability?

No; pmap hackers are in short supply and have been rather busy lately.

Regarding your other questions...

The RK3399 support is still barely out of the "glad it works" stage.
PCie and XHCI don't work yet.  That is partly due to the fact that
U-Boot support for this board isn't mature enough yet.

Speaking about the RK3399; if somebody is ordering a Firefly RK3399
and would be willing to buy an extra PCIe M.2 to SSD adapter board:

  http://shop.t-firefly.com/goods.php?id=53

and send it to me, I'd be very grateful.  The shipping cost are a bit
high to order just that board.