Re: ARM test 323 failure

On Fri, Oct 18, 2019 at 12:07 PM Daniel Stenberg via curl-library
<curl-library_at_cool.haxx.se> wrote:
>
> On Fri, 18 Oct 2019, Dan Fandrich via curl-library wrote:
>
> > I have a PR[1] to add an ARM build to Travis (ARM support is still in beta)
> > and I've noticed a few problems. For one, test 323 consistently fails (see
> > an example at [2]). curl is returning CURLE_GOT_NOTHING instead of the
> > expected CURLE_SSL_CONNECT_ERROR in that case. The PR skips that test on ARM
> > since it fails every time.
> >
> > Secondly, I noticed a couple of test runs ended in segfaults (for example
> > [3]; I thought there was another one but I can't find it now). That's a bit
> > more worrisome, but it's going to be hard to produce without ready access to
> > the hardware.
> >
> > curl is built and used on ARM all the time, so it's odd that I'm seeing
> > this. Has anyone else run into this before?
>
> I love this ability of finally getting some CI builds going on ARM!
>
> Those test problems look really unexpected and weird. They're truly really
> hard to debug from remote like this when all we have are test failures or a
> very brief "segmentation fault" output.
>
> I'll put "get myself a Raspberry Pi 4" on my TODO list to see if I can debug
> these problems with one.

GCC117 and GCC118 from the Compile Farm are ARMv8 machines with CRC and Crypto.

There are four other ARMv8 machines, but they are early Mustang
boards. The early Mustang boards are ARMv8, but lack CRC and Crypto.

I believe the CFarm machines are A-57's, not A-72's however. The
A-57's were the server cpu's (as opposed to the A-53's, which were
more dev-board quality).

If you don't have a CFarm account, you can signup at
https://cfarm.tetaneutral.net/.

Jeff
-------------------------------------------------------------------
Unsubscribe: https://cool.haxx.se/list/listinfo/curl-library
Etiquette: https://curl.haxx.se/mail/etiquette.html
Received on 2019-10-18