Re: [squeak-dev] Squeak5.3 linux ARMv6 segfaults on startup

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

Re: [squeak-dev] Squeak5.3 linux ARMv6 segfaults on startup

Eliot Miranda-2
 
Hi Nicolas,

On Sun, Apr 26, 2020 at 6:27 PM Eliot Miranda <[hidden email]> wrote:


On Apr 26, 2020, at 1:50 PM, Nicolas Cellier <[hidden email]> wrote:

Alas no.  The bug is definitely still there in both 32-bit and 64-bits in the trunk of VMMaker under the simulator.  Run with your favorite trunk image and in the simulation open up an About Squeak SystemReporter. The good news is that it is *only* in the simulator on 64-bits.  On the real VM on my Mac I see no problems.

Nicolas, can you point me to the (un)signed shifts that had to be changed in Slang to fix the problem in the real VM?  That should make it trivial to fix the simulation.  I ask because there are no senders of >>> in BitBltSimulation so I don't see an obvious place to focus on.



Thanks!!


Le dim. 26 avr. 2020 à 22:48, Nicolas Cellier <[hidden email]> a écrit :

Le dim. 26 avr. 2020 à 22:44, Eliot Miranda <[hidden email]> a écrit :
Hi Tim,

On Sun, Mar 22, 2020 at 2:50 PM Eliot Miranda <[hidden email]> wrote:
Hi Bruce,
On Mar 22, 2020, at 8:31 AM, Bruce O'Neel <[hidden email]> wrote:
Hi,

Good news, playing with

--enable-fast-bitblt


and

--disable-fast-bitblt


does work in that it builds a working VM.

Good news/bad news, it does not change the font problem.  So that's not the problem.

Great news.  We now know it is an X11 problem and can stop worrying about the fast BitBLT code.  Thanks.

Looks like the problem, or at least a version of the problem, is nothing to do with X11.  If I open an About Squeak System Reporter in the simulator (64-bits x64) I see the fixed pitch font corruption we saw on X11 on ARM.  So it looks tile the issue is actually in BitBlt itself.

<SysRepFixedPitch.png>



cheers

bruce

21 March 2020 01:07 tim Rowledge <[hidden email]> wrote:


> On 2020-03-20, at 6:16 AM, Bruce O'Neel wrote:
>
>
> I specialize in ridiculous. Good news, I guess, is that I dug up a monitor, and, walked it downstairs with a keyboard and mouse and attached it. The effect with the BitstreamVeraSans and ComicSans fonts are the same. Maybe it likes serfed fonts?

I'm completely baffled by this. I don't get this effect with any ARM vm that I have that actually runs, with any image I have.

>
> So that means that it is not some funky X11 over the wire problem with the Mac and Windows X11 servers problem. That's good.

Guess so, though it just makes life weirder.

>
> I have no idea then why. Are these fonts part of the image? Is it some funky binary format that for some reason Coq is mis-reading?

Yes, the font glyphs are in-image. We've been using these ones for goodness knows how many years.

>
> I built a stack VM and I get the same result with a Squeak 5.3 image.

There*shouldn't* be any difference between the stack & cog vms.

You could try building a VM with the fastbitblt turned off I suppose -
a quick hack is to find the BITBLT_FLAGS= -DENABLE_FAST_BLT and related lines in the Makefile in you squeak.cog.spur/build directory. I haven't actually built an ARM vm without that in years so I don't know it it even still works.


tim
--
It is easier to change the specification to fit the program than vice versa.








--
_,,,^..^,,,_
best, Eliot




--
_,,,^..^,,,_
best, Eliot
Reply | Threaded
Open this post in threaded view
|

Re: [squeak-dev] Squeak5.3 linux ARMv6 segfaults on startup

Nicolas Cellier
 
Hi Eliot,
a few mails above in the thread, it seems to be in #stepToNextBezier

Le lun. 27 avr. 2020 à 20:01, Eliot Miranda <[hidden email]> a écrit :
 
Hi Nicolas,

On Sun, Apr 26, 2020 at 6:27 PM Eliot Miranda <[hidden email]> wrote:


On Apr 26, 2020, at 1:50 PM, Nicolas Cellier <[hidden email]> wrote:

Alas no.  The bug is definitely still there in both 32-bit and 64-bits in the trunk of VMMaker under the simulator.  Run with your favorite trunk image and in the simulation open up an About Squeak SystemReporter. The good news is that it is *only* in the simulator on 64-bits.  On the real VM on my Mac I see no problems.

Nicolas, can you point me to the (un)signed shifts that had to be changed in Slang to fix the problem in the real VM?  That should make it trivial to fix the simulation.  I ask because there are no senders of >>> in BitBltSimulation so I don't see an obvious place to focus on.



Thanks!!


Le dim. 26 avr. 2020 à 22:48, Nicolas Cellier <[hidden email]> a écrit :

Le dim. 26 avr. 2020 à 22:44, Eliot Miranda <[hidden email]> a écrit :
Hi Tim,

On Sun, Mar 22, 2020 at 2:50 PM Eliot Miranda <[hidden email]> wrote:
Hi Bruce,
On Mar 22, 2020, at 8:31 AM, Bruce O'Neel <[hidden email]> wrote:
Hi,

Good news, playing with

--enable-fast-bitblt


and

--disable-fast-bitblt


does work in that it builds a working VM.

Good news/bad news, it does not change the font problem.  So that's not the problem.

Great news.  We now know it is an X11 problem and can stop worrying about the fast BitBLT code.  Thanks.

Looks like the problem, or at least a version of the problem, is nothing to do with X11.  If I open an About Squeak System Reporter in the simulator (64-bits x64) I see the fixed pitch font corruption we saw on X11 on ARM.  So it looks tile the issue is actually in BitBlt itself.

<SysRepFixedPitch.png>



cheers

bruce

21 March 2020 01:07 tim Rowledge <[hidden email]> wrote:


> On 2020-03-20, at 6:16 AM, Bruce O'Neel wrote:
>
>
> I specialize in ridiculous. Good news, I guess, is that I dug up a monitor, and, walked it downstairs with a keyboard and mouse and attached it. The effect with the BitstreamVeraSans and ComicSans fonts are the same. Maybe it likes serfed fonts?

I'm completely baffled by this. I don't get this effect with any ARM vm that I have that actually runs, with any image I have.

>
> So that means that it is not some funky X11 over the wire problem with the Mac and Windows X11 servers problem. That's good.

Guess so, though it just makes life weirder.

>
> I have no idea then why. Are these fonts part of the image? Is it some funky binary format that for some reason Coq is mis-reading?

Yes, the font glyphs are in-image. We've been using these ones for goodness knows how many years.

>
> I built a stack VM and I get the same result with a Squeak 5.3 image.

There*shouldn't* be any difference between the stack & cog vms.

You could try building a VM with the fastbitblt turned off I suppose -
a quick hack is to find the BITBLT_FLAGS= -DENABLE_FAST_BLT and related lines in the Makefile in you squeak.cog.spur/build directory. I haven't actually built an ARM vm without that in years so I don't know it it even still works.


tim
--
It is easier to change the specification to fit the program than vice versa.








--
_,,,^..^,,,_
best, Eliot




--
_,,,^..^,,,_
best, Eliot