[Vm-dev] ArmV8 linux crash in MVC issue 617

Eliot Miranda eliot.miranda at gmail.com
Sat Mar 5 03:26:08 UTC 2022

On Fri, Mar 4, 2022 at 12:09 PM Bruce O'Neel <bruce.oneel at pckswarms.ch>

> Hi all,
> So I spent a few hours last night and a few more today running the broken
> code in Arm32 and Arm64.
> I don't see the problem.  I can not decide if it is in the the BitBlt
> plugin built by cog, though I think not, nor in the lower level code called
> to actually move bits because that code is called from so many places and
> runs well.
> The bug only happens on Armv8 linux because the bug is in code that is
> only built on Arm* linux.  The code only happens in MVC with a 32 bit
> screen depth.  But this is because the code path is taken only for 32 bit
> screen depths.
> So....  We would like to release a new version of Squeak soonish.  I think
> we have a few choices:
> 1. Assume that the combo of ArmV8 and MVC will be rare enough that almost
> no one will stumble on it.  If they do there is a workaround of using some
> other screen depth in MVC.  A touch ugly but to most of today's youth I'll
> bet they already thought MVC was ugly and won't use it.
> 2.  Put my patch in place for the moment.
> In both cases I want to keep the issue open so it is not forgotten.  I
> would like to fix it I just don't see the problem.
> Thoughts?

Try and produce an expression (a BitBlt invocation on the Display object?)
that invokes the pathological path so we don't have to use the UI at all.
Then I have a reasonable chance of fixing it quickly.  i may een be able to
reproduce it on my fab M1 MPB.

> cheers
> bruce

best, Eliot
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20220304/1c14d83e/attachment.html>

More information about the Vm-dev mailing list