On Sat, 2021-01-09 at 22:49 +0000, Richard Purdie via
lists.openembedded.org wrote:
> On Sat, 2021-01-09 at 12:37 -0800, Khem Raj wrote:
> > On Sat, Jan 9, 2021 at 1:19 AM Richard Purdie
> > <[email protected]> wrote:
> > > 
> > > On Fri, 2021-01-08 at 10:00 +0000, Richard Purdie via
> > > lists.openembedded.org wrote:
> > > > On Wed, 2021-01-06 at 14:53 -0800, Alistair Francis wrote:
> > > > > On Wed, Jan 6, 2021 at 2:36 PM Richard Purdie
> > > > > <[email protected]> wrote:
> > > > > > 
> > > > > > When building with the new version of qemu we see errors like:
> > > > > > 
> > > > > > """
> > > > > > qemu-i386: Unable to reserve 0x7ffff000 bytes of virtual address 
> > > > > > space at
> > > > > > 0x1000 (Success) for use as guest address space (check your virtual 
> > > > > > memory
> > > > > > ulimit setting, min_mmap_addr or reserve less using -R option)
> > > > > > 
> > > > > > ERROR: The postinstall intercept hook 
> > > > > > 'update_gio_module_cache-nativesdk' failed
> > > > > > """
> > > > > > 
> > > > > > The VM reseration patches we're carrying look suspicious in this 
> > > > > > context.
> > > > > > Drop them since we don't appear to be seeing those issues any more 
> > > > > > on the
> > > > > > autobuilder and I suspect the patches have become broken and a 
> > > > > > liability.
> > > > > > webkitgtk builds seem to be ok now.
> > > > > 
> > > > > Yes! Getting rid of these patches is great!
> > > > > 
> > > > > > 
> > > > > > Signed-off-by: Richard Purdie <[email protected]>
> > > > > 
> > > > > Reviewed-by: Alistair Francis <[email protected]>
> > > > 
> > > > Unfortunately this issue is still present, I thought we weren't seeing
> > > > it but once other errors cleared, this one remains, only on
> > > > qemux86+musl for webkitgtk.
> > > > 
> > > > I think we need to get to the bottom of it and figure out something
> > > > which is upstreamable. This will block the qemu upgrade until we can
> > > > fix it unfortunately unless we block webkitgtk on musl on 32 bit x86.
> > > 
> > > I've sent out a couple of linux-user mmap patches for this. With those
> > > fixes applied, qemu seems fine so I've upgraded.
> > > 
> > > Khem: I do wonder whether musl's memory allocation is all ok given it
> > > loops indefinitely if it doesn't see EFAULT and only ENOMEM. That may
> > > need investigation?
> > 
> > I forwarded it to musl community as well. Other musl distros are also
> > carrying some patches in qemu
> > eg, voidlinux has this
> > https://github.com/void-linux/void-packages/blob/master/srcpkgs/qemu/patches/mmap-mremap-efault.patch
> > which actually could be forwarded upstream qemu.
> 
> Yes, the first bit of that is what I sent upstream to qemu, I came to
> the same conclusion :)
> 
> The other bit looks to optimise the looping...

Just for the archives, we ran into a problem on the centos7 builders
with this change where gobject-introspection would show:

qemu-i386: Unable to reserve 0x7ffff000 bytes of virtual address space
at 0xa000 (Success) for use as guest address space (check yourvirtual
memory ulimit setting, min_mmap_addr or reserve less using -R option) 

/proc/sys/vm/mmap_min_addr was set to 4096.

Somewhat confusingly, the way to make things work was to set this
higher to to 65536 (there is an addr != temp test in there and addr was
coming back as 0x10000 instead of 0x1000).

Cheers,

Richard

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#146597): 
https://lists.openembedded.org/g/openembedded-core/message/146597
Mute This Topic: https://lists.openembedded.org/mt/79486601/21656
Group Owner: [email protected]
Unsubscribe: https://lists.openembedded.org/g/openembedded-core/unsub 
[[email protected]]
-=-=-=-=-=-=-=-=-=-=-=-

Reply via email to