Aw: Re: how to open elf file image from inside genode? libbacktrace question

[Uwe]

> Gesendet: Donnerstag, 08. April 2021 um 15:23 Uhr
> Von: "Josef Söntgen" <josef.soentgen at genode-labs.com>
> An: "Alexander Tormasov via users" <users at lists.genode.org>
> Betreff: Re: how to open elf file image from inside genode? libbacktrace question
>
> Hallo Alexander,
> 
> > Returning to my question, could I use posix open(«/test-go») to open
> > my binary from which I run executable, if I have the following code in
> > run file:
> > […]
> 
> Yes, using 'open()' is the way to go. As libbacktrace hardcodes a few
Is that really a good idea? Depending on vfs for 
something as bas(e)ic as stack tracing seems 
to me as overly broad. Because the needed 
capability is already a (private) member of 
the environment as _binary- >cap(). I only
need to find a way to make it accessible by 
POSIX later. Or make it directly accessible by 
the stack tracing part. 
> options you could piggy-back on one of them. I picked the '/proc/self/exe'
> option:
> 
> ! <start name="test-go" caps="250">
> !   <resource name="RAM" quantum="12M"/>
> !   <config verbose="yes" ld_verbose="yes">
> !     <vfs>
> !       <dir name="dev"> <log/> </dir>
> !
> !       <dir name="proc">
> !         <dir name="self">
> !           <rom name="exe" label="test-go"/>
> !         </dir>
> !       </dir>
> !     </vfs>
> !     <libc stdout="/dev/log" stderr="/dev/log"/>
> !   </config>
> ! </start>
> 
> With this VFS configuration in place libbacktrace opens the binary
> successfully. However, you will be greeted with the by now infamous
> deadlock ahead error message when using '-smp 2':
> 
> ! [init -> test-go] Error: deadlock ahead, mutex=0x164040, return ip=0xf3383
> 
> The mutex in question is the 'Linker::mutex()' object. Again this is due
> to the fact that as side-effect of 'dl_iterate_phdr' a jump-slot
> relocation for 'mknod' takes place (*). Why that is I have not checked.
> 
> Now we are not entirely sure if taking the mutex in 'dl_iterate_phdr' is
> strictly necessary - that is something we have to investigate. Removing
> the mutex allows the 'go_app' to run through. So you are free to do the
> same. But to be clear, we do not know if this leads to other regressions.
> We are going to look into that.
> 
> 
> (*) You can instrument that by applying the patch I attached that
>     makes the linker verbose and shows you the object and index for
>     the relocation:
> 
>     ! […]
>     ! [init -> test-go] LD: SLOT libc.lib.so 0x20c
>     ! [init -> test-go] LD: SLOT binary 0xe
>     ! [init -> test-go] Error: deadlock ahead, mutex=0x164040, return ip=0xf3383
> 
>     With this information you can check to which symbol the index
>     belongs to:
> 
>     $ readelf --dyn-syms test-go|grep '\<14:'
>         14: 0000000000000000     0 FUNC    GLOBAL DEFAULT  UND mknod
> 
> 
> Regards
> Josef
> 
> -- 
> Josef Söntgen
> Genode Labs
> 
> http://www.genode-labs.com/ · http://genode.org/
> _______________________________________________
> Genode users mailing list
> users at lists.genode.org
> https://lists.genode.org/listinfo/users