jl_yield() causes segfault from c

[dioptre]

I'm inserting a jl_yield so that my hungry process permits @parallel and @spawnat - but alas it doesn't look possible. It sometimes allows me to @spawnat but mostly it just crashes.

I tried: uv_async_send using multiple threads and it was about 1000 times slower than what I'm doing at present in a single thread (must be due to lib_uv joins).

If I leave the yield out it works fine :(

Not sure about this one!

signal (11): Segmentation fault
unknown function (ip: 0x7faadaf6ea0c)
unknown function (ip: 0x7faadaf73b0c)
unknown function (ip: 0x7faadaf7e64c)
jl_apply_generic at /usr/bin/../lib/x86_64-linux-gnu/julia/libjulia.so (unknown line)
wait at ./task.jl:369
yield at task.jl:355
jlcall_yield_21543 at  (unknown line)
jl_apply_generic at /usr/bin/../lib/x86_64-linux-gnu/julia/libjulia.so (unknown line)
jl_call0 at /usr/bin/../lib/x86_64-linux-gnu/julia/libjulia.so (unknown line)
processmusesync at libmse.so.0.0.1 (unknown line)
anonymous at /home/a/projects/bxi/server/pipeline.jl:50
jl_f_apply at /usr/bin/../lib/x86_64-linux-gnu/julia/libjulia.so (unknown line)
anonymous at multi.jl:904
run_work_thunk at multi.jl:645
run_work_thunk at multi.jl:654
jlcall_run_work_thunk_21275 at  (unknown line)
jl_apply_generic at /usr/bin/../lib/x86_64-linux-gnu/julia/libjulia.so (unknown line)
anonymous at task.jl:58
unknown function (ip: 0x7faadafe0f83)
unknown function (ip: (nil))
Worker 2 terminated.
ERROR (unhandled task failure): EOFError: read end of file

Place of the call looks like:

    while(1) {
        if ((size_t)d[4] % 100 == 0) {
            jl_yield();
            usleep(30000);
        }

[yuyichao]

Please quote your code, especially macros in order to not pull in random unrelated people. Please also give a more complete picture of what you are doing.

[dioptre]

All I'm doing is writing to a buffer (float array) in RAM every 3 ms.

I pass a pointer from julia to the RAM location, and then write to it.

At this point I haven't even read the value back in Julia. Let me know if you'd like more details.

Ie.

   while(1) {
        if ((size_t)d[4] % 100 == 0) {
            jl_yield();
            usleep(30000);
        }
        d[1] = 0.5;
        d[2] = 0.5;
        usleep(3000);
}

[yuyichao]

It's still hard to tell what you are doing without seeing the complete picture.

[yuyichao]

i.e. the whole program or a stand alone repro.

[dioptre]

I run:

@spawnat 2 begin
ccall( (:process, "libms.so.0.0.1"), Csize_t, (Ptr{Float32},), ee)
end

Which runs the following code for a while above successfully, then randomly crashes.

The thing it runs is the code below:

   while(1) {
        if ((size_t)d[4] % 100 == 0) {
            jl_yield();
            usleep(30000);
        }
        d[1] = 0.5;
        d[2] = 0.5;
        usleep(3000);
}

When I comment out the jl_yield it works fine.

I'm running it so I can share volatile memory between a julia process and a few others. But as I said it never gets read and crashes.

If I remove the jl_yield I can read and write to it without issues.

I hope that's enough to understand the plight!

[StefanKarpinski]

Is jl_yield being called from a different thread?

[dioptre]

no same one that i started with ccall from spawn 2
same effect happens too when called from pid 1

[dioptre]

Bueller? Damn

[tkelman]

Can you please provide a MCVE?

[vtjnash]

if turning off COPY_STACKS in options.h fixes this, i suspect the solution is #13099

[dioptre]

Thanks guys! I'll take a look!

On Tue, Mar 8, 2016, 11:28 Jameson Nash [email protected] wrote:

if turning off COPY_STACKS in options.h fixes this, i suspect the solution
is #13099 #13099

—
Reply to this email directly or view it on GitHub
#15301 (comment).

[KristofferC]

Closing due to lack of more info.