Our port of muslc to seL4 is very minimal and does not implement errno - when it's accessed by a thread that thread will fault. If you are in debug mode and that thread does not have a fault handler, you'll see a double fault, where seL4 will print out the details of the fault. The kernel itself is still running, but just the idle thread.

With respect to setting up malloc, you can take a look at the sel4bench project, which bootstraps an allocator for the initial process here:

Note that both of these approaches are using a static malloc - where malloc is allocating memory from a fixed size array configured with CONFIG_LIB_SEL4_MUSLC_SYS_MORECORE_BYTES. It is possible to set up a dynamic pool backed by a virtual memory by setting this to value to 0.

We don't have an implementation that supports errno. You could map a page at the address muslc expects errno to be, or set up a fault handler for your thread that maps pages in.

From: Devel <devel-bounces@sel4.systems> on behalf of Fabrizio Bertocci <fabriziobertocci@gmail.com>
Sent: Tuesday, 4 July 2017 11:57 PM
To: devel@sel4.systems
Subject: [seL4] malloc & errno

Hi all,

I am facing a puzzling problem related to muslc. While running some tests with the ia32 simulator (qemu), have noticed that whenever I call malloc() and I run out of memory (because my morecore buffer is too small), I am getting some kernel crash.

I have traced the issue to the place where errno is set to -ENOMEM. Referencing to errno is platform-dependent and end up calling a function that retrieve the struct pthread associated and access errno as one of the field of the struct pthread.

What is the correct process to set up a heap from the root task in such a way that we can call malloc() and correctly handle error conditions through errno?

I am pretty sure I am doing something wrong in my initialization code (I have based my main(0 on the code from the tutorials), because I am also having other problems with malloc(). (even after I increase my heap area).

For example, in some cases a call to malloc() end up calling a cascade of functions that eventually (from inside muslc) end up calling again malloc(). As consequence it enters in an infinite loop that stop only after it exhaust the initial stack.

I could not find any clear example that show how to correctly set up a system to use muslc.

Can anyone point me to a good example that shows how to bootstrap a system for the 5.2.0 branch?

Regards,

Fabrizio Bertocci

Real-Time Innovations, Inc.

Sunnyvale, CA