On Thu, May 20, 2021 at 08:53:34PM +0200, Volker Rümelin wrote:
On Fri, May 14, 2021 at 08:03:20PM +0200, Volker Rümelin wrote:
The comment above the yield() function suggests that yield() allows interrupts for a short time. Currently this is only true if seabios was built without CONFIG_THREADS or if yield() is called from the main thread. Change the code to always call check_irqs() in yield().
I'm confused about the failure scenario you describe here, as yield() really should always allow irqs to run prior to returning. When called from a "background thread" it will not directly enable irqs, but it will always cycle through all "threads" before returning. Thus the "main thread" should always be reached and when it runs yield() it will permit irqs to run.
Am I missing something?
Hi Kevin,
the main thread calls run_thread() to create a background thread, adds it behind main thread in the thread list and immediately executes the background thread until the background thread calls yield(). All background threads in the thread list run in sequence and call yield() until the last background thread yields to main thread. The main thread has not reached wait_threads(), which means it will create another background thread and run it. This will repeat until all background threads are created. You can see until now the main thread did not call yield() but the background threads ran several times. The main thread calls yield() for the first time in wait_threads().
Thanks. It does seem the code is not doing what was intended. Could we add the check to the bottom of run_thread() though? Something like:
if (cur == &MainThread) // Permit irqs to fire check_irqs();
I think check_irqs() has to be called at the top of run_thread() before the new background thread gets called. I'll test. I can reliably reproduce this problem.
With best regards, Volker