Mutex Implementation.

[astrocrep]

Combuster:

Thank you soo much. You have given me enough a ton of excellent information! Ill be busy for weeks!

Thanks again,
Rich

[Combuster]

Thank you soo much. You have given me enough a ton of excellent information! Ill be busy for weeks!

You're welcome.

[astrocrep]

My fetch and increment:

Code: Select all

unsigned long fetch_and_increment(volatile unsigned long *mem)
{
	unsigned long val = 1;
    asm volatile ("lock; xadd %1, (%2)"                              
		: "=r"(val)          
        : "r"(val), "r"(mem) 
        : "%1","memory");    
    return val;
}

Looks ok right?

It seems to be working.

If I do:

Code: Select all

int a =0;
fetch_and_increment(&a); // Returns 0
fetch_and_increment(&a); // Returns 1
fetch_and_increment(&a); // Returns 2

Thanks,
Rich

[astrocrep]

So... no... there not working...

At first I tried to use a struct per mutex... and failed miserably...

Now I am try something much simpler:

Code: Select all

unsigned long fetch_and_increment(volatile unsigned long *mem)
{
   unsigned long val = 1;
    asm volatile ("lock; xadd %1, (%2)"                             
      : "=r"(val)         
        : "r"(val), "r"(mem)
        : "%1","memory");   
    return val;
} 


int entries = 0;
int exits = 0;

void mutex_lock(volatile mutex_t mt)
{
     int ticket = fetch_and_increment(&entries);
     kprintf("GOT TICKET: %d\n",ticket);
     while (ticket != exits)
     {
          //schedule(); // or anything else you can do while waiting for your turn
     }
}

void mutex_unlock(volatile mutex_t mt)
{
     exits = exits + 1;
     kprintf("MADE EXIT: %d\n",exits);
} 

It runs for a brief while giving and closing tickets... but it usually locks up pretty quick...

After about 10 - 100 tickets...
However, I actually think that the tickets are all on the same thread and after the switch the switch it blows up. Ignore the function headers the volatile mutex_t mt was for something else I was testing and failed... Also if I turn entries and exits volatile, I get two grabs for ticket 1 and no exits...

If you think it matters, I post my scheduler code... but I am attaching screen shots of the output.

Thanks,
Rich

[Combuster]

Have you made sure that 'entries' and 'exits' are shared among all threads?

[astrocrep]

No... each thread has its own esp0 stack allocated at run time.

However, I was assuming since these values would accessing the same physical address it shouldn't matter... I was wrong.

So do I need to do a context switch when I want to access a mutex?

Thanks,
Rich

[Combuster]

no, just shared memory (kernel?)

[astrocrep]

Yeah... the first 4mb is identity mapped and the kernel starts at 1mb. Those declarations for entries and exits where with in the kernel code.

EDIT: I think I got it working!!

Thanks!!!

Thanks,
Rich