Hi,
i need to use adeos-ipipe just at the interrupt level. In trying to
achieve my goal I've been able both to succeed and fail, somewhat randomly.
So I decided to play it in steps starting from the very basic one, i.e.
registering the HEAD domain and doing nothing with it.
Here is the simple module I used.
--------------------------------------------------------------------
#include <linux/module.h>
#include <linux/ipipe.h>
MODULE_LICENSE("GPL");
static struct ipipe_domain head_domain;
static void head_domain_entry(void)
{
printk("\n***** HEAD_DOMAIN_ENTRY has been called *****\n");
}
int domain_init(void)
{
struct ipipe_domain_attr attr;
int retval;
ipipe_init_attr(&attr);
attr.name = "MYHEAD";
attr.domid = 0xabcdef;
attr.entry = head_domain_entry;
attr.priority = IPIPE_HEAD_PRIORITY;
retval = ipipe_register_domain(&head_domain, &attr);
printk("***** IPIPE_REGISTER_DOMAIN returned %d *****\n", retval);
return 0;
}
void domain_exit(void)
{
ipipe_unregister_domain(&head_domain);
return;
}
module_init(domain_init);
module_exit(domain_exit);
--------------------------------------------------------------------
The above module works with 2.6.32.11 patched with
adeos-ipipe-2.6.32.11-x86-2.6-03.patch but when used with 2.6.35.9 and
adeos-ipipe-2.6.35.9-x86-2.8-04.patch it simply locks the machine.
(For your information it works also with kernel 2.6.32.20 with
adeos-ipipe-2.6.32.20-x86-2.7-03.patch)
Following those locks I tried a silly brute force debugging by
anticipating the exiting from ipipe_register_domain at various points.
By so doing I found that the following:
.
.
.
/*
* Finally, allow the new domain to perform its initialization
* duties.
*/
local_irq_save_hw_smp(flags);
__ipipe_current_domain = ipd;
local_irq_restore_hw_smp(flags);
attr->entry();
return 123456;
.
.
.
avoids the locking and let me read the following with dmesg:
-----------------------------------------------------------------------
I-pipe: Domain MYHEAD registered.
BUG: Unhandled exception over domain MYHEAD at 0x805d1df - switching to ROOT
***** HEAD_DOMAIN_ENTRY has been called *****
***** IPIPE_REGISTER_DOMAIN returned 123456 *****
Pid: 751, comm: udevd Not tainted 2.6.35.9-adeos #3
Call Trace:
[<c1015dbb>] __ipipe_handle_exception+0x165/0x1af
[<c13f7b6b>] error_code+0x63/0x70
-----------------------------------------------------------------------
Following it I discovered that the fault can be avoided by commenting a
single instruction as for:
.
.
.
/*
* Finally, allow the new domain to perform its initialization
* duties.
*/
local_irq_save_hw_smp(flags);
// __ipipe_current_domain = ipd;
local_irq_restore_hw_smp(flags);
attr->entry();
.
.
.
and letting the function finish normally.
I've been unable to understand why such an assignment causes the
reported trap.
So: what am I doing wrong?
My hardware is based on an AXIOM industrial grade CPU SHB212 with an Intel Atom
D425 CPU @1.8GHz.
The system chipset is an Intel ICH8M.
The kernel is compiled without SMP support configurated for a Pentium III CPU.
I tried also defining an Atom CPU but I had the same result.
Local APIC and IO-APIC support enabled.
HPET Timer disabled.
I tried also to put some barrier() and wmb() calls in the ipipe_register_domain
in many points around the critical parts without success.
Giuseppe.
_______________________________________________
Adeos-main mailing list
[email protected]
https://mail.gna.org/listinfo/adeos-main
