Re: [PATCH 5/7] KVM/userspace: Device Assignment: Support for assigning PCI devices to guests

[Anthony Liguori <anthony@codemonkey.ws>]

Muli Ben-Yehuda wrote:
> On Tue, Sep 23, 2008 at 11:30:32AM -0500, Anthony Liguori wrote:
>   
>>> +		+ (addr - r_access->e_physbase);
>>> +
>>> +	if (r_access->debug & DEVICE_ASSIGNMENT_DEBUG_PIO) {
>>> +		fprintf(stderr, "%s: r_pio=%08x e_physbase=%08x"
>>> +			" r_virtbase=%08lx value=%08x\n",
>>> +			__func__, r_pio, (int)r_access->e_physbase,
>>> +			(unsigned long)r_access->r_virtbase, value);
>>> +	}
>>> +	iopl(3);
>>> +	outb(value, r_pio);
>>>   
>>>       
>> The formatting is wrong for this entire file.  Also, you shouldn't
>> have device specific debug.  Should probably error check iopl(3).
>> It's not necessary to call it every time you do an outb, just once
>> when initialized.
>>     
>
> We tried that at first, but ran into cases where even after iopl()
> ran, pio's from qemu still failed. Does qemu do anything to drop
> iopl() privileges? In any case calling iopl() unconditionally on every
> pio fixed it, but is obviously not the right long-term solution.
>   

Make sure you issue iopl() before any of the VCPU threads are spawned.  
Otherwise, you may be running into issues when something other than 
VCPU-0 is doing PIO/MMIO and you haven't iopl()'d for that thread.

Regards,

Anthony Liguori

> Cheers,
> Muli
>