Re: [PATCH 4/6] mini-os/x86-64 entry: remove unnecessary event blocking

[Xu Zhang <xzhang@cs.uic.edu>]

Sorry about my delayed response. I just got some time to take a look and 
revise these patches.

It seems to me the bit of %gs manipulation is used for executing 
"do_hypervisor_callback" on a different stack. After it's done, it 
switches back to the original stack and resumes from there. The event 
mask is not touched during this process. I put mini-os under gdb just to 
verify, and for the runs I observed, event mask remains untouched.

Is there anything I am missing?

On 03/15/2013 03:16 PM, Konrad Rzeszutek Wilk wrote:
> On Fri, Mar 08, 2013 at 03:30:17PM -0600, Xu Zhang wrote:
>> We don't need to block events here because:
>>   - if we came from "hypervisor_callback", events are disabled at this point,
>>     no need to block again;
> Hm, I think I am not following it. It does a bit of %gs manipulation which
> looks to point to the cpu0_pda.irqcount.
>
> ?
>>   - if we came from "error_entry", we shouldn't touch event mask, for
>>     exception hanlding are meant to be interrupted by Xen events (virtual
> handling
>>     irq).
>>
>> Signed-off-by: Xu Zhang <xzhang@cs.uic.edu>
>> ---
>>   extras/mini-os/arch/x86/x86_64.S |    1 -
>>   1 files changed, 0 insertions(+), 1 deletions(-)
>>
>> diff --git a/extras/mini-os/arch/x86/x86_64.S b/extras/mini-os/arch/x86/x86_64.S
>> index 79e893f..5e0021b 100644
>> --- a/extras/mini-os/arch/x86/x86_64.S
>> +++ b/extras/mini-os/arch/x86/x86_64.S
>> @@ -185,7 +185,6 @@ ENTRY(hypervisor_callback2)
>>   
>>   error_exit:
>>   	RESTORE_REST
>> -	XEN_BLOCK_EVENTS(%rsi)
>>   
>>   retint_kernel:
>>   retint_restore_args:
>> -- 
>> 1.7.7.6
>>
>>
>> _______________________________________________
>> Xen-devel mailing list
>> Xen-devel@lists.xen.org
>> http://lists.xen.org/xen-devel
>>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel