[Xen-devel] RE: Another update of xenoprof-oprofile

["Santos, Jose Renato G" <joserenato.santos@xxxxxx>]

Thanks
Renato

>> -----Original Message-----
>> From: Yang, Xiaowei [mailto:xiaowei.yang@xxxxxxxxx] 
>> Sent: Friday, August 26, 2005 2:54 AM
>> To: Santos, Jose Renato G; Zhai, Edwin
>> Cc: xen-devel@xxxxxxxxxxxxxxxxxxx
>> Subject: Another update of xenoprof-oprofile
>> 
>> 
>> Renato,
>> This is our latest update of xenoprof-oprofile, to add the 
>> feature of mapping all IP samples for each passive domain 
>> applications to an extra entry. Now let me explains more of 
>> our enhancement. 
>> - We split Xen HV samples for each existing domain (whatever 
>> domU or VMX). There are some time we not only care about the 
>> total number of HV, but we need to know how much percent of 
>> HV works on behave of which domain.
>> - We take more advantage of IP samples for passive domains. 
>> Originally passive domain samples are presented as a whole. 
>> What we did is to sepreate them to 3 parts: 
>> Xen(HV)/kernel/apps. In the function level, samples 
>> belonging to the first 2 parts can be mapped to their symbol 
>> tables (we also passed passive domain kernel symbol to 
>> oprofiled). And in IP level, how many times of the IP 
>> happened can be seen more clearly.
>> 
>> BTW, when debug the code, I found another little bug. 
>> Instead of using cpu_has_ht, use smp_num_siblings to judge 
>> if HT is enabled. cpu_has_ht is always true on a cpu with 
>> HT. Then we can take full use of all 8 counters when HT disabled.
>> 
>> Thanks,
>> -Xiaowei
>> 
>> 

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel