ia64/linux-2.6.18-xen.hg

changeset 617:678ad99920c8

[IA64] improve response time in dom 0 at creating a guest domain

The hypercall takes several hundred mili seconds,
and it takes around 5 mili seconds with my new patch.
Time of one hypercall should be smaller than a vcpu time slice.

Signed-off-by: Akio Takebe <takebe_akio@jp.fujitsu.com>
author Isaku Yamahata <yamahata@valinux.co.jp>
date Tue Aug 05 12:19:38 2008 +0900 (2008-08-05)
parents 0deba952a6d3
children 324a5c041301
files arch/ia64/xen/xcom_privcmd.c
line diff
     1.1 --- a/arch/ia64/xen/xcom_privcmd.c	Mon Jul 28 17:24:40 2008 +0900
     1.2 +++ b/arch/ia64/xen/xcom_privcmd.c	Tue Aug 05 12:19:38 2008 +0900
     1.3 @@ -437,15 +437,18 @@ xencomm_privcmd_memory_reservation_op(pr
     1.4  		 *   may cause the soft lockup warning.
     1.5  		 *   In order to avoid the warning, we limit
     1.6  		 *   the number of extents and repeat the hypercall.
     1.7 -		 *   The following value is determined by experimentation.
     1.8 -		 *   If the following limit causes soft lockup warning,
     1.9 +		 *   The following value is determined by evaluation.
    1.10 +		 *   Time of one hypercall should be smaller than
    1.11 +		 *   a vcpu time slice. The time with current
    1.12 +		 *   MEMORYOP_MAX_EXTENTS is around 5 msec.
    1.13 +		 *   If the following limit causes some issues,
    1.14  		 *   we should decrease this value.
    1.15  		 *
    1.16  		 *   Another way would be that start with small value and
    1.17  		 *   increase adoptively measuring hypercall time.
    1.18  		 *   It might be over-kill.
    1.19  		 */
    1.20 -#define MEMORYOP_MAX_EXTENTS	(MEMORYOP_XENCOMM_LIMIT / 4)
    1.21 +#define MEMORYOP_MAX_EXTENTS	(MEMORYOP_XENCOMM_LIMIT / 512)
    1.22  
    1.23  		while (nr_extents > 0) {
    1.24  			xen_ulong_t nr_tmp = nr_extents;