2.6.7-mm1 I/O regression ?

[FabF <fabian.frederick@skynet.be>]

Hi,

	I've got strange I/O regression from 2.6.7-mm1.Here's a small fgrep
report :
	
2.6.7-mm1 (worst results ever seen)
-----------------------------------
Grepping  /usr/bin  :
23% cpu - 86.16 RT Sec - 1.77 Sec in KM
Entries scanned :
1569
90192 Kb analyzed this time

2.6.7-mm1 vm profile snippet
----------------------------

61.9360  vmlinux                  poll_idle
0.8899  vmlinux                  mark_offset_tsc
0.8857  vmlinux                  __copy_to_user_ll
0.6136  vmlinux                  mask_and_ack_8259A
0.5605  vmlinux                  do_anonymous_page
0.3445  vmlinux                  unix_poll
0.3240  vmlinux                  __switch_to
0.2016  vmlinux                  enable_8259A_irq
0.1448  vmlinux                  do_select
0.1336  vmlinux                  get_exec_dcookie
0.1128  vmlinux                  fget
0.0959  vmlinux                  schedule

2.6.7 (best results ever seen)
------------------------------
Grepping  /usr/bin  :
48% cpu - 37.61 RT Sec - 1.46 Sec in KM
Entries scanned :
1569
90184 Kb analyzed this time

2.6.7 vm Profile snippet
------------------------
80.4850  vmlinux                  poll_idle
0.9149  vmlinux                  mark_offset_tsc
0.7879  vmlinux                  mask_and_ack_8259A
0.4552  vmlinux                  do_wp_page
0.3943  vmlinux                  do_anonymous_page
0.2559  vmlinux                  enable_8259A_irq
0.2464  vmlinux                  __copy_to_user_ll
0.1721  vmlinux                  unix_poll
0.1302  vmlinux                  irq_entries_start

Well, profiles don't give a clue but fgrep RT delta is impressive.Did I missed something ?

PS : Those tests were made on ext3 partition.

Regards,
FabF