On Tue, 2009-04-28 at 22:53 -0700, Anirban Chakraborty wrote: > My reply to Douglas missed the scsi list, sorry about it. Resending it here. I really think you need a better mail tool. Your reply is almost impossible to spot in the text below. > ________________________________________ > From: Styner, Douglas W [douglas.w.styner@xxxxxxxxx] > Sent: Tuesday, April 28, 2009 2:31 PM > To: Anirban Chakraborty; andi@xxxxxxxxxxxxxx; James.Bottomley@xxxxxxxxxxxxxxxxxxxxx > Cc: linux-scsi@xxxxxxxxxxxxxxx; Linux Driver; Ma, Chinang; Wilcox, Matthew R; willy@xxxxxxxxxxxxxxx; Tripathi, Sharad C > Subject: RE: [PATCH] qla2xxx: Fix a regression due to refactoring in interrupt handler code > > > Anirban Chakraborty <anirban.chakraborty@xxxxxxxxxx> writes: > > > >From: Anirban Chakraborty [mailto:anirban.chakraborty@xxxxxxxxxx] > >Sent: Tuesday, April 28, 2009 9:28 AM > >To: Styner, Douglas W; andi@xxxxxxxxxxxxxx; > >James.Bottomley@xxxxxxxxxxxxxxxxxxxxx > >Cc: linux-scsi@xxxxxxxxxxxxxxx; linux-driver@xxxxxxxxxx > >Subject: [PATCH] qla2xxx: Fix a regression due to refactoring in interrupt > >handler code > > > >Remove unnecessary search for a response completion host when driver is > >running > >in single queue mode. This should fix the increased cycles seen in > >qla24xx_intr_handler code. > > > >Please apply. > >Thanks, > >Anirban > > > >Reported-by: Styner, Douglas W <douglas.w.styner@xxxxxxxxx> > >Signed-off-by: Anirban Chakraborty <anirban.chakraborty@xxxxxxxxxx> > > There is no significant impact from this patch. The net effect of this patch when using MSI is to replace calls to qla2x00_get_rsp_host(), which calls the in-lined function pci_get_drvdata(), with direct calls to pci_get_drvdata(). qla2x00_get_rsp_host is very inexpensive (~0.0117%), so removing does not have a significant impact. It begins here: > The number of interrupts have gone up significantly between 2.6.24.2 and 2.6.30-rc3. As a matter of fact, it has gone up from 2.6.27.2, please see below, > http://markmail.org/message/f3ut7d5tqd4xsmbm#query:+page:1+mid:ueql52x45mbgnrod+state:results > I’d think that increased interrupts is definitely a reason for increase in cycles in the ISR. However, I am looking further to see if we are missing any other holes. > Thanks, > Anirban And ends here. > Linux OLTP Performance summary > Kernel# Speedup(x) Intr/s CtxSw/s us% sys% idle% iowait% > 2.6.30-rc3 1.000 30645 43027 75 25 0 0 > 2.6.30-rc3_qla2xxx 0.999 30737 43256 75 25 0 0 > > Server configurations: > Intel Xeon Quad-core 2.0GHz 2 cpus/8 cores/8 threads > 64GB memory, 3 qle2462 FC HBA, 450 spindles (30 logical units) > > > ======oprofile CPU_CLK_UNHALTED for top 30 functions > Cycles% 2.6.30-rc3 Cycles% 2.6.30-rc3_qla2xxx > 69.1925 <database> 67.8477 <database> > 1.1314 qla24xx_intr_handler 1.2426 qla24xx_start_scsi > 1.0031 qla24xx_start_scsi 1.1148 qla24xx_intr_handler > 0.8476 __schedule 0.8508 kmem_cache_alloc > 0.6532 kmem_cache_alloc 0.8021 __schedule > 0.4490 __blockdev_direct_IO 0.5179 __blockdev_direct_IO > 0.4199 __sigsetjmp 0.4456 __sigsetjmp > 0.3946 __switch_to 0.3952 scsi_request_fn > 0.3538 __list_add 0.3817 __switch_to > 0.3499 task_rq_lock 0.3716 task_rq_lock > 0.3402 scsi_request_fn 0.3481 __list_add > 0.3382 rb_get_reader_page 0.3380 rb_get_reader_page > 0.3363 copy_user_generic_string 0.3178 ring_buffer_consume > 0.3324 aio_complete 0.2993 aio_complete > 0.3110 try_to_wake_up 0.2993 copy_user_generic_string > 0.2877 ring_buffer_consume 0.2741 try_to_wake_up > 0.2683 mod_timer 0.2623 blk_queue_end_tag > 0.2605 qla2x00_process_completed_re0.2573 lock_timer_base > 0.2566 blk_queue_end_tag 0.2522 memset_c > 0.2566 generic_make_request 0.2472 memmove > 0.2547 tcp_sendmsg 0.2472 qla2x00_process_completed_re > 0.2372 lock_timer_base 0.2270 generic_make_request > 0.2333 memmove 0.2253 mod_timer > 0.2294 memset_c 0.2186 __make_request > 0.2080 mempool_free 0.2136 scsi_device_unbusy > 0.2022 generic_file_aio_read 0.2136 kmem_cache_free > 0.1963 scsi_device_unbusy 0.2119 mempool_free > 0.1963 plist_del 0.2018 kref_get > 0.1944 dequeue_rt_stack 0.2018 plist_del > 0.1924 e1000_xmit_frame 0.1984 ipc_lock > > Doug > -- > To unsubscribe from this list: send the line "unsubscribe linux-scsi" in > the body of a message to majordomo@xxxxxxxxxxxxxxx > More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
