8000 interrupts per second when idle!

[dom]

We have the FIQ patch working on the 2.91 USB driver. Something in the update to 2.94 broke it.
We've got someone investigating.

[samsamsam]

We have the FIQ patch working on the 2.91 USB driver. Something in the update to 2.94 broke it.
We've got someone investigating.

Is there any progress with this patch?

[jamesh]

Bank holiday weekend in the UK - so probably not much work done the last three days. Patience is required!

[samsamsam]

Oki I am patient. Oki I am patient. I just asked...

[samsamsam]

Hello,
Does someone still working on this patch to port it to the new driver?

[dom]

Hello,
Does someone still working on this patch to port it to the new driver?

Good news. Costas (one of our interns) found the problem earlier today. I'm just testing, and if it looks good will push it out tomorrow.
Looks to give about 10% performance boost to the Pi, without the side effects of the SOF patch.

[dom]

FIQ patch is pushed. Get it from rpi-update. It is disabled by default, but add:
dwc_otg.fiq_fix_enable=1
to cmdline.txt to test. It should give about 10% extra performance to ARM when USB is not busy. It shouldn't change behaviour of USB.

[samsamsam]

Super news I am going to test this patch

Thank you very much!

[dom]

You can check the number of interrupts per second with and without this patch with:

Code: Select all

sh -c "cat /proc/interrupts | grep dwc && sleep 10 && cat /proc/interrupts | grep dwc" | awk '{ if (s==0) s=$2 } END { print ($2-s)/10 }'

You should get 8000+ when the patch is disabled.
You should get 600+ when the patch is enabled. This will vary based on USB devices connected and how busy they are. When USB is running flat out you may see 8000+.

[samsamsam]

Thanks for info.
I will test this patch tonight, however without patch I get:

Code: Select all

root@raspberry-pi:/etc/oscam# sh -c "cat /proc/interrupts | grep dwc && sleep 10 && cat /proc/interrupts | grep dwc" | awk '{ if (s==0) s=$2 } END { print ($2-s)/10 }'
3564

Is it possible? I use network and 4 ftdiToRs232 converters and 1GB pendrive with root partition.

vmstat give me similar results

Code: Select all

root@raspberry-pi:/etc/oscam# vmstat
procs -----------memory---------- ---swap-- -----io---- -system-- ----cpu----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa
 0  0      0 205592   3880  17240    0    0     0     0 3742  161  1  1 95  2

[dom]

Is it possible? I use network and 4 ftdiToRs232 converters and 1GB pendrive with root partition.

I don't think so. You haven't enabled SOF reduction have you?
Is it any different with the USB devices unplugged?

[samsamsam]

My command line:

Code: Select all

Kernel command line: dma.dmachans=0x3c bcm2708_fb.fbwidth=656 bcm2708_fb.fbheight=416 bcm2708.boardrev=0x2 bcm2708.serial=0xec6c4cc5 smsc95xx.macaddr=B8:27:EB:6C:4C:C5 dwc_otg.microframe_schedule=1 smsc95xx.turbo_mode=N sdhci-bcm2708.missing_status=0 sdhci-bcm2708.sync_after_dma=0 dwc_otg.lpm_enable=0 console=tty1 root=/dev/sda1 rootfstype=ext4 rootwait dwc_otg.speed=1

Maybe it is related to parameter dwc_otg.speed=1 or smsc95xx.turbo_mode=N? I can not test what happens when I unplug device now but tonight I will test this also.

I must use dwc_otg.speed=1 because without this my FTDI devices always restart. It is probably similar problem to the: http://www.mail-archive.com/linux-usb-u ... 14694.html

[dom]

My command line:

Code: Select all

Kernel command line: dma.dmachans=0x3c bcm2708_fb.fbwidth=656 bcm2708_fb.fbheight=416 bcm2708.boardrev=0x2 bcm2708.serial=0xec6c4cc5 smsc95xx.macaddr=B8:27:EB:6C:4C:C5 dwc_otg.microframe_schedule=1 smsc95xx.turbo_mode=N sdhci-bcm2708.missing_status=0 sdhci-bcm2708.sync_after_dma=0 dwc_otg.lpm_enable=0 console=tty1 root=/dev/sda1 rootfstype=ext4 rootwait dwc_otg.speed=1

Maybe it is related to parameter dwc_otg.speed=1 or smsc95xx.turbo_mode=N? I can not test what happens when I unplug device now but tonight I will test this also.

I must use dwc_otg.speed=1 because without this my FTDI devices always restart. It is probably similar problem to the: http://www.mail-archive.com/linux-usb-u ... 14694.html

Yes, dwc_otg.speed=1 limits speed to full speed (rather than USB 2's high speed) protocol, and so only 1000 interrupts/sec when idle.
This patch will have much less benefit if you are using dwc_otg.speed=1.

[samsamsam]

Yes I know However I think that the patch will make my RPI more stable. Although right now my RPI is very stable after night battles

OFFTOP

What do you think can I restore the root partition on SD card since the patch: "low-latency mode to sdcard driver"?

If yes what value should be set for the parameter sdhci-bcm2708.missing_status?

[dom]

What do you think can I restore the root partition on SD card since the patch: "low-latency mode to sdcard driver"?

If yes what value should be set for the parameter sdhci-bcm2708.missing_status?

sdhci-bcm2708.missing_status is not recommended now, just leave it off command line. The sdhci-bcm2708.enable_llm is now enabled by default and fixes the USB latency problem caused by sdcard a safer way.

[samsamsam]

Thank you very much for your help:) I will move my root partition to the SD and test these two patches:)

Thanks once again

[samsamsam]

Stats with new kernel and root partition moved to SD card:

Code: Select all

root@raspberry-pi:~# vmstat
procs -----------memory---------- ---swap-- -----io---- -system-- ----cpu----
 r  b   swpd   free   buff  cache   si   so    bi    bo   in   cs us sy id wa
 0  0      0 179380  27148  17624    0    0     0     0 3316  141  2  2 93  4

Code: Select all

Kernel command line: dma.dmachans=0x3c bcm2708_fb.fbwidth=656 bcm2708_fb.fbheight=416 bcm2708.boardrev=0x2 bcm2708.serial=0xec6c4cc5 smsc95xx.macaddr=B8:27:EB:6C:4C:C5 dwc_otg.microframe_schedule=1 smsc95xx.turbo_mode=N dwc_otg.fiq_fix_enable=1 sdhci-bcm2708.sync_after_dma=0 dwc_otg.lpm_enable=0 console=tty1 root=/dev/mmcblk0p2 rootfstype=ext4 rootwait dwc_otg.speed=1

So far, the system is stable

[Koenkk]

Not working for me, getting a kp at boot (details: http://i.imgur.com/PfJG0.jpg). I ran rpi-update. cmdline.txt is containing : dwc_otg.fiq_fix_enable=1. Kernel: 3.2.27+ #102 PREEMPT Sat Sep 1 01:00:50 BST 2012 armv6l.

[dom]

Not working for me, getting a kp at boot (details: http://i.imgur.com/PfJG0.jpg). I ran rpi-update. cmdline.txt is containing : dwc_otg.fiq_fix_enable=1. Kernel: 3.2.27+ #102 PREEMPT Sat Sep 1 01:00:50 BST 2012 armv6l.

Looks like it can't find rootfs. Is your rootfs on sdcard, or are you using nfs/usb?
Does dwc_otg.fiq_fix_enable=0 allow it to boot? Does reverting kernel allow it to boot?

[samsamsam]

cmdline.txt should contain at least:

Code: Select all

dwc_otg.fiq_fix_enable=1 sdhci-bcm2708.sync_after_dma=0 dwc_otg.lpm_enable=0 console=tty1 root=/dev/mmcblk0p2 rootfstype=ext4 rootwait

[Koenkk]

cmdline.txt should contain at least:

Code: Select all

dwc_otg.fiq_fix_enable=1 sdhci-bcm2708.sync_after_dma=0 dwc_otg.lpm_enable=0 console=tty1 root=/dev/mmcblk0p2 rootfstype=ext4 rootwait

working, thanks for this uber fast response!!

[rtramon]

FIQ patch is pushed. Get it from rpi-update. It is disabled by default, but add:
dwc_otg.fiq_fix_enable=1
to cmdline.txt to test. It should give about 10% extra performance to ARM when USB is not busy. It shouldn't change behaviour of USB.

Updated firmware with rip-update set the cmdline option, and after some time my headless system becomes inaccessible with the following error in /var/log/messages:

Sep 2 14:37:34 raspi kernel: [ 8406.167322] smsc95xx 1-1.1:1.0: eth0: Failed to w
rite HW_CFG_LRST_ bit in HW_CFG register, ret = -110
Sep 2 14:37:40 raspi kernel: [ 8412.167492] smsc95xx 1-1.1:1.0: eth0: Failed to w
rite register index 0x00000014
(repeated many times)

cmdline.txt:
dwc_otg.lpm_enable=0 root=/dev/mmcblk0p2 rootfstype=ext4 rootwait sdhci-bcm2708.sync_after_dma=0 dwc_otg.microframe_schedule=1 dwc_otg.fiq_fix_enable=1

Also interupts per seconds is 44K! like normally on my system using pl2303 usb to rs232 converter reading serial output from small measurement system.

[samsamsam]

You can try to add line

Code: Select all

vm.min_free_kbytes = 16384

at the and of file /etc/sysctl.conf

And add parameter smsc95xx.turbo_mode=N to the cmdline.txt

[dom]

Sep 2 14:37:34 raspi kernel: [ 8406.167322] smsc95xx 1-1.1:1.0: eth0: Failed to w
rite HW_CFG_LRST_ bit in HW_CFG register, ret = -110
Sep 2 14:37:40 raspi kernel: [ 8412.167492] smsc95xx 1-1.1:1.0: eth0: Failed to w
rite register index 0x00000014
(repeated many times)

cmdline.txt:
dwc_otg.lpm_enable=0 root=/dev/mmcblk0p2 rootfstype=ext4 rootwait sdhci-bcm2708.sync_after_dma=0 dwc_otg.microframe_schedule=1 dwc_otg.fiq_fix_enable=1

If you change to dwc_otg.fiq_fix_enable=0 does the problem go away?
If you change to dwc_otg.microframe_schedule=0 does the problem go away?

[bitwelder]

Hello,
I applied the patch and in my case (nothing attached to USB) interrupts went from 8000+ to 290.

Question: is there any known 'side effect' to this FIQ patch, or any conditions where it may not be convenient to activate it?