[Milhouse]

You should be able to disable FIQ using the settings you have already tried, what do you get with "dmesg | grep FIQ" when you have disabled FIQ (and still experience HDD/network problems)?

[User 203163]

In 0308 if I try to disable FIQ using:
boot=/dev/mmcblk0p1 disk=/dev/mmcblk0p2 dwc_otg.fiq_enable=0 dwc_otg.fiq_fsm_enable=0 ssh quiet
I can't use ssh. I get no IP on the RPi

---

If I don't try to disable FIQ in 0308 I get the following.

xbmc:~ # dmesg | grep FIQ
[ 1.509471] WARN::dwc_otg_hcd_init:1053: FIQ DMA bounce buffers: virt = 0xd0d51000 dma = 0x4ef48000 len=9024
[ 1.509497] FIQ FSM acceleration enabled for :
[ 1.509554] WARN::hcd_init:474: FIQ at 0xc04bb45c
[ 1.509568] WARN::hcd_init:475: FIQ ASM at 0xc04bb734 length 36
[ 1.510949] dwc_otg: FIQ enabled
[ 1.510969] dwc_otg: FIQ split-transaction FSM enabled

[Milhouse]

What if you add only

Code:

dwc_otg.fiq_fsm_enable=0

This should disable the FSM FIQ so that NOP FIQ is used instead.

[User 203163]

What if you add only

Code:

dwc_otg.fiq_fsm_enable=0

This should disable the FSM FIQ so that NOP FIQ is used instead.

That worked better! I now get IP and can ssh.

xbmc:~ # dmesg | grep FIQ
[ 1.525680] WARN::dwc_otg_hcd_init:1053: FIQ DMA bounce buffers: virt = 0xd0d51000 dma = 0x4ef4c000 len=9024
[ 1.525882] WARN::hcd_init:474: FIQ at 0xc04bb64c
[ 1.525964] WARN::hcd_init:475: FIQ ASM at 0xc04bb734 length 36
[ 1.560713] dwc_otg: FIQ enabled
[ 1.560736] dwc_otg: FIQ split-transaction FSM disabled

Will try uploading a 4GB file to the USB-hdd.

[popcornmix]

I just want to note a (for me) weired behavior. The Alsa playback for bcm2835 works, if no hdmi cable is plugged in.
I have no clue why. Power supply issues shouldn't be possible (2A output) and dmesg shows no differences.

Never use the bcm2835 ALSA device. It doesn't let you choose analogue/hdmi output and it is much less efficient than "PI: HDMI" or "PI: Analogue".
It's just enabled as part of the testing for external ALSA sound cards.

[Silo100]

What if you add only

Code:

dwc_otg.fiq_fsm_enable=0

This should disable the FSM FIQ so that NOP FIQ is used instead.

That worked better! I now get IP and can ssh.

xbmc:~ # dmesg | grep FIQ
[ 1.525680] WARN::dwc_otg_hcd_init:1053: FIQ DMA bounce buffers: virt = 0xd0d51000 dma = 0x4ef4c000 len=9024
[ 1.525882] WARN::hcd_init:474: FIQ at 0xc04bb64c
[ 1.525964] WARN::hcd_init:475: FIQ ASM at 0xc04bb734 length 36
[ 1.560713] dwc_otg: FIQ enabled
[ 1.560736] dwc_otg: FIQ split-transaction FSM disabled

Will try uploading a 4GB file to the USB-hdd.

I get this as well, network goes away (even with dwc_otg.fiq_fsm_enable=0). I can trigger it every time by running "texturecache P" on my library, where everything is stored on the network (movies on NAS, remote MySQL), so it seems somehow related to network traffic?

[Milhouse]

Are you overclocked, does restoring stock clocks help when FIQ FSM is enabled? Just one more variable to rule out.

I've tried running "texturecache.py p" several times now (MySQL with media on NFS) and so far no problems so this isn't easy to reproduce.

[User 203163]

Are you overclocked, does restoring stock clocks help when FIQ FSM is enabled? Just one more variable to rule out.

I've tried running "texturecache.py p" several times now (MySQL with media on NFS) and so far no problems so this isn't easy to reproduce.

With dwc_otg.fiq_fsm_enable=0 my 4GB file copy went smooth. No problems.

For testing purposes I went back to default clock settings and have FIQ FSM enabled.
The first thing I noticed was how slow the copy over network is when on stock clock. 1.5MB/s vs 2.99MB/s with overclock. But that's a bit off topic.

After about 5 minutes of copying the RPi network connection died so it doesn't look like overclocking is causing this.

[mildenhall]

Are you overclocked, does restoring stock clocks help when FIQ FSM is enabled? Just one more variable to rule out.

I've tried running "texturecache.py p" several times now (MySQL with media on NFS) and so far no problems so this isn't easy to reproduce.

RaspberryPi #1 @ 1GHz with FIQ FSM on works fine.
RaspberryPi #2 @ 1GHz with FIQ FSM on "sometimes works" - 25% of any transfers I tried.
RaspberryPi #2 @ 950Mhz with FIQ_FSM on "sometimes works" - 25% of any transfers I tried.
RaspberryPi #2 @ 950Mhz with FIQ_FSM off - all transfers work.

Could do some debugs if you like but not until tomorrow night - mother-in-law's birthday!

[popcornmix]

RaspberryPi #1 @ 1GHz with FIQ FSM on works fine.
RaspberryPi #2 @ 1GHz with FIQ FSM on "sometimes works" - 25% of any transfers I tried.
RaspberryPi #2 @ 950Mhz with FIQ_FSM on "sometimes works" - 25% of any transfers I tried.
RaspberryPi #2 @ 950Mhz with FIQ_FSM off - all transfers work.

Is it the Pi#2 board or sdcard image that is the problem?
i.e. can you swap sdcards and see if the problem swaps with it.

[ijsbeer79]

I also have the same issue. A build of a week ago runs smooth, but this one will have network errors if I use the texturecache command (with option C). I installed this to a new SD card (which held a Frodo version without issues).
I'm pleased with the speed of this build, but have a lot of 'gray' thumbs in my library, because of the failing texturecache
Here is a log: http://xbmclogs.com/show.php?id=145414

I use NFS and a MySQ db. Watching movies isn't an issue at all, as long as I don't use texturecache....

[anonimo2001]

@MilhouseVH: do you want me to record the issue with my mobile phone, for example? Of coruse I can. I will put my xbmc.log too.

[Milhouse]

@MilhouseVH: do you want me to record the issue with my mobile phone, for example? Of coruse I can. I will put my xbmc.log too.

No, can you upload a segment of one of the videos you are having problems with, that way it may be possible for others to reproduce and investigate further.

[anonimo2001]

@MilhouseVH
The log
http://xbmclogs.com./show.php?id=145456
The file (SPOILER: True Detective episode 2) It's 80Mb
http://www.filedropper.com/truedetective

[Silo100]

Are you overclocked, does restoring stock clocks help when FIQ FSM is enabled? Just one more variable to rule out.

I've tried running "texturecache.py p" several times now (MySQL with media on NFS) and so far no problems so this isn't easy to reproduce.

No overclock at all. Everything stock. Certified SD-CARD. I went back to the latest build where it was disabled by default, all was fine. Can't provide logs as i've got no network (and can't get to the box easily). I'm just posting in the hope that other people have the same issue and it's not just me