nV News Forums

 
 

nV News Forums (http://www.nvnews.net/vbulletin/index.php)
-   NVIDIA Linux (http://www.nvnews.net/vbulletin/forumdisplay.php?f=14)
-   -   195.36.15 weird MSI behaviour (http://www.nvnews.net/vbulletin/showthread.php?t=149171)

zbiggy 03-21-10 03:57 PM

195.36.15 weird MSI behaviour
 
1 Attachment(s)
Hello,

I have done quite interesting experiment:
I usually have #options nvidia NVreg_EnableMSI=1 commented out in /etc/modprobe.d/options.conf because it freezes Linux permanently in a short time.

Today I loaded nvidia as usual without MSI. Then I exited X, rmmod nvidia, uncommented EnableMSI in options.conf. Then loaded X again. In /proc/interrupts I see nvidia uses MSI. The older irq nvidia module used before is empty, it is listed on interrupts list but with empty name. Nvidia module is present on MSI interrupt.
xrender tests passed oK and glxgears too. Only VDPAU and Xv refused to work - mplayer hangs.

Then I rebooted PC. Nvidia jumped in to MSI. This time there is no empty remaining place where nvidia usually loads. X become frozen right after loading KDE4. I was unable to move mouse cursor.

Summary:
If nvidia loads as MSI only it freezes machine.
If nvidia loads as default, then removed and loaded as MSI it works except Xv and VDPAU because after removal something with empty name is still registered at IRQ nvidia was using before.

According to findings on the Internet only Geforce 8200 can not work in MSI mode. All recent add on Geforces work great in MSI mode.
Can you please fix MSI for Geforce 8200?

Milena 03-21-10 06:13 PM

Re: 195.36.15 weird MSI behaviour
 
Hi

Just a quick question about MSI,

if it doesn't cause trouble is it recommend to make use of it and if so how do i enable it properly ?
There is no /etc/modprobe.d/options.conf on my Fedora 12 installation. Creating this file with the entry
Code:

options nvidia NVreg_EnableMSI=1
is all i have to do ? I asked about this recently on my distributions forum > http://forums.fedoraforum.org/showthread.php?t=242523

Best Regards

Milena 03-21-10 06:25 PM

Re: 195.36.15 weird MSI behaviour
 
Well I just added an options.conf file and it seems to work as intended.

before:

Code:

          CPU0      CPU1      CPU2      CPU3     
 17:          5          4          0        52  IO-APIC-fasteoi  firewire_ohci
 19:          0          0    188043        857  IO-APIC-fasteoi  nvidia
 20:      51279      23962          1        59  IO-APIC-fasteoi  ohci_hcd:usb3
 23:          0          0          0          0  IO-APIC-fasteoi  ohci_hcd:usb4
 27:          1      27922      14110      6741  PCI-MSI-edge      ahci
 28:      23770    106265          2        135  PCI-MSI-edge      eth0

after:

Code:

bash-4.0$ cat /proc/interrupts
          CPU0      CPU1      CPU2      CPU3     
 17:          0          0          0        10  IO-APIC-fasteoi  firewire_ohci
 20:          0          1          1      16930  IO-APIC-fasteoi  ohci_hcd:usb3
 21:          1        406      2379      3638  IO-APIC-fasteoi  ehci_hcd:usb2, HDA Intel
 22:          0          0          0          4  IO-APIC-fasteoi  ehci_hcd:usb1
 23:          0          0          0          0  IO-APIC-fasteoi  ohci_hcd:usb4
 27:          1      10205        16      7515  PCI-MSI-edge      ahci
 28:      5432          0          3        154  PCI-MSI-edge      eth0
 29:          0          2      24344      10982  PCI-MSI-edge      nvidia
NMI:          0          0          0          0  Non-maskable interrupts

Compiz, XV, VDPAU work without problems so far and no X crash but it's a GTX275 here.

zbiggy 03-22-10 06:29 PM

Re: 195.36.15 weird MSI behaviour
 
Yes you are using MSI now.

I would love to use MSI too. But I can't - MSI on Geforce 8200 hangs terribly. System is so frozen that only hard reset can help. After freeze and reboot the disk may become corrupted. It happened to me after few seconds of X use. That is why I do not test MSI often. Just hope from time to time Nvidia will fix this - integrated GPUs are a little bit different than addon GPUs.

MSI is working great for me in SATA, Eth and integrated audio driver.
You can move your HDA Intel audio which is shared on INT 21 according to your logs also to MSI. This way you will avoid IRQ sharing which may make some troubles under load and improve overall performance a little bit.


Code:

options snd_hda_intel enable_msi=1

zbiggy 03-22-10 06:47 PM

Re: 195.36.15 weird MSI behaviour
 
Quote:

Originally Posted by Milena (Post 2214351)
Just a quick question about MSI,
if it doesn't cause trouble is it recommend to make use of it and if so how do i enable it properly ?
There is no /etc/modprobe.d/options.conf on my Fedora 12 installation. Creating this file with the entry
Code:

options nvidia NVreg_EnableMSI=1
is all i have to do ? I asked about this recently on my distributions forum > http://forums.fedoraforum.org/showthread.php?t=242523

Best Regards

The answer to all question is yes. Moving devices to MSI frees APIC or PIC interrupts. This way you can avoid IRQ sharing and can connect more devices. More advantages here:
http://en.wikipedia.org/wiki/Message...led_Interrupts

My Geforce 8200 (MCP78S) is treated very badly by Nvidia:
- enabling MSI for integrated GPU freezes my computer
- USB 1.1 OHCI hangs in APIC mode since I bought this mainboard.
Nvidia says they quit chipset business. After ongoing chipset success since year 2000 the last one is so crappy. How I can recommend their hardware seeing how crappy it is.
I hope they do something with these 2 ugly bugs.

zbiggy 03-22-10 07:10 PM

Re: 195.36.15 weird MSI behaviour
 
You can list all devices on your system with MSI capability:
(execute this command as root or use sudo in ubuntu like systems):
Code:

lspci -vvv | grep 'Message Signalled Interrupts'
The result will be like this:
Code:

        Capabilities: [b0] Message Signalled Interrupts: Mask- 64bit+ Queue=0/3 Enable+
        Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/1 Enable+
        Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/1 Enable+
        Capabilities: [50] Message Signalled Interrupts: Mask- 64bit+ Queue=0/1 Enable+
        Capabilities: [68] Message Signalled Interrupts: Mask- 64bit+ Queue=0/0 Enable-

Enable+ means MSI is enabled and in use
Enable- means MSI is disabled and can be enabled (if driver supports MSI) in options.conf for example.

As you see I use MSI for everything except Geforce 8200.

Milena 03-22-10 09:14 PM

Re: 195.36.15 weird MSI behaviour
 
Thank you zbiggy for the info

The command lspci -vvv | grep 'Message Signalled Interrupts' gives no output at all on Fedora 12, not as user nor root. Also thanks for the hint with snd_hda_intel but its the onboard soundchip of my ASUS M4N82 (Nvidia chipset) and it stays in IO-APIC-fasteoi mode regardless what setting, i tried to add it to options.conf as well as boot option for the kernel. My GTX275 card works fine with MSI and very stable, i hope Nvidia will fix the issue for older models soon.

Best Regards

zbiggy 03-23-10 04:25 AM

Re: 195.36.15 weird MSI behaviour
 
Quote:

Originally Posted by Milena (Post 2215425)
Thank you zbiggy for the info

The command lspci -vvv | grep 'Message Signalled Interrupts' gives no output at all on Fedora 12, not as user nor root.

Strange. I tried this on RedHat EL 5.4 and works on root. Indeed when I tried as normal user it returned 'command not found'. Will try on Fedora when come back home. Maybe you do not have lspci command installed.

Quote:

Originally Posted by Milena (Post 2215425)
Also thanks for the hint with snd_hda_intel but its the onboard soundchip of my ASUS M4N82 (Nvidia chipset) and it stays in IO-APIC-fasteoi mode regardless what setting, i tried to add it to options.conf as well as boot option for the kernel. My GTX275 card works fine with MSI and very stable, i hope Nvidia will fix the issue for older models soon.

snd_hda_intel is only driver name common for IHDA itegrated audio in new chipsets from Nvidia/AMD/Intel. I have AMD CPU and mainboard and use snd_hda_intel. Do
Code:

lsmod | grep snd_
and send me the output. Maybe you use AC97 instead of IHDA.

MSI for snd_hda_intel driver is present since alsa 1.0.13 if your alsa is older you do not have MSI support yet.

Milena 03-23-10 08:52 PM

Re: 195.36.15 weird MSI behaviour
 
Well the output of lsmod | grep snd_ is:

Code:

bash-4.0$ lsmod | grep snd_
snd_hda_codec_realtek  279457  1
snd_hda_intel          23712  2
snd_hda_codec          71956  2 snd_hda_codec_realtek,snd_hda_intel
snd_seq                52773  0
snd_usb_audio          85310  1
snd_usb_lib            16364  1 snd_usb_audio
snd_pcm                78375  3 snd_hda_intel,snd_hda_codec,snd_usb_audio
snd_rawmidi            20374  1 snd_usb_lib
snd_seq_device          6151  2 snd_seq,snd_rawmidi
snd_hwdep              6446  2 snd_hda_codec,snd_usb_audio
snd_timer              19840  2 snd_seq,snd_pcm
snd                    62376  17 snd_hda_codec_realtek,snd_hda_intel,snd_hda_codec,snd_seq,snd_usb_audio,snd_usb_lib,snd_pcm,snd_rawmidi,snd_seq_device,snd_hwdep,snd_timer
snd_page_alloc          7389  2 snd_hda_intel,snd_pcm

It is the snd_hda_intel module, a few kernels ago i had to pass snd_hda_intel.powersave=0 to the grub kernel options to avoid crackling sounds and that worked too. I will try again to set it up for MSI.

stecklum 03-25-10 03:06 AM

Re: 195.36.15 weird MSI behaviour
 
Thanks for bringing the NVreg_EnableMSI option to my attention. I was struggling with SATA timeout/exception issues on my M1330 equipped with a GeForce 8400M GS from the beginning.I tried all recipes I found so far but none worked. Since this option was activated in /etc/modprobe.d/options no more timeouts were encountered (with driver 195.36.15). There seem to be quite a number of people facing this problem. For those having NVIDIA graphics cards a hint on this option in the driver README to enable MSI would perhaps be helpful.

zbiggy 03-25-10 06:00 AM

Re: 195.36.15 weird MSI behaviour
 
I hope in future Nvidia driver releases I will also be able to use MSI on my Geforce 8200. I hope this may fix or make less frequent my OHCI usb 1.1 timeouts/hangs/crashes. I'm glad I could help you.

stecklum 04-15-10 02:09 PM

Re: 195.36.15 weird MSI behaviour
 
Again, I was too hastily with my conclusion. The timeouts are not gone but less frequent (still an improvement). Further googling brought this to my attention which keeps my hope alive that sometime soon the problem will be solved

LKML: Linus Torvalds: Re: [git patches] libata updates for 2.6.34
http://lkml.org/lkml/2010/3/9/345


All times are GMT -5. The time now is 01:59 AM.

Powered by vBulletin® Version 3.7.1
Copyright ©2000 - 2014, Jelsoft Enterprises Ltd.
Copyright 1998 - 2014, nV News.