nvidia card and kernel 3.1

[Dmitriy Tochansky]

Hello!
I bough new videocard yesterday and have some problems with it.
After inserting it to my PC, I found in dmesg:
[    9.877721] PERCPU: allocation failed, size=240 align=4, failed to
allocate new chunk
[    9.877729] Pid: 755, comm: modprobe Not tainted 3.1.0-7.fc16.i686 #1
[    9.877732] Call Trace:
[    9.877743]  [<c0808e3e>] ? printk+0x2d/0x2f
[    9.877752]  [<c04c6904>] pcpu_alloc+0x659/0x6c1
[    9.877759]  [<c0468e19>] ? sys_init_module+0x5b/0x15c7
[    9.877767]  [<c04d3233>] ? __vmalloc_node+0x54/0x5b
[    9.877773]  [<c0429d7c>] ? should_resched+0xd/0x27
[    9.877788]  [<c04c6b69>] __alloc_reserved_percpu+0x12/0x14
[    9.877794]  [<c0469169>] sys_init_module+0x3ab/0x15c7
[    9.877801]  [<c0813ad9>] ? do_page_fault+0x2a9/0x304
[    9.877807]  [<c04ff24c>] ? mntput_no_expire+0x27/0xb9
[    9.877816]  [<c0810b0c>] syscall_call+0x7/0xb
[    9.877823]  [<c044007b>] ? sys_waitid+0x4a/0x141
[    9.877827] nf_conntrack: Could not allocate 240 bytes percpu data
[    9.878335] modprobe[755]: FATAL: Error inserting xt_state
(/lib/modules/3.1.0-7.fc16.i686/kernel/net/netfilter/xt_state.ko): Cannot
allocate memory
[    9.878664] iptables.init[738]: iptables: Applying firewall rules:
iptables-restore: line 12 failed
[    9.883352] iptables.init[738]: [FAILED]
[    9.909336] ip6_tables: (C) 2000-2006 Netfilter Core Team
[    9.917425] PERCPU: allocation failed, size=240 align=4, failed to
allocate new chunk
[    9.917433] Pid: 777, comm: modprobe Not tainted 3.1.0-7.fc16.i686 #1
[    9.917437] Call Trace:
[    9.917448]  [<c0808e3e>] ? printk+0x2d/0x2f
[    9.917456]  [<c04c6904>] pcpu_alloc+0x659/0x6c1
[    9.917463]  [<c0468e19>] ? sys_init_module+0x5b/0x15c7
[    9.917471]  [<c04d3233>] ? __vmalloc_node+0x54/0x5b
[    9.917477]  [<c0429d7c>] ? should_resched+0xd/0x27
[    9.917492]  [<c04c6b69>] __alloc_reserved_percpu+0x12/0x14
[    9.917497]  [<c0469169>] sys_init_module+0x3ab/0x15c7
[    9.917504]  [<c0813ad9>] ? do_page_fault+0x2a9/0x304
[    9.917511]  [<c04ff24c>] ? mntput_no_expire+0x27/0xb9
[    9.917520]  [<c0810b0c>] syscall_call+0x7/0xb
[    9.917527]  [<c044007b>] ? sys_waitid+0x4a/0x141
[    9.917531] nf_conntrack: Could not allocate 240 bytes percpu data
[    9.917802] modprobe[777]: FATAL: Error inserting xt_state
(/lib/modules/3.1.0-7.fc16.i686/kernel/net/netfilter/xt_state.ko): Cannot
allocate memory

Much other problems like:
[root@home toch]# modprobe fuse
FATAL: Error inserting fuse
(/lib/modules/3.1.0-7.fc16.i686/kernel/fs/fuse/fuse.ko): Cannot allocate
memory

Boot with kernel 3.3 is failed with blackscreen.

[Dave Jones]

On Mon, Mar 26, 2012 at 09:11:25AM +0400, Dmitriy Tochansky wrote:
> Much other problems like:
> [root@home toch]# modprobe fuse
> FATAL: Error inserting fuse
> (/lib/modules/3.1.0-7.fc16.i686/kernel/fs/fuse/fuse.ko): Cannot allocate
> memory
>
> Boot with kernel 3.3 is failed with blackscreen.

Looks like something ate the whole vmalloc space.
Maybe nouveau is over-allocating. Can you file a bug, and attach the whole dmesg please ?

Dave

[Dmitriy Tochansky]

Hello!
> > Much other problems like:
> > [root@home toch]# modprobe fuse
> > FATAL: Error inserting fuse
> > (/lib/modules/3.1.0-7.fc16.i686/kernel/fs/fuse/fuse.ko): Cannot
>
> allocate
> > memory
> > Boot with kernel 3.3 is failed with blackscreen.
>
> Looks like something ate the whole vmalloc space.
> Maybe nouveau is over-allocating. Can you file a bug, and attach the whole
> dmesg please ?

I worked around this problem with "acpi=off" parameter in grub. I can
reproduce this problem later, when come back home.

[Ben Skeggs]

On Mon, 2012-03-26 at 13:36 -0400, Dave Jones wrote:
> On Mon, Mar 26, 2012 at 09:11:25AM +0400, Dmitriy Tochansky wrote:
> > Much other problems like:
> > [root@home toch]# modprobe fuse
> > FATAL: Error inserting fuse
> > (/lib/modules/3.1.0-7.fc16.i686/kernel/fs/fuse/fuse.ko): Cannot allocate
> > memory
> >
> > Boot with kernel 3.3 is failed with blackscreen.
>
> Looks like something ate the whole vmalloc space.
> Maybe nouveau is over-allocating. Can you file a bug, and attach the whole dmesg please ?

I doubt we're over-allocating.  However, NVIDIA GPUs have *huge*
register BARs that we need to map.  We already limit the size of the
mappings to what we can reasonably manage without making the code
extremely convoluted.

The problem will be worse with the NVIDIA binary driver, they map the
entire BAR0 (16MiB usually, 64MiB on some cards) and BAR3 (16MiB/32MiB
depending on the card), always.

Ben.
> Dave
>
> kernel mailing list
> [e-mail]
> https://admin.fedoraproject.org/mailman/listinfo/kernel