* Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Patrick McHardy @ 2010-02-02 16:48 UTC (permalink / raw)
To: Jon Masters
Cc: Eric Dumazet, Alexey Dobriyan, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <1265129192.2861.141.camel@tonnant>
Jon Masters wrote:
> On Tue, 2010-02-02 at 06:35 -0500, Jon Masters wrote:
>
>> I think there's something more fundamental going on here.
>
> What happens is the conntrack code attempts to free
> nf_conntrack_untracked back into the SL[U]B cache from which it
> allocates other ct's.
That shouldn't happen, the untracked conntrack is initialized to a
refcount of 1, which is never released.
> There's just one problem...that's a static struct
> not from the cache. So, this is why we end up with the SLAB being
> corrupted and the address immediately following the
> nf_conntrack_untracked being corrupted.
>
> I shoved some debug comments into the destroy code to see if we were
> trying to free nf_conntrack_untracked, and bingo. I have shoved a panic
> in there now, will send you a backtrace.
Thanks.
^ permalink raw reply
* PROBLEM with summary: Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Jon Masters @ 2010-02-02 16:58 UTC (permalink / raw)
To: Eric Dumazet
Cc: Alexey Dobriyan, Patrick McHardy, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <1265129192.2861.141.camel@tonnant>
On Tue, 2010-02-02 at 11:46 -0500, Jon Masters wrote:
> On Tue, 2010-02-02 at 06:35 -0500, Jon Masters wrote:
>
> > I think there's something more fundamental going on here.
>
> What happens is the conntrack code attempts to free
> nf_conntrack_untracked back into the SL[U]B cache from which it
> allocates other ct's. There's just one problem...that's a static struct
> not from the cache. So, this is why we end up with the SLAB being
> corrupted and the address immediately following the
> nf_conntrack_untracked being corrupted.
>
> I shoved some debug comments into the destroy code to see if we were
> trying to free nf_conntrack_untracked, and bingo. I have shoved a panic
> in there now, will send you a backtrace.
So we attach after starting a VM due to icmpv6_error setting the ct in
some incoming skb to the "untracked" catchall one. Then we panic when I
force a panic if attempting to free the nf_conntrack_untracked static
struct with this backtrace:
#5 0xffffffff81455884 in panic (
fmt=0xffffffff81ec51e0 "JCM: nf_conntrack_destroy: trying to destroy
nf_conntrack_untracked!\n")
at kernel/panic.c:73
#6 0xffffffff813d266c in nf_conntrack_destroy (nfct=<value optimized
out>) at net/netfilter/core.c:244
#7 0xffffffff813abd97 in nf_conntrack_put (skb=0xffff880223b8b700) at
include/linux/skbuff.h:1924
#8 skb_release_head_state (skb=0xffff880223b8b700) at
net/core/skbuff.c:402
#9 0xffffffff813abaf9 in skb_release_all (skb=0xffff880223b8b700) at
net/core/skbuff.c:420
#10 __kfree_skb (skb=0xffff880223b8b700) at net/core/skbuff.c:435
#11 0xffffffff813abbfe in kfree_skb (skb=0xffff880223b8b700) at
net/core/skbuff.c:456
Could you please add (or recommend for me to test) some logic to catch
attempts to free the nf_conntrack_untracked and prevent it? Also, maybe
in the init_net code you could remove the re-initialization of the
untracked conntrack each time a namespace is created?
Thanks!
Jon.
^ permalink raw reply
* Re: [PATCH net-next-2.6] net: use helpers to access mc list
From: David Miller @ 2010-02-02 17:03 UTC (permalink / raw)
To: jpirko; +Cc: netdev
In-Reply-To: <20100202163107.GC2601@psychotron.lab.eng.brq.redhat.com>
From: Jiri Pirko <jpirko@redhat.com>
Date: Tue, 2 Feb 2010 17:31:08 +0100
> Tue, Feb 02, 2010 at 05:06:01PM CET, davem@davemloft.net wrote:
>>From: Jiri Pirko <jpirko@redhat.com>
>>Date: Tue, 2 Feb 2010 17:03:48 +0100
>>
>>> You mean "struct dev_mc_list"? But that would solve nothing. If I would still
>>> use current structure in drivers, then still the migration to struct_hw_addr
>>> would be all-at-once for all drivers :( This patch was exacly made to avoid
>>> this.
>>
>>I think changing the iterator type will have to be done
>>wholesale in one changeset, there is no reasonable way
>>to avoid it.
>
> So what you are proposing is to change drivers' code to use macro to iterate
> through lists as step n1 and then change the iterator type as step n2 right?
>
> If we would use my patch, we would have this done in single step. But I understand
> that a kind of ugliness patch introduces is simply not acceptable not even for a
> short time period, right?
Right.
^ permalink raw reply
* Re: PROBLEM with summary: Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Patrick McHardy @ 2010-02-02 17:04 UTC (permalink / raw)
To: Jon Masters
Cc: Eric Dumazet, Alexey Dobriyan, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <1265129903.2861.150.camel@tonnant>
Jon Masters wrote:
> On Tue, 2010-02-02 at 11:46 -0500, Jon Masters wrote:
>> On Tue, 2010-02-02 at 06:35 -0500, Jon Masters wrote:
>>
>>> I think there's something more fundamental going on here.
>> What happens is the conntrack code attempts to free
>> nf_conntrack_untracked back into the SL[U]B cache from which it
>> allocates other ct's. There's just one problem...that's a static struct
>> not from the cache. So, this is why we end up with the SLAB being
>> corrupted and the address immediately following the
>> nf_conntrack_untracked being corrupted.
>>
>> I shoved some debug comments into the destroy code to see if we were
>> trying to free nf_conntrack_untracked, and bingo. I have shoved a panic
>> in there now, will send you a backtrace.
>
> So we attach after starting a VM due to icmpv6_error setting the ct in
> some incoming skb to the "untracked" catchall one. Then we panic when I
> force a panic if attempting to free the nf_conntrack_untracked static
> struct with this backtrace:
>
> #5 0xffffffff81455884 in panic (
> fmt=0xffffffff81ec51e0 "JCM: nf_conntrack_destroy: trying to destroy
> nf_conntrack_untracked!\n")
> at kernel/panic.c:73
> #6 0xffffffff813d266c in nf_conntrack_destroy (nfct=<value optimized
> out>) at net/netfilter/core.c:244
> #7 0xffffffff813abd97 in nf_conntrack_put (skb=0xffff880223b8b700) at
> include/linux/skbuff.h:1924
> #8 skb_release_head_state (skb=0xffff880223b8b700) at
> net/core/skbuff.c:402
> #9 0xffffffff813abaf9 in skb_release_all (skb=0xffff880223b8b700) at
> net/core/skbuff.c:420
> #10 __kfree_skb (skb=0xffff880223b8b700) at net/core/skbuff.c:435
> #11 0xffffffff813abbfe in kfree_skb (skb=0xffff880223b8b700) at
> net/core/skbuff.c:456
>
> Could you please add (or recommend for me to test) some logic to catch
> attempts to free the nf_conntrack_untracked and prevent it? Also, maybe
> in the init_net code you could remove the re-initialization of the
> untracked conntrack each time a namespace is created?
Ah nice catch, that seems to be the problem. When the untracked
conntrack is already attached to an skb and thus has refcnt > 1
and we re-initalize the refcnt, it will get freed.
The question is whether the ct_net pointer of the untracked conntrack
is actually required. If so, we need one instance per namespace,
otherwise we can just move initialization and cleanup to the init_net
init/cleanup functions. Alexey, do you happen to know this?
^ permalink raw reply
* Re: netfilter 00/02: netfilter fixes
From: David Miller @ 2010-02-02 17:05 UTC (permalink / raw)
To: kaber; +Cc: netdev, netfilter-devel
In-Reply-To: <20100202162736.24867.67227.sendpatchset@x2.localnet>
From: Patrick McHardy <kaber@trash.net>
Date: Tue, 2 Feb 2010 17:27:37 +0100 (MET)
> the following two patches fix two bugs in netfilter:
>
> - an off-by-one in SIP conntrack short header parsing, causing mismatches
> with UAs not inserting a space after the colon
>
> - a missing initialization in ctnetlink when dumping an expectation mask,
> causing an invalid layer 4 protocol number to be used
>
> Please apply or pull from:
>
> git://git.kernel.org/pub/scm/linux/kernel/git/kaber/nf-2.6.git master
Pulled, thanks Patrick.
^ permalink raw reply
* Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Jon Masters @ 2010-02-02 17:07 UTC (permalink / raw)
To: Patrick McHardy
Cc: Eric Dumazet, Alexey Dobriyan, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <4B685756.8010107@trash.net>
On Tue, 2010-02-02 at 17:48 +0100, Patrick McHardy wrote:
> Jon Masters wrote:
> > On Tue, 2010-02-02 at 06:35 -0500, Jon Masters wrote:
> >
> >> I think there's something more fundamental going on here.
> >
> > What happens is the conntrack code attempts to free
> > nf_conntrack_untracked back into the SL[U]B cache from which it
> > allocates other ct's.
>
> That shouldn't happen, the untracked conntrack is initialized to a
> refcount of 1, which is never released.
Ah, but I think it is :) It's also re-initialized (with an atomic_set)
every time a new namespace is created, whereas this should probably only
be done in the init_init_net code, not in init_net :)
> > There's just one problem...that's a static struct
> > not from the cache. So, this is why we end up with the SLAB being
> > corrupted and the address immediately following the
> > nf_conntrack_untracked being corrupted.
> >
> > I shoved some debug comments into the destroy code to see if we were
> > trying to free nf_conntrack_untracked, and bingo. I have shoved a panic
> > in there now, will send you a backtrace.
>
> Thanks.
No problem. And thanks for your help. I'm sorry if I sound frustrated at
this, it's just causing all of my test machines running KVM guests to
fall over :)
Jon.
^ permalink raw reply
* Re: PROBLEM with summary: Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Eric Dumazet @ 2010-02-02 17:16 UTC (permalink / raw)
To: Patrick McHardy
Cc: Jon Masters, Alexey Dobriyan, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <4B685B14.1040207@trash.net>
Le mardi 02 février 2010 à 18:04 +0100, Patrick McHardy a écrit :
> Ah nice catch, that seems to be the problem. When the untracked
> conntrack is already attached to an skb and thus has refcnt > 1
> and we re-initalize the refcnt, it will get freed.
>
> The question is whether the ct_net pointer of the untracked conntrack
> is actually required. If so, we need one instance per namespace,
> otherwise we can just move initialization and cleanup to the init_net
> init/cleanup functions. Alexey, do you happen to know this?
>
One untracked per netns seems the way to go, and move it outside of
read_mostly area too, we obviously can modify its refcount frequently...
--
To unsubscribe from this list: send the line "unsubscribe netfilter-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply
* Re: PROBLEM with summary: Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Jon Masters @ 2010-02-02 17:23 UTC (permalink / raw)
To: Eric Dumazet
Cc: Patrick McHardy, Alexey Dobriyan, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <1265130990.2953.176.camel@edumazet-laptop>
On Tue, 2010-02-02 at 18:16 +0100, Eric Dumazet wrote:
> Le mardi 02 février 2010 à 18:04 +0100, Patrick McHardy a écrit :
>
> > Ah nice catch, that seems to be the problem. When the untracked
> > conntrack is already attached to an skb and thus has refcnt > 1
> > and we re-initalize the refcnt, it will get freed.
> >
> > The question is whether the ct_net pointer of the untracked conntrack
> > is actually required. If so, we need one instance per namespace,
> > otherwise we can just move initialization and cleanup to the init_net
> > init/cleanup functions. Alexey, do you happen to know this?
> >
>
> One untracked per netns seems the way to go, and move it outside of
> read_mostly area too, we obviously can modify its refcount frequently...
Sure, that will work. Also, rather than just the NF_CT_ASSERT on the use
count, maybe worth catching the specific case of trying to free the
untracked ct, but that's only if it's not a horrible fast path.
Anyway, thanks. If you want to send me a patch, I'll try it.
Jon.
^ permalink raw reply
* Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Alexey Dobriyan @ 2010-02-02 17:58 UTC (permalink / raw)
To: Jon Masters
Cc: Patrick McHardy, Eric Dumazet, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <1265130426.2861.158.camel@tonnant>
On Tue, Feb 2, 2010 at 7:07 PM, Jon Masters <jonathan@jonmasters.org> wrote:
> No problem. And thanks for your help. I'm sorry if I sound frustrated at
> this, it's just causing all of my test machines running KVM guests to
> fall over :)
So it's my bug either way. :-(
Yes, moving to init_net-only function is fine.
^ permalink raw reply
* Re: [PATCH v2 2/2] virtio_net: Defer skb allocation in receive path
From: Shirley Ma @ 2010-02-02 18:07 UTC (permalink / raw)
To: Michael S. Tsirkin; +Cc: Amit Shah, Rusty Russell, Avi Kivity, netdev, kvm
In-Reply-To: <20100129140553.GA4413@redhat.com>
On Fri, 2010-01-29 at 16:05 +0200, Michael S. Tsirkin wrote:
>
> Now that's in, how does the send patch look?
Thanks. I will submit it today. It's a simple patch.
Shirley
^ permalink raw reply
* Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Jon Masters @ 2010-02-02 18:16 UTC (permalink / raw)
To: Alexey Dobriyan
Cc: Patrick McHardy, Eric Dumazet, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <b6fcc0a1002020958u17315624u4d54d06a413be6bb@mail.gmail.com>
On Tue, 2010-02-02 at 19:58 +0200, Alexey Dobriyan wrote:
> Yes, moving to init_net-only function is fine.
So moving the "setup up fake conntrack" bits to init_init_net from
init_net still results in the panic, which means that the use count
really is dropping to zero and we really are trying to free it when
using multiple namespaces. Per ns is probably an easier way to go.
Just for kicks, I'll have it error out on attempting to free to see if I
can get this box to stay up for a while.
Jon.
^ permalink raw reply
* Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Jon Masters @ 2010-02-02 18:34 UTC (permalink / raw)
To: Alexey Dobriyan
Cc: Patrick McHardy, Eric Dumazet, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <1265134598.2861.191.camel@tonnant>
On Tue, 2010-02-02 at 13:16 -0500, Jon Masters wrote:
> On Tue, 2010-02-02 at 19:58 +0200, Alexey Dobriyan wrote:
>
> > Yes, moving to init_net-only function is fine.
>
> So moving the "setup up fake conntrack" bits to init_init_net from
> init_net still results in the panic, which means that the use count
> really is dropping to zero and we really are trying to free it when
> using multiple namespaces. Per ns is probably an easier way to go.
>
> Just for kicks, I'll have it error out on attempting to free to see if I
> can get this box to stay up for a while.
Confirmed. It boots and the hashsize is not inadvertedly corrupted if I
do the following:
void nf_conntrack_destroy(struct nf_conntrack *nfct)
{
void (*destroy)(struct nf_conntrack *);
if ((struct nf_conn *)nfct == &nf_conntrack_untracked) {
printk("JCM: nf_conntrack_destroy: trying to destroy
nf_conntrack_untracked! CONTINUING...\n");
//panic("JCM: nf_conntrack_destroy: trying to destroy
nf_conntrack_untracked!\n");
return; /* refuse to free nf_conntrack_untracked */
}
rcu_read_lock();
destroy = rcu_dereference(nf_ct_destroy);
BUG_ON(destroy == NULL);
destroy(nfct);
rcu_read_unlock();
}
EXPORT_SYMBOL(nf_conntrack_destroy);
Clearly that's just a hack (though catching the specific attempt to free
the untracked conntrack sounds like a very good idea in general). I will
leave this running for a while, but so far no problems:
[jcm@perihelion jcm_26]$ dmesg|grep JCM
[ 29.952717] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 30.207091] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 30.403248] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 31.403319] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 32.977106] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 33.347100] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 33.966092] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 34.967111] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 35.404323] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 35.911430] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 35.912442] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 38.967061] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 39.403342] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 41.288392] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 42.966063] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 44.036451] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 95.621174] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 95.673061] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 95.741180] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 96.273429] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 96.741296] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 97.246108] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 98.246486] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 100.747170] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 102.252067] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 104.747066] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 105.762433] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 106.252084] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 147.260998] JCM: nf_conntrack_destroy: trying to destroy
nf_conntrack_untracked! CONTINUING...
[ 155.485160] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 156.403661] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 157.402928] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 161.402477] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 163.270622] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 163.277368] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 167.718677] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 176.901877] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 182.554024] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 182.616693] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 183.616932] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 187.617026] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 188.362002] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 191.617001] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 193.574899] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 197.455338] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 214.163491] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 214.311528] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 215.311577] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 219.317496] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 220.693493] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 223.317095] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 240.326198] JCM: nf_conntrack_destroy: trying to destroy
nf_conntrack_untracked! CONTINUING...
[ 252.481116] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 252.881087] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 253.881251] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 257.922816] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 258.882989] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 261.413093] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 261.921943] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
[ 266.350983] JCM: icmpv6_error: attaching to nf_conntrack_untracked.
So, over to you :)
Jon.
^ permalink raw reply
* Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Patrick McHardy @ 2010-02-02 18:36 UTC (permalink / raw)
To: Jon Masters
Cc: Alexey Dobriyan, Eric Dumazet, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <1265134598.2861.191.camel@tonnant>
Jon Masters wrote:
> On Tue, 2010-02-02 at 19:58 +0200, Alexey Dobriyan wrote:
>
>> Yes, moving to init_net-only function is fine.
>
> So moving the "setup up fake conntrack" bits to init_init_net from
> init_net still results in the panic, which means that the use count
> really is dropping to zero and we really are trying to free it when
> using multiple namespaces. Per ns is probably an easier way to go.
Agreed, that will also avoid problems in the future with the
ct_net pointer pointing to &init_net. I'll take care of this
tommorrow.
^ permalink raw reply
* Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Jon Masters @ 2010-02-02 18:39 UTC (permalink / raw)
To: Patrick McHardy
Cc: Alexey Dobriyan, Eric Dumazet, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <4B6870AF.6060109@trash.net>
On Tue, 2010-02-02 at 19:36 +0100, Patrick McHardy wrote:
> Jon Masters wrote:
> > On Tue, 2010-02-02 at 19:58 +0200, Alexey Dobriyan wrote:
> >
> >> Yes, moving to init_net-only function is fine.
> >
> > So moving the "setup up fake conntrack" bits to init_init_net from
> > init_net still results in the panic, which means that the use count
> > really is dropping to zero and we really are trying to free it when
> > using multiple namespaces. Per ns is probably an easier way to go.
>
> Agreed, that will also avoid problems in the future with the
> ct_net pointer pointing to &init_net. I'll take care of this
> tommorrow.
Ok. I'll leave this box running with the hack. I think at the very least
that this specific issue needs to get fixed and in the stable tree, then
the other bits (per namespace cachep...) are probably a good idea at the
same time but that's up to you.
Thanks for your help!
Jon.
^ permalink raw reply
* Re: [PATCH] netfilter: per netns nf_conntrack_cachep
From: Jon Masters @ 2010-02-02 18:42 UTC (permalink / raw)
To: Patrick McHardy
Cc: Alexey Dobriyan, Eric Dumazet, linux-kernel, netdev,
netfilter-devel, Paul E. McKenney
In-Reply-To: <1265135982.2861.199.camel@tonnant>
On Tue, 2010-02-02 at 13:39 -0500, Jon Masters wrote:
> On Tue, 2010-02-02 at 19:36 +0100, Patrick McHardy wrote:
> > Jon Masters wrote:
> > > On Tue, 2010-02-02 at 19:58 +0200, Alexey Dobriyan wrote:
> > >
> > >> Yes, moving to init_net-only function is fine.
> > >
> > > So moving the "setup up fake conntrack" bits to init_init_net from
> > > init_net still results in the panic, which means that the use count
> > > really is dropping to zero and we really are trying to free it when
> > > using multiple namespaces. Per ns is probably an easier way to go.
> >
> > Agreed, that will also avoid problems in the future with the
> > ct_net pointer pointing to &init_net. I'll take care of this
> > tommorrow.
>
> Ok. I'll leave this box running with the hack. I think at the very least
> that this specific issue needs to get fixed and in the stable tree, then
> the other bits (per namespace cachep...) are probably a good idea at the
> same time but that's up to you.
FYI, my box has the quick don't free untracked hack *and* per-ns cachep.
I don't think the latter has anything specific to do with this (though
it needs fixing also), but worth knowing my test is using both.
Back to the podcasts tonight instead of this ;)
Jon.
^ permalink raw reply
* Re: IP: Send an ICMP "Fragment Reassembly Timeout" message when enabling connection track
From: David Miller @ 2010-02-02 19:47 UTC (permalink / raw)
To: kaber; +Cc: jamie.iles, netdev, shanwei
In-Reply-To: <4B6856FF.3050904@trash.net>
From: Patrick McHardy <kaber@trash.net>
Date: Tue, 02 Feb 2010 17:46:55 +0100
> ipv4: ip_fragment: fix unbalanced rcu_read_unlock()
>
> Signed-off-by: Patrick McHardy <kaber@trash.net>
>
Applied to net-next-2.6, thanks patrick.
^ permalink raw reply
* [PATCH] seastar - SeaStar Ethernet driver
From: Kevin Pedretti @ 2010-02-02 20:58 UTC (permalink / raw)
To: netdev; +Cc: linux-kernel
[PATCH] seastar - SeaStar Ethernet driver
The following patch introduces the seastar driver for the
SeaStar network interface in Cray XT3/XT4/XT5 systems. The
driver is called 'seastar'. This patch is against 2.6.32.7.
The driver uses a simple datagram interface exported by the
SeaStar network interface to encapsulate Ethernet frames
on the Cray XT high speed network. The driver has been tested
to function correctly and is in use on Cray XT4 development
systems at Sandia.
Signed-off-by: Kevin Pedretti <ktpedre@sandia.gov>
diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/Kconfig linux-2.6.32.7/drivers/net/Kconfig
--- linux-2.6.32.7-vanilla/drivers/net/Kconfig 2010-02-02 09:10:55.000000000 -0700
+++ linux-2.6.32.7/drivers/net/Kconfig 2010-02-02 09:12:04.000000000 -0700
@@ -2760,6 +2760,17 @@ config QLGE
To compile this driver as a module, choose M here: the module
will be called qlge.
+config SEASTAR
+ tristate "Cray XT SeaStar Ethernet driver"
+ depends on PCI
+ depends on HT_IRQ
+ ---help---
+ This driver supports the Cray XT SeaStar network interface in
+ Ethernet mode.
+
+ To compile this driver as a module, choose M here. The module
+ will be called seastar.
+
source "drivers/net/sfc/Kconfig"
source "drivers/net/benet/Kconfig"
diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/Makefile linux-2.6.32.7/drivers/net/Makefile
--- linux-2.6.32.7-vanilla/drivers/net/Makefile 2010-02-02 09:10:55.000000000 -0700
+++ linux-2.6.32.7/drivers/net/Makefile 2010-02-02 09:12:04.000000000 -0700
@@ -149,6 +149,7 @@ obj-$(CONFIG_XILINX_LL_TEMAC) += ll_tema
obj-$(CONFIG_XILINX_EMACLITE) += xilinx_emaclite.o
obj-$(CONFIG_QLA3XXX) += qla3xxx.o
obj-$(CONFIG_QLGE) += qlge/
+obj-$(CONFIG_SEASTAR) += seastar/
obj-$(CONFIG_PPP) += ppp_generic.o
obj-$(CONFIG_PPP_ASYNC) += ppp_async.o
diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/seastar/firmware.c linux-2.6.32.7/drivers/net/seastar/firmware.c
--- linux-2.6.32.7-vanilla/drivers/net/seastar/firmware.c 1969-12-31 17:00:00.000000000 -0700
+++ linux-2.6.32.7/drivers/net/seastar/firmware.c 2010-02-02 09:13:44.000000000 -0700
@@ -0,0 +1,236 @@
+/*******************************************************************************
+ SeaStar NIC Linux Driver
+
+ Copyright 2009-2010 Sandia Corporation. Under the terms of Contract
+ DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government
+ retains certain rights in this software.
+
+ Copyright (c) 2009-2010 Cray Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 2 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program; if not, write to the Free Software
+ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+
+ Contact Information:
+ Kevin Pedretti <ktpedre@sandia.gov>
+ Sandia National Laboratories
+ P.O. Box 5800
+ Albuquerque, NM 87185-1319
+
+*******************************************************************************/
+
+#include <linux/netdevice.h>
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/pci.h>
+#include "firmware.h"
+#include "seastar.h"
+
+
+/**
+ * Maps a region of host memory into the SeaStar.
+ */
+static void seastar_map_host_region(struct ss_priv *ssp, const void *addr)
+{
+ /* Round addr to the nearest 128 MB */
+ unsigned long raw_paddr = __pa(addr);
+ unsigned long paddr = raw_paddr & ~((1 << 28) - 1);
+
+ htb_map[8] = 0x8000 | ((paddr >> 28) + 0);
+ htb_map[9] = 0x8000 | ((paddr >> 28) + 1);
+
+ ssp->host_region_phys = paddr;
+}
+
+
+/**
+ * Converts a kernel virtual address to a SeaStar address.
+ */
+static uint32_t virt_to_fw(struct ss_priv *ssp, void *addr)
+{
+ unsigned long saddr;
+
+ saddr = __pa(addr) - ssp->host_region_phys;
+ saddr &= (2 << 28) - 1;
+ saddr += (8 << 28);
+
+ return saddr;
+}
+
+
+/**
+ * Send a command to the Seastar.
+ */
+static uint32_t seastar_cmd(struct ss_priv *ssp, const struct command *cmd,
+ int wait_for_result)
+{
+ struct mailbox *mbox = ssp->mailbox;
+ unsigned int next_write;
+ uint32_t tail, result;
+
+ /* Copy the command into the mailbox */
+ mbox->commandq[ssp->mailbox_cached_write] = *cmd;
+ next_write = ssp->mailbox_cached_write + 1;
+ if (next_write == COMMAND_Q_LENGTH)
+ next_write = 0;
+
+ /* Wait until it is safe to advance the write pointer */
+ while (next_write == ssp->mailbox_cached_read)
+ ssp->mailbox_cached_read = mbox->commandq_read;
+
+ /* Advance the write pointer */
+ mbox->commandq_write = next_write;
+ ssp->mailbox_cached_write = next_write;
+
+ if (!wait_for_result)
+ return 0;
+
+ /* Wait for the result to arrive */
+ tail = mbox->resultq_read;
+ while (tail == mbox->resultq_write)
+ ;
+
+ /* Read the result */
+ result = mbox->resultq[tail];
+ mbox->resultq_read = (tail >= RESULT_Q_LENGTH - 1) ? 0 : tail + 1;
+
+ return result;
+}
+
+
+/**
+ * Sends a datagram transmit command to the SeaStar.
+ */
+void seastar_ip_tx_cmd(struct ss_priv *ssp, uint16_t nid, uint16_t length,
+ uint64_t address, uint16_t pending_index)
+{
+ struct command_ip_tx tx_cmd = {
+ .op = COMMAND_IP_TX,
+ .nid = nid,
+ .length = length,
+ .address = address,
+ .pending_index = pending_index,
+ };
+
+ seastar_cmd(ssp, (struct command *) &tx_cmd, 0);
+}
+
+
+/**
+ * Programs the SeaStar's HTB_BI register.
+ */
+void seastar_setup_htb_bi(uint32_t idr)
+{
+ /* Mask the APIC dest setup by Linux, causes problems with SeaStar */
+ idr &= 0xFFFF0000;
+
+ *htb_bi = 0xFD000000 | (idr >> 8);
+}
+
+
+/**
+ * Brings up the low-level Seastar hardware.
+ */
+int seastar_hw_init(struct ss_priv *ssp)
+{
+ uint32_t lower_memory = SEASTAR_HOST_BASE;
+ const int num_eq = 1;
+ uint32_t lower_pending;
+ uint32_t lower_eqcb;
+ uint32_t result;
+ struct command_init init_cmd;
+ struct command_init_eqcb eqcb_cmd;
+ struct command_mark_alive alive_cmd;
+
+ /* Read our NID from SeaStar and write it to the NIC control block */
+ niccb->local_nid = *tx_source;
+
+ printk(KERN_INFO "%s: nid %d (0x%x) version %x built %x\n",
+ __func__,
+ niccb->local_nid,
+ niccb->local_nid,
+ niccb->version,
+ niccb->build_time
+ );
+
+ /* Allocate the PPC memory */
+ lower_pending = lower_memory;
+ lower_memory += NUM_PENDINGS * FW_PENDING_SIZE;
+
+ lower_eqcb = lower_memory;
+ lower_memory = num_eq * FW_EQCB_SIZE;
+
+ /* Initialize the HTB map so that the Seastar can see our memory.
+ * Since we are only doing upper pendings, we just use the
+ * upper_pending_phys instead of the host_phys area. */
+ seastar_map_host_region(ssp, ssp);
+
+ ssp->mailbox = &seastar_mailbox[0];
+ ssp->mailbox_cached_read = ssp->mailbox->commandq_read;
+ ssp->mailbox_cached_write = ssp->mailbox->commandq_write;
+
+ /* Attempt to send a setup command to the NIC */
+ init_cmd.op = COMMAND_INIT;
+ init_cmd.process_index = 1;
+ init_cmd.uid = 0;
+ init_cmd.jid = 0;
+
+ init_cmd.num_pendings = NUM_PENDINGS;
+ init_cmd.pending_tx_limit = NUM_TX_PENDINGS;
+ init_cmd.pending_table_addr = lower_pending;
+ init_cmd.up_pending_table_addr = virt_to_fw(ssp, ssp->pending_table);
+ init_cmd.up_pending_table_ht_addr = 0;
+
+ init_cmd.num_memds = 0;
+ init_cmd.memd_table_addr = 0;
+
+ init_cmd.num_eqcbs = num_eq;
+ init_cmd.eqcb_table_addr = lower_eqcb;
+ init_cmd.eqheap_addr = virt_to_fw(ssp, ssp->eq);
+ init_cmd.eqheap_length = NUM_EQ_ENTRIES * sizeof(ssp->eq[0]);
+
+ init_cmd.shdr_table_ht_addr = 0;
+ init_cmd.result_block_addr = 0;
+ init_cmd.smb_table_addr = 0;
+
+ result = seastar_cmd(ssp, (struct command *) &init_cmd, 1);
+ if (result != 0) {
+ dev_err(&ssp->pdev->dev,
+ "init command failed, result=%d.\n", result);
+ return -1;
+ }
+
+ eqcb_cmd.op = COMMAND_INIT_EQCB;
+ eqcb_cmd.eqcb_index = 0;
+ eqcb_cmd.base = virt_to_fw(ssp, ssp->eq);
+ eqcb_cmd.count = NUM_EQ_ENTRIES;
+
+ result = seastar_cmd(ssp, (struct command *) &eqcb_cmd, 1);
+ if (result != 1) {
+ dev_err(&ssp->pdev->dev,
+ "init_eqcb command failed, result=%d.\n", result);
+ return -1;
+ }
+
+ alive_cmd.op = COMMAND_MARK_ALIVE;
+ alive_cmd.index = 1;
+
+ result = seastar_cmd(ssp, (struct command *) &alive_cmd, 1);
+ if (result != 0) {
+ dev_err(&ssp->pdev->dev,
+ "mark_alive command failed, result=%d\n", result);
+ return -1;
+ }
+
+ return 0;
+}
diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/seastar/firmware.h linux-2.6.32.7/drivers/net/seastar/firmware.h
--- linux-2.6.32.7-vanilla/drivers/net/seastar/firmware.h 1969-12-31 17:00:00.000000000 -0700
+++ linux-2.6.32.7/drivers/net/seastar/firmware.h 2010-02-02 09:15:02.000000000 -0700
@@ -0,0 +1,329 @@
+/*******************************************************************************
+ SeaStar NIC Linux Driver
+
+ Copyright 2009-2010 Sandia Corporation. Under the terms of Contract
+ DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government
+ retains certain rights in this software.
+
+ Copyright (c) 2009-2010 Cray Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 2 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program; if not, write to the Free Software
+ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+
+ Contact Information:
+ Kevin Pedretti <ktpedre@sandia.gov>
+ Sandia National Laboratories
+ P.O. Box 5800
+ Albuquerque, NM 87185-1319
+
+*******************************************************************************/
+
+#ifndef _SEASTAR_FIRMWARE_H
+#define _SEASTAR_FIRMWARE_H
+
+
+/**
+ * Number of entries in Host -> SeaStar command queue.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+#define COMMAND_Q_LENGTH 63
+
+
+/**
+ * Number of entries in SeaStar -> Host result queue.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+#define RESULT_Q_LENGTH 2
+
+
+/**
+ * SeaStar -> Host event types.
+ *
+ * WARNING: These must match the definitions used by the
+ * closed-source SeaStar firmware.
+ */
+#define EVENT_TX_END 125
+#define EVENT_RX 126
+#define EVENT_RX_EMPTY 127
+
+
+/**
+ * Host -> SeaStar command types.
+ *
+ * WARNING: These must match the definitions used by the
+ * closed-source SeaStar firmware.
+ */
+#define COMMAND_INIT 0
+#define COMMAND_MARK_ALIVE 1
+#define COMMAND_INIT_EQCB 2
+#define COMMAND_IP_TX 13
+
+
+/**
+ * Number of entries in the incoming datagram buffer table.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+#define NUM_SKBS 64
+
+
+/**
+ * Size of the pending structure used by the SeaStar firmware.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+#define FW_PENDING_SIZE 32
+
+
+/**
+ * Size of the event queue control block structure used by the SeaStar firmware.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+#define FW_EQCB_SIZE 32
+
+
+/**
+ * SeaStar addresses of important structures in SeaStar memory.
+ *
+ * WARNING: These must match the definitions used by the
+ * closed-source SeaStar firmware.
+ */
+#define SEASTAR_SCRATCH_BASE 0xFFFA0000
+#define SEASTAR_TX_SOURCE 0xFFE00108
+#define SEASTAR_MAILBOX_BASE 0xFFFA0000
+#define SEASTAR_SKB_BASE 0xFFFA4000
+#define SEASTAR_HOST_BASE 0xFFFA5000
+#define SEASTAR_HTB_BASE 0xFFE20000
+#define SEASTAR_HTB_BI 0xFFE20048
+#define SEASTAR_NICCB_BASE 0xFFFFE000
+
+
+/**
+ * Kernel virtual address where the SeaStar memory is mapped.
+ */
+#define SEASTAR_VIRT_BASE (0xFFFFFFFFull << 32)
+
+
+/**
+ * Kernel virtual address of the SeaStar's NIC control block.
+ */
+static volatile struct niccb * const niccb
+ = (void *)(SEASTAR_VIRT_BASE + SEASTAR_NICCB_BASE);
+
+
+/**
+ * Kernel virtual address of the SeaStar's HTB_BI register.
+ */
+static volatile uint32_t * const htb_bi
+ = (void *)(SEASTAR_VIRT_BASE + SEASTAR_HTB_BI);
+
+
+/**
+ * Kernel virtual address of the SeaStar's HyperTransport map.
+ */
+static volatile uint32_t * const htb_map
+ = (void *)(SEASTAR_VIRT_BASE + SEASTAR_HTB_BASE);
+
+
+/**
+ * Kernel virtual address of the Host <-> SeaStar mailbox.
+ */
+static struct mailbox * const seastar_mailbox
+ = (void *)(SEASTAR_VIRT_BASE + SEASTAR_MAILBOX_BASE);
+
+
+/**
+ * Kernel virtual address of the incoming datagram buffer table.
+ */
+static volatile uint64_t * const seastar_skb
+ = (void *)(SEASTAR_VIRT_BASE + SEASTAR_SKB_BASE);
+
+
+/**
+ * Kernel virtual address of the SeaStar TX Source register.
+ */
+static volatile uint16_t * const tx_source
+ = (void *)(SEASTAR_VIRT_BASE + SEASTAR_TX_SOURCE);
+
+
+/**
+ * The SeaStar NIC Control Block.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+struct niccb {
+ uint32_t version; /* 0 */
+ uint8_t pad[24];
+ uint32_t build_time; /* 28 */
+ uint8_t pad2[68];
+ uint32_t ip_tx; /* 100 */
+ uint32_t ip_tx_drop; /* 104 */
+ uint32_t ip_rx; /* 108 */
+ uint32_t ip_rx_drop; /* 112 */
+ uint8_t pad3[52];
+ uint16_t local_nid; /* 168 */
+} __attribute__((packed, aligned));
+
+
+/**
+ * SeaStar datagram packet wire header.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+struct sshdr {
+ uint16_t length; /* 0 */
+ uint8_t lo_macs; /* 2 */
+ uint8_t hdr_type; /* 3 */
+} __attribute__((packed));
+
+
+/**
+ * Generic Host -> SeaStar command structure.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+struct command {
+ uint8_t op; /* 0 */
+ uint8_t pad[63]; /* [1,63] */
+} __attribute__((packed));
+
+
+/**
+ * Initialize firmware command.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+struct command_init {
+ uint8_t op; /* 0 */
+ uint8_t process_index; /* 1 */
+ uint16_t pad; /* 2 */
+ uint16_t pid; /* 4 */
+ uint16_t jid; /* 6 */
+ uint16_t num_pendings; /* 8 */
+ uint16_t num_memds; /* 10 */
+ uint16_t num_eqcbs; /* 12 */
+ uint16_t pending_tx_limit; /* 14 */
+ uint32_t pending_table_addr; /* 16 */
+ uint32_t up_pending_table_addr; /* 20 */
+ uint32_t up_pending_table_ht_addr; /* 24 */
+ uint32_t memd_table_addr; /* 28 */
+ uint32_t eqcb_table_addr; /* 32 */
+ uint32_t shdr_table_ht_addr; /* 36 */
+ uint32_t result_block_addr; /* 40 */
+ uint32_t eqheap_addr; /* 44 */
+ uint32_t eqheap_length; /* 48 */
+ uint32_t smb_table_addr; /* 52 */
+ uint32_t uid; /* 56 */
+} __attribute__((packed));
+
+
+/**
+ * Start firmware running command.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+struct command_mark_alive {
+ uint8_t op; /* 0 */
+ uint8_t index; /* 1 */
+} __attribute__((packed));
+
+
+/**
+ * Initialize event queue command.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+struct command_init_eqcb {
+ uint8_t op; /* 0 */
+ uint8_t pad; /* 1 */
+ uint16_t eqcb_index; /* 2 */
+ uint32_t base; /* 4 */
+ uint32_t count; /* 8 */
+} __attribute__((packed));
+
+
+/**
+ * Send datagram command.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+struct command_ip_tx {
+ uint8_t op; /* 0 */
+ uint8_t pad; /* 1 */
+ uint16_t nid; /* 2 */
+ uint16_t length; /* 4 */
+ uint16_t pad2; /* 6 */
+ uint64_t address; /* 8 */
+ uint16_t pending_index; /* 16 */
+} __attribute__((packed));
+
+
+/**
+ * Host <-> SeaStar Mailbox structure.
+ *
+ * WARNING: This must match the definition used by the
+ * closed-source SeaStar firmware.
+ */
+struct mailbox {
+ volatile struct command commandq[COMMAND_Q_LENGTH]; /* 0 */
+ volatile uint32_t resultq[RESULT_Q_LENGTH]; /* 4032 */
+
+ volatile uint32_t resultq_read; /* 4040 */
+ volatile uint32_t resultq_write; /* 4044 */
+ volatile uint32_t commandq_write; /* 4048 */
+ volatile uint32_t commandq_read; /* 4052 */
+} __attribute__((packed, aligned(PAGE_SIZE)));
+
+
+struct ss_priv;
+
+
+extern void
+seastar_ip_tx_cmd(
+ struct ss_priv *ssp,
+ uint16_t nid,
+ uint16_t length,
+ uint64_t address,
+ uint16_t pending_index
+);
+
+
+void
+seastar_setup_htb_bi(
+ uint32_t idr
+);
+
+
+extern int
+seastar_hw_init(
+ struct ss_priv *ssp
+);
+
+
+#endif
diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/seastar/main.c linux-2.6.32.7/drivers/net/seastar/main.c
--- linux-2.6.32.7-vanilla/drivers/net/seastar/main.c 1969-12-31 17:00:00.000000000 -0700
+++ linux-2.6.32.7/drivers/net/seastar/main.c 2010-02-02 12:49:20.000000000 -0700
@@ -0,0 +1,584 @@
+/*******************************************************************************
+ SeaStar NIC Linux Driver
+
+ Copyright 2009-2010 Sandia Corporation. Under the terms of Contract
+ DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government
+ retains certain rights in this software.
+
+ Copyright (c) 2009-2010 Cray Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 2 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program; if not, write to the Free Software
+ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+
+ Contact Information:
+ Kevin Pedretti <ktpedre@sandia.gov>
+ Sandia National Laboratories
+ P.O. Box 5800
+ Albuquerque, NM 87185-1319
+
+*******************************************************************************/
+
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/if_arp.h>
+#include <linux/ip.h>
+#include <linux/htirq.h>
+#include <linux/io.h>
+#include <linux/uaccess.h>
+#include <net/arp.h>
+#include "firmware.h"
+#include "seastar.h"
+
+
+#define SEASTAR_VERSION_STR "1.0"
+
+
+MODULE_DESCRIPTION("Cray SeaStar Native IP driver");
+MODULE_AUTHOR("Maintainer: Kevin Pedretti <ktpedre@sandia.gov>");
+MODULE_VERSION(SEASTAR_VERSION_STR);
+MODULE_LICENSE("GPL");
+
+
+static struct pending *alloc_tx_pending(struct ss_priv *ssp)
+{
+ struct pending *pending = ssp->tx_pending_free_list;
+ if (!pending)
+ return NULL;
+
+ ssp->tx_pending_free_list = pending->next;
+ pending->next = 0;
+
+ return pending;
+}
+
+
+static void free_tx_pending(struct ss_priv *ssp, struct pending *pending)
+{
+ pending->next = ssp->tx_pending_free_list;
+ ssp->tx_pending_free_list = pending;
+}
+
+
+static uint16_t pending_to_index(struct ss_priv *ssp, struct pending *pending)
+{
+ return pending - ssp->pending_table;
+}
+
+
+static struct pending *index_to_pending(struct ss_priv *ssp, unsigned int index)
+{
+ return &ssp->pending_table[index];
+}
+
+
+static void refill_skb(struct net_device *netdev, int i)
+{
+ struct ss_priv *ssp = netdev_priv(netdev);
+ struct sk_buff *skb;
+
+ skb = dev_alloc_skb(netdev->mtu + SKB_PAD);
+ if (!skb) {
+ dev_err(&ssp->pdev->dev, "dev_alloc_skb() failed.\n");
+ return;
+ }
+
+ skb->dev = netdev;
+ skb_reserve(skb, SKB_PAD);
+
+ /* Push it down to the PPC as a quadbyte address */
+ ssp->skb_table_phys[i] = virt_to_phys(skb->data) >> 2;
+ ssp->skb_table_virt[i] = skb;
+}
+
+
+static int ss_open(struct net_device *netdev)
+{
+ struct ss_priv *ssp = netdev_priv(netdev);
+ int i;
+
+ netif_start_queue(netdev);
+
+ for (i = 0; i < NUM_SKBS; i++) {
+ ssp->skb_table_phys[i] = 0;
+ ssp->skb_table_virt[i] = 0;
+ refill_skb(netdev, i);
+ }
+
+ return 0;
+}
+
+
+static int eth2ss(struct ss_priv *ssp, struct sk_buff *skb)
+{
+ struct ethhdr *ethhdr;
+ struct sshdr *sshdr;
+ uint8_t source_lo_mac, dest_lo_mac;
+ uint32_t qb_len;
+
+ /* Read the "low" bytes of the source and destination MAC addresses */
+ ethhdr = (struct ethhdr *)skb->data;
+ source_lo_mac = ethhdr->h_source[5];
+ dest_lo_mac = ethhdr->h_dest[5];
+
+ /* Drop anything not IPv4 */
+ if (ethhdr->h_proto != ntohs(ETH_P_IP)) {
+ dev_err(&ssp->pdev->dev, "squashing non-IPv4 packet.");
+ return -1;
+ }
+
+ /* Squash broadcast packets, SeaStar doesn't support broadcast */
+ if (dest_lo_mac == 0xFF) {
+ dev_err(&ssp->pdev->dev, "squashing broadcast packet.");
+ return -1;
+ }
+
+ /* We only support 4 bits of virtual hosts per physical node */
+ if ((source_lo_mac & ~0xF) || (dest_lo_mac & ~0xF)) {
+ dev_err(&ssp->pdev->dev, "lo_mac out of range.");
+ return -1;
+ }
+
+ /* Move ahead to allow sshdr to be filled in overtop of the ethhdr */
+ sshdr = (struct sshdr *)
+ skb_pull(skb, (unsigned int)(ETH_HLEN - sizeof(struct sshdr)));
+
+ /* The length in quad bytes, rounded up to the nearest quad byte.
+ * SS header is already counted in skb->len as per skb_pull() above */
+ qb_len = (ROUNDUP4(skb->len) >> 2) - 1;
+
+ /* Build the SeaStar header */
+ sshdr->length = qb_len;
+ sshdr->lo_macs = (source_lo_mac << 4) | dest_lo_mac;
+ sshdr->hdr_type = (2 << 5); /* Datagram 2, type 0 == IP */
+
+ return 0;
+}
+
+
+static int ss2eth(struct sk_buff *skb)
+{
+ struct sshdr *sshdr;
+ struct ethhdr *ethhdr;
+ uint8_t source_lo_mac, dest_lo_mac;
+
+ /* Read the "low" bytes of the source and destination MAC addresses */
+ sshdr = (struct sshdr *)skb->data;
+ source_lo_mac = (sshdr->lo_macs >> 4);
+ dest_lo_mac = sshdr->lo_macs & 0xF;
+
+ /* Make room for the rest of the ethernet header and zero it */
+ ethhdr = (struct ethhdr *)
+ skb_push(skb, (unsigned int)(ETH_HLEN - sizeof(struct sshdr)));
+ memset(ethhdr, 0x00, ETH_HLEN);
+
+ /* h_proto and h_dest[] are available. Just 0xff h_source[2-5] */
+ ethhdr->h_proto = htons(ETH_P_IP);
+
+ /* We're assuming the source MAC is the same as the local
+ * host's MAC in order to support loopback in promiscous mode */
+ memcpy(ðhdr->h_source, &skb->dev->dev_addr, ETH_ALEN);
+ memcpy(ðhdr->h_dest, &skb->dev->dev_addr, ETH_ALEN);
+ ethhdr->h_source[5] = source_lo_mac;
+ ethhdr->h_dest[5] = dest_lo_mac;
+
+ return 0;
+}
+
+
+static int ss_tx(struct sk_buff *skb, struct net_device *netdev)
+{
+ unsigned long flags;
+ struct ss_priv *ssp = netdev_priv(netdev);
+ struct ethhdr *eh = (struct ethhdr *)skb->data;
+ struct sshdr *sshdr;
+ uint32_t dest_nid = ntohl(*(uint32_t *)eh->h_dest);
+ struct pending *pending = NULL;
+ void *msg;
+
+ spin_lock_irqsave(&ssp->lock, flags);
+
+ if (netif_queue_stopped(netdev)) {
+ spin_unlock_irqrestore(&ssp->lock, flags);
+ return NETDEV_TX_BUSY;
+ }
+
+ /* Convert the SKB from an ethernet frame to a seastar frame */
+ if (eth2ss(ssp, skb)) {
+ netdev->stats.tx_errors++;
+ goto drop;
+ }
+
+ sshdr = (struct sshdr *)skb->data;
+
+ /* Get a tx_pending so that we can track the completion of this SKB */
+ pending = alloc_tx_pending(ssp);
+ if (!pending) {
+ netif_stop_queue(netdev);
+ spin_unlock_irqrestore(&ssp->lock, flags);
+ return NETDEV_TX_BUSY;
+ }
+
+ /* Stash skb away in the pending, will be needed in ss_tx_end() */
+ pending->skb = skb;
+
+ /* Make sure buffer we pass to SeaStar is quad-byte aligned */
+ if (((unsigned long)skb->data & 0x3) == 0) {
+ pending->bounce = NULL;
+ msg = skb->data;
+ } else {
+ /* Need to use bounce buffer to get quad-byte alignment */
+ pending->bounce = kmalloc(skb->len, GFP_KERNEL);
+ if (!pending->bounce) {
+ dev_err(&ssp->pdev->dev, "dev_alloc_skb() failed.\n");
+ goto drop;
+ }
+ memcpy(pending->bounce, skb->data, skb->len);
+ msg = pending->bounce;
+ }
+
+ seastar_ip_tx_cmd(
+ ssp,
+ dest_nid,
+ sshdr->length,
+ virt_to_phys(msg) >> 2,
+ pending_to_index(ssp, pending)
+ );
+
+ netdev->stats.tx_packets++;
+ netdev->stats.tx_bytes += skb->len;
+
+ spin_unlock_irqrestore(&ssp->lock, flags);
+ return 0;
+
+drop:
+ dev_kfree_skb_any(skb);
+ if (pending)
+ free_tx_pending(ssp, pending);
+ spin_unlock_irqrestore(&ssp->lock, flags);
+ return 0;
+}
+
+
+static void ss_tx_end(struct net_device *netdev, unsigned int pending_index)
+{
+ unsigned long flags;
+ struct ss_priv *ssp = netdev_priv(netdev);
+ struct pending *pending = index_to_pending(ssp, pending_index);
+
+ spin_lock_irqsave(&ssp->lock, flags);
+
+ if (pending->skb)
+ dev_kfree_skb_any(pending->skb);
+
+ kfree(pending->bounce);
+
+ free_tx_pending(ssp, pending);
+
+ if (netif_queue_stopped(netdev))
+ netif_wake_queue(netdev);
+
+ spin_unlock_irqrestore(&ssp->lock, flags);
+}
+
+
+static void ss_rx_skb(struct net_device *netdev, struct sk_buff *skb)
+{
+ struct sshdr *sshdr = (struct sshdr *)skb_tail_pointer(skb);
+
+ const uint32_t qb_len = sshdr->length;
+ const uint32_t len = (qb_len + 1) << 2;
+
+ skb_put(skb, len);
+ ss2eth(skb);
+
+ skb->protocol = htons(ETH_P_IP);
+ skb->ip_summed = CHECKSUM_UNNECESSARY;
+ skb_set_mac_header(skb, 0);
+
+ /* Skip past the ethernet header we just built */
+ skb_pull(skb, ETH_HLEN);
+
+ netdev->stats.rx_packets++;
+ netdev->stats.rx_bytes += len;
+
+ netif_rx(skb);
+}
+
+
+static void ss_rx(struct net_device *netdev, unsigned int skb_index)
+{
+ struct ss_priv *ssp = netdev_priv(netdev);
+ struct sk_buff *skb = ssp->skb_table_virt[skb_index];
+
+ ssp->skb_table_virt[skb_index] = 0;
+ ss_rx_skb(netdev, skb);
+
+ refill_skb(netdev, skb_index);
+}
+
+
+static int ss_header_create(struct sk_buff *skb, struct net_device *netdev,
+ unsigned short type, const void *daddr,
+ const void *saddr, unsigned int length)
+{
+ struct ethhdr *eh;
+
+ /* Make room for the ethernet header and zero it */
+ eh = (struct ethhdr *)skb_push(skb, ETH_HLEN);
+ memset(eh, 0, ETH_HLEN);
+
+ /* Although we can only do IPv4, build other packets correctly for
+ * now and drop it in the ndo_start_xmit hook. This way the fact that
+ * these packets are being generated is not invisible. */
+ eh->h_proto = htons(type);
+
+ /* Set the source hardware address */
+ if (!saddr)
+ saddr = netdev->dev_addr;
+ memcpy(eh->h_source, saddr, ETH_ALEN);
+
+ /* Set the destination hardware address */
+ if (daddr) {
+ memcpy(eh->h_dest, daddr, ETH_ALEN);
+ return ETH_HLEN;
+ }
+
+ /* No destination address supplied !?! */
+ return -ETH_HLEN;
+}
+
+
+static uint32_t next_event(struct ss_priv *ssp)
+{
+ uint32_t ev = ssp->eq[ssp->eq_read];
+ if (!ev)
+ return 0;
+
+ ssp->eq[ssp->eq_read] = 0;
+ ssp->eq_read = (ssp->eq_read + 1) % NUM_EQ_ENTRIES;
+
+ return ev;
+}
+
+
+static void ss_rx_refill(struct net_device *netdev)
+{
+ struct ss_priv *ssp = netdev_priv(netdev);
+ int i;
+
+ for (i = 0; i < NUM_SKBS; i++) {
+ if (ssp->skb_table_virt[i] == 0)
+ refill_skb(netdev, i);
+ }
+}
+
+
+static irqreturn_t ss_interrupt(int irq, void *dev)
+{
+ struct net_device *netdev = (struct net_device *)dev;
+ struct ss_priv *ssp = netdev_priv(netdev);
+ uint32_t ev;
+ unsigned int type, index;
+
+ while (1) {
+ ev = next_event(ssp);
+ if (!ev)
+ break;
+
+ type = (ev >> 16) & 0xFFFF;
+ index = (ev >> 0) & 0xFFFF;
+
+ switch (type) {
+
+ case EVENT_TX_END:
+ ss_tx_end(netdev, index);
+ break;
+
+ case EVENT_RX:
+ ss_rx(netdev, index);
+ break;
+
+ case EVENT_RX_EMPTY:
+ ss_rx_refill(netdev);
+ break;
+
+ default:
+ dev_err(&ssp->pdev->dev,
+ "unknown event type (type=%u, index=%u).\n",
+ type, index);
+ }
+ }
+
+ return IRQ_HANDLED;
+}
+
+
+static const struct net_device_ops ss_netdev_ops = {
+ .ndo_open = ss_open,
+ .ndo_start_xmit = ss_tx,
+ .ndo_set_mac_address = eth_mac_addr,
+};
+
+
+static const struct header_ops ss_header_ops = {
+ .create = ss_header_create,
+};
+
+
+static void ss_ht_irq_update(struct pci_dev *dev, int irq,
+ struct ht_irq_msg *msg)
+{
+ seastar_setup_htb_bi(msg->address_lo);
+}
+
+
+static int __devinit ss_probe(struct pci_dev *pdev,
+ const struct pci_device_id *id)
+{
+ struct net_device *netdev;
+ struct ss_priv *ssp;
+ int i, irq, err = 0;
+
+ err = pci_enable_device(pdev);
+ if (err != 0) {
+ dev_err(&pdev->dev, "Could not enable PCI device.\n");
+ return -ENODEV;
+ }
+
+ netdev = alloc_etherdev(sizeof(*ssp));
+ if (netdev == NULL) {
+ dev_err(&pdev->dev, "Could not allocate ethernet device.\n");
+ return -ENOMEM;
+ }
+
+ SET_NETDEV_DEV(netdev, &pdev->dev);
+
+ strcpy(netdev->name, "ss");
+ netdev->netdev_ops = &ss_netdev_ops;
+ netdev->header_ops = &ss_header_ops;
+ netdev->mtu = 16000;
+ netdev->flags = IFF_NOARP;
+
+ /* Setup private state */
+ ssp = netdev_priv(netdev);
+ memset(ssp, 0, sizeof(*ssp));
+
+ spin_lock_init(&ssp->lock);
+ ssp->skb_table_phys = seastar_skb;
+ ssp->eq_read = 0;
+ ssp->pdev = pdev;
+
+ /* Build the TX pending free list */
+ ssp->tx_pending_free_list = 0;
+ for (i = 0; i < NUM_TX_PENDINGS; i++)
+ free_tx_pending(ssp, index_to_pending(ssp, i));
+
+ irq = __ht_create_irq(pdev, 0, ss_ht_irq_update);
+ if (irq < 0) {
+ dev_err(&pdev->dev, "__ht_create_irq() failed, err=%d.\n", err);
+ goto err_out;
+ }
+
+ err = request_irq(irq, ss_interrupt, IRQF_NOBALANCING,
+ "seastar", netdev);
+ if (err != 0) {
+ dev_err(&pdev->dev, "request_irq() failed, err=%d.\n", err);
+ goto err_out;
+ }
+
+ err = seastar_hw_init(netdev_priv(netdev));
+ if (err != 0) {
+ dev_err(&pdev->dev, "seastar_hw_init() failed, err=%d.\n", err);
+ goto err_out;
+ }
+
+ err = register_netdev(netdev);
+ if (err != 0) {
+ dev_err(&pdev->dev, "register_netdev() failed, err=%d.\n", err);
+ goto err_out;
+ }
+
+ return 0;
+
+err_out:
+ free_netdev(netdev);
+ return err;
+}
+
+
+static void __devexit ss_remove(struct pci_dev *pdev)
+{
+ struct net_device *netdev = pci_get_drvdata(pdev);
+
+ unregister_netdev(netdev);
+ free_netdev(netdev);
+ pci_disable_device(pdev);
+}
+
+
+#define PCI_VENDOR_ID_CRAY 0x17DB
+#define PCI_DEVICE_ID_SEASTAR 0x0101
+
+
+static struct pci_device_id ss_pci_tbl[] = {
+ {PCI_DEVICE(PCI_VENDOR_ID_CRAY, PCI_DEVICE_ID_SEASTAR)},
+ {0},
+};
+
+
+#ifdef CONFIG_PM
+static int ss_suspend(struct pci_dev *pdev, pm_message_t state)
+{
+ return -ENOSYS;
+}
+
+
+static int ss_resume(struct pci_dev *pdev)
+{
+ return -ENOSYS;
+}
+#endif
+
+
+static struct pci_driver ss_driver = {
+ .name = "seastar",
+ .probe = ss_probe,
+ .remove = __devexit_p(ss_remove),
+ .id_table = ss_pci_tbl,
+#ifdef CONFIG_PM
+ .suspend = ss_suspend,
+ .resume = ss_resume,
+#endif
+};
+
+
+static __init int ss_init_module(void)
+{
+ printk(KERN_INFO "%s: module loaded (version %s)\n",
+ ss_driver.name, SEASTAR_VERSION_STR);
+
+ return pci_register_driver(&ss_driver);
+}
+
+
+static __exit void ss_cleanup_module(void)
+{
+ pci_unregister_driver(&ss_driver);
+}
+
+
+module_init(ss_init_module);
+module_exit(ss_cleanup_module);
diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/seastar/Makefile linux-2.6.32.7/drivers/net/seastar/Makefile
--- linux-2.6.32.7-vanilla/drivers/net/seastar/Makefile 1969-12-31 17:00:00.000000000 -0700
+++ linux-2.6.32.7/drivers/net/seastar/Makefile 2010-02-02 09:12:04.000000000 -0700
@@ -0,0 +1,3 @@
+obj-$(CONFIG_SEASTAR) += seastar.o
+
+seastar-y := main.o firmware.o
diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/seastar/seastar.h linux-2.6.32.7/drivers/net/seastar/seastar.h
--- linux-2.6.32.7-vanilla/drivers/net/seastar/seastar.h 1969-12-31 17:00:00.000000000 -0700
+++ linux-2.6.32.7/drivers/net/seastar/seastar.h 2010-02-02 09:14:33.000000000 -0700
@@ -0,0 +1,104 @@
+/*******************************************************************************
+ SeaStar NIC Linux Driver
+
+ Copyright 2009-2010 Sandia Corporation. Under the terms of Contract
+ DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government
+ retains certain rights in this software.
+
+ Copyright (c) 2009-2010 Cray Inc.
+
+ This program is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 2 of the License, or
+ (at your option) any later version.
+
+ This program is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ You should have received a copy of the GNU General Public License
+ along with this program; if not, write to the Free Software
+ Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
+
+ Contact Information:
+ Kevin Pedretti <ktpedre@sandia.gov>
+ Sandia National Laboratories
+ P.O. Box 5800
+ Albuquerque, NM 87185-1319
+
+*******************************************************************************/
+
+#ifndef _SEASTAR_H
+#define _SEASTAR_H
+
+
+/**
+ * Rounds up to the nearest quadbyte.
+ */
+#define ROUNDUP4(val) ((val + (4-1)) & ~(4-1))
+
+
+/**
+ * SeaStar datagram packet maximum transfer unit size in bytes.
+ */
+#define SEASTAR_MTU 8192
+
+
+/**
+ * Number of transmit and receive pending structures.
+ */
+#define NUM_TX_PENDINGS 64
+#define NUM_RX_PENDINGS 64
+#define NUM_PENDINGS (NUM_TX_PENDINGS + NUM_RX_PENDINGS)
+
+
+/**
+ * Number of entries in the SeaStar -> Host event queue.
+ */
+#define NUM_EQ_ENTRIES 1024
+
+
+/**
+ * When allocating an SKB, allocate this many bytes extra.
+ */
+#define SKB_PAD (16 - sizeof(struct sshdr))
+
+
+/**
+ * Pending structure.
+ * One of these is used to track each in progress transmit.
+ */
+struct pending {
+ struct sk_buff *skb;
+ struct pending *next;
+ void *bounce;
+};
+
+
+/**
+ * SeaStar driver private data.
+ */
+struct ss_priv {
+ spinlock_t lock;
+
+ unsigned long host_region_phys;
+
+ volatile uint64_t *skb_table_phys;
+ struct sk_buff *skb_table_virt[NUM_SKBS];
+
+ struct pending pending_table[NUM_PENDINGS];
+ struct pending *tx_pending_free_list;
+
+ uint32_t eq[NUM_EQ_ENTRIES];
+ unsigned int eq_read;
+
+ struct mailbox *mailbox;
+ unsigned int mailbox_cached_read;
+ unsigned int mailbox_cached_write;
+
+ struct pci_dev *pdev;
+};
+
+
+#endif
^ permalink raw reply
* Re: [PATCH net-next-2.6] bonding: Remove net_device_stats from bonding struct
From: Jay Vosburgh @ 2010-02-02 21:40 UTC (permalink / raw)
To: Ajit Khaparde; +Cc: David Miller, netdev
In-Reply-To: <20100202000648.GA30286@serverengines.com>
Ajit Khaparde <ajitkhaparde@gmail.com> wrote:
>There is no need to maintain stats in the bonding structure.
>Use the instance of net_device_stats in netdevice.
>
>Signed-off-by: Ajit Khaparde <ajitk@serverengines.com>
Note that the accumulation of stats across the slaves is done
into a separate local instance of net_device_stats, so partial results
shouldn't be visible.
Signed-off-by: Jay Vosburgh <fubar@us.ibm.com>
-J
>---
> drivers/net/bonding/bond_main.c | 2 +-
> drivers/net/bonding/bonding.h | 1 -
> 2 files changed, 1 insertions(+), 2 deletions(-)
>
>diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
>index 6221936..1787e3c 100644
>--- a/drivers/net/bonding/bond_main.c
>+++ b/drivers/net/bonding/bond_main.c
>@@ -3742,7 +3742,7 @@ static int bond_close(struct net_device *bond_dev)
> static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
> {
> struct bonding *bond = netdev_priv(bond_dev);
>- struct net_device_stats *stats = &bond->stats;
>+ struct net_device_stats *stats = &bond_dev->stats;
> struct net_device_stats local_stats;
> struct slave *slave;
> int i;
>diff --git a/drivers/net/bonding/bonding.h b/drivers/net/bonding/bonding.h
>index 558ec13..257a7a4 100644
>--- a/drivers/net/bonding/bonding.h
>+++ b/drivers/net/bonding/bonding.h
>@@ -197,7 +197,6 @@ struct bonding {
> s8 send_grat_arp;
> s8 send_unsol_na;
> s8 setup_by_slave;
>- struct net_device_stats stats;
> #ifdef CONFIG_PROC_FS
> struct proc_dir_entry *proc_entry;
> char proc_file_name[IFNAMSIZ];
>--
>1.6.3.3
>
>--
>To unsubscribe from this list: send the line "unsubscribe netdev" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at http://vger.kernel.org/majordomo-info.html
^ permalink raw reply
* Re: [PATCH] seastar - SeaStar Ethernet driver
From: David Miller @ 2010-02-02 21:42 UTC (permalink / raw)
To: ktpedre; +Cc: netdev, linux-kernel
In-Reply-To: <20100202205845.GE5246@hawkeye.sandia.gov>
From: "Kevin Pedretti" <ktpedre@sandia.gov>
Date: Tue, 2 Feb 2010 13:58:45 -0700
> +void seastar_setup_htb_bi(uint32_t idr)
Please use the in-kernel sized types "u32", "u16", etc.
instead of "uint32_t" et al.
> +extern void
> +seastar_ip_tx_cmd(
> + struct ss_priv *ssp,
> + uint16_t nid,
> + uint16_t length,
> + uint64_t address,
> + uint16_t pending_index
> +);
> +
> +
> +void
> +seastar_setup_htb_bi(
> + uint32_t idr
> +);
> +
> +
> +extern int
> +seastar_hw_init(
> + struct ss_priv *ssp
> +);
Please fix the formatting of these function declarations,
something like:
extern void seastar_ip_tx_cmd(struct ss_priv *ssp,
uint16_t nid,
uint16_t length,
uint64_t address,
uint16_t pending_index);
extern void seastar_setup_htb_bi(uint32_t idr);
extern int seastar_hw_init(struct ss_priv *ssp);
And again use "u16" instead of "uint16_t" etc.
There are many bad code formatting cases like this in your
driver, lease fix them all up.
> +static int ss_open(struct net_device *netdev)
> +{
> + struct ss_priv *ssp = netdev_priv(netdev);
> + int i;
> +
> + netif_start_queue(netdev);
> +
> + for (i = 0; i < NUM_SKBS; i++) {
> + ssp->skb_table_phys[i] = 0;
> + ssp->skb_table_virt[i] = 0;
> + refill_skb(netdev, i);
> + }
> +
> + return 0;
> +}
You shouldn't call netif_start_queue() until you are completely
done initializing the chip. Packets can start being transmitted
to the driver the exact moment that function returns.
> +static int eth2ss(struct ss_priv *ssp, struct sk_buff *skb)
...
> +static int ss2eth(struct sk_buff *skb)
This device can only transmit IPv4 packets and can only receive IPv4
packets?
> +#ifdef CONFIG_PM
> +static int ss_suspend(struct pci_dev *pdev, pm_message_t state)
> +{
> + return -ENOSYS;
> +}
> +
> +
> +static int ss_resume(struct pci_dev *pdev)
> +{
> + return -ENOSYS;
> +}
> +#endif
If you don't support suspend and resume, simply leave the
method pointers unassigned, there is no need to provide
NOP routines like this.
^ permalink raw reply
* Re: [PATCH] seastar - SeaStar Ethernet driver
From: Randy Dunlap @ 2010-02-02 21:59 UTC (permalink / raw)
To: Kevin Pedretti; +Cc: netdev, linux-kernel
In-Reply-To: <20100202205845.GE5246@hawkeye.sandia.gov>
On 02/02/10 12:58, Kevin Pedretti wrote:
> [PATCH] seastar - SeaStar Ethernet driver
>
> The following patch introduces the seastar driver for the
> SeaStar network interface in Cray XT3/XT4/XT5 systems. The
> driver is called 'seastar'. This patch is against 2.6.32.7.
>
> The driver uses a simple datagram interface exported by the
> SeaStar network interface to encapsulate Ethernet frames
> on the Cray XT high speed network. The driver has been tested
> to function correctly and is in use on Cray XT4 development
> systems at Sandia.
>
> Signed-off-by: Kevin Pedretti<ktpedre@sandia.gov>
> obj-$(CONFIG_PPP_ASYNC) += ppp_async.o
> diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/seastar/firmware.c linux-2.6.32.7/drivers/net/seastar/firmware.c
> --- linux-2.6.32.7-vanilla/drivers/net/seastar/firmware.c 1969-12-31 17:00:00.000000000 -0700
> +++ linux-2.6.32.7/drivers/net/seastar/firmware.c 2010-02-02 09:13:44.000000000 -0700
> @@ -0,0 +1,236 @@
> +
> +/**
> + * Maps a region of host memory into the SeaStar.
> + */
Please note that "/**" in Linux kernel sources means "beginning of a kernel-doc comment",
and none of these is in kernel-doc notation (format), so please change all of them
to a simple "/*".
(throughout all .c/.h files)
> +static void seastar_map_host_region(struct ss_priv *ssp, const void *addr)
> +{
> + /* Round addr to the nearest 128 MB */
> + unsigned long raw_paddr = __pa(addr);
> + unsigned long paddr = raw_paddr& ~((1<< 28) - 1);
> +
> + htb_map[8] = 0x8000 | ((paddr>> 28) + 0);
> + htb_map[9] = 0x8000 | ((paddr>> 28) + 1);
space before <<, &, and >>
> +
> + ssp->host_region_phys = paddr;
> +}
> +
> +
> +/**
> + * Converts a kernel virtual address to a SeaStar address.
> + */
> +static uint32_t virt_to_fw(struct ss_priv *ssp, void *addr)
> +{
> + unsigned long saddr;
> +
> + saddr = __pa(addr) - ssp->host_region_phys;
> + saddr&= (2<< 28) - 1;
> + saddr += (8<< 28);
space before << (2x)
> +
> + return saddr;
> +}
> +
> +
> +/**
> + * Send a command to the Seastar.
> + */
> +static uint32_t seastar_cmd(struct ss_priv *ssp, const struct command *cmd,
> + int wait_for_result)
> +{
> + struct mailbox *mbox = ssp->mailbox;
> + unsigned int next_write;
> + uint32_t tail, result;
> +
> + /* Copy the command into the mailbox */
> + mbox->commandq[ssp->mailbox_cached_write] = *cmd;
> + next_write = ssp->mailbox_cached_write + 1;
> + if (next_write == COMMAND_Q_LENGTH)
> + next_write = 0;
> +
> + /* Wait until it is safe to advance the write pointer */
> + while (next_write == ssp->mailbox_cached_read)
> + ssp->mailbox_cached_read = mbox->commandq_read;
> +
> + /* Advance the write pointer */
> + mbox->commandq_write = next_write;
> + ssp->mailbox_cached_write = next_write;
> +
> + if (!wait_for_result)
> + return 0;
> +
> + /* Wait for the result to arrive */
> + tail = mbox->resultq_read;
> + while (tail == mbox->resultq_write)
> + ;
I would limit that while loop somehow (not allowing it to continue forever).
> +
> + /* Read the result */
> + result = mbox->resultq[tail];
> + mbox->resultq_read = (tail>= RESULT_Q_LENGTH - 1) ? 0 : tail + 1;
> +
> + return result;
> +}
> +
> +
> +/**
> + * Sends a datagram transmit command to the SeaStar.
> + */
> +void seastar_ip_tx_cmd(struct ss_priv *ssp, uint16_t nid, uint16_t length,
> + uint64_t address, uint16_t pending_index)
> +{
> + struct command_ip_tx tx_cmd = {
> + .op = COMMAND_IP_TX,
> + .nid = nid,
> + .length = length,
> + .address = address,
> + .pending_index = pending_index,
> + };
> +
> + seastar_cmd(ssp, (struct command *)&tx_cmd, 0);
> +}
> +
> +
> +/**
> + * Programs the SeaStar's HTB_BI register.
> + */
> +void seastar_setup_htb_bi(uint32_t idr)
> +{
> + /* Mask the APIC dest setup by Linux, causes problems with SeaStar */
> + idr&= 0xFFFF0000;
space after '&'
> +
> + *htb_bi = 0xFD000000 | (idr>> 8);
space before >>
> +}
> +
> +
> +/**
> + * Brings up the low-level Seastar hardware.
> + */
> +int seastar_hw_init(struct ss_priv *ssp)
> +{
> + uint32_t lower_memory = SEASTAR_HOST_BASE;
> + const int num_eq = 1;
> + uint32_t lower_pending;
> + uint32_t lower_eqcb;
> + uint32_t result;
> + struct command_init init_cmd;
> + struct command_init_eqcb eqcb_cmd;
> + struct command_mark_alive alive_cmd;
> +
> + /* Read our NID from SeaStar and write it to the NIC control block */
> + niccb->local_nid = *tx_source;
> +
> + printk(KERN_INFO "%s: nid %d (0x%x) version %x built %x\n",
> + __func__,
> + niccb->local_nid,
> + niccb->local_nid,
> + niccb->version,
> + niccb->build_time
> + );
> +
> + /* Allocate the PPC memory */
> + lower_pending = lower_memory;
> + lower_memory += NUM_PENDINGS * FW_PENDING_SIZE;
> +
> + lower_eqcb = lower_memory;
> + lower_memory = num_eq * FW_EQCB_SIZE;
> +
> + /* Initialize the HTB map so that the Seastar can see our memory.
> + * Since we are only doing upper pendings, we just use the
> + * upper_pending_phys instead of the host_phys area. */
> + seastar_map_host_region(ssp, ssp);
> +
> + ssp->mailbox =&seastar_mailbox[0];
space before '&'
> + ssp->mailbox_cached_read = ssp->mailbox->commandq_read;
> + ssp->mailbox_cached_write = ssp->mailbox->commandq_write;
> +
> + /* Attempt to send a setup command to the NIC */
> + init_cmd.op = COMMAND_INIT;
> + init_cmd.process_index = 1;
> + init_cmd.uid = 0;
> + init_cmd.jid = 0;
> +
> + init_cmd.num_pendings = NUM_PENDINGS;
> + init_cmd.pending_tx_limit = NUM_TX_PENDINGS;
> + init_cmd.pending_table_addr = lower_pending;
> + init_cmd.up_pending_table_addr = virt_to_fw(ssp, ssp->pending_table);
> + init_cmd.up_pending_table_ht_addr = 0;
> +
> + init_cmd.num_memds = 0;
> + init_cmd.memd_table_addr = 0;
> +
> + init_cmd.num_eqcbs = num_eq;
> + init_cmd.eqcb_table_addr = lower_eqcb;
> + init_cmd.eqheap_addr = virt_to_fw(ssp, ssp->eq);
> + init_cmd.eqheap_length = NUM_EQ_ENTRIES * sizeof(ssp->eq[0]);
> +
> + init_cmd.shdr_table_ht_addr = 0;
> + init_cmd.result_block_addr = 0;
> + init_cmd.smb_table_addr = 0;
> +
> + result = seastar_cmd(ssp, (struct command *)&init_cmd, 1);
> + if (result != 0) {
> + dev_err(&ssp->pdev->dev,
> + "init command failed, result=%d.\n", result);
> + return -1;
> + }
> +
> + eqcb_cmd.op = COMMAND_INIT_EQCB;
> + eqcb_cmd.eqcb_index = 0;
> + eqcb_cmd.base = virt_to_fw(ssp, ssp->eq);
> + eqcb_cmd.count = NUM_EQ_ENTRIES;
> +
> + result = seastar_cmd(ssp, (struct command *)&eqcb_cmd, 1);
> + if (result != 1) {
> + dev_err(&ssp->pdev->dev,
> + "init_eqcb command failed, result=%d.\n", result);
> + return -1;
> + }
> +
> + alive_cmd.op = COMMAND_MARK_ALIVE;
> + alive_cmd.index = 1;
> +
> + result = seastar_cmd(ssp, (struct command *)&alive_cmd, 1);
> + if (result != 0) {
> + dev_err(&ssp->pdev->dev,
> + "mark_alive command failed, result=%d\n", result);
> + return -1;
> + }
> +
> + return 0;
> +}
> diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/seastar/firmware.h linux-2.6.32.7/drivers/net/seastar/firmware.h
> --- linux-2.6.32.7-vanilla/drivers/net/seastar/firmware.h 1969-12-31 17:00:00.000000000 -0700
> +++ linux-2.6.32.7/drivers/net/seastar/firmware.h 2010-02-02 09:15:02.000000000 -0700
> @@ -0,0 +1,329 @@
> +
> +#ifndef _SEASTAR_FIRMWARE_H
> +#define _SEASTAR_FIRMWARE_H
> +
> +
> +/**
> + * Number of entries in Host -> SeaStar command queue.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +#define COMMAND_Q_LENGTH 63
> +
> +
> +/**
> + * Number of entries in SeaStar -> Host result queue.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +#define RESULT_Q_LENGTH 2
> +
> +
> +/**
> + * SeaStar -> Host event types.
> + *
> + * WARNING: These must match the definitions used by the
> + * closed-source SeaStar firmware.
> + */
> +#define EVENT_TX_END 125
> +#define EVENT_RX 126
> +#define EVENT_RX_EMPTY 127
> +
> +
> +/**
> + * Host -> SeaStar command types.
> + *
> + * WARNING: These must match the definitions used by the
> + * closed-source SeaStar firmware.
> + */
> +#define COMMAND_INIT 0
> +#define COMMAND_MARK_ALIVE 1
> +#define COMMAND_INIT_EQCB 2
> +#define COMMAND_IP_TX 13
> +
> +
> +/**
> + * Number of entries in the incoming datagram buffer table.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +#define NUM_SKBS 64
> +
> +
> +/**
> + * Size of the pending structure used by the SeaStar firmware.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +#define FW_PENDING_SIZE 32
> +
> +
> +/**
> + * Size of the event queue control block structure used by the SeaStar firmware.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +#define FW_EQCB_SIZE 32
> +
> +
> +/**
> + * SeaStar addresses of important structures in SeaStar memory.
> + *
> + * WARNING: These must match the definitions used by the
> + * closed-source SeaStar firmware.
> + */
> +#define SEASTAR_SCRATCH_BASE 0xFFFA0000
> +#define SEASTAR_TX_SOURCE 0xFFE00108
> +#define SEASTAR_MAILBOX_BASE 0xFFFA0000
> +#define SEASTAR_SKB_BASE 0xFFFA4000
> +#define SEASTAR_HOST_BASE 0xFFFA5000
> +#define SEASTAR_HTB_BASE 0xFFE20000
> +#define SEASTAR_HTB_BI 0xFFE20048
> +#define SEASTAR_NICCB_BASE 0xFFFFE000
> +
> +
> +/**
> + * Kernel virtual address where the SeaStar memory is mapped.
> + */
> +#define SEASTAR_VIRT_BASE (0xFFFFFFFFull<< 32)
> +
> +
> +/**
> + * Kernel virtual address of the SeaStar's NIC control block.
> + */
> +static volatile struct niccb * const niccb
> + = (void *)(SEASTAR_VIRT_BASE + SEASTAR_NICCB_BASE);
> +
> +
> +/**
> + * Kernel virtual address of the SeaStar's HTB_BI register.
> + */
> +static volatile uint32_t * const htb_bi
> + = (void *)(SEASTAR_VIRT_BASE + SEASTAR_HTB_BI);
> +
> +
> +/**
> + * Kernel virtual address of the SeaStar's HyperTransport map.
> + */
> +static volatile uint32_t * const htb_map
> + = (void *)(SEASTAR_VIRT_BASE + SEASTAR_HTB_BASE);
> +
> +
> +/**
> + * Kernel virtual address of the Host<-> SeaStar mailbox.
> + */
> +static struct mailbox * const seastar_mailbox
> + = (void *)(SEASTAR_VIRT_BASE + SEASTAR_MAILBOX_BASE);
> +
> +
> +/**
> + * Kernel virtual address of the incoming datagram buffer table.
> + */
> +static volatile uint64_t * const seastar_skb
> + = (void *)(SEASTAR_VIRT_BASE + SEASTAR_SKB_BASE);
> +
> +
> +/**
> + * Kernel virtual address of the SeaStar TX Source register.
> + */
> +static volatile uint16_t * const tx_source
> + = (void *)(SEASTAR_VIRT_BASE + SEASTAR_TX_SOURCE);
> +
> +
> +/**
> + * The SeaStar NIC Control Block.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +struct niccb {
> + uint32_t version; /* 0 */
> + uint8_t pad[24];
> + uint32_t build_time; /* 28 */
> + uint8_t pad2[68];
> + uint32_t ip_tx; /* 100 */
> + uint32_t ip_tx_drop; /* 104 */
> + uint32_t ip_rx; /* 108 */
> + uint32_t ip_rx_drop; /* 112 */
> + uint8_t pad3[52];
> + uint16_t local_nid; /* 168 */
> +} __attribute__((packed, aligned));
> +
> +
> +/**
> + * SeaStar datagram packet wire header.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +struct sshdr {
> + uint16_t length; /* 0 */
> + uint8_t lo_macs; /* 2 */
> + uint8_t hdr_type; /* 3 */
> +} __attribute__((packed));
> +
> +
> +/**
> + * Generic Host -> SeaStar command structure.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +struct command {
> + uint8_t op; /* 0 */
> + uint8_t pad[63]; /* [1,63] */
> +} __attribute__((packed));
> +
> +
> +/**
> + * Initialize firmware command.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +struct command_init {
> +
> +/**
> + * Start firmware running command.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +struct command_mark_alive {
> + uint8_t op; /* 0 */
> + uint8_t index; /* 1 */
> +} __attribute__((packed));
> +
> +
> +/**
> + * Initialize event queue command.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +struct command_init_eqcb {
> + uint8_t op; /* 0 */
> + uint8_t pad; /* 1 */
> + uint16_t eqcb_index; /* 2 */
> + uint32_t base; /* 4 */
> + uint32_t count; /* 8 */
> +} __attribute__((packed));
> +
> +
> +/**
> + * Send datagram command.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +struct command_ip_tx {
> + uint8_t op; /* 0 */
> + uint8_t pad; /* 1 */
> + uint16_t nid; /* 2 */
> + uint16_t length; /* 4 */
> + uint16_t pad2; /* 6 */
> + uint64_t address; /* 8 */
> + uint16_t pending_index; /* 16 */
> +} __attribute__((packed));
> +
> +
> +/**
> + * Host<-> SeaStar Mailbox structure.
> + *
> + * WARNING: This must match the definition used by the
> + * closed-source SeaStar firmware.
> + */
> +struct mailbox {
> diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/seastar/main.c linux-2.6.32.7/drivers/net/seastar/main.c
> --- linux-2.6.32.7-vanilla/drivers/net/seastar/main.c 1969-12-31 17:00:00.000000000 -0700
> +++ linux-2.6.32.7/drivers/net/seastar/main.c 2010-02-02 12:49:20.000000000 -0700
> @@ -0,0 +1,584 @@
> +
> +static void refill_skb(struct net_device *netdev, int i)
> +{
> + struct ss_priv *ssp = netdev_priv(netdev);
> + struct sk_buff *skb;
> +
> + skb = dev_alloc_skb(netdev->mtu + SKB_PAD);
> + if (!skb) {
> + dev_err(&ssp->pdev->dev, "dev_alloc_skb() failed.\n");
> + return;
> + }
> +
> + skb->dev = netdev;
> + skb_reserve(skb, SKB_PAD);
> +
> + /* Push it down to the PPC as a quadbyte address */
> + ssp->skb_table_phys[i] = virt_to_phys(skb->data)>> 2;
space before >>
> + ssp->skb_table_virt[i] = skb;
> +}
> +
> +
> +static int ss_open(struct net_device *netdev)
> +{
> + struct ss_priv *ssp = netdev_priv(netdev);
> + int i;
> +
> + netif_start_queue(netdev);
> +
> + for (i = 0; i< NUM_SKBS; i++) {
space before '<'
> + ssp->skb_table_phys[i] = 0;
> + ssp->skb_table_virt[i] = 0;
> + refill_skb(netdev, i);
> + }
> +
> + return 0;
> +}
> +
> +
> +static int eth2ss(struct ss_priv *ssp, struct sk_buff *skb)
> +{
> + struct ethhdr *ethhdr;
> + struct sshdr *sshdr;
> + uint8_t source_lo_mac, dest_lo_mac;
> + uint32_t qb_len;
> +
> + /* Read the "low" bytes of the source and destination MAC addresses */
> + ethhdr = (struct ethhdr *)skb->data;
> + source_lo_mac = ethhdr->h_source[5];
> + dest_lo_mac = ethhdr->h_dest[5];
> +
> + /* Drop anything not IPv4 */
> + if (ethhdr->h_proto != ntohs(ETH_P_IP)) {
> + dev_err(&ssp->pdev->dev, "squashing non-IPv4 packet.");
> + return -1;
> + }
> +
> + /* Squash broadcast packets, SeaStar doesn't support broadcast */
> + if (dest_lo_mac == 0xFF) {
> + dev_err(&ssp->pdev->dev, "squashing broadcast packet.");
> + return -1;
> + }
> +
> + /* We only support 4 bits of virtual hosts per physical node */
> + if ((source_lo_mac& ~0xF) || (dest_lo_mac& ~0xF)) {
space before '&' (2x)
> + dev_err(&ssp->pdev->dev, "lo_mac out of range.");
> + return -1;
> + }
> +
> + /* Move ahead to allow sshdr to be filled in overtop of the ethhdr */
> + sshdr = (struct sshdr *)
> + skb_pull(skb, (unsigned int)(ETH_HLEN - sizeof(struct sshdr)));
> +
> + /* The length in quad bytes, rounded up to the nearest quad byte.
> + * SS header is already counted in skb->len as per skb_pull() above */
Is that second line supposed to explain the "- 1" below?
> + qb_len = (ROUNDUP4(skb->len)>> 2) - 1;
space before >>
> +
> + /* Build the SeaStar header */
> + sshdr->length = qb_len;
> + sshdr->lo_macs = (source_lo_mac<< 4) | dest_lo_mac;
> + sshdr->hdr_type = (2<< 5); /* Datagram 2, type 0 == IP */
(2 << 5); ...
> +
> + return 0;
> +}
> +
> +
> +
> +static int ss_tx(struct sk_buff *skb, struct net_device *netdev)
> +{
> + unsigned long flags;
> + struct ss_priv *ssp = netdev_priv(netdev);
> + struct ethhdr *eh = (struct ethhdr *)skb->data;
> + struct sshdr *sshdr;
> + uint32_t dest_nid = ntohl(*(uint32_t *)eh->h_dest);
> + struct pending *pending = NULL;
> + void *msg;
> +
> + spin_lock_irqsave(&ssp->lock, flags);
> +
> + if (netif_queue_stopped(netdev)) {
> + spin_unlock_irqrestore(&ssp->lock, flags);
> + return NETDEV_TX_BUSY;
> + }
> +
> + /* Convert the SKB from an ethernet frame to a seastar frame */
> + if (eth2ss(ssp, skb)) {
> + netdev->stats.tx_errors++;
> + goto drop;
> + }
> +
> + sshdr = (struct sshdr *)skb->data;
> +
> + /* Get a tx_pending so that we can track the completion of this SKB */
> + pending = alloc_tx_pending(ssp);
> + if (!pending) {
> + netif_stop_queue(netdev);
> + spin_unlock_irqrestore(&ssp->lock, flags);
> + return NETDEV_TX_BUSY;
> + }
> +
> + /* Stash skb away in the pending, will be needed in ss_tx_end() */
> + pending->skb = skb;
> +
> + /* Make sure buffer we pass to SeaStar is quad-byte aligned */
> + if (((unsigned long)skb->data& 0x3) == 0) {
> + pending->bounce = NULL;
> + msg = skb->data;
> + } else {
> + /* Need to use bounce buffer to get quad-byte alignment */
> + pending->bounce = kmalloc(skb->len, GFP_KERNEL);
> + if (!pending->bounce) {
> + dev_err(&ssp->pdev->dev, "dev_alloc_skb() failed.\n");
> + goto drop;
> + }
> + memcpy(pending->bounce, skb->data, skb->len);
> + msg = pending->bounce;
> + }
> +
> + seastar_ip_tx_cmd(
> + ssp,
> + dest_nid,
> + sshdr->length,
> + virt_to_phys(msg)>> 2,
(msg) >> 2,
> + pending_to_index(ssp, pending)
> + );
> +
> + netdev->stats.tx_packets++;
> + netdev->stats.tx_bytes += skb->len;
> +
> + spin_unlock_irqrestore(&ssp->lock, flags);
> + return 0;
> +
> +drop:
> + dev_kfree_skb_any(skb);
> + if (pending)
> + free_tx_pending(ssp, pending);
> + spin_unlock_irqrestore(&ssp->lock, flags);
> + return 0;
> +}
> +
> +
> +
> +static irqreturn_t ss_interrupt(int irq, void *dev)
> +{
> + struct net_device *netdev = (struct net_device *)dev;
> + struct ss_priv *ssp = netdev_priv(netdev);
> + uint32_t ev;
> + unsigned int type, index;
> +
> + while (1) {
> + ev = next_event(ssp);
> + if (!ev)
> + break;
There is usually some condition that limits how long a while (1) can continue.
Does this environment not need that?
> +
> + type = (ev>> 16)& 0xFFFF;
> + index = (ev>> 0)& 0xFFFF;
> +
> + switch (type) {
> +
> + case EVENT_TX_END:
> + ss_tx_end(netdev, index);
> + break;
> +
> + case EVENT_RX:
> + ss_rx(netdev, index);
> + break;
> +
> + case EVENT_RX_EMPTY:
> + ss_rx_refill(netdev);
> + break;
> +
> + default:
> + dev_err(&ssp->pdev->dev,
> + "unknown event type (type=%u, index=%u).\n",
> + type, index);
> + }
> + }
> +
> + return IRQ_HANDLED;
> +}
> +
> +
> +static int __devinit ss_probe(struct pci_dev *pdev,
> + const struct pci_device_id *id)
> +{
> + struct net_device *netdev;
> + struct ss_priv *ssp;
> + int i, irq, err = 0;
> +
> + err = pci_enable_device(pdev);
> + if (err != 0) {
> + dev_err(&pdev->dev, "Could not enable PCI device.\n");
> + return -ENODEV;
> + }
> +
> + netdev = alloc_etherdev(sizeof(*ssp));
> + if (netdev == NULL) {
> + dev_err(&pdev->dev, "Could not allocate ethernet device.\n");
> + return -ENOMEM;
> + }
> +
> + SET_NETDEV_DEV(netdev,&pdev->dev);
> +
> + strcpy(netdev->name, "ss");
> + netdev->netdev_ops =&ss_netdev_ops;
> + netdev->header_ops =&ss_header_ops;
= &ss...
> + netdev->mtu = 16000;
> + netdev->flags = IFF_NOARP;
> +
> + /* Setup private state */
> + ssp = netdev_priv(netdev);
> + memset(ssp, 0, sizeof(*ssp));
> +
> + spin_lock_init(&ssp->lock);
> + ssp->skb_table_phys = seastar_skb;
> + ssp->eq_read = 0;
> + ssp->pdev = pdev;
> +
> + /* Build the TX pending free list */
> + ssp->tx_pending_free_list = 0;
> + for (i = 0; i< NUM_TX_PENDINGS; i++)
; i < NUM_TX...
> + free_tx_pending(ssp, index_to_pending(ssp, i));
> +
> + irq = __ht_create_irq(pdev, 0, ss_ht_irq_update);
> + if (irq< 0) {
if (irq < 0) {
> + dev_err(&pdev->dev, "__ht_create_irq() failed, err=%d.\n", err);
> + goto err_out;
> + }
> +
> + err = request_irq(irq, ss_interrupt, IRQF_NOBALANCING,
> + "seastar", netdev);
> + if (err != 0) {
> + dev_err(&pdev->dev, "request_irq() failed, err=%d.\n", err);
> + goto err_out;
> + }
> +
> + err = seastar_hw_init(netdev_priv(netdev));
> + if (err != 0) {
> + dev_err(&pdev->dev, "seastar_hw_init() failed, err=%d.\n", err);
> + goto err_out;
> + }
> +
> + err = register_netdev(netdev);
> + if (err != 0) {
> + dev_err(&pdev->dev, "register_netdev() failed, err=%d.\n", err);
> + goto err_out;
> + }
> +
> + return 0;
> +
> +err_out:
> + free_netdev(netdev);
> + return err;
> +}
> +
> +
> +static void __devexit ss_remove(struct pci_dev *pdev)
> +{
> + struct net_device *netdev = pci_get_drvdata(pdev);
> +
> + unregister_netdev(netdev);
> + free_netdev(netdev);
> + pci_disable_device(pdev);
> +}
> +
> +
> +#define PCI_VENDOR_ID_CRAY 0x17DB
> +#define PCI_DEVICE_ID_SEASTAR 0x0101
> +
> +
> +static struct pci_device_id ss_pci_tbl[] = {
> + {PCI_DEVICE(PCI_VENDOR_ID_CRAY, PCI_DEVICE_ID_SEASTAR)},
> + {0},
> +};
> +
> +
> diff -uprN -X linux-2.6.32.7-vanilla/Documentation/dontdiff linux-2.6.32.7-vanilla/drivers/net/seastar/seastar.h linux-2.6.32.7/drivers/net/seastar/seastar.h
> --- linux-2.6.32.7-vanilla/drivers/net/seastar/seastar.h 1969-12-31 17:00:00.000000000 -0700
> +++ linux-2.6.32.7/drivers/net/seastar/seastar.h 2010-02-02 09:14:33.000000000 -0700
> @@ -0,0 +1,104 @@
> +
> +#ifndef _SEASTAR_H
> +#define _SEASTAR_H
> +
> +
> +/**
> + * Rounds up to the nearest quadbyte.
> + */
> +#define ROUNDUP4(val) ((val + (4-1))& ~(4-1))
> +
> +
> +/**
> + * SeaStar datagram packet maximum transfer unit size in bytes.
> + */
> +#define SEASTAR_MTU 8192
> +
> +
> +/**
> + * Number of transmit and receive pending structures.
> + */
> +#define NUM_TX_PENDINGS 64
> +#define NUM_RX_PENDINGS 64
> +#define NUM_PENDINGS (NUM_TX_PENDINGS + NUM_RX_PENDINGS)
> +
> +
> +/**
> + * Number of entries in the SeaStar -> Host event queue.
> + */
> +#define NUM_EQ_ENTRIES 1024
> +
> +
> +/**
> + * When allocating an SKB, allocate this many bytes extra.
> + */
> +#define SKB_PAD (16 - sizeof(struct sshdr))
> +
> +
> +/**
> + * Pending structure.
> + * One of these is used to track each in progress transmit.
> + */
> +struct pending {
> + struct sk_buff *skb;
> + struct pending *next;
> + void *bounce;
> +};
> +
> +
> +/**
> + * SeaStar driver private data.
> + */
--
~Randy
^ permalink raw reply
* Re: [PATCH] seastar - SeaStar Ethernet driver
From: Ben Hutchings @ 2010-02-02 22:12 UTC (permalink / raw)
To: Randy Dunlap; +Cc: Kevin Pedretti, netdev, linux-kernel
In-Reply-To: <4B68A03E.1010403@xenotime.net>
On Tue, 2010-02-02 at 13:59 -0800, Randy Dunlap wrote:
> On 02/02/10 12:58, Kevin Pedretti wrote:
[...]
> > +static void seastar_map_host_region(struct ss_priv *ssp, const void *addr)
> > +{
> > + /* Round addr to the nearest 128 MB */
> > + unsigned long raw_paddr = __pa(addr);
> > + unsigned long paddr = raw_paddr& ~((1<< 28) - 1);
> > +
> > + htb_map[8] = 0x8000 | ((paddr>> 28) + 0);
> > + htb_map[9] = 0x8000 | ((paddr>> 28) + 1);
>
> space before <<, &, and >>
[...]
The spacing looked correct here.
Ben.
--
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.
^ permalink raw reply
* Re: [Bugme-new] [Bug 14711] New: usbnet ax88172 mac address change broken
From: Karsten König @ 2010-02-02 22:21 UTC (permalink / raw)
To: Andrew Morton; +Cc: netdev, bugzilla-daemon, bugme-daemon, linux-usb
In-Reply-To: <20091203133845.4977551c.akpm@linux-foundation.org>
No one around to tackle this?
In any case it should reject the MAC adress change as unsupported
Thanks,
Karsten
Am Donnerstag, 3. Dezember 2009 22:38:45 schrieb Andrew Morton:
> (switched to email. Please respond via emailed reply-to-all, not via the
> bugzilla web interface).
>
> On Tue, 1 Dec 2009 15:41:10 GMT
>
> bugzilla-daemon@bugzilla.kernel.org wrote:
> > http://bugzilla.kernel.org/show_bug.cgi?id=14711
> >
> > Summary: usbnet ax88172 mac address change broken
> > Product: Drivers
> > Version: 2.5
> > Kernel Version: 2.6.31rc8
> > Platform: All
> > OS/Version: Linux
> > Tree: Mainline
> > Status: NEW
> > Severity: normal
> > Priority: P1
> > Component: Network
> > AssignedTo: drivers_network@kernel-bugs.osdl.org
> > ReportedBy: remur@gmx.net
> > Regression: No
> >
> >
> > When changing the MAC address on a NETGEAR FA120 (because an old one is
> > registered with my provider) the replies to the traffic don't reach the
> > system anymore, the NIC is dropping them, putting the NIC into promiscous
> > mode will show they are properly replied to.
> >
> > Searching around I found this patch from the openBSD folks
> > http://www.mail-archive.com/misc@openbsd.org/msg76418.html
> > http://www.openbsd.org/cgi-bin/cvsweb/src/sys/dev/usb/if_axe.c.diff?r1=1.
> >91;r2=1.92
> >
> > Looks like it is not working with the ax88172, but only with newer
> > models. Still an error message would be nice when setting another
> > physical address
>
> It would be nice if some asix-competent person could take a look at porting
> that openbsd patch across, see if it fixes Karsten's hardware.
>
^ permalink raw reply
* [patch 3/4] isdn: mISDN, don't compile unused stuff
From: akpm @ 2010-02-02 22:43 UTC (permalink / raw)
To: isdn; +Cc: netdev, akpm, jirislaby
From: Jiri Slaby <jirislaby@gmail.com>
Remove these compiler warnings:
drivers/isdn/hardware/mISDN/w6692.c:534: warning: `setvolume' defined but not used
drivers/isdn/hardware/mISDN/w6692.c:561: warning: `enable_pots' defined but not used
by moving the functions inside #if 0 ... #endif. And an alternative is
to remove them completely if nobody has plans to use them.
Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Cc: Karsten Keil <isdn@linux-pingi.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---
drivers/isdn/hardware/mISDN/w6692.c | 2 ++
1 file changed, 2 insertions(+)
diff -puN drivers/isdn/hardware/mISDN/w6692.c~isdn-misdn-dont-compile-unused-stuff drivers/isdn/hardware/mISDN/w6692.c
--- a/drivers/isdn/hardware/mISDN/w6692.c~isdn-misdn-dont-compile-unused-stuff
+++ a/drivers/isdn/hardware/mISDN/w6692.c
@@ -529,6 +529,7 @@ W6692_fill_Bfifo(struct w6692_ch *wch)
}
}
+#if 0
static int
setvolume(struct w6692_ch *wch, int mic, struct sk_buff *skb)
{
@@ -571,6 +572,7 @@ enable_pots(struct w6692_ch *wch)
WriteW6692(card, W_PCTL, card->pctl);
return 0;
}
+#endif
static int
disable_pots(struct w6692_ch *wch)
_
^ permalink raw reply
* [patch 2/4] hardware/mISDN/mISDNinfineon.c: bail out of loop on error
From: akpm @ 2010-02-02 22:43 UTC (permalink / raw)
To: isdn; +Cc: netdev, akpm, darrenrjenkins
From: Darren Jenkins <darrenrjenkins@gmail.com>
If setup_instance() fails we kfree() the card, and then use it in the next
loop iteration. So lets bail out of the loop instead.
Coverity CID: 13357
Signed-off-by: Darren Jenkins <darrenrjenkins@gmail.com>
Cc: Karsten Keil <isdn@linux-pingi.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---
drivers/isdn/hardware/mISDN/mISDNinfineon.c | 1 +
1 file changed, 1 insertion(+)
diff -puN drivers/isdn/hardware/mISDN/mISDNinfineon.c~hardware-misdn-misdninfineonc-bail-out-of-loop-on-error drivers/isdn/hardware/mISDN/mISDNinfineon.c
--- a/drivers/isdn/hardware/mISDN/mISDNinfineon.c~hardware-misdn-misdninfineonc-bail-out-of-loop-on-error
+++ a/drivers/isdn/hardware/mISDN/mISDNinfineon.c
@@ -1133,6 +1133,7 @@ inf_probe(struct pci_dev *pdev, const st
if (err) {
kfree(sc);
release_card(card);
+ break;
} else
card->sc[i - 1] = sc;
}
_
^ permalink raw reply
* [patch 1/4] hisax: timeout off by one in waitrecmsg()
From: akpm @ 2010-02-02 22:43 UTC (permalink / raw)
To: isdn; +Cc: netdev, akpm, roel.kluin
From: Roel Kluin <roel.kluin@gmail.com>
With `while (timeout++ < maxdelay)' timeout reaches maxdelay + 1 after the
loop This is probably unlikely a problem in practice.
Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Cc: Karsten Keil <isdn@linux-pingi.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---
drivers/isdn/hisax/isar.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff -puN drivers/isdn/hisax/isar.c~hisax-timeout-off-by-one-in-waitrecmsg drivers/isdn/hisax/isar.c
--- a/drivers/isdn/hisax/isar.c~hisax-timeout-off-by-one-in-waitrecmsg
+++ a/drivers/isdn/hisax/isar.c
@@ -138,7 +138,7 @@ waitrecmsg(struct IsdnCardState *cs, u_c
while((!(cs->BC_Read_Reg(cs, 0, ISAR_IRQBIT) & ISAR_IRQSTA)) &&
(timeout++ < maxdelay))
udelay(1);
- if (timeout >= maxdelay) {
+ if (timeout > maxdelay) {
printk(KERN_WARNING"isar recmsg IRQSTA timeout\n");
return(0);
}
_
^ permalink raw reply
page: next (older) | prev (newer) | latest
- recent:[subjects (threaded)|topics (new)|topics (active)]
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox