On Thu, 2015-01-08 at 17:47 +, Erik Grinaker wrote:
> FWIW, I've done a bisection, and it’s triggered by this change:
>
> https://github.com/torvalds/linux/commit/4e4f1fc226816905c937f9b29dabe351075dfe0f
This totally makes sense, thanks for doing the bisection !
>
> > We are not going to
On Mon, 2015-01-26 at 13:12 -0800, Iyappan Subramanian wrote:
> On Thu, Jan 22, 2015 at 2:50 PM, Eric Dumazet wrote:
> > On Thu, 2015-01-22 at 12:03 -0800, Iyappan Subramanian wrote:
> >> This patch fixes the following kernel crash,
> >>
> >> WARNING: C
On Mon, 2015-01-26 at 13:27 -0800, Eric Dumazet wrote:
> What happens if you compile a kernel with CONFIG_SMP=n ?
>
>
> Most drivers in drivers/net use rmb() in this case, not smp_rmb() or
> barrier()
Note that dma_rmb() was recently added as well.
--
To unsubscribe from this
On Wed, 2015-01-28 at 00:13 +0100, Karl Beldan wrote:
> Here however I don't assume that a is "small", however I assume it has
> never overflowed, which is trivial to verify since we only add 3 32bits
> values and 2 16 bits values to a 64bits.
> Now we just want (a + b + carry(a + b)) % 2^32, and
On Tue, 2014-09-23 at 05:15 +, Sharma, Karun wrote:
> Hello:
>
> I am facing an issue wherein kernel 3.17 crashes while loading IPoIB
> module. I guess the issue discussed in this thread
> (https://www.mail-archive.com/linux-rdma@vger.kernel.org/msg20963.html) is
> similar.
>
> We were able
e in registers. This also has
> impact,
> especially on RISC processors. In the same spirit as the change done by
> Eric Dumazet on csum_replace2(), this patch rewrites
> inet_proto_csum_replace4()
> taking into account RFC1624.
>
> I spotted during a NATted tcp transfert that cs
e in registers. This also has
> impact,
> especially on RISC processors. In the same spirit as the change done by
> Eric Dumazet on csum_replace2(), this patch rewrites
> inet_proto_csum_replace4()
> taking into account RFC1624.
>
> I spotted during a NATted tcp transfert that cs
On Fri, 2014-09-19 at 14:38 +0800, Jason Wang wrote:
> Commit ce93718fb7cdbc064c3000ff59e4d3200bdfa744 ("net: Don't keep
> around original SKB when we software segment GSO frames") frees the
> original skb after software GSO even for dodgy gso skbs. This breaks
> the stream throughput from untruste
44 ("net: Don't keep
> around original SKB when we software segment GSO frames.")
>
> Cc: David S. Miller
> Cc: Eric Dumazet
> Signed-off-by: Jason Wang
> ---
Acked-by: Eric Dumazet
--
To unsubscribe from this list: send the line "unsubscribe linux-kern
On Wed, 2014-10-15 at 15:25 +0800, Jason Wang wrote:
> Accumulate the sent packets and sent bytes in local variables and perform a
> single u64_stats_update_begin/end() after.
>
> Cc: Rusty Russell
> Cc: Michael S. Tsirkin
> Signed-off-by: Jason Wang
> ---
> drivers/net/virtio_net.c | 12 +++
On Wed, 2014-10-15 at 15:25 +0800, Jason Wang wrote:
...
> +static int free_old_xmit_skbs(struct send_queue *sq, int budget)
> +{
> + struct sk_buff *skb;
> + unsigned int len;
> + struct virtnet_info *vi = sq->vq->vdev->priv;
> + struct virtnet_stats *stats = this_cpu_ptr(vi->sta
On Wed, 2014-10-15 at 14:57 -0700, Benjamin Poirier wrote:
> On 2014/10/13 12:52, Lluís Batlle i Rossell wrote:
> > Hello,
> >
> > on the 7th of January 2014 ths patch was applied:
> > https://lkml.org/lkml/2014/1/7/307
> >
> > [PATCH v2] net: Do not enable tx-nocache-copy by default
> >
On Thu, 2014-10-16 at 10:34 -0700, Benjamin Poirier wrote:
> On 2014/10/15 15:45, Eric Dumazet wrote:
> > kmap_atomic()/kunmap_atomic() is missing, so we lack
> > __cpuc_flush_dcache_area() operations.
> >
>
> You lost me there.
> 1) I don't see the link
>
On Fri, 2014-10-17 at 07:25 +0100, Thomas Graf wrote:
> On 10/17/14 at 02:34am, Steinar H. Gunderson wrote:
> > On Fri, Oct 17, 2014 at 02:21:32AM +0200, Steinar H. Gunderson wrote:
> > > Hi,
> > >
> > > We recently upgraded a machine from 3.14.5 to 3.17.1, and a Perl script
> > > we're
> > > run
On Fri, 2014-10-17 at 12:30 -0400, David Miller wrote:
> Can I ask a serious question? What is the synchronize_net() in AF_NETLINK
> exactly needed for?
__netlink_lookup() calls rhashtable_lookup_compare()
So you really want that any object found in this lookup respects
rcu grace period before
On Sat, 2014-10-18 at 21:37 -0700, Guenter Roeck wrote:
> Hi,
>
> I am getting lots of build failures with the latest upstream kernel.
>
> In file included from net/core/sock.c:140:0:
> include/net/tcp.h: In function 'tcp_v6_iif':
> include/net/tcp.h:738:32: error: 'union ' has no member named 'h
On Wed, 2014-10-15 at 16:23 +0300, Michael S. Tsirkin wrote:
> commit 0b725a2ca61bedc33a2a63d0451d528b268cf975
> net: Remove ndo_xmit_flush netdev operation, use signalling instead.
>
> added code that looks at skb->xmit_more after the skb has
> been put in TX VQ. Since some paths process the
On Fri, 2014-10-31 at 14:07 +0800, Jason Wang wrote:
> Since they are called before the possible free_old_xmit_skbs(), skb
> won't get freed at this time.
Oh right, I forgot there is no completion handler yet, timer based or
whatever.
Thanks.
--
To unsubscribe from this list: send the line "uns
On Fri, 2014-10-31 at 09:24 -0700, Alexei Starovoitov wrote:
> cc-ing netdev
>
> On Fri, Oct 31, 2014 at 7:50 AM, Chen Weilong wrote:
> > From: Weilong Chen
> >
> > We got a problem like this:
> > [8801c1a05570] machine_kexec at 81025039
> > [8801c1a055d0] crash_kexec at ff
From: Eric Dumazet
Do not reuse skb if it was pfmemalloc tainted, otherwise
future frame might be dropped anyway.
Signed-off-by: Eric Dumazet
---
net/core/dev.c |4
1 file changed, 4 insertions(+)
diff --git a/net/core/dev.c b/net/core/dev.c
index b793e3521a36..945bbd001359 100644
On Fri, 2014-10-24 at 19:38 -0700, H. Peter Anvin wrote:
> On 10/24/2014 05:22 PM, Andy Lutomirski wrote:
> > Is there any good reason not to use vmalloc for x86_64 stacks?
>
> Additional TLB pressure if anything else.
It seems TLB pressure gets less and less interest these days...
Is it still w
On Wed, 2014-11-19 at 17:37 +1100, Jon Maxwell wrote:
> Prerequisites for this to happen:
> 1) The local tcp port range must be exhausted.
> 2) A process must have called bind() followed by connect() for all
> local ports.
How the bind() is done exactly ? How SO_REUSEADDR is used ?
> 3) A diffe
On Thu, 2014-11-20 at 14:44 +1100, Jonathan Maxwell wrote:
> > > Prerequisites for this to happen:
> > > 1) The local tcp port range must be exhausted.
> > > 2) A process must have called bind() followed by connect() for all
> > > local ports.
> >
> > How the bind() is done exactly ? How SO_REUSEA
On Thu, 2014-11-20 at 21:47 +, Al Viro wrote:
> As far as I can see, these retries on the send side are simply broken -
> normally we are talking to TCP sockets there and tcp_sendmsg() does *not*
> modify iovec in normal case.
Arg... I sent this morning something doing this (against net-next
On Thu, 2014-11-20 at 22:25 +, Al Viro wrote:
> Yes, it is. You are breaking several _other_ kernel_sendmsg() users.
> They are already slightly broken, but that'll make breakage much more
> common.
>
> Please, don't - the right thing to do is to have iov_iter in msghdr
> (we already have th
On Thu, 2014-11-20 at 15:09 -0800, Calvin Owens wrote:
> Commit c3ae62af8e755 ("tcp: should drop incoming frames without ACK
> flag set") was created to mitigate a security vulnerability in which a
> local attacker is able to inject data into locally-opened sockets by
> using TCP protocol statistic
counter
> only if it is set when a valid RST packet is seen.
Seems tricky, a Challenge ACK do not necessarily gives an RST.
Anyway this certainly can wait, as we already have a sysctl to
eventually work around the issue.
Acked-by: Eric Dumazet
Thanks !
--
To unsubscribe from this list: se
On Fri, 2014-11-21 at 08:49 +, Al Viro wrote:
> Another thing is tcp_sendmsg_fastopen() and tcp_send_rcvq(). The latter
> should just use copy_from_iter() instead of memcpy_from_iovec(), the former
> is dealt with by making tcp_send_syn_data() use the same copy_from_iter()
> instead of memcpy
On Fri, 2014-11-21 at 10:06 -0500, Pranith Kumar wrote:
> Recently lockless_dereference() was added which can be used in place of
> hard-coding smp_read_barrier_depends(). The following PATCH makes the change.
>
> Signed-off-by: Pranith Kumar
> ---
> net/ipv4/netfilter/arp_tables.c | 3 +--
> ne
rse errors):
> net/phonet/pn_dev.c:278:25: error: incompatible types in comparison
> expression (different address spaces)
> net/phonet/pn_dev.c:391:17: error: incompatible types in comparison
> expression (different address spaces)
>
> Suggested-by: Eric Dumazet
> Signed-of
On Mon, 2014-10-06 at 14:51 -0700, Joe Perches wrote:
> Reduce noise when compiling W=1.
>
> All the variables are unused.
> The functions are not called outside of the file so static
> is preferred.
>
> Signed-off-by: Joe Perches
> ---
>
> John, can you please verify that these gen_stats acces
On Tue, 2014-10-07 at 15:19 +0200, Lothar Waßmann wrote:
> commit 1b7bde6d659d ("net: fec: implement rx_copybreak to improve rx
> performance")
> introduced a regression for i.MX28. The swap_buffer() function doing
> the endian conversion of the received data on i.MX28 may access memory
> beyond t
On Tue, 2014-10-07 at 14:23 +, David Laight wrote:
> The point I was making is that if you have to do a read-write of the received
> data (to byteswap it) then you might as well always copy it into a new skb
> that
> is just big enough for the actual receive frame.
+1
--
To unsubscribe fro
On Wed, 2014-10-08 at 07:42 -0700, Jim Davis wrote:
> Building with the attached random configuration file,
>
> drivers/net/ethernet/mellanox/mlx4/en_tx.c: In function
> ‘mlx4_en_process_tx_cq’:
> drivers/net/ethernet/mellanox/mlx4/en_tx.c:395:27: error: ‘struct
> netdev_queue’
> has no member n
On Wed, 2014-10-08 at 19:59 +0400, Sergei Shtylyov wrote:
> Hello.
>
> On 10/08/2014 06:53 PM, Masanari Iida wrote:
>
> > In commit 7bced397510ab569d31de4c70b39e13355046387,
>
> Please also specify that commit's summary line in parens.
Note that this information is useful for backports and
On Fri, 2014-11-07 at 09:25 +, Zoltan Kiss wrote:
Please do not top post.
> Hi,
>
> AFAIK in this scenario your skb frag is wrong. The page pointer should
> point to the original compound page (not a member of it), and offset
> should be set accordingly.
> For example, if your compound pag
From: Eric Dumazet
It is not uncommon on busy servers to get stuck hundred of ms in
vmalloc() calls (like file descriptor expansions).
Add a cond_resched() to __vmalloc_area_node() to be gentle to
other tasks.
Signed-off-by: Eric Dumazet
Cc: Hugh Dickins
Cc: David Rientjes
---
mm/vmalloc.c
On Tue, 2014-06-24 at 21:03 -0700, Tom Herbert wrote:
>
> It looks like a likely culprit is that SKB_GSO_CB()->csum_start is
> not set correctly when doing non-scatter gather. We are using
> offset as opposed to doffset.
>
> Reported-by: Dave Jones
> Signed-off-by: Tom Herbert
> ---
> net/cor
On Sun, 2014-07-06 at 14:56 +0300, Andrey Utkin wrote:
> The sock ref counting is off so there is a kernel panic when you run
> `atalkd`. See https://bugzilla.kernel.org/show_bug.cgi?id=79441
> This fix is similar to 0ae89beb283a ('can: add destructor for self
> generated skbs')
> should
> Report
On Mon, 2014-07-07 at 11:03 +0300, Andrey Utkin wrote:
> 2014-07-07 0:42 GMT+03:00 Eric Dumazet :
> >> /* Queue packet (standard) */
> >> + sock_hold(sock);
> >> + skb->destructor = atalk_skb_destructor;
> >> skb->sk = sock;
> &g
On Mon, 2014-07-07 at 10:57 +0200, Eric Dumazet wrote:
> On Mon, 2014-07-07 at 11:03 +0300, Andrey Utkin wrote:
> > 2014-07-07 0:42 GMT+03:00 Eric Dumazet :
> > >> /* Queue packet (standard) */
> > >> + sock_hold(sock);
> > >>
On Mon, 2014-07-07 at 13:02 +0300, Andrey Utkin wrote:
> 2014-07-07 12:26 GMT+03:00 Eric Dumazet :
> > Reading again this code, I think all you need is to remove the 2 buggy
> > lines.
> >
> > No need for setup destructors.
>
> Reviewing the code again, i find you
;sk.
> Thanks to Eric Dumazet for correct solution.
>
> Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=79441
> Reported-by: Ed Martin
> Signed-off-by: Andrey Utkin
> ---
Thanks !
Signed-off-by: Eric Dumazet
--
To unsubscribe from this list: send the line &quo
On Tue, 2014-07-15 at 17:41 +0800, Jason Wang wrote:
> Add basic support for rx busy polling.
>
> 1 byte netperf tcp_rr on mlx4 shows 116% improvement: the transaction
> rate was increased from 9151.94 to 19787.37.
This is a misleading changelog.
You forgot to describe how you allowed busy polli
On Fri, 2014-09-05 at 16:35 -0700, Prashant Sreedharan wrote:
> fyi.. Initially the driver was doing a skb_copy()
> (tigon3_dma_hwbug_workaround()) for LSO skb that met HW bug conditions
> but users started reporting page allocation failures due to copying of
> large LSO skbs. To avoid this Commit
On Sat, 2014-09-13 at 11:35 -0700, shakil A Khan wrote:
> On Saturday, September 13, 2014 04:50:22 AM Eric Dumazet wrote:
> > On Sat, 2014-09-13 at 01:27 -0700, Shakil A Khan wrote:
> > > Signed-off-by: Shakil A Khan
> > > ---
> > >
> > > net/
On Mon, 2014-09-15 at 14:18 +0200, Thomas Graf wrote:
> As the expansion/shrinking is moved to a worker thread, no allocations
> will be performed anymore.
>
You meant : no GFP_ATOMIC allocations ?
I would rephrase using something like :
Because hash resizes are potentially time consuming, they
On Mon, 2014-09-15 at 13:49 +0100, Thomas Graf wrote:
> Agreed. Will introduce this through a table parameter option when
> converting the inet hash table.
I am not sure you covered the /proc/net/tcp problem yet ? (or inet_diag)
--
To unsubscribe from this list: send the line "unsubscribe linux
On Mon, 2014-09-15 at 14:18 +0200, Thomas Graf wrote:
> +static int alloc_bucket_locks(struct rhashtable *ht, struct bucket_table
> *tbl)
> +{
> + unsigned int i, size;
> +#if defined(CONFIG_PROVE_LOCKING)
> + unsigned int nr_pcpus = 2;
> +#else
> + unsigned int nr_pcpus = num_possibl
On Wed, 2014-07-02 at 20:54 +0100, Zoltan Kiss wrote:
> This is a prototype patch to enable sending TCP packets with pktgen. The
> original motivation is to test TCP GSO with xen-netback/netfront, but I'm not
> sure about how the checksum should be set up, and also someone should verify
> the
> GS
On Tue, 2014-08-26 at 09:10 +, David Laight wrote:
> From: Arnd Bergmann
> > While this seems correct, I wonder why you don't do the normal approach of
> > dequeuing the skb from the chain and adding a newly allocated skb to it to
> > save the memcpy.
>
> Because the receive buffer area isn't
On Wed, 2014-08-27 at 17:18 +0300, Andreea-Cristina Bernat wrote:
> The "curr_active_slave" local variable obtained through the rcu_dereference()
> call it is not dereferenced in the rest of the function.
> Therefore, it is recommended to use rcu_access_pointer() instead of
> rcu_dereference().
> T
On Tue, 2014-09-02 at 11:04 -0700, Cong Wang wrote:
> On Tue, Sep 2, 2014 at 10:58 AM, Hannes Frederic Sowa
> > I definitely don't have a problem cleaning this up in net-next. I wanted
> > a minimal patch for stable because I didn't check history where and when
> > additional users of dev_get_by_f
On Tue, 2014-09-02 at 11:15 -0700, Cong Wang wrote:
> That is what we do when backporting patches, I can do that if David asks
> me to backport it, but you know for netdev that is David's work.
>
> (I am not saying I don't want to help him, I just want to point out the fact.
> I am very pleased t
On Tue, 2014-09-02 at 14:31 -0700, Alexei Starovoitov wrote:
> > +static inline void bpf_prog_unlock_ro(struct bpf_prog *fp)
> > +{
> > + set_memory_rw((unsigned long)fp, fp->pages);
>
> why rw is needed?
> since fp is allocated with vmalloc, vfree doesn't need
> to touch the pages to free
On Sat, 2014-09-13 at 01:27 -0700, Shakil A Khan wrote:
> Signed-off-by: Shakil A Khan
> ---
> net/core/dst.c | 5 -
> 1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/net/core/dst.c b/net/core/dst.c
> index a028409..6a848b0 100644
> --- a/net/core/dst.c
> +++ b/net/core/dst.c
etlink_seq_stop().
> --
Yes, two places use rht_dereference() instead of rht_dereference_rcu()
[PATCH net-next] netlink: fix lockdep splats
With netlink_lookup() conversion to RCU, we need to use appropriate
rcu dereference in netlink_seq_socket_idx() & netlink_seq_next()
Reported-by: Sasha L
On Thu, 2014-08-07 at 11:48 +0530, Deepak wrote:
> on SMP system, del_timer() might return even if the timer function
> is running on other cpu so sk_stop_timer() will execute __sock_put()
> while timer is accessing the socket on other cpu causing
> "use-after-free".
>
> This commi
On Thu, 2014-08-07 at 15:15 +, Das, Deepak wrote:
Please do not top post on netdev, thanks.
> I apologies for not explaining the scenario previously.
>
> sk_stop_timer() is used to stop the tcp timers with expiry callback
> tcp_write_timer(), tcp_delack_timer(), tcp_keepalive_timer(), ...
>
On Tue, 2015-04-21 at 22:06 +0200, Mateusz Guzik wrote:
> On Tue, Apr 21, 2015 at 11:05:43AM -0700, Eric Dumazet wrote:
> > 3) I avoid multiple threads doing a resize and then only one wins the
> > deal.
> >
>
> One could argue this last bit could be committed separa
On Tue, 2015-04-21 at 22:12 +0200, Mateusz Guzik wrote:
> in dup_fd:
>for (i = open_files; i != 0; i--) {
> struct file *f = *old_fds++;
> if (f) {
> get_file(f);
>
I see no new requirement here. f is either NULL or not.
multi threa
On Tue, 2015-04-21 at 23:32 +0200, Thomas Gleixner wrote:
>
> Are you realizing that __mod_timer() is a massive hotpath for network
> heavy workloads?
BTW I was considering using mod_timer_pinned() from these networking
timers (ie sk_reset_timer())
get_nohz_timer_target() sounds cool for laptop
From: Eric Dumazet
Mateusz Guzik reported :
Currently obtaining a new file descriptor results in locking fdtable
twice - once in order to reserve a slot and second time to fill it.
Holding the spinlock in __fd_install() is needed in case a resize is
done, or to prevent a resize.
Mateusz
On Wed, 2015-04-22 at 15:31 +0200, Mateusz Guzik wrote:
> On Tue, Apr 21, 2015 at 02:06:53PM -0700, Eric Dumazet wrote:
> > On Tue, 2015-04-21 at 22:12 +0200, Mateusz Guzik wrote:
> >
> > > in dup_fd:
> > >for (i = open_files; i != 0; i--) {
> > >
On Wed, 2015-04-22 at 17:29 +0200, Peter Zijlstra wrote:
> Hmm, that sounds unfortunate, this would wreck life for the power aware
> laptop/tablet etc.. people.
>
> There is already a sysctl to manage this, is that not enough to mitigate
> this problem on the server side of things?
The thing is
On Wed, 2015-04-22 at 20:56 +0200, Thomas Gleixner wrote:
> On Wed, 22 Apr 2015, Eric Dumazet wrote:
> > Check commit 4a8e320c929991c9480 ("net: sched: use pinned timers")
> > for a specific example of the problems that can be raised.
>
> If you have a problem w
On Wed, 2015-04-22 at 23:56 +0200, Thomas Gleixner wrote:
> -int get_nohz_timer_target(int pinned)
> +int get_nohz_timer_target(void)
> {
> - int cpu = smp_processor_id();
> - int i;
> + int i, cpu = smp_processor_id();
> struct sched_domain *sd;
>
> - if (pinned || !get_
On Fri, 2015-04-24 at 17:28 -0400, Sasha Levin wrote:
> Hey Eric,
>
> Your commit 79930f5892e ("net: do not deplete pfmemalloc reserve") assumes
> that
> build_skb() will only handle contiguous allocations because of the
> virt_to_head_page().
>
> However, netlink_sendmsg() calls build_skb() wit
On Fri, 2015-04-24 at 15:02 -0700, Eric Dumazet wrote:
> On Fri, 2015-04-24 at 17:28 -0400, Sasha Levin wrote:
> > Hey Eric,
> >
> > Your commit 79930f5892e ("net: do not deplete pfmemalloc reserve") assumes
> > that
> > build_skb() will only h
On Tue, 2015-04-14 at 14:25 +0200, Denys Vlasenko wrote:
> On x86 allyesconfig build:
> The function compiles to 130 bytes of machine code.
> It has 493 callsites.
> Total reduction of vmlinux size: 27906 bytes.
>
>textdata bss dec hex filename
> 82447071 22255384 20
On Thu, 2015-04-02 at 11:47 -0700, tip-bot for Peter Zijlstra wrote:
> Commit-ID: b337a9380f7effd60d082569dd7e0b97a7549730
> Gitweb: http://git.kernel.org/tip/b337a9380f7effd60d082569dd7e0b97a7549730
> Author: Peter Zijlstra
> AuthorDate: Tue, 31 Mar 2015 20:49:00 +0530
> Committer: Ingo
On Tue, 2015-04-14 at 15:57 +0200, Denys Vlasenko wrote:
> My allyesconfig, with BUG_ON's commented out:
>
Right. But I can tell you nobody uses lockdep on a production kernel.
Here, at Google, we get what I described.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel"
On Tue, 2015-04-14 at 17:04 +0200, Denys Vlasenko wrote:
> On 04/14/2015 04:21 PM, Eric Dumazet wrote:
> > On Tue, 2015-04-14 at 15:57 +0200, Denys Vlasenko wrote:
> >
> >> My allyesconfig, with BUG_ON's commented out:
> >>
> >
> > Right. But
On Tue, 2015-04-14 at 23:20 +0800, Honggang Li wrote:
> Starting monitoring for VG vg_rdma01: 3 logical volume(s) in volume
> group "vg_rdma01" monitored
> [ OK ]
> CR2: 0120
> ---[ end trace a8610f6e9640eb85 ]---
>
> Signed-off-by: Honggang Li
When was this bug added ?
Please
On Tue, 2015-04-14 at 23:53 +0800, Honggang LI wrote:
> On Tue, Apr 14, 2015 at 05:49:55PM +0200, Nicolas Dichtel wrote:
> > Le 14/04/2015 17:44, Honggang LI a écrit :
> > >On Tue, Apr 14, 2015 at 08:34:33AM -0700, Eric Dumazet wrote:
> > >>On Tue, 2015-04-14 at
On Wed, 2015-04-15 at 14:43 +0100, George Dunlap wrote:
> On Mon, Apr 13, 2015 at 2:49 PM, Eric Dumazet wrote:
> > On Mon, 2015-04-13 at 11:56 +0100, George Dunlap wrote:
> >
> >> Is the problem perhaps that netback/netfront delays TX completion?
> >> Would i
On Wed, 2015-04-15 at 18:23 +0100, George Dunlap wrote:
> On 04/15/2015 05:38 PM, Eric Dumazet wrote:
> > My thoughts that instead of these long talks you should guys read the
> > code :
> >
> > /* TCP Small Queues :
> > * C
On Wed, 2015-04-15 at 18:23 +0100, George Dunlap wrote:
> Which means that max(2*skb->truesize, sk->sk_pacing_rate >>10) is
> *already* larger for Xen; that calculation mentioned in the comment is
> *already* doing the right thing.
Sigh.
1ms of traffic at 40Gbit is 5 MBytes
The reason for the c
On Wed, 2015-04-15 at 18:41 +0100, George Dunlap wrote:
> So you'd be OK with a patch like this? (With perhaps a better changelog?)
>
> -George
>
> ---
> TSQ: Raise default static TSQ limit
>
> A new dynamic TSQ limit was introduced in c/s 605ad7f18 based on the
> size of actual packets and t
On Wed, 2015-04-15 at 10:55 -0700, Rick Jones wrote:
> >
> > Have you tested this patch on a NIC without GSO/TSO ?
> >
> > This would allow more than 500 packets for a single flow.
> >
> > Hello bufferbloat.
>
> Woudln't the fq_codel qdisc on that interface address that problem?
Last time I check
On Wed, 2015-04-15 at 18:58 +0100, Stefano Stabellini wrote:
> On Wed, 15 Apr 2015, Eric Dumazet wrote:
> > On Wed, 2015-04-15 at 18:23 +0100, George Dunlap wrote:
> >
> > > Which means that max(2*skb->truesize, sk->sk_pacing_rate >>10) is
> > &g
On Wed, 2015-04-15 at 19:04 +0100, George Dunlap wrote:
> Maybe you should stop wasting all of our time and just tell us what
> you're thinking.
I think you make me wasting my time.
I already gave all the hints in prior discussions.
Rome was not built in one day.
--
To unsubscribe from this l
On Wed, 2015-04-15 at 11:19 -0700, Rick Jones wrote:
> Well, I'm not sure that it is George and Jonathan themselves who don't
> want to change a sysctl, but the customers who would have to tweak that
> in their VMs?
Keep in mind some VM users install custom qdisc, or even custom TCP
sysctls.
On Wed, 2015-04-15 at 23:48 +0100, Al Viro wrote:
> On Wed, Apr 15, 2015 at 03:28:58PM -0700, Andy Lutomirski wrote:
> > On Wed, Apr 15, 2015 at 3:18 PM, Al Viro wrote:
> > > On Wed, Apr 15, 2015 at 03:11:17PM -0700, Andy Lutomirski wrote:
> > >
> > >> This is functionally identical to passing AF_
On Thu, 2015-04-16 at 12:20 +0800, Herbert Xu wrote:
> Eric Dumazet wrote:
> >
> > We already have netdev->gso_max_size and netdev->gso_max_segs
> > which are cached into sk->sk_gso_max_size & sk->sk_gso_max_segs
>
> It is quite dangerous to attempt tri
On Thu, 2015-04-16 at 12:39 +0100, George Dunlap wrote:
> On 04/15/2015 07:17 PM, Eric Dumazet wrote:
> > Do not expect me to fight bufferbloat alone. Be part of the challenge,
> > instead of trying to get back to proven bad solutions.
>
> I tried that. I wrote a descript
On Thu, 2015-04-16 at 13:14 +0200, Denys Vlasenko wrote:
> However, without BUG_ONs, function is still a bit big
> on PREEMPT configs.
Only on allyesconfig builds, that nobody use but to prove some points
about code size.
If you look at net_generic(), it is mostly used from code that is
normally
On Thu, 2015-04-16 at 11:01 +0100, George Dunlap wrote:
> He suggested that after he'd been prodded by 4 more e-mails in which two
> of us guessed what he was trying to get at. That's what I was
> complaining about.
My big complain is that I suggested to test to double the sysctl, which
gave goo
On Thu, 2015-04-16 at 14:16 +0200, Mateusz Guzik wrote:
> Hi,
>
> Currently obtaining a new file descriptor results in locking fdtable
> twice - once in order to reserve a slot and second time to fill it.
>
> Hack below gets rid of the second lock usage.
>
> It gives me a ~30% speedup (~300k ops
On Thu, 2015-04-16 at 19:09 +0100, Al Viro wrote:
> On Thu, Apr 16, 2015 at 02:16:31PM +0200, Mateusz Guzik wrote:
> > @@ -165,8 +165,10 @@ static int expand_fdtable(struct files_struct *files,
> > int nr)
> > cur_fdt = files_fdtable(files);
> > if (nr >= cur_fdt->max_fds) {
> >
On Thu, 2015-04-16 at 13:42 -0700, Eric Dumazet wrote:
> On Thu, 2015-04-16 at 19:09 +0100, Al Viro wrote:
> > On Thu, Apr 16, 2015 at 02:16:31PM +0200, Mateusz Guzik wrote:
> > > @@ -165,8 +165,10 @@ static int expand_fdtable(struct files_struct
> > > *fil
On Fri, 2015-04-17 at 00:00 +0200, Mateusz Guzik wrote:
> On Thu, Apr 16, 2015 at 01:55:39PM -0700, Eric Dumazet wrote:
> > On Thu, 2015-04-16 at 13:42 -0700, Eric Dumazet wrote:
> > > On Thu, 2015-04-16 at 19:09 +0100, Al Viro wrote:
> > > > On Thu, Apr 16, 2015 at 0
On Thu, 2015-04-23 at 14:45 +0200, Thomas Gleixner wrote:
> You definitely have a point from the high throughput networking
> perspective.
>
> Though in a power optimizing scenario with minimal network traffic
> this might be the wrong decision. We have to gather data from the
> power maniacs whe
wrote:
> Hi Greg,
>
> On Sun, Apr 26, 2015 at 03:46:26PM +0200, Greg Kroah-Hartman wrote:
>> 3.10-stable review patch. If anyone has any objections, please let me know.
>>
>> ------
>>
>> From: Eric Dumazet
>>
>> [ Upstream commit
On Mon, 2015-04-27 at 21:05 +0200, Mateusz Guzik wrote:
> On Tue, Apr 21, 2015 at 09:59:28PM -0700, Eric Dumazet wrote:
> > From: Eric Dumazet
> >
> > Mateusz Guzik reported :
> >
> > Currently obtaining a new file descriptor results in locking fdtable
> &
From: Eric Dumazet
Mateusz Guzik reported :
Currently obtaining a new file descriptor results in locking fdtable
twice - once in order to reserve a slot and second time to fill it.
Holding the spinlock in __fd_install() is needed in case a resize is
done, or to prevent a resize.
Mateusz
On Tue, 2015-04-07 at 15:57 -0700, Tuan Bui wrote:
> Hi all,
>
> I am consistently seeing this kernel panic on a 16 sockets machine
> running Spark PageRank workload using Docker. I am running RHEL 7.0
> stock kernel which is 3.10.0-123.el7.x86_64.
Have you tried a recent upstream kernel ?
--
On Tue, 2015-04-07 at 17:01 -0700, Tuan Bui wrote:
> On Tue, 2015-04-07 at 16:33 -0700, Eric Dumazet wrote:
> > On Tue, 2015-04-07 at 15:57 -0700, Tuan Bui wrote:
> > > Hi all,
> > >
> > > I am consistently seeing this kernel panic on a 16 sockets machine
&
; Through bisection I found that the perf regression is caused by the
> prensence of the following commit in the guest kernel:
>
>
> commit 605ad7f184b60cfaacbc038aa6c55ee68dee3c89
> Author: Eric Dumazet
> Date: Sun Dec 7 12:22:18 2014 -0800
>
> tcp: refine TSO autosizin
On Tue, 2015-05-05 at 15:00 +0200, Thomas Gleixner wrote:
> On Sat, 25 Apr 2015, Eric Dumazet wrote:
> > On Thu, 2015-04-23 at 14:45 +0200, Thomas Gleixner wrote:
> >
> > > You definitely have a point from the high throughput networking
> > > perspective.
> >
1301 - 1400 of 2671 matches
Mail list logo