* Re: [Patch 3/3] net: reserve ports for applications using fixed port numbers
From: Cong Wang @ 2010-04-13 8:48 UTC (permalink / raw)
To: Tetsuo Handa
Cc: opurdila, eric.dumazet, netdev, nhorman, davem, ebiederm,
linux-kernel
In-Reply-To: <4BC41994.7030707@redhat.com>
Cong Wang wrote:
> Tetsuo Handa wrote:
>> Hello.
>>
>>> --- linux-2.6.orig/drivers/infiniband/core/cma.c
>>> +++ linux-2.6/drivers/infiniband/core/cma.c
>>> @@ -1980,6 +1980,8 @@ retry:
>>> /* FIXME: add proper port randomization per like inet_csk_get_port */
>>> do {
>>> ret = idr_get_new_above(ps, bind_list, next_port, &port);
>>> + if (!ret && inet_is_reserved_local_port(port))
>>> + ret = -EAGAIN;
>>> } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
>>>
>>> if (ret)
>>>
>> I think above part is wrong. Below program
> ...
>> This result suggests that above loop will continue until idr_pre_get() fails
>> due to out of memory if all ports were reserved.
>>
>> Also, if idr_get_new_above() returned 0, bind_list (which is a kmalloc()ed
>> pointer) is already installed into a free slot (see comment on
>> idr_get_new_above_int()). Thus, simply calling idr_get_new_above() again will
>> install the same pointer into multiple slots. I guess it will malfunction later.
>
> Thanks for testing!
>
> How about:
>
> + if (!ret && inet_is_reserved_local_port(port))
> + ret = -EBUSY;
>
> ? So that it will break the loop and return error.
>
Or use the similar trick:
int tries = 10;
...
if(!ret && inet_is_reserved_local_port(port)) {
if (tries--)
ret = -EAGAIN;
else
ret = -EBUSY;
}
Any comments?
^ permalink raw reply
* Re: [PATCH] Fix some #includes in CAN drivers
From: David Miller @ 2010-04-13 8:36 UTC (permalink / raw)
To: hjk
Cc: socketcan-core, netdev, kernel, 21cnbao, celston, chripell, wg,
haas, per.dalen, P.B.Cheblakov, oliver.hartkopp, anantgole
In-Reply-To: <20100408192545.GD2004@bluebox.local>
From: "Hans J. Koch" <hjk@linutronix.de>
Date: Thu, 8 Apr 2010 21:25:45 +0200
> In the current implementation, CAN drivers need to #include <linux/can.h>
> _before_ they #include <linux/can/dev.h>, which is both ugly and
> unnecessary.
>
> Fix this by including <linux/can.h> in <linux/can/dev.h> and remove the
> #include <linux/can.h> lines from drivers.
>
> Signed-off-by: Hans J. Koch <hjk@linutronix.de>
This doesn't apply cleanly to net-next-2.6, please respin your patch
and resubmit.
Thanks.
^ permalink raw reply
* Re: [PATCH 2/2] [V5] Add non-Virtex5 support for LL TEMAC driver
From: David Miller @ 2010-04-13 8:34 UTC (permalink / raw)
To: grant.likely
Cc: john.linn, netdev, linuxppc-dev, jwboyer, eric.dumazet,
john.williams, michal.simek, jtyner
In-Reply-To: <x2sfa686aa41004091110l87cd2234rc68906080fc4fe03@mail.gmail.com>
From: Grant Likely <grant.likely@secretlab.ca>
Date: Fri, 9 Apr 2010 12:10:21 -0600
> On Thu, Apr 8, 2010 at 11:08 AM, John Linn <john.linn@xilinx.com> wrote:
>> This patch adds support for using the LL TEMAC Ethernet driver on
>> non-Virtex 5 platforms by adding support for accessing the Soft DMA
>> registers as if they were memory mapped instead of solely through the
>> DCR's (available on the Virtex 5).
>>
>> The patch also updates the driver so that it runs on the MicroBlaze.
>> The changes were tested on the PowerPC 440, PowerPC 405, and the
>> MicroBlaze platforms.
>>
>> Signed-off-by: John Tyner <jtyner@cs.ucr.edu>
>> Signed-off-by: John Linn <john.linn@xilinx.com>
>
> Picked up and build tested both patches on 405, 440, 60x and ppc64.
> No build problems found either built-in or as a module.
>
> for both:
> Acked-by: Grant Likely <grant.likely@secretlab.ca>
Ok, both applied to net-next-2.6, thanks everyone for sorting this
out.
^ permalink raw reply
* Re: Linux arp flux problem
From: Julian Anastasov @ 2010-04-13 8:26 UTC (permalink / raw)
To: Ming-Ching Tiew; +Cc: Net Dev
In-Reply-To: <908408.23515.qm@web31502.mail.mud.yahoo.com>
Hello,
On Sun, 11 Apr 2010, Ming-Ching Tiew wrote:
> The following link explains the Linux arp flux problem pretty well, and I myself have been burnt badly by a life site where the "arp_filter" does not help at all.
>
> http://linux-ip.net/html/ether-arp.html
>
> And I tested the kernel patch by Julian Anastasov, and it works 100% reliably :-
>
> http://www.ssi.bg/~ja/#hidden
>
> My question is the patches has been around for many years, why has it not been included into the kernel ? Is it that Linux is supposed to have this "side effects" of arp linux on purpose ?
The "hidden" flag is obsolete, kernel already has arp_ignore
and arp_announce vars, see Documentation/networking/ip-sysctl.txt
for more information. May be what you need is arp_announce=1/2
and (arp_ignore=1/2 or arp_filter=1 or rp_filter=1).
Regards
--
Julian Anastasov <ja@ssi.bg>
^ permalink raw reply
* Re: [PATCH v2] packet: support for TX time stamps on RAW sockets
From: David Miller @ 2010-04-13 8:30 UTC (permalink / raw)
To: richardcochran; +Cc: netdev
In-Reply-To: <20100408084128.GA17098@riccoc20.at.omicron.at>
From: Richard Cochran <richardcochran@gmail.com>
Date: Thu, 8 Apr 2010 10:41:28 +0200
> Enable the SO_TIMESTAMPING socket infrastructure for raw packet sockets.
> We introduce PACKET_TX_TIMESTAMP for the control message cmsg_type.
>
> Similar support for UDP and CAN sockets was added in commit
> 51f31cabe3ce5345b51e4a4f82138b38c4d5dc91
>
> Signed-off-by: Richard Cochran <richard.cochran@omicron.at>
Looks good, applied to net-next-2.6, thanks.
^ permalink raw reply
* Re: [PATCH] r6040: use (dev|netdev)_<level> macros helpers
From: David Miller @ 2010-04-13 8:29 UTC (permalink / raw)
To: florian; +Cc: netdev
In-Reply-To: <201004080939.27917.florian@openwrt.org>
From: Florian Fainelli <florian@openwrt.org>
Date: Thu, 8 Apr 2010 09:39:27 +0200
>
> Signed-off-by: Florian Fainelli <florian@openwrt.org>
Applied to net-next-2.6, thanks.
^ permalink raw reply
* Re: KS8851: Possible NULL dereferenced in ks8851_rx_pkts
From: David Miller @ 2010-04-13 8:28 UTC (permalink / raw)
To: abraham.arce.moreno; +Cc: netdev
In-Reply-To: <z2ocb8016981004080003h7f5ca358r7a1a4106809909ce@mail.gmail.com>
From: Abraham Arce <abraham.arce.moreno@gmail.com>
Date: Thu, 8 Apr 2010 02:03:57 -0500
> Hi,
>
> These changes avoid a possible dereference in skb_reserve when skb is
> NULL. I am increasing rx dropped packet count but not sure about how
> to handle the dump of frames. Any advice is appreciated.
This isn't sufficient to handle the NULL pointer.
At a minimum you're going to have to do something about
the fact that the chip has already been told to start
bringing the packet into the RX fifo.
If we just return without finishing up the hardware
operation, the chip is probably going to hang the next
time we come in here and tell it to start another RX
DMA operation without having completed the previous one.
The bug definitely needs to be fixed, however, but someone who knows
this hardware and has access to it for testing will need to implement
the fix.
^ permalink raw reply
* Re: [PATCH v2] can: Add esd board support to plx_pci CAN driver
From: David Miller @ 2010-04-13 8:23 UTC (permalink / raw)
To: wg; +Cc: matthias.fuchs, netdev, Socketcan-core
In-Reply-To: <4BC1E1C6.1050405@grandegger.com>
From: Wolfgang Grandegger <wg@grandegger.com>
Date: Sun, 11 Apr 2010 16:50:46 +0200
> Matthias Fuchs wrote:
>> This patch adds support for SJA1000 based PCI CAN interface cards
>> from electronic system design gmbh.
>>
>> Some changes have been done on the common code:
>> - esd boards must not have the 2nd local interupt enabled (PLX9030/9050)
>> - a new path for PLX9056/PEX8311 chips has been added
>> - new plx9056 reset function has been implemented
>> - struct plx_card_info got a reset function entry
>>
>> In detail the following additional boards are now supported:
>>
>> CAN-PCI/200 (PCI)
>> CAN-PCI/266 (PCI)
>> CAN-PMC266 (PMC module)
>> CAN-PCIe/2000 (PCI Express)
>> CAN-CPCI/200 (Compact PCI, 3U)
>> CAN-PCI104 (PCI104)
>>
>> Signed-off-by: Matthias Fuchs <matthias.fuchs@esd.eu>
>> ---
>> v2:
>> - update Kconfig
>> - add proper plx9056 reset function
>> - add reset function pointer to plx_card_info structure
>> - use card's reset function in plx_pci_del_card()
>
> Acked-by: Wolfgang Grandegger <wg@grandegger.com>
Applied to net-next-2.6, thanks.
^ permalink raw reply
* Re: [PATCH net-next-2.6 1/3 (TAKE 3 RESENT)] ipv6 mcast: Introduce include/net/mld.h for MLD definitions.
From: David Miller @ 2010-04-13 8:21 UTC (permalink / raw)
To: shemminger; +Cc: yoshfuji, netdev
In-Reply-To: <20100408094753.596bb61c@nehalam>
From: Stephen Hemminger <shemminger@vyatta.com>
Date: Thu, 8 Apr 2010 09:47:53 -0700
> Is there a piece missing. I tried this against net-next-2.6
>
>
> --
> CC [M] net/ipv6/mcast.o
> net/ipv6/mcast.c: In function ‘igmp6_event_query’:
> net/ipv6/mcast.c:1134: error: implicit declaration of function ‘mld_msg’
> net/ipv6/mcast.c:1134: warning: assignment makes pointer from integer without a cast
> net/ipv6/mcast.c: In function ‘igmp6_event_report’:
> net/ipv6/mcast.c:1254: warning: assignment makes pointer from integer without a cast
Yeah I also can't see how this patch set ever compiled successfully.
I'm tossing this entire set.
^ permalink raw reply
* Re: [PATCH Resubmission] drivers/net/usb: Add new driver ipheth
From: David Miller @ 2010-04-13 8:15 UTC (permalink / raw)
To: agimenez
Cc: linux-kernel, dgiagio, dborca, gregkh, jonas.sjoquist,
steve.glendinning, torgny.johansson, dbrownell, omar.oberthur,
linux-usb, netdev
In-Reply-To: <1270678281-20750-1-git-send-email-agimenez@sysvalve.es>
From: L. Alberto Giménez <agimenez@sysvalve.es>
Date: Thu, 8 Apr 2010 00:11:20 +0200
> From: dborca@yahoo.com
>
> Add new driver to use tethering with an iPhone device. After initial submission,
> apply fixes to fit the new driver into the kernel standards.
>
> There are still a couple of minor (almost cosmetic-level) issues, but the driver
> is fully functional right now.
>
> Signed-off-by: L. Alberto Giménez <agimenez@sysvalve.es>
I can't apply this.
Unless you add a rule to drivers/net/Makefile, the build won't
actually get to your driver unless one of the other USB networking
devices are configured.
Please fix this up and resubmit.
^ permalink raw reply
* Re: [PATCH v2] net: batch skb dequeueing from softnet input_pkt_queue
From: Eric Dumazet @ 2010-04-13 8:08 UTC (permalink / raw)
To: Changli Gao; +Cc: David S. Miller, netdev
In-Reply-To: <1271173102-2980-1-git-send-email-xiaosuo@gmail.com>
Le mardi 13 avril 2010 à 23:38 +0800, Changli Gao a écrit :
> batch skb dequeueing from softnet input_pkt_queue
>
> batch skb dequeueing from softnet input_pkt_queue to reduce potential lock
> contention and irq disabling/enabling.
>
Very interesting idea, but implementation is too complex, and probably
buggy, in a area that too few people understand today.
Could you keep it as simple as possible ?
> Signed-off-by: Changli Gao <xiaosuo@gmail.com>
> ----
> include/linux/netdevice.h | 2 ++
> net/core/dev.c | 45 +++++++++++++++++++++++++++++++++++++++------
> 2 files changed, 41 insertions(+), 6 deletions(-)
> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> index d1a21b5..bc7a0d7 100644
> --- a/include/linux/netdevice.h
> +++ b/include/linux/netdevice.h
> @@ -1335,6 +1335,8 @@ struct softnet_data {
> struct call_single_data csd ____cacheline_aligned_in_smp;
> #endif
> struct sk_buff_head input_pkt_queue;
> + struct sk_buff_head processing_queue;
Probably not necessary.
> + volatile bool flush_processing_queue;
Use of 'volatile' is strongly discouraged, I would say, forbidden.
Its usually a sign of 'I dont exactly what memory ordering I need, so I
throw a volatile just in case'. We live in a world full of RCU, read ,
write, full barriers. And these apis are well documented.
> struct napi_struct backlog;
> };
>
> diff --git a/net/core/dev.c b/net/core/dev.c
> index a10a216..ac24293 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -2324,6 +2324,11 @@ static void trigger_softirq(void *data)
> }
> #endif /* CONFIG_SMP */
>
> +static inline u32 softnet_input_qlen(struct softnet_data *queue)
> +{
> + return queue->input_pkt_queue.qlen + queue->processing_queue.qlen;
> +}
> +
> /*
> * enqueue_to_backlog is called to queue an skb to a per CPU backlog
> * queue (may be a remote CPU queue).
> @@ -2339,8 +2344,8 @@ static int enqueue_to_backlog(struct sk_buff *skb, int cpu)
> __get_cpu_var(netdev_rx_stat).total++;
>
> rps_lock(queue);
> - if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
> - if (queue->input_pkt_queue.qlen) {
> + if (softnet_input_qlen(queue) <= netdev_max_backlog) {
> + if (softnet_input_qlen(queue)) {
> enqueue:
> __skb_queue_tail(&queue->input_pkt_queue, skb);
> rps_unlock(queue);
> @@ -2803,6 +2808,7 @@ static void flush_backlog(void *arg)
> __skb_unlink(skb, &queue->input_pkt_queue);
> kfree_skb(skb);
> }
> + queue->flush_processing_queue = true;
Probably not necessary
> rps_unlock(queue);
> }
>
> @@ -3112,14 +3118,23 @@ static int process_backlog(struct napi_struct *napi, int quota)
> struct softnet_data *queue = &__get_cpu_var(softnet_data);
> unsigned long start_time = jiffies;
>
> + if (queue->flush_processing_queue) {
Really... this is bloat IMHO
> + struct sk_buff *skb;
> +
> + queue->flush_processing_queue = false;
> + while ((skb = __skb_dequeue(&queue->processing_queue)))
> + kfree_skb(skb);
> + }
> +
> napi->weight = weight_p;
> do {
> struct sk_buff *skb;
>
> local_irq_disable();
> rps_lock(queue);
> - skb = __skb_dequeue(&queue->input_pkt_queue);
> - if (!skb) {
> + skb_queue_splice_tail_init(&queue->input_pkt_queue,
> + &queue->processing_queue);
> + if (skb_queue_empty(&queue->processing_queue)) {
> __napi_complete(napi);
> rps_unlock(queue);
> local_irq_enable();
> @@ -3128,9 +3143,22 @@ static int process_backlog(struct napi_struct *napi, int quota)
> rps_unlock(queue);
> local_irq_enable();
>
> - __netif_receive_skb(skb);
> - } while (++work < quota && jiffies == start_time);
> + while ((skb = __skb_dequeue(&queue->processing_queue))) {
> + __netif_receive_skb(skb);
> + if (++work < quota && jiffies == start_time)
> + continue;
> + if (!queue->flush_processing_queue)
> + goto out;
> + queue->flush_processing_queue = false;
once again, ... so much code for a unlikely event...
> + while ((skb = __skb_dequeue(&queue->processing_queue))) {
> + __netif_receive_skb(skb);
> + ++work;
> + }
> + goto out;
> + }
> + } while (1);
>
> +out:
> return work;
> }
>
> @@ -5487,6 +5515,9 @@ static int dev_cpu_callback(struct notifier_block *nfb,
> raise_softirq_irqoff(NET_TX_SOFTIRQ);
> local_irq_enable();
>
> + while ((skb = __skb_dequeue(&oldsd->processing_queue)))
> + netif_rx(skb);
> +
> /* Process offline CPU's input_pkt_queue */
> while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
> netif_rx(skb);
> @@ -5709,6 +5740,8 @@ static int __init net_dev_init(void)
>
> queue = &per_cpu(softnet_data, i);
> skb_queue_head_init(&queue->input_pkt_queue);
> + skb_queue_head_init(&queue->processing_queue);
> + queue->flush_processing_queue = false;
> queue->completion_queue = NULL;
> INIT_LIST_HEAD(&queue->poll_list);
>
I advise to keep it simple.
My suggestion would be to limit this patch only to process_backlog().
Really if you touch other areas, there is too much risk.
Perform sort of skb_queue_splice_tail_init() into a local (stack) queue,
but the trick is to not touch input_pkt_queue.qlen, so that we dont slow
down enqueue_to_backlog().
Process at most 'quota' skbs (or jiffies limit).
relock queue.
input_pkt_queue.qlen -= number_of_handled_skbs;
In the unlikely event we have unprocessed skbs in local queue,
re-insert the remaining skbs at head of input_pt_queue.
Consider if input_pkt_queue.qlen is 0 or not, to call
__napi_complete(napi); or not :)
^ permalink raw reply
* Re: Strange packet drops with heavy firewalling
From: Benny Amorsen @ 2010-04-13 7:56 UTC (permalink / raw)
To: Eric Dumazet; +Cc: Changli Gao, zhigang gong, netdev
In-Reply-To: <1271138186.16881.168.camel@edumazet-laptop>
Eric Dumazet <eric.dumazet@gmail.com> writes:
> Benny, I am not sure your irqbalance is up2date with multiqueue devices,
> you might need to disable it and manually irqaffine each interrupt
True, that would probably help. Irqbalance might just believe that the
load is so low that it isn't worth rebalancing. The CPU's are spending
more than 90% of their time idling.
I'll keep monitoring the server, and if it starts dropping packets again
or load increases I'll check whether irqbalanced does the right thing,
and if not I'll implement your suggestion.
Thank you very much!
/Benny
^ permalink raw reply
* [PATCH v2] net: batch skb dequeueing from softnet input_pkt_queue
From: Changli Gao @ 2010-04-13 15:38 UTC (permalink / raw)
To: David S. Miller; +Cc: Eric Dumazet, netdev, Changli Gao
batch skb dequeueing from softnet input_pkt_queue
batch skb dequeueing from softnet input_pkt_queue to reduce potential lock
contention and irq disabling/enabling.
Signed-off-by: Changli Gao <xiaosuo@gmail.com>
----
include/linux/netdevice.h | 2 ++
net/core/dev.c | 45 +++++++++++++++++++++++++++++++++++++++------
2 files changed, 41 insertions(+), 6 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index d1a21b5..bc7a0d7 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1335,6 +1335,8 @@ struct softnet_data {
struct call_single_data csd ____cacheline_aligned_in_smp;
#endif
struct sk_buff_head input_pkt_queue;
+ struct sk_buff_head processing_queue;
+ volatile bool flush_processing_queue;
struct napi_struct backlog;
};
diff --git a/net/core/dev.c b/net/core/dev.c
index a10a216..ac24293 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2324,6 +2324,11 @@ static void trigger_softirq(void *data)
}
#endif /* CONFIG_SMP */
+static inline u32 softnet_input_qlen(struct softnet_data *queue)
+{
+ return queue->input_pkt_queue.qlen + queue->processing_queue.qlen;
+}
+
/*
* enqueue_to_backlog is called to queue an skb to a per CPU backlog
* queue (may be a remote CPU queue).
@@ -2339,8 +2344,8 @@ static int enqueue_to_backlog(struct sk_buff *skb, int cpu)
__get_cpu_var(netdev_rx_stat).total++;
rps_lock(queue);
- if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
- if (queue->input_pkt_queue.qlen) {
+ if (softnet_input_qlen(queue) <= netdev_max_backlog) {
+ if (softnet_input_qlen(queue)) {
enqueue:
__skb_queue_tail(&queue->input_pkt_queue, skb);
rps_unlock(queue);
@@ -2803,6 +2808,7 @@ static void flush_backlog(void *arg)
__skb_unlink(skb, &queue->input_pkt_queue);
kfree_skb(skb);
}
+ queue->flush_processing_queue = true;
rps_unlock(queue);
}
@@ -3112,14 +3118,23 @@ static int process_backlog(struct napi_struct *napi, int quota)
struct softnet_data *queue = &__get_cpu_var(softnet_data);
unsigned long start_time = jiffies;
+ if (queue->flush_processing_queue) {
+ struct sk_buff *skb;
+
+ queue->flush_processing_queue = false;
+ while ((skb = __skb_dequeue(&queue->processing_queue)))
+ kfree_skb(skb);
+ }
+
napi->weight = weight_p;
do {
struct sk_buff *skb;
local_irq_disable();
rps_lock(queue);
- skb = __skb_dequeue(&queue->input_pkt_queue);
- if (!skb) {
+ skb_queue_splice_tail_init(&queue->input_pkt_queue,
+ &queue->processing_queue);
+ if (skb_queue_empty(&queue->processing_queue)) {
__napi_complete(napi);
rps_unlock(queue);
local_irq_enable();
@@ -3128,9 +3143,22 @@ static int process_backlog(struct napi_struct *napi, int quota)
rps_unlock(queue);
local_irq_enable();
- __netif_receive_skb(skb);
- } while (++work < quota && jiffies == start_time);
+ while ((skb = __skb_dequeue(&queue->processing_queue))) {
+ __netif_receive_skb(skb);
+ if (++work < quota && jiffies == start_time)
+ continue;
+ if (!queue->flush_processing_queue)
+ goto out;
+ queue->flush_processing_queue = false;
+ while ((skb = __skb_dequeue(&queue->processing_queue))) {
+ __netif_receive_skb(skb);
+ ++work;
+ }
+ goto out;
+ }
+ } while (1);
+out:
return work;
}
@@ -5487,6 +5515,9 @@ static int dev_cpu_callback(struct notifier_block *nfb,
raise_softirq_irqoff(NET_TX_SOFTIRQ);
local_irq_enable();
+ while ((skb = __skb_dequeue(&oldsd->processing_queue)))
+ netif_rx(skb);
+
/* Process offline CPU's input_pkt_queue */
while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
netif_rx(skb);
@@ -5709,6 +5740,8 @@ static int __init net_dev_init(void)
queue = &per_cpu(softnet_data, i);
skb_queue_head_init(&queue->input_pkt_queue);
+ skb_queue_head_init(&queue->processing_queue);
+ queue->flush_processing_queue = false;
queue->completion_queue = NULL;
INIT_LIST_HEAD(&queue->poll_list);
^ permalink raw reply related
* Re: [Patch 1/3] sysctl: refactor integer handling proc code
From: Cong Wang @ 2010-04-13 7:35 UTC (permalink / raw)
To: Alexey Dobriyan
Cc: linux-kernel, Octavian Purdila, Eric Dumazet, penguin-kernel,
netdev, Neil Horman, ebiederm, David Miller
In-Reply-To: <20100413111814.GB4396@x200>
Alexey Dobriyan wrote:
> On Mon, Apr 12, 2010 at 06:04:04AM -0400, Amerigo Wang wrote:
>> As we are about to add another integer handling proc function a little
>> bit of cleanup is in order: add a few helper functions to improve code
>> readability and decrease code duplication.
>>
>> In the process a bug is also fixed: if the user specifies a number
>> with more then 20 digits it will be interpreted as two integers
>> (e.g. 10000...13 will be interpreted as 100.... and 13).
>
> ULONG_MAX is not 22 digits always.
>
> The fix is to not rely on simple_strtoul()
>
> I guess it's time to finally remove it. :-(
Or use strict_strtoul()?
>
> Also, it's better to copy_from user stuff once.
> Without looking at non-trivial users, one page should be enough.
It seems that all proc code assumes that the input buffer will
not exceed one page size.
>
>> Behavior for EFAULT handling was changed as well. Previous to this
>> patch, when an EFAULT error occurred in the middle of a write
>> operation, although some of the elements were set, that was not
>> acknowledged to the user (by shorting the write and returning the
>> number of bytes accepted). EFAULT is now treated just like any other
>> errors by acknowledging the amount of bytes accepted.
>
>> +static int proc_skip_wspace(char __user **buf, size_t *size)
>> +{
>> + char c;
>> +
>> + while (*size) {
>> + if (get_user(c, *buf))
>> + return -EFAULT;
>> + if (!isspace(c))
>> + break;
>> + (*size)--;
>> + (*buf)++;
>> + }
>> +
>> + return 0;
>> +}
>
> yeah, copy_from_user once, so we won't have this.
Ok.
>
>> +static bool isanyof(char c, const char *v, unsigned len)
>
> A what?
> this is memchr()
>
Hmm, right, it should be memchr(v, c, len).
Thanks!
^ permalink raw reply
* Re: BUG: using smp_processor_id() in preemptible [00000000] code: avahi-daemon: caller is netif_rx
From: Eric Dumazet @ 2010-04-13 7:14 UTC (permalink / raw)
To: Tom Herbert; +Cc: Eric Paris, netdev, David Miller
In-Reply-To: <m2m65634d661004121354ta0189542l80473b82911e43f4@mail.gmail.com>
Le lundi 12 avril 2010 à 13:54 -0700, Tom Herbert a écrit :
> Would it be better to disable preemption in netif_rx? Also note that
> with RFS we would be taking rcu_read_lock in netif_rx anyway and that
> could cover all the instances of smp_processor_id().
>
Ok that makes sense.
What do you think applying a small fix before RFS integration, it is
better to have smaller patches anyway :)
[PATCH net-next-2.6] net: netif_rx() must disable preemption
Eric Paris reported netif_rx() is calling smp_processor_id() from
preemptible context, in particular when caller is
ip_dev_loopback_xmit().
RPS commit added this smp_processor_id() call, this patch makes sure
preemption is disabled. rps_get_cpus() wants rcu_read_lock() anyway, we
can dot it a bit earlier.
Reported-by: Eric Paris <eparis@redhat.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
---
net/core/dev.c | 19 ++++++++++++-------
1 file changed, 12 insertions(+), 7 deletions(-)
diff --git a/net/core/dev.c b/net/core/dev.c
index a10a216..a96ea6a 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2206,6 +2206,7 @@ DEFINE_PER_CPU(struct netif_rx_stats, netdev_rx_stat) = { 0, };
/*
* get_rps_cpu is called from netif_receive_skb and returns the target
* CPU from the RPS map of the receiving queue for a given skb.
+ * rcu_read_lock must be held on entry.
*/
static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb)
{
@@ -2217,8 +2218,6 @@ static int get_rps_cpu(struct net_device *dev, struct sk_buff *skb)
u8 ip_proto;
u32 addr1, addr2, ports, ihl;
- rcu_read_lock();
-
if (skb_rx_queue_recorded(skb)) {
u16 index = skb_get_rx_queue(skb);
if (unlikely(index >= dev->num_rx_queues)) {
@@ -2296,7 +2295,6 @@ got_hash:
}
done:
- rcu_read_unlock();
return cpu;
}
@@ -2392,7 +2390,7 @@ enqueue:
int netif_rx(struct sk_buff *skb)
{
- int cpu;
+ int ret;
/* if netpoll wants it, pretend we never saw it */
if (netpoll_rx(skb))
@@ -2402,14 +2400,21 @@ int netif_rx(struct sk_buff *skb)
net_timestamp(skb);
#ifdef CONFIG_RPS
+ {
+ int cpu;
+
+ rcu_read_lock();
cpu = get_rps_cpu(skb->dev, skb);
if (cpu < 0)
cpu = smp_processor_id();
+ ret = enqueue_to_backlog(skb, cpu);
+ rcu_read_unlock();
+ }
#else
- cpu = smp_processor_id();
+ ret = enqueue_to_backlog(skb, get_cpu());
+ put_cpu();
#endif
-
- return enqueue_to_backlog(skb, cpu);
+ return ret;
}
EXPORT_SYMBOL(netif_rx);
^ permalink raw reply related
* Re: [Patch 3/3] net: reserve ports for applications using fixed port numbers
From: Cong Wang @ 2010-04-13 7:13 UTC (permalink / raw)
To: Tetsuo Handa
Cc: opurdila, eric.dumazet, netdev, nhorman, davem, ebiederm,
linux-kernel
In-Reply-To: <201004130121.o3D1Lhh7099571@www262.sakura.ne.jp>
Tetsuo Handa wrote:
> Hello.
>
>> --- linux-2.6.orig/drivers/infiniband/core/cma.c
>> +++ linux-2.6/drivers/infiniband/core/cma.c
>> @@ -1980,6 +1980,8 @@ retry:
>> /* FIXME: add proper port randomization per like inet_csk_get_port */
>> do {
>> ret = idr_get_new_above(ps, bind_list, next_port, &port);
>> + if (!ret && inet_is_reserved_local_port(port))
>> + ret = -EAGAIN;
>> } while ((ret == -EAGAIN) && idr_pre_get(ps, GFP_KERNEL));
>>
>> if (ret)
>>
> I think above part is wrong. Below program
...
>
> This result suggests that above loop will continue until idr_pre_get() fails
> due to out of memory if all ports were reserved.
>
> Also, if idr_get_new_above() returned 0, bind_list (which is a kmalloc()ed
> pointer) is already installed into a free slot (see comment on
> idr_get_new_above_int()). Thus, simply calling idr_get_new_above() again will
> install the same pointer into multiple slots. I guess it will malfunction later.
Thanks for testing!
How about:
+ if (!ret && inet_is_reserved_local_port(port))
+ ret = -EBUSY;
? So that it will break the loop and return error.
^ permalink raw reply
* Re: [PATCH] CONFIG_SMP should be CONFIG_RPS
From: Eric Dumazet @ 2010-04-13 6:59 UTC (permalink / raw)
To: Changli Gao; +Cc: David S. Miller, netdev, Tom Herbert
In-Reply-To: <1271169406-8115-1-git-send-email-xiaosuo@gmail.com>
Le mardi 13 avril 2010 à 22:36 +0800, Changli Gao a écrit :
> CONFIG_SMP should be CONFIG_RPS
>
> CONFIG_SMP should be CONFIG_RPS
>
> Signed-off-by: Changli Gao <xiaosuo@gmail.com>
> ----
> include/linux/netdevice.h | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
> diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
> index d1a21b5..0efb36e 100644
> --- a/include/linux/netdevice.h
> +++ b/include/linux/netdevice.h
> @@ -1331,7 +1331,7 @@ struct softnet_data {
> struct sk_buff *completion_queue;
>
> /* Elements below can be accessed between CPUs for RPS */
> -#ifdef CONFIG_SMP
> +#ifdef CONFIG_RPS
> struct call_single_data csd ____cacheline_aligned_in_smp;
> #endif
> struct sk_buff_head input_pkt_queue;
Thanks Changli, this is part of RFS patches :)
^ permalink raw reply
* [PATCH] CONFIG_SMP should be CONFIG_RPS
From: Changli Gao @ 2010-04-13 14:36 UTC (permalink / raw)
To: David S. Miller; +Cc: Eric Dumazet, netdev, Tom Herbert, Changli Gao
CONFIG_SMP should be CONFIG_RPS
CONFIG_SMP should be CONFIG_RPS
Signed-off-by: Changli Gao <xiaosuo@gmail.com>
----
include/linux/netdevice.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index d1a21b5..0efb36e 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1331,7 +1331,7 @@ struct softnet_data {
struct sk_buff *completion_queue;
/* Elements below can be accessed between CPUs for RPS */
-#ifdef CONFIG_SMP
+#ifdef CONFIG_RPS
struct call_single_data csd ____cacheline_aligned_in_smp;
#endif
struct sk_buff_head input_pkt_queue;
^ permalink raw reply related
* Re: Strange packet drops with heavy firewalling
From: Eric Dumazet @ 2010-04-13 5:56 UTC (permalink / raw)
To: Changli Gao; +Cc: Benny Amorsen, zhigang gong, netdev
In-Reply-To: <u2y412e6f7f1004121618p6d6eff30q8a45a03faa59a912@mail.gmail.com>
Le mardi 13 avril 2010 à 07:18 +0800, Changli Gao a écrit :
> On Tue, Apr 13, 2010 at 1:06 AM, Benny Amorsen <benny+usenet@amorsen.dk> wrote:
> >
> > 99: 24 1306226 3 2 PCI-MSI-edge eth1-tx-0
> > 100: 15735 1648774 3 7 PCI-MSI-edge eth1-tx-1
> > 101: 8 11 9 1083022 PCI-MSI-edge eth1-tx-2
> > 102: 0 0 0 0 PCI-MSI-edge eth1-tx-3
> > 103: 18 15 6131 1095383 PCI-MSI-edge eth1-rx-0
> > 104: 217 32 46544 1335325 PCI-MSI-edge eth1-rx-1
> > 105: 154 1305595 218 16 PCI-MSI-edge eth1-rx-2
> > 106: 17 16 8229 1467509 PCI-MSI-edge eth1-rx-3
> > 107: 0 0 1 0 PCI-MSI-edge eth1
> > 108: 2 14 15 1003053 PCI-MSI-edge eth0-tx-0
> > 109: 8226 1668924 478 487 PCI-MSI-edge eth0-tx-1
> > 110: 3 1188874 17 12 PCI-MSI-edge eth0-tx-2
> > 111: 0 0 0 0 PCI-MSI-edge eth0-tx-3
> > 112: 203 185 5324 1015263 PCI-MSI-edge eth0-rx-0
> > 113: 4141 1600793 153 159 PCI-MSI-edge eth0-rx-1
> > 114: 16242 1210108 436 3124 PCI-MSI-edge eth0-rx-2
> > 115: 267 4173 19471 1321252 PCI-MSI-edge eth0-rx-3
> > 116: 0 1 0 0 PCI-MSI-edge eth0
> >
> >
> > irqbalanced seems to have picked CPU1 and CPU3 for all the interrupts,
> > which to my mind should cause the same problem as before (where CPU1 and
> > CPU3 was handling all packets). Yet the box clearly works much better
> > than before.
>
> irqbalanced? I don't think it can work properly. Try RPS in netdev and
> linux-next tree, and if cpu load isn't even, try this patch:
> http://patchwork.ozlabs.org/patch/49915/ .
>
>
Dont try RPS on multiqueue devices !
If number of queues matches CPU numbers, it brings nothing but extra
latencies !
Benny, I am not sure your irqbalance is up2date with multiqueue devices,
you might need to disable it and manually irqaffine each interrupt
echo 01 >/proc/irq/100/smp_affinity
echo 02 >/proc/irq/101/smp_affinity
echo 04 >/proc/irq/102/smp_affinity
echo 08 >/proc/irq/103/smp_affinity
echo 10 >/proc/irq/104/smp_affinity
echo 20 >/proc/irq/105/smp_affinity
echo 40 >/proc/irq/106/smp_affinity
echo 80 >/proc/irq/107/smp_affinity
echo 01 >/proc/irq/108/smp_affinity
echo 02 >/proc/irq/109/smp_affinity
echo 04 >/proc/irq/110/smp_affinity
echo 08 >/proc/irq/111/smp_affinity
echo 10 >/proc/irq/112/smp_affinity
echo 20 >/proc/irq/113/smp_affinity
echo 40 >/proc/irq/114/smp_affinity
echo 80 >/proc/irq/115/smp_affinity
^ permalink raw reply
* Re: [PATCH] ARM: dmabounce: fix partial sync in dma_sync_single_* API
From: FUJITA Tomonori @ 2010-04-13 5:27 UTC (permalink / raw)
To: linux; +Cc: fujita.tomonori, netdev, davem, linux-arm-kernel, linux-kernel
In-Reply-To: <20100412193536.GO3048@n2100.arm.linux.org.uk>
On Mon, 12 Apr 2010 20:35:36 +0100
Russell King - ARM Linux <linux@arm.linux.org.uk> wrote:
> On Mon, Apr 05, 2010 at 12:39:32PM +0900, FUJITA Tomonori wrote:
> > I don't have arm hardware that uses dmabounce so I can't confirm the
> > problem but seems that dmabounce doesn't work for some drivers...
>
> Patch reviews fine, except for one niggle. I too don't have hardware
> I can test (well, I do except the kernel stopped supporting the UDA1341
> audio codec on the SA1110 Neponset.)
Thanks for reviewing.
> > @@ -171,10 +172,17 @@ find_safe_buffer(struct dmabounce_device_info *device_info, dma_addr_t safe_dma_
> > read_lock_irqsave(&device_info->lock, flags);
> >
> > list_for_each_entry(b, &device_info->safe_buffers, node)
> > - if (b->safe_dma_addr == safe_dma_addr) {
> > - rb = b;
> > - break;
> > - }
> > + if (for_sync) {
> > + if (b->safe_dma_addr <= safe_dma_addr &&
> > + safe_dma_addr < b->safe_dma_addr + b->size) {
> > + rb = b;
> > + break;
> > + }
> > + } else
> > + if (b->safe_dma_addr == safe_dma_addr) {
> > + rb = b;
> > + break;
> > + }
>
> This is the niggle; I don't like this indentation style. If you want to
> indent this if () statement, then please format like this:
>
> } else {
> if (b->safe...) {
> ...
> }
> }
>
> or format it as:
>
> } else if (b->safe...) {
> ...
> }
ok, here's the fixed patch.
=
From: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Subject: [PATCH] ARM: dmabounce: fix partial sync in dma_sync_single_* API
Some network drivers do a partial sync with
dma_sync_single_for_{device|cpu}. The dma_addr argument might not be
the same as one as passed into the mapping API.
This adds some tricks to find_safe_buffer() for
dma_sync_single_for_{device|cpu}.
Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
---
arch/arm/common/dmabounce.c | 30 +++++++++++++++++++++---------
1 files changed, 21 insertions(+), 9 deletions(-)
diff --git a/arch/arm/common/dmabounce.c b/arch/arm/common/dmabounce.c
index cc0a932..2e6deec 100644
--- a/arch/arm/common/dmabounce.c
+++ b/arch/arm/common/dmabounce.c
@@ -163,7 +163,8 @@ alloc_safe_buffer(struct dmabounce_device_info *device_info, void *ptr,
/* determine if a buffer is from our "safe" pool */
static inline struct safe_buffer *
-find_safe_buffer(struct dmabounce_device_info *device_info, dma_addr_t safe_dma_addr)
+find_safe_buffer(struct dmabounce_device_info *device_info, dma_addr_t safe_dma_addr,
+ int for_sync)
{
struct safe_buffer *b, *rb = NULL;
unsigned long flags;
@@ -171,9 +172,17 @@ find_safe_buffer(struct dmabounce_device_info *device_info, dma_addr_t safe_dma_
read_lock_irqsave(&device_info->lock, flags);
list_for_each_entry(b, &device_info->safe_buffers, node)
- if (b->safe_dma_addr == safe_dma_addr) {
- rb = b;
- break;
+ if (for_sync) {
+ if (b->safe_dma_addr <= safe_dma_addr &&
+ safe_dma_addr < b->safe_dma_addr + b->size) {
+ rb = b;
+ break;
+ }
+ } else {
+ if (b->safe_dma_addr == safe_dma_addr) {
+ rb = b;
+ break;
+ }
}
read_unlock_irqrestore(&device_info->lock, flags);
@@ -205,7 +214,8 @@ free_safe_buffer(struct dmabounce_device_info *device_info, struct safe_buffer *
/* ************************************************** */
static struct safe_buffer *find_safe_buffer_dev(struct device *dev,
- dma_addr_t dma_addr, const char *where)
+ dma_addr_t dma_addr, const char *where,
+ int for_sync)
{
if (!dev || !dev->archdata.dmabounce)
return NULL;
@@ -216,7 +226,7 @@ static struct safe_buffer *find_safe_buffer_dev(struct device *dev,
pr_err("unknown device: Trying to %s invalid mapping\n", where);
return NULL;
}
- return find_safe_buffer(dev->archdata.dmabounce, dma_addr);
+ return find_safe_buffer(dev->archdata.dmabounce, dma_addr, for_sync);
}
static inline dma_addr_t map_single(struct device *dev, void *ptr, size_t size,
@@ -286,7 +296,7 @@ static inline dma_addr_t map_single(struct device *dev, void *ptr, size_t size,
static inline void unmap_single(struct device *dev, dma_addr_t dma_addr,
size_t size, enum dma_data_direction dir)
{
- struct safe_buffer *buf = find_safe_buffer_dev(dev, dma_addr, "unmap");
+ struct safe_buffer *buf = find_safe_buffer_dev(dev, dma_addr, "unmap", 0);
if (buf) {
BUG_ON(buf->size != size);
@@ -398,7 +408,7 @@ int dmabounce_sync_for_cpu(struct device *dev, dma_addr_t addr,
dev_dbg(dev, "%s(dma=%#x,off=%#lx,sz=%zx,dir=%x)\n",
__func__, addr, off, sz, dir);
- buf = find_safe_buffer_dev(dev, addr, __func__);
+ buf = find_safe_buffer_dev(dev, addr, __func__, 1);
if (!buf)
return 1;
@@ -411,6 +421,8 @@ int dmabounce_sync_for_cpu(struct device *dev, dma_addr_t addr,
DO_STATS(dev->archdata.dmabounce->bounce_count++);
if (dir == DMA_FROM_DEVICE || dir == DMA_BIDIRECTIONAL) {
+ if (addr != buf->safe_dma_addr)
+ off = addr - buf->safe_dma_addr;
dev_dbg(dev, "%s: copy back safe %p to unsafe %p size %d\n",
__func__, buf->safe + off, buf->ptr + off, sz);
memcpy(buf->ptr + off, buf->safe + off, sz);
@@ -427,7 +439,7 @@ int dmabounce_sync_for_device(struct device *dev, dma_addr_t addr,
dev_dbg(dev, "%s(dma=%#x,off=%#lx,sz=%zx,dir=%x)\n",
__func__, addr, off, sz, dir);
- buf = find_safe_buffer_dev(dev, addr, __func__);
+ buf = find_safe_buffer_dev(dev, addr, __func__, 1);
if (!buf)
return 1;
--
1.6.5
^ permalink raw reply related
* Re: [PATCH] net: batch skb dequeueing from softnet input_pkt_queue
From: Changli Gao @ 2010-04-13 5:19 UTC (permalink / raw)
To: David Miller; +Cc: eric.dumazet, netdev
In-Reply-To: <20100412.214453.57473458.davem@davemloft.net>
On Tue, Apr 13, 2010 at 12:44 PM, David Miller <davem@davemloft.net> wrote:
> From: Changli Gao <xiaosuo@gmail.com>
> Date: Tue, 13 Apr 2010 18:41:08 +0800
>
>> batch skb dequeueing from softnet input_pkt_queue to reduce potential lock
>> contention and irq disabling/enabling.
>
> In exchange for a bunch of new atomic operations.
>
> No, thanks.
>
Oh, I can eliminate atomic operations totally, if I use another
variable instead. I'll submit another version later.
^ permalink raw reply
* Re: [PATCH] net: batch skb dequeueing from softnet input_pkt_queue
From: David Miller @ 2010-04-13 4:44 UTC (permalink / raw)
To: xiaosuo; +Cc: eric.dumazet, netdev
In-Reply-To: <1271155268-2999-1-git-send-email-xiaosuo@gmail.com>
From: Changli Gao <xiaosuo@gmail.com>
Date: Tue, 13 Apr 2010 18:41:08 +0800
> batch skb dequeueing from softnet input_pkt_queue to reduce potential lock
> contention and irq disabling/enabling.
In exchange for a bunch of new atomic operations.
No, thanks.
^ permalink raw reply
* Re: [Bugme-new] [Bug 15720] New: IPv6's ipv4-compatibility addresses don't bind
From: Andrew Morton @ 2010-04-13 1:22 UTC (permalink / raw)
To: netdev; +Cc: bugzilla-daemon, bugme-daemon, charles
In-Reply-To: <bug-15720-10286@https.bugzilla.kernel.org/>
(switched to email. Please respond via emailed reply-to-all, not via the
bugzilla web interface).
On Wed, 7 Apr 2010 23:17:32 GMT bugzilla-daemon@bugzilla.kernel.org wrote:
> https://bugzilla.kernel.org/show_bug.cgi?id=15720
>
> Summary: IPv6's ipv4-compatibility addresses don't bind
A 2.6.9 -> 2.6.32 regression ;)
> Product: Networking
> Version: 2.5
> Kernel Version: 2.6.32-2-686-bigmem
> Platform: All
> OS/Version: Linux
> Tree: Mainline
> Status: NEW
> Severity: normal
> Priority: P1
> Component: IPV6
> AssignedTo: yoshfuji@linux-ipv6.org
> ReportedBy: charles@kde.org
> Regression: Yes
>
>
> When attempting to bind to an address using ipv4-compatibility, for example,
> "::ffff:127.0.0.1", Linux refuses to bind to that address when
> /proc/sys/net/ipv6/bindv6only is set.
>
> Yes, you could say "but you specifically told ipv6 to not bind to ipv4
> addresses!" However, ::ffff:127.0.0.1 is *clearly* an ipv4 address, it's not an
> alternate representation of an ipv6 address, it's an ipv4 address and only
> ipv4.
>
> This seems to not have been the case as of linux 2.6.9, although I'm not sure
> at what version this changed.
>
> It seems to me that the intent of "bindv6only" was to not bind to the ipv4
> address when you bind to all addresses (specifically ipv6 address "::"). So
> when you bind to ::, an ipv4 client connects to you, and it appears to be
> connecting from ::ffff:192.168.5.5. I don't think its intent was to effectively
> disable binding to ::ffff:x.x.x.x addresses - just breaking that feature makes
> no sense.
>
> The Linux 2.6.9 approach seems to match MacOS's (and I'm pretty sure Solaris's,
> too).
^ permalink raw reply
* Re: [PATCH] rps: add flow director support
From: Changli Gao @ 2010-04-13 3:11 UTC (permalink / raw)
To: Tom Herbert; +Cc: David S. Miller, netdev
In-Reply-To: <h2m65634d661004121013uf2c86b81ndded3bb138dee7a9@mail.gmail.com>
On Tue, Apr 13, 2010 at 1:13 AM, Tom Herbert <therbert@google.com> wrote:
> On Mon, Apr 12, 2010 at 7:27 AM, Changli Gao <xiaosuo@gmail.com> wrote:
>> On Mon, Apr 12, 2010 at 9:34 PM, Tom Herbert <therbert@google.com> wrote:
>>
>
> Ideally, this should replace rps_cpus if it's a better interface....
> right now these would be conflicting interfaces.
>
How about sw-rxs and sw-rx-$ (SoftWare Receive queue). It is a
software emulation of hardware receive queue, as softnet to NIC.
sw-rx-$ is equivalent of /proc/irq/$/smp_affinity, only cpuid is
instead of cpumask.
Does anyone support this interface replacing the current rps_cpus? I do. :)
>
> It's probably a little more work, but the CPU->weight mappings could
> be implemented to cause minimal disruption in the rps_map. Also, if
> OOO is an issue, then the mitigation technique in RFS could be applied
> (this will work best when hash table is larger I believe).
>
I am thinking about the cost of keeping packets in order. Is it really
worth for this kind of random migration?
--
Regards，
Changli Gao(xiaosuo@gmail.com)
^ permalink raw reply
* [PATCH] net: batch skb dequeueing from softnet input_pkt_queue
From: Changli Gao @ 2010-04-13 10:41 UTC (permalink / raw)
To: David S. Miller; +Cc: Eric Dumazet, netdev, Changli Gao
batch skb dequeueing from softnet input_pkt_queue
batch skb dequeueing from softnet input_pkt_queue to reduce potential lock
contention and irq disabling/enabling.
Signed-off-by: Changli Gao <xiaosuo@gmail.com>
----
include/linux/netdevice.h | 1 +
net/core/dev.c | 36 +++++++++++++++++++++++++++---------
2 files changed, 28 insertions(+), 9 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index d1a21b5..f3f8cca 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -1336,6 +1336,7 @@ struct softnet_data {
#endif
struct sk_buff_head input_pkt_queue;
struct napi_struct backlog;
+ atomic_t input_qlen;
};
DECLARE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
diff --git a/net/core/dev.c b/net/core/dev.c
index a10a216..8816204 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2339,10 +2339,11 @@ static int enqueue_to_backlog(struct sk_buff *skb, int cpu)
__get_cpu_var(netdev_rx_stat).total++;
rps_lock(queue);
- if (queue->input_pkt_queue.qlen <= netdev_max_backlog) {
- if (queue->input_pkt_queue.qlen) {
+ if (atomic_read(&queue->input_qlen) <= netdev_max_backlog) {
+ if (atomic_read(&queue->input_qlen)) {
enqueue:
__skb_queue_tail(&queue->input_pkt_queue, skb);
+ atomic_inc(&queue->input_qlen);
rps_unlock(queue);
local_irq_restore(flags);
return NET_RX_SUCCESS;
@@ -2801,6 +2802,7 @@ static void flush_backlog(void *arg)
skb_queue_walk_safe(&queue->input_pkt_queue, skb, tmp)
if (skb->dev == dev) {
__skb_unlink(skb, &queue->input_pkt_queue);
+ atomic_dec(&queue->input_qlen);
kfree_skb(skb);
}
rps_unlock(queue);
@@ -3111,25 +3113,38 @@ static int process_backlog(struct napi_struct *napi, int quota)
int work = 0;
struct softnet_data *queue = &__get_cpu_var(softnet_data);
unsigned long start_time = jiffies;
+ struct sk_buff_head skb_queue;
+ __skb_queue_head_init(&skb_queue);
napi->weight = weight_p;
do {
struct sk_buff *skb;
local_irq_disable();
rps_lock(queue);
- skb = __skb_dequeue(&queue->input_pkt_queue);
- if (!skb) {
+ skb_queue_splice_tail_init(&queue->input_pkt_queue, &skb_queue);
+ if (skb_queue_empty(&skb_queue)) {
__napi_complete(napi);
- rps_unlock(queue);
- local_irq_enable();
break;
}
rps_unlock(queue);
local_irq_enable();
- __netif_receive_skb(skb);
- } while (++work < quota && jiffies == start_time);
+ while ((skb = __skb_dequeue(&skb_queue))) {
+ atomic_dec(&queue->input_qlen);
+ __netif_receive_skb(skb);
+ if (++work < quota && jiffies == start_time)
+ continue;
+ local_irq_disable();
+ rps_lock(queue);
+ skb_queue_splice(&skb_queue, &queue->input_pkt_queue);
+ goto out;
+ }
+ } while (1);
+
+out:
+ rps_unlock(queue);
+ local_irq_enable();
return work;
}
@@ -5488,8 +5503,10 @@ static int dev_cpu_callback(struct notifier_block *nfb,
local_irq_enable();
/* Process offline CPU's input_pkt_queue */
- while ((skb = __skb_dequeue(&oldsd->input_pkt_queue)))
+ while ((skb = __skb_dequeue(&oldsd->input_pkt_queue))) {
+ atomic_dec(&oldsd->input_qlen);
netif_rx(skb);
+ }
return NOTIFY_OK;
}
@@ -5709,6 +5726,7 @@ static int __init net_dev_init(void)
queue = &per_cpu(softnet_data, i);
skb_queue_head_init(&queue->input_pkt_queue);
+ atomic_set(&queue->input_qlen, 0);
queue->completion_queue = NULL;
INIT_LIST_HEAD(&queue->poll_list);
^ permalink raw reply related
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