* Re: [PATCH v2 1/1] tcp: fixing TLP's FIN recovery
From: Josh Hunt @ 2014-10-07 15:03 UTC (permalink / raw)
To: David Miller
Cc: per.hurtig, Eric Dumazet, panweiping3, nanditad, netdev,
anna.brunstrom, mohammad.rajiullah, ncardwell, sergei.shtylyov
In-Reply-To: <20140612.110611.584155630422154338.davem@davemloft.net>
On Thu, Jun 12, 2014 at 1:06 PM, David Miller <davem@davemloft.net> wrote:
> From: Per Hurtig <per.hurtig@kau.se>
> Date: Thu, 12 Jun 2014 17:08:32 +0200
>
>> Fix to a problem observed when losing a FIN segment that does not
>> contain data. In such situations, TLP is unable to recover from
>> *any* tail loss and instead adds at least PTO ms to the
>> retransmission process, i.e., RTO = RTO + PTO.
>>
>> Signed-off-by: Per Hurtig <per.hurtig@kau.se>
>
> Applied, thanks.
Can we queue this up for stable? 2cd0d743b05e87 (tcp: fix
tcp_match_skb_to_sack() for unaligned SACK at end of an skb) is
already in stable and based on the changelog was put in place to fix a
case that this patch introduced:
"This was visible now because the recently simplified TLP logic in
bef1909ee3ed1c ("tcp: fixing TLP's FIN recovery") could find that 0-byte
skb at the end of the write queue, and now that we do not check that
skb's length we could send it as a TLP probe."
However, the patch to fix TLP's FIN recovery is not in -stable.
Thanks
--
Josh
^ permalink raw reply
* Re: Quota in __qdisc_run() (was: qdisc: validate skb without holding lock)
From: Eric Dumazet @ 2014-10-07 15:01 UTC (permalink / raw)
To: Hannes Frederic Sowa
Cc: Jesper Dangaard Brouer, David Miller, netdev, therbert, fw,
dborkman, jhs, alexander.duyck, john.r.fastabend, dave.taht, toke
In-Reply-To: <1412693013.4140057.176149461.0FBF6CBD@webmail.messagingengine.com>
On Tue, 2014-10-07 at 16:43 +0200, Hannes Frederic Sowa wrote:
> This needs to be:
>
> do
> ...
> while ((iskb = iskb->next))
I do not feel needed to break the bulk dequeue at precise quota
boundary. These quotas are advisory, and bql prefers to get its full
budget for appropriate feedback from TX completion.
Quota was a packet quota, which was quite irrelevant if segmentation had
to be done, so I would just let the dequeue be done so that we benefit
from optimal xmit_more.
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index 2b349a4de3c8e3491fad210a9400d26bda5b52fe..581ba0dcc2474f325d1c0b3e1dc957648f11992f 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -58,7 +58,8 @@ static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
static void try_bulk_dequeue_skb(struct Qdisc *q,
struct sk_buff *skb,
- const struct netdev_queue *txq)
+ const struct netdev_queue *txq,
+ int *packets)
{
int bytelimit = qdisc_avail_bulklimit(txq) - skb->len;
@@ -71,6 +72,7 @@ static void try_bulk_dequeue_skb(struct Qdisc *q,
bytelimit -= nskb->len; /* covers GSO len */
skb->next = nskb;
skb = nskb;
+ (*packets)++;
}
skb->next = NULL;
}
@@ -78,11 +80,13 @@ static void try_bulk_dequeue_skb(struct Qdisc *q,
/* Note that dequeue_skb can possibly return a SKB list (via skb->next).
* A requeued skb (via q->gso_skb) can also be a SKB list.
*/
-static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate)
+static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate,
+ int *packets)
{
struct sk_buff *skb = q->gso_skb;
const struct netdev_queue *txq = q->dev_queue;
+ *packets = 1; /* yes, this might be not accurate, only if BQL is wrong */
*validate = true;
if (unlikely(skb)) {
/* check the reason of requeuing without tx lock first */
@@ -99,7 +103,7 @@ static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate)
!netif_xmit_frozen_or_stopped(txq)) {
skb = q->dequeue(q);
if (skb && qdisc_may_bulk(q))
- try_bulk_dequeue_skb(q, skb, txq);
+ try_bulk_dequeue_skb(q, skb, txq, packets);
}
}
return skb;
@@ -205,7 +209,7 @@ int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
* >0 - queue is not empty.
*
*/
-static inline int qdisc_restart(struct Qdisc *q)
+static inline int qdisc_restart(struct Qdisc *q, int *packets)
{
struct netdev_queue *txq;
struct net_device *dev;
@@ -214,7 +218,7 @@ static inline int qdisc_restart(struct Qdisc *q)
bool validate;
/* Dequeue packet */
- skb = dequeue_skb(q, &validate);
+ skb = dequeue_skb(q, &validate, packets);
if (unlikely(!skb))
return 0;
@@ -230,14 +234,16 @@ static inline int qdisc_restart(struct Qdisc *q)
void __qdisc_run(struct Qdisc *q)
{
int quota = weight_p;
+ int packets;
- while (qdisc_restart(q)) {
+ while (qdisc_restart(q, &packets)) {
/*
* Ordered by possible occurrence: Postpone processing if
* 1. we've exceeded packet quota
* 2. another process needs the CPU;
*/
- if (--quota <= 0 || need_resched()) {
+ quota -= packets;
+ if (quota <= 0 || need_resched()) {
__netif_schedule(q);
break;
}
^ permalink raw reply related
* Re: [PATCH 2/9] netfilter: move nf_send_resetX() code to nf_reject_ipvX modules
From: Eric Dumazet @ 2014-10-07 14:45 UTC (permalink / raw)
To: Pablo Neira Ayuso; +Cc: netfilter-devel, davem, netdev
In-Reply-To: <1412356824-6362-3-git-send-email-pablo@netfilter.org>
On Fri, 2014-10-03 at 19:20 +0200, Pablo Neira Ayuso wrote:
> Move nf_send_reset() and nf_send_reset6() to nf_reject_ipv4 and
> nf_reject_ipv6 respectively. This code is shared by x_tables and
> nf_tables.
>
> Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org>
> ---
> include/net/netfilter/ipv4/nf_reject.h | 118 +----------------------
> net/ipv4/netfilter/Kconfig | 6 ++
> net/ipv4/netfilter/Makefile | 3 +
> net/ipv4/netfilter/nf_reject_ipv4.c | 127 +++++++++++++++++++++++++
> net/ipv6/netfilter/Kconfig | 6 ++
> net/ipv6/netfilter/Makefile | 3 +
> net/ipv6/netfilter/nf_reject_ipv6.c | 163 ++++++++++++++++++++++++++++++++
> 7 files changed, 309 insertions(+), 117 deletions(-)
> create mode 100644 net/ipv4/netfilter/nf_reject_ipv4.c
> create mode 100644 net/ipv6/netfilter/nf_reject_ipv6.c
I am confused
We now have two copies of nf_send_reset6()
$ git grep -n nf_send_reset6
include/net/netfilter/ipv6/nf_reject.h:21:static void nf_send_reset6(struct net *net, struct sk_buff *oldskb, int hook)
net/bridge/netfilter/nft_reject_bridge.c:49: nf_send_reset6(net, pkt->skb, pkt->ops->hooknum);
net/ipv6/netfilter/ip6t_REJECT.c:66: nf_send_reset6(net, skb, par->hooknum);
net/ipv6/netfilter/nf_reject_ipv6.c:14:void nf_send_reset6(struct net *net, struct sk_buff *oldskb, int hook)
net/ipv6/netfilter/nf_reject_ipv6.c:163:EXPORT_SYMBOL_GPL(nf_send_reset6);
net/ipv6/netfilter/nft_reject_ipv6.c:35: nf_send_reset6(net, pkt->skb, pkt->ops->hooknum);
net/netfilter/nft_reject_inet.c:49: nf_send_reset6(net, pkt->skb, pkt->ops->hooknum);
^ permalink raw reply
* Re: pull request: wireless-next 2014-10-03
From: Larry Finger @ 2014-10-07 14:45 UTC (permalink / raw)
To: David Miller, linville; +Cc: linux-wireless, netdev, linux-kernel
In-Reply-To: <20141007.004724.187788111776188480.davem@davemloft.net>
On 10/06/2014 11:47 PM, David Miller wrote:
> From: David Miller <davem@davemloft.net>
> Date: Sun, 05 Oct 2014 21:38:53 -0400 (EDT)
>
>> From: David Miller <davem@davemloft.net>
>> Date: Sun, 05 Oct 2014 21:35:11 -0400 (EDT)
>>
>>> From: "John W. Linville" <linville@tuxdriver.com>
>>> Date: Fri, 3 Oct 2014 14:01:52 -0400
>>>
>>>> Please pull tihs batch of updates intended for the 3.18 stream!
>>>
>>> Pulled, thanks for the stellar pull request text, as always.
>>
>> John, what's the deal with the following? Will it be resolved by the
>> driver being removed from the staging tree?
>>
>> WARNING: drivers/staging/rtl8192ee/r8192ee: 'rtl_evm_dbm_jaguar' exported twice. Previous export was in drivers/net/wireless/rtlwifi/rtlwifi.ko
>
> John, ping?
David,
Yes, that message will go away once the removal of the staging driver, which was
handled by GregKH, gets merged with the changes from John.
Larry
^ permalink raw reply
* Re: Quota in __qdisc_run() (was: qdisc: validate skb without holding lock)
From: Hannes Frederic Sowa @ 2014-10-07 14:43 UTC (permalink / raw)
To: Jesper Dangaard Brouer, Eric Dumazet
Cc: David Miller, netdev, therbert, fw, dborkman, jhs,
alexander.duyck, john.r.fastabend, dave.taht, toke
In-Reply-To: <20141007153050.792c9743@redhat.com>
On Tue, Oct 7, 2014, at 15:30, Jesper Dangaard Brouer wrote:
> On Tue, 07 Oct 2014 05:47:18 -0700
> Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
> > On Tue, 2014-10-07 at 09:34 +0200, Jesper Dangaard Brouer wrote:
> > > On Fri, 03 Oct 2014 16:30:44 -0700 Eric Dumazet <eric.dumazet@gmail.com> wrote:
> > >
> > > > Another problem we need to address is the quota in __qdisc_run()
> > > > is no longer meaningfull, if each qdisc_restart() can pump many packets.
> > >
> > > I fully agree. My earlier "magic" packet limit was covering/pampering
> > > over this issue.
> >
> > Although quota was multiplied by 7 or 8 in worst case ?
>
> Yes, exactly not a very elegant solution ;-)
>
>
> > > > An idea would be to use the bstats (or cpu_qstats if applicable)
> > >
> > > Please elaborate some more, as I don't completely follow (feel free to
> > > show with a patch ;-)).
> > >
> >
> > I was hoping John could finish the percpu stats before I do that.
> >
> > Problem with q->bstats.packets is that TSO packets with 45 MSS add 45 to
> > this counter.
> >
> > Using a time quota would be better, but : jiffies is too big, and
> > local_clock() might be too expensive.
>
> Hannes hacked up this patch for me... (didn't finish testing)
>
> The basic idea is we want keep/restore the quota fairness between
> qdisc's , that we sort of broke with commit 5772e9a346 ("qdisc: bulk
> dequeue support for qdiscs with TCQ_F_ONETXQUEUE").
>
> We choose not to account for the number of packets inside the TSO/GSO
> packets ("skb_gso_segs"). As the previous fairness also had this
> "defect".
>
> You might view this as a short term solution, until you can fix it with
> your q->bstats.packets or time quota?
>
>
> [RFC PATCH] net_sched: restore quota limits after bulk dequeue
>
> --- a/net/sched/sch_generic.c
> +++ b/net/sched/sch_generic.c
> @@ -57,17 +57,19 @@ static inline int dev_requeue_skb(struct sk_buff
> *skb, struct Qdisc *q)
>
> static void try_bulk_dequeue_skb(struct Qdisc *q,
> struct sk_buff *skb,
> - const struct netdev_queue *txq)
> + const struct netdev_queue *txq,
> + int *quota)
> {
> int bytelimit = qdisc_avail_bulklimit(txq) - skb->len;
>
> - while (bytelimit > 0) {
> + while (bytelimit > 0 && *quota > 0) {
> struct sk_buff *nskb = q->dequeue(q);
>
> if (!nskb)
> break;
>
> bytelimit -= nskb->len; /* covers GSO len */
> + --*quota;
> skb->next = nskb;
> skb = nskb;
> }
> @@ -77,7 +79,7 @@ static void try_bulk_dequeue_skb(struct Qdisc *q,
> /* Note that dequeue_skb can possibly return a SKB list (via skb->next).
> * A requeued skb (via q->gso_skb) can also be a SKB list.
> */
> -static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate)
> +static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate, int
> *quota)
> {
> struct sk_buff *skb = q->gso_skb;
> const struct netdev_queue *txq = q->dev_queue;
> @@ -87,18 +89,25 @@ static struct sk_buff *dequeue_skb(struct Qdisc *q,
> bool *validate)
> /* check the reason of requeuing without tx lock first */
> txq = skb_get_tx_queue(txq->dev, skb);
> if (!netif_xmit_frozen_or_stopped(txq)) {
> + struct sk_buff *iskb = skb;
> +
> q->gso_skb = NULL;
> q->q.qlen--;
> - } else
> + do
> + --*quota;
> + while ((iskb = skb->next));
This needs to be:
do
...
while ((iskb = iskb->next))
Bye,
Hannes
^ permalink raw reply
* RE: [PATCH] net: fec: fix regression on i.MX28 introduced by rx_copybreak support
From: David Laight @ 2014-10-07 14:41 UTC (permalink / raw)
To: 'Eric Dumazet', Lothar Waßmann
Cc: netdev@vger.kernel.org, David S. Miller, Russell King, Frank Li,
Fabio Estevam, linux-kernel@vger.kernel.org
In-Reply-To: <1412692034.11091.122.camel@edumazet-glaptop2.roam.corp.google.com>
From: Eric Dumazet
> On Tue, 2014-10-07 at 15:19 +0200, Lothar Wamann 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 the actual packet size in the DMA buffer. fec_enet_copybreak()
> > does not copy those bytes, so that the last bytes of a packet may be
> > filled with invalid data after swapping.
> > This will likely lead to checksum errors on received packets.
> > E.g. when trying to mount an NFS rootfs:
> > UDP: bad checksum. From 192.168.1.225:111 to 192.168.100.73:44662 ulen 36
> >
> > Do the byte swapping and copying to the new skb in one go if
> > necessary.
> >
> > Signed-off-by: Lothar Wamann <LW@KARO-electronics.de>
> > ---
> > drivers/net/ethernet/freescale/fec_main.c | 25 +++++++++++++++++++++----
> > 1 file changed, 21 insertions(+), 4 deletions(-)
> >
> > diff --git a/drivers/net/ethernet/freescale/fec_main.c b/drivers/net/ethernet/freescale/fec_main.c
> > index 87975b5..eaaebad 100644
> > --- a/drivers/net/ethernet/freescale/fec_main.c
> > +++ b/drivers/net/ethernet/freescale/fec_main.c
> > @@ -339,6 +339,18 @@ static void *swap_buffer(void *bufaddr, int len)
> > return bufaddr;
> > }
> >
> > +static void *swap_buffer2(void *dst_buf, void *src_buf, int len)
> > +{
> > + int i;
> > + unsigned int *src = src_buf;
> > + unsigned int *dst = dst_buf;
> > +
> > + for (i = 0; i < DIV_ROUND_UP(len, 4); i++, src++, dst++)
> > + *dst = cpu_to_be32(*src);
>
> No need for the DIV :
>
> for (i = 0; i < len; i += sizeof(*dst), src++, dst++)
> *dst = cpu_to_be32(*src);
>
> Also are you sure both src/dst are aligned to word boundaries, or is
> this architecture OK with possible misalignment ?
I wondered about that as well.
I wouldn't have expected ppc to support misaligned transfers, and you'd also
want to make sure that cpu_to_be(*src) was using a byte-swapping instruction.
Hmmm... cpu_to_be() doesn't sound like the right 'swap' macro name.
It looks as though this is processing an entire ethernet frame, in which
case the data (destination address) would normally be on an 4n+2 boundary.
The two bytes before the data can be safely copied (as can the rounding bytes at
the end) - so provided the pointers allow for this it should be fine.
OTOH I've NFI what the actual data is.
David
^ permalink raw reply
* RE: [PATCH] net: fec: fix regression on i.MX28 introduced by rx_copybreak support
From: Eric Dumazet @ 2014-10-07 14:34 UTC (permalink / raw)
To: David Laight
Cc: 'Lothar Waßmann', netdev@vger.kernel.org,
David S. Miller, Russell King, Frank Li, Fabio Estevam,
linux-kernel@vger.kernel.org
In-Reply-To: <063D6719AE5E284EB5DD2968C1650D6D174C60B1@AcuExch.aculab.com>
On Tue, 2014-10-07 at 14:23 +0000, 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
^ permalink raw reply
* Re: [PATCH] net: fec: fix regression on i.MX28 introduced by rx_copybreak support
From: Eric Dumazet @ 2014-10-07 14:27 UTC (permalink / raw)
To: Lothar Waßmann
Cc: netdev, David S. Miller, Russell King, Frank Li, Fabio Estevam,
linux-kernel
In-Reply-To: <1412687977-11742-1-git-send-email-LW@KARO-electronics.de>
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 the actual packet size in the DMA buffer. fec_enet_copybreak()
> does not copy those bytes, so that the last bytes of a packet may be
> filled with invalid data after swapping.
> This will likely lead to checksum errors on received packets.
> E.g. when trying to mount an NFS rootfs:
> UDP: bad checksum. From 192.168.1.225:111 to 192.168.100.73:44662 ulen 36
>
> Do the byte swapping and copying to the new skb in one go if
> necessary.
>
> Signed-off-by: Lothar Waßmann <LW@KARO-electronics.de>
> ---
> drivers/net/ethernet/freescale/fec_main.c | 25 +++++++++++++++++++++----
> 1 file changed, 21 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/net/ethernet/freescale/fec_main.c b/drivers/net/ethernet/freescale/fec_main.c
> index 87975b5..eaaebad 100644
> --- a/drivers/net/ethernet/freescale/fec_main.c
> +++ b/drivers/net/ethernet/freescale/fec_main.c
> @@ -339,6 +339,18 @@ static void *swap_buffer(void *bufaddr, int len)
> return bufaddr;
> }
>
> +static void *swap_buffer2(void *dst_buf, void *src_buf, int len)
> +{
> + int i;
> + unsigned int *src = src_buf;
> + unsigned int *dst = dst_buf;
> +
> + for (i = 0; i < DIV_ROUND_UP(len, 4); i++, src++, dst++)
> + *dst = cpu_to_be32(*src);
No need for the DIV :
for (i = 0; i < len; i += sizeof(*dst), src++, dst++)
*dst = cpu_to_be32(*src);
Also are you sure both src/dst are aligned to word boundaries, or is
this architecture OK with possible misalignment ?
^ permalink raw reply
* RE: [PATCH] net: fec: fix regression on i.MX28 introduced by rx_copybreak support
From: David Laight @ 2014-10-07 14:23 UTC (permalink / raw)
To: 'Lothar Waßmann'
Cc: netdev@vger.kernel.org, David S. Miller, Russell King, Frank Li,
Fabio Estevam, linux-kernel@vger.kernel.org
In-Reply-To: <20141007161702.7cd1e800@ipc1.ka-ro>
From: Lothar
> David Laight wrote:
> > From: Lothar Waßmann
> > > 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 the actual packet size in the DMA buffer. fec_enet_copybreak()
> > > does not copy those bytes, so that the last bytes of a packet may be
> > > filled with invalid data after swapping.
> > > This will likely lead to checksum errors on received packets.
> > > E.g. when trying to mount an NFS rootfs:
> > > UDP: bad checksum. From 192.168.1.225:111 to 192.168.100.73:44662 ulen 36
> > >
> > > Do the byte swapping and copying to the new skb in one go if
> > > necessary.
> >
> > ISTM that if you need to do the 'swap' you should copy the data regardless
> > of the length.
> >
> The swap function has to look at at most 3 bytes beyond the actual
> packet length. That is what the original swap_buffer() function does and
> what the new function swap_buffer2(), that does the endian swapping
> while copying to the new buffer, also does.
I understood the bug.
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.
David
^ permalink raw reply
* Re: Quota in __qdisc_run() (was: qdisc: validate skb without holding lock)
From: Jesper Dangaard Brouer @ 2014-10-07 13:30 UTC (permalink / raw)
To: Eric Dumazet
Cc: David Miller, netdev, therbert, hannes, fw, dborkman, jhs,
alexander.duyck, john.r.fastabend, dave.taht, toke, brouer
In-Reply-To: <1412686038.11091.111.camel@edumazet-glaptop2.roam.corp.google.com>
On Tue, 07 Oct 2014 05:47:18 -0700
Eric Dumazet <eric.dumazet@gmail.com> wrote:
> On Tue, 2014-10-07 at 09:34 +0200, Jesper Dangaard Brouer wrote:
> > On Fri, 03 Oct 2014 16:30:44 -0700 Eric Dumazet <eric.dumazet@gmail.com> wrote:
> >
> > > Another problem we need to address is the quota in __qdisc_run()
> > > is no longer meaningfull, if each qdisc_restart() can pump many packets.
> >
> > I fully agree. My earlier "magic" packet limit was covering/pampering
> > over this issue.
>
> Although quota was multiplied by 7 or 8 in worst case ?
Yes, exactly not a very elegant solution ;-)
> > > An idea would be to use the bstats (or cpu_qstats if applicable)
> >
> > Please elaborate some more, as I don't completely follow (feel free to
> > show with a patch ;-)).
> >
>
> I was hoping John could finish the percpu stats before I do that.
>
> Problem with q->bstats.packets is that TSO packets with 45 MSS add 45 to
> this counter.
>
> Using a time quota would be better, but : jiffies is too big, and
> local_clock() might be too expensive.
Hannes hacked up this patch for me... (didn't finish testing)
The basic idea is we want keep/restore the quota fairness between
qdisc's , that we sort of broke with commit 5772e9a346 ("qdisc: bulk
dequeue support for qdiscs with TCQ_F_ONETXQUEUE").
We choose not to account for the number of packets inside the TSO/GSO
packets ("skb_gso_segs"). As the previous fairness also had this "defect".
You might view this as a short term solution, until you can fix it with
your q->bstats.packets or time quota?
[RFC PATCH] net_sched: restore quota limits after bulk dequeue
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -57,17 +57,19 @@ static inline int dev_requeue_skb(struct sk_buff *skb, struct Qdisc *q)
static void try_bulk_dequeue_skb(struct Qdisc *q,
struct sk_buff *skb,
- const struct netdev_queue *txq)
+ const struct netdev_queue *txq,
+ int *quota)
{
int bytelimit = qdisc_avail_bulklimit(txq) - skb->len;
- while (bytelimit > 0) {
+ while (bytelimit > 0 && *quota > 0) {
struct sk_buff *nskb = q->dequeue(q);
if (!nskb)
break;
bytelimit -= nskb->len; /* covers GSO len */
+ --*quota;
skb->next = nskb;
skb = nskb;
}
@@ -77,7 +79,7 @@ static void try_bulk_dequeue_skb(struct Qdisc *q,
/* Note that dequeue_skb can possibly return a SKB list (via skb->next).
* A requeued skb (via q->gso_skb) can also be a SKB list.
*/
-static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate)
+static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate, int *quota)
{
struct sk_buff *skb = q->gso_skb;
const struct netdev_queue *txq = q->dev_queue;
@@ -87,18 +89,25 @@ static struct sk_buff *dequeue_skb(struct Qdisc *q, bool *validate)
/* check the reason of requeuing without tx lock first */
txq = skb_get_tx_queue(txq->dev, skb);
if (!netif_xmit_frozen_or_stopped(txq)) {
+ struct sk_buff *iskb = skb;
+
q->gso_skb = NULL;
q->q.qlen--;
- } else
+ do
+ --*quota;
+ while ((iskb = skb->next));
+ } else {
skb = NULL;
+ }
/* skb in gso_skb were already validated */
*validate = false;
} else {
if (!(q->flags & TCQ_F_ONETXQUEUE) ||
!netif_xmit_frozen_or_stopped(txq)) {
skb = q->dequeue(q);
+ --*quota;
if (skb && qdisc_may_bulk(q))
- try_bulk_dequeue_skb(q, skb, txq);
+ try_bulk_dequeue_skb(q, skb, txq, quota);
}
}
return skb;
@@ -204,7 +213,7 @@ int sch_direct_xmit(struct sk_buff *skb, struct Qdisc *q,
* >0 - queue is not empty.
*
*/
-static inline int qdisc_restart(struct Qdisc *q)
+static inline int qdisc_restart(struct Qdisc *q, int *quota)
{
struct netdev_queue *txq;
struct net_device *dev;
@@ -213,7 +222,7 @@ static inline int qdisc_restart(struct Qdisc *q)
bool validate;
/* Dequeue packet */
- skb = dequeue_skb(q, &validate);
+ skb = dequeue_skb(q, &validate, quota);
if (unlikely(!skb))
return 0;
@@ -230,13 +239,13 @@ void __qdisc_run(struct Qdisc *q)
{
int quota = weight_p;
- while (qdisc_restart(q)) {
+ while (qdisc_restart(q, "a)) {
/*
* Ordered by possible occurrence: Postpone processing if
* 1. we've exceeded packet quota
* 2. another process needs the CPU;
*/
- if (--quota <= 0 || need_resched()) {
+ if (quota <= 0 || need_resched()) {
__netif_schedule(q);
break;
}
^ permalink raw reply
* Re: pull request: wireless-next 2014-10-03
From: John W. Linville @ 2014-10-07 14:18 UTC (permalink / raw)
To: David Miller; +Cc: linux-wireless, netdev, linux-kernel, larry.finger
In-Reply-To: <20141007.004724.187788111776188480.davem@davemloft.net>
On Tue, Oct 07, 2014 at 12:47:24AM -0400, David Miller wrote:
> From: David Miller <davem@davemloft.net>
> Date: Sun, 05 Oct 2014 21:38:53 -0400 (EDT)
>
> > From: David Miller <davem@davemloft.net>
> > Date: Sun, 05 Oct 2014 21:35:11 -0400 (EDT)
> >
> >> From: "John W. Linville" <linville@tuxdriver.com>
> >> Date: Fri, 3 Oct 2014 14:01:52 -0400
> >>
> >>> Please pull tihs batch of updates intended for the 3.18 stream!
> >>
> >> Pulled, thanks for the stellar pull request text, as always.
> >
> > John, what's the deal with the following? Will it be resolved by the
> > driver being removed from the staging tree?
> >
> > WARNING: drivers/staging/rtl8192ee/r8192ee: 'rtl_evm_dbm_jaguar' exported twice. Previous export was in drivers/net/wireless/rtlwifi/rtlwifi.ko
>
> John, ping?
Sorry, Dave -- sendmail died on my local machine and I didn't notice
until now... :-(
Anyway...yes, the staging driver should be removed by a patch in
Greg's queue.
John
--
John W. Linville Someday the world will need a hero, and you
linville@tuxdriver.com might be all we have. Be ready.
^ permalink raw reply
* Re: [PATCH] net: fec: fix regression on i.MX28 introduced by rx_copybreak support
From: Lothar Waßmann @ 2014-10-07 14:17 UTC (permalink / raw)
To: David Laight
Cc: netdev@vger.kernel.org, David S. Miller, Russell King, Frank Li,
Fabio Estevam, linux-kernel@vger.kernel.org
In-Reply-To: <063D6719AE5E284EB5DD2968C1650D6D174C6025@AcuExch.aculab.com>
Hi,
David Laight wrote:
> From: Lothar Waßmann
> > 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 the actual packet size in the DMA buffer. fec_enet_copybreak()
> > does not copy those bytes, so that the last bytes of a packet may be
> > filled with invalid data after swapping.
> > This will likely lead to checksum errors on received packets.
> > E.g. when trying to mount an NFS rootfs:
> > UDP: bad checksum. From 192.168.1.225:111 to 192.168.100.73:44662 ulen 36
> >
> > Do the byte swapping and copying to the new skb in one go if
> > necessary.
>
> ISTM that if you need to do the 'swap' you should copy the data regardless
> of the length.
>
The swap function has to look at at most 3 bytes beyond the actual
packet length. That is what the original swap_buffer() function does and
what the new function swap_buffer2(), that does the endian swapping
while copying to the new buffer, also does.
Lothar Waßmann
--
___________________________________________________________
Ka-Ro electronics GmbH | Pascalstraße 22 | D - 52076 Aachen
Phone: +49 2408 1402-0 | Fax: +49 2408 1402-10
Geschäftsführer: Matthias Kaussen
Handelsregistereintrag: Amtsgericht Aachen, HRB 4996
www.karo-electronics.de | info@karo-electronics.de
___________________________________________________________
^ permalink raw reply
* Re: [PATCH] net: fec: fix regression on i.MX28 introduced by rx_copybreak support
From: Fabio Estevam @ 2014-10-07 14:02 UTC (permalink / raw)
To: Lothar Waßmann
Cc: netdev@vger.kernel.org, David S. Miller, Russell King, Frank Li,
Fabio Estevam, linux-kernel
In-Reply-To: <1412687977-11742-1-git-send-email-LW@KARO-electronics.de>
Hi Lothar,
On Tue, Oct 7, 2014 at 10:19 AM, Lothar Waßmann <LW@karo-electronics.de> 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 the actual packet size in the DMA buffer. fec_enet_copybreak()
> does not copy those bytes, so that the last bytes of a packet may be
> filled with invalid data after swapping.
> This will likely lead to checksum errors on received packets.
> E.g. when trying to mount an NFS rootfs:
> UDP: bad checksum. From 192.168.1.225:111 to 192.168.100.73:44662 ulen 36
>
> Do the byte swapping and copying to the new skb in one go if
> necessary.
>
> Signed-off-by: Lothar Waßmann <LW@KARO-electronics.de>
Yesterday night I was using linux-next on mx28evk and could not boot from NFS.
I haven't had a chance to debug it, but I am glad you found a fix.
Won't be able to have access to my mx28evk until Thursday to test it though.
Thanks
^ permalink raw reply
* [PATCH RESEND v3] 3c59x: fix bad split of cpu_to_le32(pci_map_single())
From: Sylvain 'ythier' Hitier @ 2014-10-07 13:40 UTC (permalink / raw)
To: linux-kernel; +Cc: nhorman, mroos, David Miller, netdev, Steffen Klassert
From: Sylvain "ythier" Hitier <sylvain.hitier@gmail.com>
In commit 6f2b6a3005b2c34c39f207a87667564f64f2f91a,
# 3c59x: Add dma error checking and recovery
the intent is to split out the mapping from the byte-swapping in order to
insert a dma_mapping_error() check.
Kinda this semantic patch:
// See http://coccinelle.lip6.fr/
//
// Beware, grouik-and-dirty!
@@
expression DEV, X, Y, Z;
@@
- cpu_to_le32(pci_map_single(DEV, X, Y, Z))
+ dma_addr_t addr = pci_map_single(DEV, X, Y, Z);
+ if (dma_mapping_error(&DEV->dev, addr))
+ /* snip */;
+ cpu_to_le32(addr)
However, the #else part (of the #if DO_ZEROCOPY test) is changed this way:
- cpu_to_le32(pci_map_single(DEV, X, Y, Z))
+ dma_addr_t addr = cpu_to_le32(pci_map_single(DEV, X, Y, Z));
// ^^^^^^^^^^^
// That mismatches the 3 other changes!
+ if (dma_mapping_error(&DEV->dev, addr))
+ /* snip */;
+ cpu_to_le32(addr)
Let's remove the leftover cpu_to_le32() for coherency.
v2: Better changelog.
v3: Add Acked-by
Fixes: 6f2b6a3005b2c34c39f207a87667564f64f2f91a
# 3c59x: Add dma error checking and recovery
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: Sylvain "ythier" Hitier <sylvain.hitier@gmail.com>
---
[Resent with maintainer and mailing-list Cc-ed]
drivers/net/ethernet/3com/3c59x.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/net/ethernet/3com/3c59x.c b/drivers/net/ethernet/3com/3c59x.c
index 8ca49f04..0a3108b3 100644
--- a/drivers/net/ethernet/3com/3c59x.c
+++ b/drivers/net/ethernet/3com/3c59x.c
@@ -2214,7 +2214,7 @@ boomerang_start_xmit(struct sk_buff *skb, struct net_device *dev)
}
}
#else
- dma_addr = cpu_to_le32(pci_map_single(VORTEX_PCI(vp), skb->data, skb->len, PCI_DMA_TODEVICE));
+ dma_addr = pci_map_single(VORTEX_PCI(vp), skb->data, skb->len, PCI_DMA_TODEVICE);
if (dma_mapping_error(&VORTEX_PCI(vp)->dev, dma_addr))
goto out_dma_err;
vp->tx_ring[entry].addr = cpu_to_le32(dma_addr);
--
1.7.10.4
Regards,
Sylvain "ythier" Hitier
--
Business is about being busy, not being rich...
Lived 777 days in a Debian package => http://en.wikipedia.org/wiki/Apt,_Vaucluse
There's THE room for ideals in this mechanical place!
^ permalink raw reply related
* RE: [PATCH] net: fec: fix regression on i.MX28 introduced by rx_copybreak support
From: David Laight @ 2014-10-07 13:31 UTC (permalink / raw)
To: 'Lothar Waßmann', netdev@vger.kernel.org
Cc: David S. Miller, Russell King, Frank Li, Fabio Estevam,
linux-kernel@vger.kernel.org
In-Reply-To: <1412687977-11742-1-git-send-email-LW@KARO-electronics.de>
From: Lothar Waßmann
> 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 the actual packet size in the DMA buffer. fec_enet_copybreak()
> does not copy those bytes, so that the last bytes of a packet may be
> filled with invalid data after swapping.
> This will likely lead to checksum errors on received packets.
> E.g. when trying to mount an NFS rootfs:
> UDP: bad checksum. From 192.168.1.225:111 to 192.168.100.73:44662 ulen 36
>
> Do the byte swapping and copying to the new skb in one go if
> necessary.
ISTM that if you need to do the 'swap' you should copy the data regardless
of the length.
David
>
> Signed-off-by: Lothar Wamann <LW@KARO-electronics.de>
> ---
> drivers/net/ethernet/freescale/fec_main.c | 25 +++++++++++++++++++++----
> 1 file changed, 21 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/net/ethernet/freescale/fec_main.c b/drivers/net/ethernet/freescale/fec_main.c
> index 87975b5..eaaebad 100644
> --- a/drivers/net/ethernet/freescale/fec_main.c
> +++ b/drivers/net/ethernet/freescale/fec_main.c
> @@ -339,6 +339,18 @@ static void *swap_buffer(void *bufaddr, int len)
> return bufaddr;
> }
>
> +static void *swap_buffer2(void *dst_buf, void *src_buf, int len)
> +{
> + int i;
> + unsigned int *src = src_buf;
> + unsigned int *dst = dst_buf;
> +
> + for (i = 0; i < DIV_ROUND_UP(len, 4); i++, src++, dst++)
> + *dst = cpu_to_be32(*src);
> +
> + return dst_buf;
> +}
> +
> static void fec_dump(struct net_device *ndev)
> {
> struct fec_enet_private *fep = netdev_priv(ndev);
> @@ -1348,7 +1360,7 @@ fec_enet_new_rxbdp(struct net_device *ndev, struct bufdesc *bdp, struct sk_buff
> }
>
> static bool fec_enet_copybreak(struct net_device *ndev, struct sk_buff **skb,
> - struct bufdesc *bdp, u32 length)
> + struct bufdesc *bdp, u32 length, int swap)
> {
> struct fec_enet_private *fep = netdev_priv(ndev);
> struct sk_buff *new_skb;
> @@ -1363,7 +1375,10 @@ static bool fec_enet_copybreak(struct net_device *ndev, struct sk_buff **skb,
> dma_sync_single_for_cpu(&fep->pdev->dev, bdp->cbd_bufaddr,
> FEC_ENET_RX_FRSIZE - fep->rx_align,
> DMA_FROM_DEVICE);
> - memcpy(new_skb->data, (*skb)->data, length);
> + if (!swap)
> + memcpy(new_skb->data, (*skb)->data, length);
> + else
> + swap_buffer2(new_skb->data, (*skb)->data, length);
> *skb = new_skb;
>
> return true;
> @@ -1393,6 +1408,7 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
> u16 vlan_tag;
> int index = 0;
> bool is_copybreak;
> + bool need_swap = id_entry->driver_data & FEC_QUIRK_SWAP_FRAME;
>
> #ifdef CONFIG_M532x
> flush_cache_all();
> @@ -1456,7 +1472,8 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
> * include that when passing upstream as it messes up
> * bridging applications.
> */
> - is_copybreak = fec_enet_copybreak(ndev, &skb, bdp, pkt_len - 4);
> + is_copybreak = fec_enet_copybreak(ndev, &skb, bdp, pkt_len - 4,
> + need_swap);
> if (!is_copybreak) {
> skb_new = netdev_alloc_skb(ndev, FEC_ENET_RX_FRSIZE);
> if (unlikely(!skb_new)) {
> @@ -1471,7 +1488,7 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
> prefetch(skb->data - NET_IP_ALIGN);
> skb_put(skb, pkt_len - 4);
> data = skb->data;
> - if (id_entry->driver_data & FEC_QUIRK_SWAP_FRAME)
> + if (!is_copybreak && need_swap)
> swap_buffer(data, pkt_len);
>
> /* Extract the enhanced buffer descriptor */
> --
> 1.7.10.4
>
> --
> 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
* [PATCH] net: fec: fix regression on i.MX28 introduced by rx_copybreak support
From: Lothar Waßmann @ 2014-10-07 13:19 UTC (permalink / raw)
To: netdev
Cc: David S. Miller, Russell King, Frank Li, Fabio Estevam,
linux-kernel, Lothar Waßmann
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 the actual packet size in the DMA buffer. fec_enet_copybreak()
does not copy those bytes, so that the last bytes of a packet may be
filled with invalid data after swapping.
This will likely lead to checksum errors on received packets.
E.g. when trying to mount an NFS rootfs:
UDP: bad checksum. From 192.168.1.225:111 to 192.168.100.73:44662 ulen 36
Do the byte swapping and copying to the new skb in one go if
necessary.
Signed-off-by: Lothar Waßmann <LW@KARO-electronics.de>
---
drivers/net/ethernet/freescale/fec_main.c | 25 +++++++++++++++++++++----
1 file changed, 21 insertions(+), 4 deletions(-)
diff --git a/drivers/net/ethernet/freescale/fec_main.c b/drivers/net/ethernet/freescale/fec_main.c
index 87975b5..eaaebad 100644
--- a/drivers/net/ethernet/freescale/fec_main.c
+++ b/drivers/net/ethernet/freescale/fec_main.c
@@ -339,6 +339,18 @@ static void *swap_buffer(void *bufaddr, int len)
return bufaddr;
}
+static void *swap_buffer2(void *dst_buf, void *src_buf, int len)
+{
+ int i;
+ unsigned int *src = src_buf;
+ unsigned int *dst = dst_buf;
+
+ for (i = 0; i < DIV_ROUND_UP(len, 4); i++, src++, dst++)
+ *dst = cpu_to_be32(*src);
+
+ return dst_buf;
+}
+
static void fec_dump(struct net_device *ndev)
{
struct fec_enet_private *fep = netdev_priv(ndev);
@@ -1348,7 +1360,7 @@ fec_enet_new_rxbdp(struct net_device *ndev, struct bufdesc *bdp, struct sk_buff
}
static bool fec_enet_copybreak(struct net_device *ndev, struct sk_buff **skb,
- struct bufdesc *bdp, u32 length)
+ struct bufdesc *bdp, u32 length, int swap)
{
struct fec_enet_private *fep = netdev_priv(ndev);
struct sk_buff *new_skb;
@@ -1363,7 +1375,10 @@ static bool fec_enet_copybreak(struct net_device *ndev, struct sk_buff **skb,
dma_sync_single_for_cpu(&fep->pdev->dev, bdp->cbd_bufaddr,
FEC_ENET_RX_FRSIZE - fep->rx_align,
DMA_FROM_DEVICE);
- memcpy(new_skb->data, (*skb)->data, length);
+ if (!swap)
+ memcpy(new_skb->data, (*skb)->data, length);
+ else
+ swap_buffer2(new_skb->data, (*skb)->data, length);
*skb = new_skb;
return true;
@@ -1393,6 +1408,7 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
u16 vlan_tag;
int index = 0;
bool is_copybreak;
+ bool need_swap = id_entry->driver_data & FEC_QUIRK_SWAP_FRAME;
#ifdef CONFIG_M532x
flush_cache_all();
@@ -1456,7 +1472,8 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
* include that when passing upstream as it messes up
* bridging applications.
*/
- is_copybreak = fec_enet_copybreak(ndev, &skb, bdp, pkt_len - 4);
+ is_copybreak = fec_enet_copybreak(ndev, &skb, bdp, pkt_len - 4,
+ need_swap);
if (!is_copybreak) {
skb_new = netdev_alloc_skb(ndev, FEC_ENET_RX_FRSIZE);
if (unlikely(!skb_new)) {
@@ -1471,7 +1488,7 @@ fec_enet_rx_queue(struct net_device *ndev, int budget, u16 queue_id)
prefetch(skb->data - NET_IP_ALIGN);
skb_put(skb, pkt_len - 4);
data = skb->data;
- if (id_entry->driver_data & FEC_QUIRK_SWAP_FRAME)
+ if (!is_copybreak && need_swap)
swap_buffer(data, pkt_len);
/* Extract the enhanced buffer descriptor */
--
1.7.10.4
^ permalink raw reply related
* Re: [Xen-devel] [PATCHv1] xen-netfront: always keep the Rx ring full of requests
From: annie li @ 2014-10-07 13:12 UTC (permalink / raw)
To: David Vrabel; +Cc: David Miller, netdev, boris.ostrovsky, xen-devel
In-Reply-To: <5433B5C8.9060309@citrix.com>
On 2014/10/7 5:43, David Vrabel wrote:
> On 06/10/14 22:07, David Miller wrote:
>> From: annie li <annie.li@oracle.com>
>> Date: Mon, 06 Oct 2014 14:41:48 -0400
>>
>>> On 2014/10/6 12:00, David Vrabel wrote:
>>>>>> + queue->rx.req_prod_pvt = req_prod;
>>>>>> +
>>>>>> + /* Not enough requests? Try again later. */
>>>>>> + if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN) {
>>>>>> + mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
>>>>>> + return;
>>>>> If the previous for loop breaks because of failure of
>>>>> xennet_alloc_one_rx_buffer, then notify_remote_via_irq is missed here
>>>>> if
>>>>> the code returns directly.
>>>> This is deliberate -- there's no point notifying the backend if there
>>>> aren't enough requests for the next packet. Since we don't know what
>>>> the next packet might be we assume it's the largest possible.
>>> That makes sense.
>>> However, the largest packet case does not happen so
>>> frequently. Moreover, netback checks the slots every incoming skb
>>> requires in xenvif_rx_ring_slots_available, not only concerning the
>>> largest case.
> An upcoming change to netback will cause it to wait for enough slots for
> the largest possible packet.
Netback knows the exact slot number that incoming skb will consumes, is
there any reason to let it wait for the largest possible packets?
Thanks
Annie
^ permalink raw reply
* [PATCH 2/2] net: fs_enet: Add NAPI TX
From: Christophe Leroy @ 2014-10-07 13:05 UTC (permalink / raw)
To: Pantelis Antoniou, Vitaly Bordug; +Cc: linux-kernel, linuxppc-dev, netdev
When using a MPC8xx as a router, 'perf' shows a significant time spent in
fs_enet_interrupt() and fs_enet_start_xmit().
'perf annotate' shows that the time spent in fs_enet_start_xmit is indeed spent
between spin_unlock_irqrestore() and the following instruction, hence in
interrupt handling. This is due to the TX complete interrupt that fires after
each transmitted packet.
This patch modifies the handling of TX complete to use NAPI.
With this patch, my NAT router offers a throughput improved by 21%
Original performance:
[root@localhost tmp]# scp toto pgs:/tmp
toto 100% 256MB 2.8MB/s 01:31
Performance with the patch:
[root@localhost tmp]# scp toto pgs:/tmp
toto 100% 256MB 3.4MB/s 01:16
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
.../net/ethernet/freescale/fs_enet/fs_enet-main.c | 47 +++++++++++++++++-----
drivers/net/ethernet/freescale/fs_enet/fs_enet.h | 9 ++++-
drivers/net/ethernet/freescale/fs_enet/mac-fcc.c | 29 +++++++++++++
drivers/net/ethernet/freescale/fs_enet/mac-fec.c | 29 +++++++++++++
drivers/net/ethernet/freescale/fs_enet/mac-scc.c | 29 +++++++++++++
5 files changed, 132 insertions(+), 11 deletions(-)
diff --git a/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c b/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c
index 71a25b4..c92c3b7 100644
--- a/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c
+++ b/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c
@@ -215,17 +215,23 @@ static int fs_enet_rx_napi(struct napi_struct *napi, int budget)
return received;
}
-static void fs_enet_tx(struct net_device *dev)
+static int fs_enet_tx_napi(struct napi_struct *napi, int budget)
{
- struct fs_enet_private *fep = netdev_priv(dev);
+ struct fs_enet_private *fep = container_of(napi, struct fs_enet_private,
+ napi_tx);
+ struct net_device *dev = fep->ndev;
cbd_t __iomem *bdp;
struct sk_buff *skb;
int dirtyidx, do_wake, do_restart;
u16 sc;
+ int has_tx_work = 0;
spin_lock(&fep->tx_lock);
bdp = fep->dirty_tx;
+ /* clear TX status bits for napi*/
+ (*fep->ops->napi_clear_tx_event)(dev);
+
do_wake = do_restart = 0;
while (((sc = CBDR_SC(bdp)) & BD_ENET_TX_READY) == 0) {
dirtyidx = bdp - fep->tx_bd_base;
@@ -278,7 +284,7 @@ static void fs_enet_tx(struct net_device *dev)
/*
* Free the sk buffer associated with this last transmit.
*/
- dev_kfree_skb_irq(skb);
+ dev_kfree_skb(skb);
fep->tx_skbuff[dirtyidx] = NULL;
/*
@@ -295,6 +301,7 @@ static void fs_enet_tx(struct net_device *dev)
*/
if (!fep->tx_free++)
do_wake = 1;
+ has_tx_work = 1;
}
fep->dirty_tx = bdp;
@@ -302,10 +309,19 @@ static void fs_enet_tx(struct net_device *dev)
if (do_restart)
(*fep->ops->tx_restart)(dev);
+ if (!has_tx_work) {
+ napi_complete(napi);
+ (*fep->ops->napi_enable_tx)(dev);
+ }
+
spin_unlock(&fep->tx_lock);
if (do_wake)
netif_wake_queue(dev);
+
+ if (has_tx_work)
+ return budget;
+ return 0;
}
/*
@@ -350,8 +366,17 @@ fs_enet_interrupt(int irq, void *dev_id)
__napi_schedule(&fep->napi);
}
- if (int_events & fep->ev_tx)
- fs_enet_tx(dev);
+ if (int_events & fep->ev_tx) {
+ napi_ok = napi_schedule_prep(&fep->napi_tx);
+
+ (*fep->ops->napi_disable_tx)(dev);
+ (*fep->ops->clear_int_events)(dev, fep->ev_napi_tx);
+
+ /* NOTE: it is possible for FCCs in NAPI mode */
+ /* to submit a spurious interrupt while in poll */
+ if (napi_ok)
+ __napi_schedule(&fep->napi_tx);
+ }
}
handled = nr > 0;
@@ -484,7 +509,6 @@ static int fs_enet_start_xmit(struct sk_buff *skb, struct net_device *dev)
cbd_t __iomem *bdp;
int curidx;
u16 sc;
- unsigned long flags;
#ifdef CONFIG_FS_ENET_MPC5121_FEC
if (((unsigned long)skb->data) & 0x3) {
@@ -499,7 +523,7 @@ static int fs_enet_start_xmit(struct sk_buff *skb, struct net_device *dev)
}
}
#endif
- spin_lock_irqsave(&fep->tx_lock, flags);
+ spin_lock(&fep->tx_lock);
/*
* Fill in a Tx ring entry
@@ -508,7 +532,7 @@ static int fs_enet_start_xmit(struct sk_buff *skb, struct net_device *dev)
if (!fep->tx_free || (CBDR_SC(bdp) & BD_ENET_TX_READY)) {
netif_stop_queue(dev);
- spin_unlock_irqrestore(&fep->tx_lock, flags);
+ spin_unlock(&fep->tx_lock);
/*
* Ooops. All transmit buffers are full. Bail out.
@@ -564,7 +588,7 @@ static int fs_enet_start_xmit(struct sk_buff *skb, struct net_device *dev)
(*fep->ops->tx_kickstart)(dev);
- spin_unlock_irqrestore(&fep->tx_lock, flags);
+ spin_unlock(&fep->tx_lock);
return NETDEV_TX_OK;
}
@@ -685,6 +709,7 @@ static int fs_enet_open(struct net_device *dev)
fs_init_bds(fep->ndev);
napi_enable(&fep->napi);
+ napi_enable(&fep->napi_tx);
/* Install our interrupt handler. */
r = request_irq(fep->interrupt, fs_enet_interrupt, IRQF_SHARED,
@@ -692,6 +717,7 @@ static int fs_enet_open(struct net_device *dev)
if (r != 0) {
dev_err(fep->dev, "Could not allocate FS_ENET IRQ!");
napi_disable(&fep->napi);
+ napi_disable(&fep->napi_tx);
return -EINVAL;
}
@@ -699,6 +725,7 @@ static int fs_enet_open(struct net_device *dev)
if (err) {
free_irq(fep->interrupt, dev);
napi_disable(&fep->napi);
+ napi_disable(&fep->napi_tx);
return err;
}
phy_start(fep->phydev);
@@ -716,6 +743,7 @@ static int fs_enet_close(struct net_device *dev)
netif_stop_queue(dev);
netif_carrier_off(dev);
napi_disable(&fep->napi);
+ napi_disable(&fep->napi_tx);
phy_stop(fep->phydev);
spin_lock_irqsave(&fep->lock, flags);
@@ -971,6 +999,7 @@ static int fs_enet_probe(struct platform_device *ofdev)
ndev->netdev_ops = &fs_enet_netdev_ops;
ndev->watchdog_timeo = 2 * HZ;
netif_napi_add(ndev, &fep->napi, fs_enet_rx_napi, fpi->napi_weight);
+ netif_napi_add(ndev, &fep->napi_tx, fs_enet_tx_napi, 2);
ndev->ethtool_ops = &fs_ethtool_ops;
diff --git a/drivers/net/ethernet/freescale/fs_enet/fs_enet.h b/drivers/net/ethernet/freescale/fs_enet/fs_enet.h
index 1ece4b1..3a4b49e 100644
--- a/drivers/net/ethernet/freescale/fs_enet/fs_enet.h
+++ b/drivers/net/ethernet/freescale/fs_enet/fs_enet.h
@@ -84,6 +84,9 @@ struct fs_ops {
void (*napi_clear_rx_event)(struct net_device *dev);
void (*napi_enable_rx)(struct net_device *dev);
void (*napi_disable_rx)(struct net_device *dev);
+ void (*napi_clear_tx_event)(struct net_device *dev);
+ void (*napi_enable_tx)(struct net_device *dev);
+ void (*napi_disable_tx)(struct net_device *dev);
void (*rx_bd_done)(struct net_device *dev);
void (*tx_kickstart)(struct net_device *dev);
u32 (*get_int_events)(struct net_device *dev);
@@ -119,6 +122,7 @@ struct phy_info {
struct fs_enet_private {
struct napi_struct napi;
+ struct napi_struct napi_tx;
struct device *dev; /* pointer back to the device (must be initialized first) */
struct net_device *ndev;
spinlock_t lock; /* during all ops except TX pckt processing */
@@ -149,6 +153,7 @@ struct fs_enet_private {
/* event masks */
u32 ev_napi_rx; /* mask of NAPI rx events */
+ u32 ev_napi_tx; /* mask of NAPI rx events */
u32 ev_rx; /* rx event mask */
u32 ev_tx; /* tx event mask */
u32 ev_err; /* error event mask */
@@ -191,8 +196,8 @@ void fs_cleanup_bds(struct net_device *dev);
#define DRV_MODULE_NAME "fs_enet"
#define PFX DRV_MODULE_NAME ": "
-#define DRV_MODULE_VERSION "1.0"
-#define DRV_MODULE_RELDATE "Aug 8, 2005"
+#define DRV_MODULE_VERSION "1.1"
+#define DRV_MODULE_RELDATE "Sep 22, 2014"
/***************************************************************************/
diff --git a/drivers/net/ethernet/freescale/fs_enet/mac-fcc.c b/drivers/net/ethernet/freescale/fs_enet/mac-fcc.c
index f5383ab..2c578db 100644
--- a/drivers/net/ethernet/freescale/fs_enet/mac-fcc.c
+++ b/drivers/net/ethernet/freescale/fs_enet/mac-fcc.c
@@ -125,6 +125,7 @@ out:
}
#define FCC_NAPI_RX_EVENT_MSK (FCC_ENET_RXF | FCC_ENET_RXB)
+#define FCC_NAPI_TX_EVENT_MSK (FCC_ENET_TXF | FCC_ENET_TXB)
#define FCC_RX_EVENT (FCC_ENET_RXF)
#define FCC_TX_EVENT (FCC_ENET_TXB)
#define FCC_ERR_EVENT_MSK (FCC_ENET_TXE)
@@ -137,6 +138,7 @@ static int setup_data(struct net_device *dev)
return -EINVAL;
fep->ev_napi_rx = FCC_NAPI_RX_EVENT_MSK;
+ fep->ev_napi_tx = FCC_NAPI_TX_EVENT_MSK;
fep->ev_rx = FCC_RX_EVENT;
fep->ev_tx = FCC_TX_EVENT;
fep->ev_err = FCC_ERR_EVENT_MSK;
@@ -446,6 +448,30 @@ static void napi_disable_rx(struct net_device *dev)
C16(fccp, fcc_fccm, FCC_NAPI_RX_EVENT_MSK);
}
+static void napi_clear_tx_event(struct net_device *dev)
+{
+ struct fs_enet_private *fep = netdev_priv(dev);
+ fcc_t __iomem *fccp = fep->fcc.fccp;
+
+ W16(fccp, fcc_fcce, FCC_NAPI_TX_EVENT_MSK);
+}
+
+static void napi_enable_tx(struct net_device *dev)
+{
+ struct fs_enet_private *fep = netdev_priv(dev);
+ fcc_t __iomem *fccp = fep->fcc.fccp;
+
+ S16(fccp, fcc_fccm, FCC_NAPI_TX_EVENT_MSK);
+}
+
+static void napi_disable_tx(struct net_device *dev)
+{
+ struct fs_enet_private *fep = netdev_priv(dev);
+ fcc_t __iomem *fccp = fep->fcc.fccp;
+
+ C16(fccp, fcc_fccm, FCC_NAPI_TX_EVENT_MSK);
+}
+
static void rx_bd_done(struct net_device *dev)
{
/* nothing */
@@ -572,6 +598,9 @@ const struct fs_ops fs_fcc_ops = {
.napi_clear_rx_event = napi_clear_rx_event,
.napi_enable_rx = napi_enable_rx,
.napi_disable_rx = napi_disable_rx,
+ .napi_clear_tx_event = napi_clear_tx_event,
+ .napi_enable_tx = napi_enable_tx,
+ .napi_disable_tx = napi_disable_tx,
.rx_bd_done = rx_bd_done,
.tx_kickstart = tx_kickstart,
.get_int_events = get_int_events,
diff --git a/drivers/net/ethernet/freescale/fs_enet/mac-fec.c b/drivers/net/ethernet/freescale/fs_enet/mac-fec.c
index 1eedfba2..3d4e08b 100644
--- a/drivers/net/ethernet/freescale/fs_enet/mac-fec.c
+++ b/drivers/net/ethernet/freescale/fs_enet/mac-fec.c
@@ -110,6 +110,7 @@ static int do_pd_setup(struct fs_enet_private *fep)
}
#define FEC_NAPI_RX_EVENT_MSK (FEC_ENET_RXF | FEC_ENET_RXB)
+#define FEC_NAPI_TX_EVENT_MSK (FEC_ENET_TXF | FEC_ENET_TXB)
#define FEC_RX_EVENT (FEC_ENET_RXF)
#define FEC_TX_EVENT (FEC_ENET_TXF)
#define FEC_ERR_EVENT_MSK (FEC_ENET_HBERR | FEC_ENET_BABR | \
@@ -126,6 +127,7 @@ static int setup_data(struct net_device *dev)
fep->fec.htlo = 0;
fep->ev_napi_rx = FEC_NAPI_RX_EVENT_MSK;
+ fep->ev_napi_tx = FEC_NAPI_TX_EVENT_MSK;
fep->ev_rx = FEC_RX_EVENT;
fep->ev_tx = FEC_TX_EVENT;
fep->ev_err = FEC_ERR_EVENT_MSK;
@@ -415,6 +417,30 @@ static void napi_disable_rx(struct net_device *dev)
FC(fecp, imask, FEC_NAPI_RX_EVENT_MSK);
}
+static void napi_clear_tx_event(struct net_device *dev)
+{
+ struct fs_enet_private *fep = netdev_priv(dev);
+ struct fec __iomem *fecp = fep->fec.fecp;
+
+ FW(fecp, ievent, FEC_NAPI_TX_EVENT_MSK);
+}
+
+static void napi_enable_tx(struct net_device *dev)
+{
+ struct fs_enet_private *fep = netdev_priv(dev);
+ struct fec __iomem *fecp = fep->fec.fecp;
+
+ FS(fecp, imask, FEC_NAPI_TX_EVENT_MSK);
+}
+
+static void napi_disable_tx(struct net_device *dev)
+{
+ struct fs_enet_private *fep = netdev_priv(dev);
+ struct fec __iomem *fecp = fep->fec.fecp;
+
+ FC(fecp, imask, FEC_NAPI_TX_EVENT_MSK);
+}
+
static void rx_bd_done(struct net_device *dev)
{
struct fs_enet_private *fep = netdev_priv(dev);
@@ -487,6 +513,9 @@ const struct fs_ops fs_fec_ops = {
.napi_clear_rx_event = napi_clear_rx_event,
.napi_enable_rx = napi_enable_rx,
.napi_disable_rx = napi_disable_rx,
+ .napi_clear_tx_event = napi_clear_tx_event,
+ .napi_enable_tx = napi_enable_tx,
+ .napi_disable_tx = napi_disable_tx,
.rx_bd_done = rx_bd_done,
.tx_kickstart = tx_kickstart,
.get_int_events = get_int_events,
diff --git a/drivers/net/ethernet/freescale/fs_enet/mac-scc.c b/drivers/net/ethernet/freescale/fs_enet/mac-scc.c
index 90b3b19..41aa0b4 100644
--- a/drivers/net/ethernet/freescale/fs_enet/mac-scc.c
+++ b/drivers/net/ethernet/freescale/fs_enet/mac-scc.c
@@ -116,6 +116,7 @@ static int do_pd_setup(struct fs_enet_private *fep)
}
#define SCC_NAPI_RX_EVENT_MSK (SCCE_ENET_RXF | SCCE_ENET_RXB)
+#define SCC_NAPI_TX_EVENT_MSK (SCCE_ENET_TXF | SCCE_ENET_TXB)
#define SCC_RX_EVENT (SCCE_ENET_RXF)
#define SCC_TX_EVENT (SCCE_ENET_TXB)
#define SCC_ERR_EVENT_MSK (SCCE_ENET_TXE | SCCE_ENET_BSY)
@@ -130,6 +131,7 @@ static int setup_data(struct net_device *dev)
fep->scc.htlo = 0;
fep->ev_napi_rx = SCC_NAPI_RX_EVENT_MSK;
+ fep->ev_napi_tx = SCC_NAPI_TX_EVENT_MSK;
fep->ev_rx = SCC_RX_EVENT;
fep->ev_tx = SCC_TX_EVENT | SCCE_ENET_TXE;
fep->ev_err = SCC_ERR_EVENT_MSK;
@@ -398,6 +400,30 @@ static void napi_disable_rx(struct net_device *dev)
C16(sccp, scc_sccm, SCC_NAPI_RX_EVENT_MSK);
}
+static void napi_clear_tx_event(struct net_device *dev)
+{
+ struct fs_enet_private *fep = netdev_priv(dev);
+ scc_t __iomem *sccp = fep->scc.sccp;
+
+ W16(sccp, scc_scce, SCC_NAPI_TX_EVENT_MSK);
+}
+
+static void napi_enable_tx(struct net_device *dev)
+{
+ struct fs_enet_private *fep = netdev_priv(dev);
+ scc_t __iomem *sccp = fep->scc.sccp;
+
+ S16(sccp, scc_sccm, SCC_NAPI_TX_EVENT_MSK);
+}
+
+static void napi_disable_tx(struct net_device *dev)
+{
+ struct fs_enet_private *fep = netdev_priv(dev);
+ scc_t __iomem *sccp = fep->scc.sccp;
+
+ C16(sccp, scc_sccm, SCC_NAPI_TX_EVENT_MSK);
+}
+
static void rx_bd_done(struct net_device *dev)
{
/* nothing */
@@ -471,6 +497,9 @@ const struct fs_ops fs_scc_ops = {
.napi_clear_rx_event = napi_clear_rx_event,
.napi_enable_rx = napi_enable_rx,
.napi_disable_rx = napi_disable_rx,
+ .napi_clear_tx_event = napi_clear_tx_event,
+ .napi_enable_tx = napi_enable_tx,
+ .napi_disable_tx = napi_disable_tx,
.rx_bd_done = rx_bd_done,
.tx_kickstart = tx_kickstart,
.get_int_events = get_int_events,
--
2.1.0
^ permalink raw reply related
* [PATCH 1/2] net: fs_enet: Remove non NAPI RX
From: Christophe Leroy @ 2014-10-07 13:04 UTC (permalink / raw)
To: Pantelis Antoniou, Vitaly Bordug; +Cc: linux-kernel, linuxppc-dev, netdev
In the probe function, use_napi is inconditionnaly set to 1. This patch removes
all the code which is conditional to !use_napi, and removes use_napi which has
then become useless.
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
.../net/ethernet/freescale/fs_enet/fs_enet-main.c | 164 ++-------------------
include/linux/fs_enet_pd.h | 1 -
2 files changed, 15 insertions(+), 150 deletions(-)
diff --git a/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c b/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c
index 748fd24..71a25b4 100644
--- a/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c
+++ b/drivers/net/ethernet/freescale/fs_enet/fs_enet-main.c
@@ -215,128 +215,6 @@ static int fs_enet_rx_napi(struct napi_struct *napi, int budget)
return received;
}
-/* non NAPI receive function */
-static int fs_enet_rx_non_napi(struct net_device *dev)
-{
- struct fs_enet_private *fep = netdev_priv(dev);
- const struct fs_platform_info *fpi = fep->fpi;
- cbd_t __iomem *bdp;
- struct sk_buff *skb, *skbn, *skbt;
- int received = 0;
- u16 pkt_len, sc;
- int curidx;
- /*
- * First, grab all of the stats for the incoming packet.
- * These get messed up if we get called due to a busy condition.
- */
- bdp = fep->cur_rx;
-
- while (((sc = CBDR_SC(bdp)) & BD_ENET_RX_EMPTY) == 0) {
-
- curidx = bdp - fep->rx_bd_base;
-
- /*
- * Since we have allocated space to hold a complete frame,
- * the last indicator should be set.
- */
- if ((sc & BD_ENET_RX_LAST) == 0)
- dev_warn(fep->dev, "rcv is not +last\n");
-
- /*
- * Check for errors.
- */
- if (sc & (BD_ENET_RX_LG | BD_ENET_RX_SH | BD_ENET_RX_CL |
- BD_ENET_RX_NO | BD_ENET_RX_CR | BD_ENET_RX_OV)) {
- fep->stats.rx_errors++;
- /* Frame too long or too short. */
- if (sc & (BD_ENET_RX_LG | BD_ENET_RX_SH))
- fep->stats.rx_length_errors++;
- /* Frame alignment */
- if (sc & (BD_ENET_RX_NO | BD_ENET_RX_CL))
- fep->stats.rx_frame_errors++;
- /* CRC Error */
- if (sc & BD_ENET_RX_CR)
- fep->stats.rx_crc_errors++;
- /* FIFO overrun */
- if (sc & BD_ENET_RX_OV)
- fep->stats.rx_crc_errors++;
-
- skb = fep->rx_skbuff[curidx];
-
- dma_unmap_single(fep->dev, CBDR_BUFADDR(bdp),
- L1_CACHE_ALIGN(PKT_MAXBUF_SIZE),
- DMA_FROM_DEVICE);
-
- skbn = skb;
-
- } else {
-
- skb = fep->rx_skbuff[curidx];
-
- dma_unmap_single(fep->dev, CBDR_BUFADDR(bdp),
- L1_CACHE_ALIGN(PKT_MAXBUF_SIZE),
- DMA_FROM_DEVICE);
-
- /*
- * Process the incoming frame.
- */
- fep->stats.rx_packets++;
- pkt_len = CBDR_DATLEN(bdp) - 4; /* remove CRC */
- fep->stats.rx_bytes += pkt_len + 4;
-
- if (pkt_len <= fpi->rx_copybreak) {
- /* +2 to make IP header L1 cache aligned */
- skbn = netdev_alloc_skb(dev, pkt_len + 2);
- if (skbn != NULL) {
- skb_reserve(skbn, 2); /* align IP header */
- skb_copy_from_linear_data(skb,
- skbn->data, pkt_len);
- /* swap */
- skbt = skb;
- skb = skbn;
- skbn = skbt;
- }
- } else {
- skbn = netdev_alloc_skb(dev, ENET_RX_FRSIZE);
-
- if (skbn)
- skb_align(skbn, ENET_RX_ALIGN);
- }
-
- if (skbn != NULL) {
- skb_put(skb, pkt_len); /* Make room */
- skb->protocol = eth_type_trans(skb, dev);
- received++;
- netif_rx(skb);
- } else {
- fep->stats.rx_dropped++;
- skbn = skb;
- }
- }
-
- fep->rx_skbuff[curidx] = skbn;
- CBDW_BUFADDR(bdp, dma_map_single(fep->dev, skbn->data,
- L1_CACHE_ALIGN(PKT_MAXBUF_SIZE),
- DMA_FROM_DEVICE));
- CBDW_DATLEN(bdp, 0);
- CBDW_SC(bdp, (sc & ~BD_ENET_RX_STATS) | BD_ENET_RX_EMPTY);
-
- /*
- * Update BD pointer to next entry.
- */
- if ((sc & BD_ENET_RX_WRAP) == 0)
- bdp++;
- else
- bdp = fep->rx_bd_base;
-
- (*fep->ops->rx_bd_done)(dev);
- }
-
- fep->cur_rx = bdp;
-
- return 0;
-}
-
static void fs_enet_tx(struct net_device *dev)
{
struct fs_enet_private *fep = netdev_priv(dev);
@@ -453,8 +331,7 @@ fs_enet_interrupt(int irq, void *dev_id)
nr++;
int_clr_events = int_events;
- if (fpi->use_napi)
- int_clr_events &= ~fep->ev_napi_rx;
+ int_clr_events &= ~fep->ev_napi_rx;
(*fep->ops->clear_int_events)(dev, int_clr_events);
@@ -462,19 +339,15 @@ fs_enet_interrupt(int irq, void *dev_id)
(*fep->ops->ev_error)(dev, int_events);
if (int_events & fep->ev_rx) {
- if (!fpi->use_napi)
- fs_enet_rx_non_napi(dev);
- else {
- napi_ok = napi_schedule_prep(&fep->napi);
-
- (*fep->ops->napi_disable_rx)(dev);
- (*fep->ops->clear_int_events)(dev, fep->ev_napi_rx);
-
- /* NOTE: it is possible for FCCs in NAPI mode */
- /* to submit a spurious interrupt while in poll */
- if (napi_ok)
- __napi_schedule(&fep->napi);
- }
+ napi_ok = napi_schedule_prep(&fep->napi);
+
+ (*fep->ops->napi_disable_rx)(dev);
+ (*fep->ops->clear_int_events)(dev, fep->ev_napi_rx);
+
+ /* NOTE: it is possible for FCCs in NAPI mode */
+ /* to submit a spurious interrupt while in poll */
+ if (napi_ok)
+ __napi_schedule(&fep->napi);
}
if (int_events & fep->ev_tx)
@@ -811,24 +684,21 @@ static int fs_enet_open(struct net_device *dev)
/* not doing this, will cause a crash in fs_enet_rx_napi */
fs_init_bds(fep->ndev);
- if (fep->fpi->use_napi)
- napi_enable(&fep->napi);
+ napi_enable(&fep->napi);
/* Install our interrupt handler. */
r = request_irq(fep->interrupt, fs_enet_interrupt, IRQF_SHARED,
"fs_enet-mac", dev);
if (r != 0) {
dev_err(fep->dev, "Could not allocate FS_ENET IRQ!");
- if (fep->fpi->use_napi)
- napi_disable(&fep->napi);
+ napi_disable(&fep->napi);
return -EINVAL;
}
err = fs_init_phy(dev);
if (err) {
free_irq(fep->interrupt, dev);
- if (fep->fpi->use_napi)
- napi_disable(&fep->napi);
+ napi_disable(&fep->napi);
return err;
}
phy_start(fep->phydev);
@@ -845,8 +715,7 @@ static int fs_enet_close(struct net_device *dev)
netif_stop_queue(dev);
netif_carrier_off(dev);
- if (fep->fpi->use_napi)
- napi_disable(&fep->napi);
+ napi_disable(&fep->napi);
phy_stop(fep->phydev);
spin_lock_irqsave(&fep->lock, flags);
@@ -1022,7 +891,6 @@ static int fs_enet_probe(struct platform_device *ofdev)
fpi->rx_ring = 32;
fpi->tx_ring = 32;
fpi->rx_copybreak = 240;
- fpi->use_napi = 1;
fpi->napi_weight = 17;
fpi->phy_node = of_parse_phandle(ofdev->dev.of_node, "phy-handle", 0);
if (!fpi->phy_node && of_phy_is_fixed_link(ofdev->dev.of_node)) {
@@ -1102,9 +970,7 @@ static int fs_enet_probe(struct platform_device *ofdev)
ndev->netdev_ops = &fs_enet_netdev_ops;
ndev->watchdog_timeo = 2 * HZ;
- if (fpi->use_napi)
- netif_napi_add(ndev, &fep->napi, fs_enet_rx_napi,
- fpi->napi_weight);
+ netif_napi_add(ndev, &fep->napi, fs_enet_rx_napi, fpi->napi_weight);
ndev->ethtool_ops = &fs_ethtool_ops;
diff --git a/include/linux/fs_enet_pd.h b/include/linux/fs_enet_pd.h
index efb0596..77d783f 100644
--- a/include/linux/fs_enet_pd.h
+++ b/include/linux/fs_enet_pd.h
@@ -139,7 +139,6 @@ struct fs_platform_info {
int rx_ring, tx_ring; /* number of buffers on rx */
__u8 macaddr[ETH_ALEN]; /* mac address */
int rx_copybreak; /* limit we copy small frames */
- int use_napi; /* use NAPI */
int napi_weight; /* NAPI weight */
int use_rmii; /* use RMII mode */
--
2.1.0
^ permalink raw reply related
* [PATCH 0/2] net: fs_enet: Remove non NAPI RX and add NAPI for TX
From: Christophe Leroy @ 2014-10-07 13:04 UTC (permalink / raw)
To: Pantelis Antoniou, Vitaly Bordug; +Cc: linux-kernel, linuxppc-dev, netdev
When using a MPC8xx as a router, 'perf' shows a significant time spent in
fs_enet_interrupt() and fs_enet_start_xmit().
'perf annotate' shows that the time spent in fs_enet_start_xmit is indeed spent
between spin_unlock_irqrestore() and the following instruction, hence in
interrupt handling. This is due to the TX complete interrupt that fires after
each transmitted packet.
This patchset first remove all non NAPI handling as NAPI has become the only
mode for RX, then adds NAPI for handling TX complete.
This improves NAT TCP throughput by 21% on MPC885 with FEC.
Tested on MPC885 with FEC.
[PATCH 1/2] net: fs_enet: Remove non NAPI RX
[PATCH 2/2] net: fs_enet: Add NAPI TX
Signed-off-by: Christophe Leroy <christophe.leroy@c-s.fr>
---
.../net/ethernet/freescale/fs_enet/fs_enet-main.c | 211 ++++++---------------
.../net/ethernet/freescale/fs_enet/fs_enet.h | 9 +-
.../net/ethernet/freescale/fs_enet/mac-fcc.c | 29 +++
.../net/ethernet/freescale/fs_enet/mac-fec.c | 29 +++
.../net/ethernet/freescale/fs_enet/mac-scc.c | 29 +++
linux/include/linux/fs_enet_pd.h | 1 -
6 files changed, 147 insertions(+), 161 deletions(-)
^ permalink raw reply
* Re: Quota in __qdisc_run() (was: qdisc: validate skb without holding lock)
From: Eric Dumazet @ 2014-10-07 12:47 UTC (permalink / raw)
To: Jesper Dangaard Brouer
Cc: David Miller, netdev, therbert, hannes, fw, dborkman, jhs,
alexander.duyck, john.r.fastabend, dave.taht, toke
In-Reply-To: <20141007093441.35ce3a02@redhat.com>
On Tue, 2014-10-07 at 09:34 +0200, Jesper Dangaard Brouer wrote:
> On Fri, 03 Oct 2014 16:30:44 -0700 Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
> > Another problem we need to address is the quota in __qdisc_run()
> > is no longer meaningfull, if each qdisc_restart() can pump many packets.
>
> I fully agree. My earlier "magic" packet limit was covering/pampering
> over this issue.
Although quota was multiplied by 7 or 8 in worst case ?
>
> > An idea would be to use the bstats (or cpu_qstats if applicable)
>
> Please elaborate some more, as I don't completely follow (feel free to
> show with a patch ;-)).
>
I was hoping John could finish the percpu stats before I do that.
Problem with q->bstats.packets is that TSO packets with 45 MSS add 45 to
this counter.
Using a time quota would be better, but : jiffies is too big, and
local_clock() might be too expensive.
^ permalink raw reply
* Re: [PATCH net-next] net/mlx4_en: remove NETDEV_TX_BUSY
From: Eric Dumazet @ 2014-10-07 12:38 UTC (permalink / raw)
To: amirv
Cc: David S. Miller, Eric Dumazet, netdev, Yevgeny Petrilin,
Or Gerlitz, Ido Shamay
In-Reply-To: <5433A48E.1040505@gmail.com>
On Tue, 2014-10-07 at 11:30 +0300, Amir Vadai wrote:
> Reviewed. Also verified that it fixes the deadlock (by sending a large
> burst - larger than ring size). Before this fix, last packet of the
> burst wasn't sent, therefore no doorbell was rang, and the queue was
> stalled.
>
> BTW, another nice optimization that we hope to send soon, is not to arm
> the CQ unless ringing the doorbell.
>
> Acked-by: Amir Vadai <amirv@mellanox.com>
This sounds great, thanks Amir
^ permalink raw reply
* Re: randconfig build error with next-20141001, in drivers/i2c/algos/i2c-algo-bit.c
From: Stephane Grosjean @ 2014-10-07 12:37 UTC (permalink / raw)
To: Oliver Hartkopp
Cc: Randy Dunlap, Jim Davis, Stephen Rothwell, linux-next, linux-i2c,
netdev@vger.kernel.org, linux-can
In-Reply-To: <5433AB31.9090603@hartkopp.net>
Le 07/10/2014 10:58, Oliver Hartkopp a écrit :
> On 10/06/2014 08:09 PM, Randy Dunlap wrote:
>> On 10/06/14 10:39, Oliver Hartkopp wrote:
>>> AFAICS there is 'just' a style problem as 'configs should not enable entire
>>> subsystems'. But it finally is a correct and valid Kconfig, right?
>> Yes, right.
> (..)
>
>> In the unlikely case that I2C is not enabled, the user should have to enable
>> it instead of a solitary driver enabling it. IOW, if a subsystem is disabled,
>> the user probably wanted it that way and a single driver should not override
>> that setting.
> Due to the fact that a change to 'depends on I2C' would make the config option
> invisible (and therefore not selectable) in the case I2C was (unlikely)
> disabled I would finally vote to leave it as-is.
>
> The current Kconfig entry already contains a description that points to the
> requirement to have I2C and I2C_ALGOBIT to be enabled to compile this driver:
>
> config CAN_PEAK_PCIEC
> bool "PEAK PCAN-ExpressCard Cards"
> depends on CAN_PEAK_PCI
> select I2C
> select I2C_ALGOBIT
> default y
> ---help---
> Say Y here if you want to use a PCAN-ExpressCard from PEAK-System
> Technik. This will also automatically select I2C and I2C_ALGO
> configuration options.
>
> AFAIK the PEAK PCAN-ExpressCard is usually used in x86 architecture Laptops,
> so it's near to an academic discussion as x86 usually selects I2C ;-)
>
> @Stephane: When updating the help text to introduce the PCAN-ExpressCard 34
> support anyway you might probably add some more information *why* the I2C
> support is needed (for CAN transceiver settings and status LED).
>
> And /s/I2C_ALGO/I2C_ALGOBIT/ :-)
Ok! (FYI, I had already prepared the help text for introducing the
PCIEC34. I will subst I2C_ALGO as well. I'll prepare the patch asap...)
Regards,
Stéphane
> Tnx & best regards,
> Oliver
--
PEAK-System Technik GmbH
Sitz der Gesellschaft Darmstadt
Handelsregister Darmstadt HRB 9183
Geschaeftsfuehrung: Alexander Gach, Uwe Wilhelm
--
^ permalink raw reply
* Re: [net v2 3/8] net/fsl_pq_mdio: Replace spin_event_timeout() with arch independent
From: Claudiu Manoil @ 2014-10-07 12:16 UTC (permalink / raw)
To: David Laight, netdev@vger.kernel.org
Cc: David S. Miller, Xiubo Li, Shruti Kanetkar, Kim Phillips
In-Reply-To: <063D6719AE5E284EB5DD2968C1650D6D174C5BAC@AcuExch.aculab.com>
On 10/7/2014 11:12 AM, David Laight wrote:
> From: netdev-owner@vger.kernel.
>> spin_event_timeout() is PPC dependent, use an arch independent
>> equivalent instead.
>
> I think you should white a local function/#define that expands to spin_event_timeout()
> on ppc and to the code below you are substituting on other architectures.
>
>> /* Wait for the transaction to finish */
>> - status = spin_event_timeout(!(ioread32be(®s->miimind) &
>> - MIIMIND_BUSY), MII_TIMEOUT, 0);
>> + timeout = MII_TIMEOUT;
>> + while ((ioread32be(®s->miimind) & MIIMIND_BUSY) && timeout) {
>> + cpu_relax();
>> + timeout--;
>> + }
>
> David
>
>
>
Hi David,
This point is debatable. Still better than adding a new local function
would be to provide an implementation for spin_even_timeout() for ARM.
But it is not the place of the ethernet driver to provide an
implementation of spin_event_timeout() on ARM.
Instead, I opted to simplify the busy wait/ timeout mechanism
in the driver using a simple and arch independent implementation
replacing the PPC specific spin_event_timeout(). This timeout
implementation, open-coded as a while() loop, is commonly used by other
drivers too and I think that for this particular driver (fsl_pq_mdio) it
is good enough to do the job, while keeping the code simple and
portable. It may not be as precise as spin_event_timeout() on PPC, but
good enough.
Thanks,
Claudiu
^ permalink raw reply
* Re: [Patch net] net_sched: copy exts->type in tcf_exts_change()
From: Jamal Hadi Salim @ 2014-10-07 11:33 UTC (permalink / raw)
To: Cong Wang, netdev; +Cc: John Fastabend
In-Reply-To: <1412641314-17335-1-git-send-email-xiyou.wangcong@gmail.com>
On 10/06/14 20:21, Cong Wang wrote:
> We need to copy exts->type when committing the change, otherwise
> it would be always 0. This is a quick fix for -net and -stable,
> for net-next tcf_exts will be removed.
>
> Fixes: commit 33be627159913b094bb578e83 ("net_sched: act: use standard struct list_head")
> Reported-by: Jamal Hadi Salim <jhs@mojatatu.com>
> Cc: Jamal Hadi Salim <jhs@mojatatu.com>
> Cc: John Fastabend <john.fastabend@gmail.com>
> Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jamal Hadi Salim <jhs@mojatatu.com>
cheers,
jamal
^ 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