From: Benjamin LaHaise <[EMAIL PROTECTED]>
Date: Fri, 9 Dec 2005 17:53:02 -0500
> Similar to the previous patch, the atomic xchg in sock_error() results in
> extra pipeline flushes due to the need to perform full synchronization as
> a memory barrier. Avoid calling sock_error() in a couple of p
From: Benjamin LaHaise <[EMAIL PROTECTED]>
Date: Fri, 9 Dec 2005 17:31:48 -0500
> AF_UNIX stream socket performance on P4 CPUs tends to suffer due to a lot
> of pipeline flushes from atomic operations. The patch below removes the
> sock_hold() and sock_put() in unix_stream_sendmsg(). This shou
From: jamal <[EMAIL PROTECTED]>
Date: Fri, 09 Dec 2005 16:30:24 -0500
> indeed sounds interesting until you start hitting clones ;->
> so dont run a sniffer or do anything of the sort if you want to see
> some good numbers - otherwise I suspect you will get worse numbers than
> the case of current
Similar to the previous patch, the atomic xchg in sock_error() results in
extra pipeline flushes due to the need to perform full synchronization as
a memory barrier. Avoid calling sock_error() in a couple of places where
it is safe to do a racy version of the test. Combined with the refcountin
AF_UNIX stream socket performance on P4 CPUs tends to suffer due to a lot
of pipeline flushes from atomic operations. The patch below removes the
sock_hold() and sock_put() in unix_stream_sendmsg(). This should be safe
as the socket still holds a reference to its peer which is only released
a
On Thu, 2005-08-12 at 21:25 +0100, Robert Olsson wrote:
> David S. Miller writes:
>
>
> > BTW, this is all related to SKB recycling.
> >
> > For example, if this is just a TCP ACK, we can do better than
> > copybreak and just let the driver use the SKB again upon
> > return from netif_recei
On Thu, 2005-08-12 at 17:57 +0100, Eric Dumazet wrote:
> > right, after i did this code, i realized that, and it is demonstrable
> > that #4 hurts, if only a little.
> > I'm sticking with my suggestion we go to #1,#2,#5
>
> I would try another thing : #1,#2,#4bis
>
> #4bis prefetch(&nex
On Fri, Dec 09, 2005 at 09:08:47PM +0100, Francois Romieu wrote:
> John W. Linville <[EMAIL PROTECTED]> :
> [...]
> > __prism2_translate_scan uses kmalloc to dynamically allocates a buffer
> > for holding some textual diagnostic output, and it does so while holding a
> > spinlock. With CONFIG_DEBU
John W. Linville <[EMAIL PROTECTED]> :
[...]
> __prism2_translate_scan uses kmalloc to dynamically allocates a buffer
> for holding some textual diagnostic output, and it does so while holding a
> spinlock. With CONFIG_DEBUG_SPINLOCK_SLEEP defined, this causes debug
> messages to appear in /var/lo
Adjust interrupt deferral values for better performance and avoid
starvation issues under stress.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -2115,9 +2115,8 @@ static int sky2_reset(struct sky2_hw *hw
Put more logic in to try and handle transmit timeout recovery.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -661,7 +661,7 @@ static void sky2_qset(struct sky2_hw *hw
/* Setup prefetch unit registers. This is t
Make the copy threshold a module parameter for those who
may want to turn it off.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -76,7 +76,6 @@
#define RX_LE_BYTES(RX_LE_SIZE*sizeof(struct sky2_rx_le)
Do phy processing in a work queue rather than a tasklet.
This means we can let bottom halves run.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -43,6 +43,7 @@
#include
#include
#include
+#include
#includ
Split Tx and Rx ring into two different data structures.
Tx needs the next value (to handle partial status), and
Rx always needs the mapaddr (to handle resubmitting same buffer).
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/
Some transmit side small fixes:
* When computing number of list elements per transmit, do full
comparision to check for checksuming.
* Get rid of racy check for tx_complete
* Change stop test to match wake condition.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig
Optimize poll routine by not clearing interrupt till after
processing, and checking for more work before re-enable.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -1736,7 +1736,6 @@ static int sky2_poll(struct ne
Hand optimize transmit completion routine.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -1227,6 +1227,8 @@ out_unlock:
static void sky2_tx_complete(struct sky2_port *sky2, u16 done)
{
struct net_device
Add support for get/set the interrupt coalescing settings.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-2.6.orig/drivers/net/sky2.c
+++ sky2-2.6/drivers/net/sky2.c
@@ -24,9 +24,6 @@
*/
/*
- * TODO
- * - coalescing setting?
- *
* TOTEST
* - speed setting
*
Don't print ring full message if we lose race.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -1081,11 +1081,16 @@ static int sky2_xmit_frame(struct sk_buf
return NETDEV_TX_LOCKED;
if (un
Faster way to compute ring distance.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -1039,7 +1039,7 @@ err_out:
/* Modular subtraction in ring */
static inline int tx_dist(unsigned tail, unsigned head)
{
-
Turn symmetric flow control on by default. This was an oversight
in the initial versions.
Signed-off-by: Stephen HEmminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -2921,7 +2921,7 @@ static __devinit struct net_device *sky2
spin_loc
Change transmit complete handling to use the status values
in the poll list because they are in-cache, rather than reading
non-cached memory for chips status.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -1274,
Update to sky2 driver incorporating a number of bug fixes for
stress test failures and other small optimizations.
Includes new ethtool interrupt coalescing support.
--
Stephen Hemminger <[EMAIL PROTECTED]>
OSDL http://developer.osdl.org/~shemminger
-
To unsubscribe from this list: send the line "
Add a couple more prefetches to where we walk the rings.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -45,6 +45,7 @@
#include
#include
#include
+#include
#include
#include
@@ -1248,6 +1249,7 @@ sta
There maybe new versions of Yukon2 in the future, so make the
driver more robust to handle this.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-2.6.orig/drivers/net/sky2.c
+++ sky2-2.6/drivers/net/sky2.c
@@ -128,17 +128,15 @@ MODULE_DEVICE_TABLE(pci, sky2_id_table);
static const
Change version number. Still don't have enough history on this
to call it 1.0 yet.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -57,7 +57,7 @@
#include "sky2.h"
#define DRV_NAME "sky2"
-#defin
Don't need to keep track of mapping length in ring structure
because we can get the same info from other info.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -729,23 +729,23 @@ static inline u32 high32(dma_addr_t
It is a useful optimization to rearrange the structure
slightly to align receive and transmit portions.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-2.6.orig/drivers/net/sky2.h
+++ sky2-2.6/drivers/net/sky2.h
@@ -1790,17 +1790,17 @@ struct sky2_port {
unsigned
Don't need to disable interrupts during the transmit routine.
Signed-off-by: Stephen Hemminger <[EMAIL PROTECTED]>
--- sky2-netdev.orig/drivers/net/sky2.c
+++ sky2-netdev/drivers/net/sky2.c
@@ -868,15 +868,14 @@ static void sky2_vlan_rx_register(struct
struct sky2_port *sky2 = netdev_pri
On Fri, Dec 09, 2005 at 07:12:10PM +0100, Olaf Hering wrote:
>
> The CSR contains garbage after a coldboot on RS/6000.
> One some systems (like my 44p 270) the MAC address is all FF,
> on others (like my B50) it is ff:ff:ff:fd:ff:6b.
>
> It can eventually be fixed by loading pcnet32, set the inte
The CSR contains garbage after a coldboot on RS/6000.
One some systems (like my 44p 270) the MAC address is all FF,
on others (like my B50) it is ff:ff:ff:fd:ff:6b.
It can eventually be fixed by loading pcnet32, set the interface
into the UP state, rmmod pcnet32 and load it again. But this worked
From: John W. Linville <[EMAIL PROTECTED]>
__prism2_translate_scan uses kmalloc to dynamically allocates a buffer
for holding some textual diagnostic output, and it does so while holding a
spinlock. With CONFIG_DEBUG_SPINLOCK_SLEEP defined, this causes debug
messages to appear in /var/log/message
From: John W. Linville <[EMAIL PROTECTED]>
At least some versions of the via-velocity hardware only support
checksumming IPv4 frames in hardware. However, the driver is currently
setting the NETIF_F_HW_CSUM flag, which indicates support for more than
just IPv4. This results in errors when trying
Hello David,
David S. Miller wrote:
From: Kazunori Miyazawa <[EMAIL PROTECTED]>
Date: Fri, 09 Dec 2005 13:20:14 +0900
I send a patch to fix pmtu calculation of IPv6 ESP.
It is a simple bug which uses the wrong member.
This bug does not seriously affect ordinary use of IPsec.
But it is import
On Tue, Dec 06, 2005 at 10:56:19PM +0100, Francois Romieu wrote:
> netif_rx_schedule_prep return netif_running(dev) &&
> dev_close clear_bit(__LINK_STATE_START, &dev->state);
Oh, of course - thanks for bearing wth me. Will fix that too and
resubmit.
--
"You grabbed my hand and we
In article <[EMAIL PROTECTED]> (at Fri, 9 Dec 2005 12:02:03 +0200 (EET)), Pekka
Savola <[EMAIL PROTECTED]> says:
> How do you differentiate between these two cases? One is valid while
> the other is not, but both seem valid to me.
>
> default via fe80::20a:5eff:fe51:7d58 dev eth1 proto kernel
In article <[EMAIL PROTECTED]> (at Fri, 9 Dec 2005 11:45:04 +0200 (EET)), Pekka
Savola <[EMAIL PROTECTED]> says:
> 1) after reboot (card in slot), default route shows up, but does not
> get renewed. After timer expiry, the expiry timer seems to wrap
> around, and after a couple of seconds, the
Hi,
This issue has been bugging me on and off for a while, so I thought to
write it up and test it a bit more extensively.
I'm running FC4 on a Thinkpad laptop and an external old Orinoco WLAN
card. I have observed four, probably somewhat related phenomena:
A) when router lifetime is abou
38 matches
Mail list logo