Bug 15014 - Poor performance on O2 Micro carbus bridge (OZ6933)
Summary: Poor performance on O2 Micro carbus bridge (OZ6933)
Status: CLOSED CODE_FIX
Alias: None
Product: Drivers
Classification: Unclassified
Component: PCMCIA (show other bugs)
Hardware: All Linux
: P1 normal
Assignee: Dominik Brodowski
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2010-01-09 14:09 UTC by frodone
Modified: 2016-01-20 23:35 UTC (History)
2 users (show)

See Also:
Kernel Version: 2.6.32
Subsystem:
Regression: No
Bisected commit-id:


Attachments
lspci -vv (7.93 KB, text/plain)
2010-01-09 14:09 UTC, frodone
Details

Description frodone 2010-01-09 14:09:33 UTC
Created attachment 24493 [details]
lspci -vv

Since kernel 2.6.32 i noticed degraded performance on network file transfers from my 8139too pcmcia card (3MB/sec instead of 11MB/sec). I bisected till this:

1ff84890b62b20823b3697a6041bbec1b5280cee is the first bad commit
commit 1ff84890b62b20823b3697a6041bbec1b5280cee
Author: Tomas Kovacik <nail@nodomain.sk>
Date:   Sun Jul 26 22:04:58 2009 +0200

    pcmcia: disable prefetch/burst for OZ6933

    Problems have been reported [1], so disable prefetch/burst, to be on the safe
    side.

    [1] http://www.mail-archive.com/linux-pcmcia@lists.infradead.org/msg02048.html

    Signed-off-by: Tomáš KováÄik <nail@nodomain.sk>
    Signed-off-by: Wolfram Sang <w.sang@pengutronix.de>
    Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

:040000 040000 bc4841745f3cba3e1cf1c952d4ce987572fa40f8 8b5f282e917ce1b11b499f81ad219a71487faccd M      drivers


I agree with the "be on the safe side" statement, but...
Bye
Comment 1 Wolfram Sang 2010-01-09 15:01:21 UTC
> I agree with the "be on the safe side" statement, but...

:)

Hmm, I'm thinking about a module parameter which can force this on/off while
the default behaviour is the currently implemented one.

Regards,

   Wolfram

-- 
Pengutronix e.K.                           | Wolfram Sang                |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Comment 2 Wolfram Sang 2010-01-09 16:29:18 UTC
O2-bridges can do read prefetch and write burst. However, for some combinations
of older bridges and cards, this causes problems, so it is disabled for those
bridges. Now, as some users know their setup works with the speedups enabled, a
new parameter is introduced to the driver. Now, a user can specifically enable
or disable these features, while the default is what we have today: detect the
bridge and decide accordingly.

Simplify and unify the printouts while we are here.

Signed-off-by: Wolfram Sang <w.sang@pengutronix.de>
Cc: frodone@gmail.com
Cc: Dominik Brodowski <linux@dominikbrodowski.net>
---

frodone: Can you please test this patch? It is just compile-tested as I don't
have access to the hardware I'd need this weekend.

 drivers/pcmcia/o2micro.h      |   38 +++++++++++++++++++++++++-------------
 drivers/pcmcia/yenta_socket.c |    5 +++++
 2 files changed, 30 insertions(+), 13 deletions(-)

diff --git a/drivers/pcmcia/o2micro.h b/drivers/pcmcia/o2micro.h
index 624442f..d98cc01 100644
--- a/drivers/pcmcia/o2micro.h
+++ b/drivers/pcmcia/o2micro.h
@@ -116,13 +116,12 @@ static int o2micro_override(struct yenta_socket *socket)
 	 * from Eric Still, 02Micro.
 	 */
 	u8 a, b;
+	bool use_speedup;
 
 	if (PCI_FUNC(socket->dev->devfn) == 0) {
 		a = config_readb(socket, O2_RESERVED1);
 		b = config_readb(socket, O2_RESERVED2);
-
-		dev_printk(KERN_INFO, &socket->dev->dev,
-			   "O2: res at 0x94/0xD4: %02x/%02x\n", a, b);
+		dev_dbg(&socket->dev->dev, "O2: 0x94/0xD4: %02x/%02x\n", a, b);
 
 		switch (socket->dev->device) {
 		/*
@@ -136,22 +135,35 @@ static int o2micro_override(struct yenta_socket *socket)
 		case PCI_DEVICE_ID_O2_6832:
 		case PCI_DEVICE_ID_O2_6836:
  		case PCI_DEVICE_ID_O2_6933:
-			dev_printk(KERN_INFO, &socket->dev->dev,
-				   "Yenta O2: old bridge, disabling read "
-				   "prefetch/write burst\n");
-			config_writeb(socket, O2_RESERVED1,
-			              a & ~(O2_RES_READ_PREFETCH | O2_RES_WRITE_BURST));
-			config_writeb(socket, O2_RESERVED2,
-			              b & ~(O2_RES_READ_PREFETCH | O2_RES_WRITE_BURST));
+			use_speedup = false;
 			break;
-
 		default:
-			dev_printk(KERN_INFO , &socket->dev->dev,
-				   "O2: enabling read prefetch/write burst\n");
+			use_speedup = true;
+			break;
+		}
+
+		if (strcasecmp(o2_speedup, "on") == 0)
+			use_speedup = true;
+		else if (strcasecmp(o2_speedup, "off") == 0)
+			use_speedup = false;
+		else if (strcasecmp(o2_speedup, "default") != 0)
+			dev_warn(&socket->dev->dev,
+				"O2: Unknown parameter, using 'default'");
+
+		if (use_speedup) {
+			dev_info(&socket->dev->dev,
+				"O2: enabling read prefetch/write burst\n");
 			config_writeb(socket, O2_RESERVED1,
 			              a | O2_RES_READ_PREFETCH | O2_RES_WRITE_BURST);
 			config_writeb(socket, O2_RESERVED2,
 			              b | O2_RES_READ_PREFETCH | O2_RES_WRITE_BURST);
+		} else {
+			dev_info(&socket->dev->dev,
+				"O2: disabling read prefetch/write burst\n");
+			config_writeb(socket, O2_RESERVED1,
+			              a & ~(O2_RES_READ_PREFETCH | O2_RES_WRITE_BURST));
+			config_writeb(socket, O2_RESERVED2,
+			              b & ~(O2_RES_READ_PREFETCH | O2_RES_WRITE_BURST));
 		}
 	}
 
diff --git a/drivers/pcmcia/yenta_socket.c b/drivers/pcmcia/yenta_socket.c
index e4d12ac..041a75a 100644
--- a/drivers/pcmcia/yenta_socket.c
+++ b/drivers/pcmcia/yenta_socket.c
@@ -37,6 +37,11 @@ static int pwr_irqs_off;
 module_param(pwr_irqs_off, bool, 0644);
 MODULE_PARM_DESC(pwr_irqs_off, "Force IRQs off during power-on of slot. Use only when seeing IRQ storms!");
 
+static char o2_speedup[] = "default";
+module_param_string(o2_speedup, o2_speedup, sizeof(o2_speedup), 0444);
+MODULE_PARM_DESC(o2_speedup, "Use prefetch/burst for O2-bridges: 'on', 'off' "
+	"or 'default' (uses recommended behaviour for the detected bridge)");
+
 #define debug(x, s, args...) dev_dbg(&s->dev->dev, x, ##args)
 
 /* Don't ask.. */
Comment 3 frodone 2010-01-09 22:15:06 UTC
Works as expected, thank you very much. Quick and clean! :)
Comment 4 Wolfram Sang 2010-09-22 21:46:10 UTC
Commited as 35169529093be3bbef70afd3c4125e35cece7e03. Can be closed now.

Note You need to log in before you can comment on or make changes to this bug.