[Patch 0/4] Accelerated network drivers

[Patch 0/4] Accelerated network drivers

[Kieran Mansley]

This patch set provides the standard net driver and accelerated network
driver plug-ins for Solarflare range of 10G network cards.  The
accelerated network plug-ins make use of the modifications contributed
to netfront and netback last year.  

The four patches are as follows:

Patch 1: standard Linux net driver (provides module "sfc")
Patch 2: resource manager driver (provides module "sfc_resource")
Patch 3: netfront plug-in module (provides module "sfc_netfront" and
"sfc_netutil")
Patch 4: netback plug-in module (provides module "sfc_netback")

Signed-off-by: Kieran Mansley <kmansley@solarflare.com>

Thanks,

Kieran

Re: [Patch 0/4] Accelerated network drivers

[Keir Fraser]

On 15/2/08 08:47, "Kieran Mansley" <kmansley@solarflare.com> wrote:

> This patch set provides the standard net driver and accelerated network
> driver plug-ins for Solarflare range of 10G network cards.  The
> accelerated network plug-ins make use of the modifications contributed
> to netfront and netback last year.

Patches 1 and 2 haven't materialised. Are they very large?

 -- Keir

Re: [Patch 0/4] Accelerated network drivers

[Kieran Mansley]

On Fri, 2008-02-15 at 09:47 +0000, Keir Fraser wrote:
> On 15/2/08 08:47, "Kieran Mansley" <kmansley@solarflare.com> wrote:
> 
> > This patch set provides the standard net driver and accelerated network
> > driver plug-ins for Solarflare range of 10G network cards.  The
> > accelerated network plug-ins make use of the modifications contributed
> > to netfront and netback last year.
> 
> Patches 1 and 2 haven't materialised. Are they very large?

Yes, I'd assumed they're queued for list moderator approval.

Kieran

Re: [Patch 0/4] Accelerated network drivers

[Kieran Mansley]

On Fri, 2008-02-15 at 08:47 +0000, Kieran Mansley wrote:
> This patch set provides the standard net driver and accelerated network
> driver plug-ins for Solarflare range of 10G network cards.  The
> accelerated network plug-ins make use of the modifications contributed
> to netfront and netback last year.  
> 
> The four patches are as follows:
> 
> Patch 1: standard Linux net driver (provides module "sfc")
> Patch 2: resource manager driver (provides module "sfc_resource")
> Patch 3: netfront plug-in module (provides module "sfc_netfront" and
> "sfc_netutil")
> Patch 4: netback plug-in module (provides module "sfc_netback")

As the first two patches haven't made it to the list (probably above the
mailing list's configured attachment size) I've put a tar of all four
patches up here:
 
https://support.solarflare.com/xen/sfc_patches.tar.gz
https://support.solarflare.com/xen/MD5SUMS

Kieran

Re: [Patch 0/4] Accelerated network drivers

[Alex Williamson]

On Mon, 2008-02-18 at 10:17 +0000, Kieran Mansley wrote:
> On Fri, 2008-02-15 at 08:47 +0000, Kieran Mansley wrote:
> > This patch set provides the standard net driver and accelerated network
> > driver plug-ins for Solarflare range of 10G network cards.  The
> > accelerated network plug-ins make use of the modifications contributed
> > to netfront and netback last year.  
> > 
> > The four patches are as follows:
> > 
> > Patch 1: standard Linux net driver (provides module "sfc")
> > Patch 2: resource manager driver (provides module "sfc_resource")
> > Patch 3: netfront plug-in module (provides module "sfc_netfront" and
> > "sfc_netutil")
> > Patch 4: netback plug-in module (provides module "sfc_netback")

Hi Kieran,

   I attempted to build this on ia64 and ran into several issues:

drivers/net/sfc/txc43128_phy.c:txc43128_phy_check_hw()
	doesn't build w/o CONFIG_SRC_DEBUGFS - not ia64 specific

drivers/net/sfc/sfc_resource/falcon.c
	has warnings using %llx for uint64_t

drivers/xen/sfc_netback/accel_solarflare.c:get_victim_filter()
	uses x86 specific rdtscl(), apparently for random number

drivers/xen/sfc_netfront/sysdep.h
	ef_vi_dma_addr_t undefined for ia64 w/ GNUC (the default)

sfc_netfront and sfc_netback apparently both define max_pages and
debugfs_create_u64(), making it impossible to build them both
statically.

Thanks,

	Alex

-- 
Alex Williamson                             HP Open Source & Linux Org.

Re: [Patch 0/4] Accelerated network drivers

[Kieran Mansley]

[-- Attachment #1: Type: text/plain, Size: 2011 bytes --]

On Tue, 2008-02-19 at 07:00 -0700, Alex Williamson wrote:
> On Mon, 2008-02-18 at 10:17 +0000, Kieran Mansley wrote:
> > On Fri, 2008-02-15 at 08:47 +0000, Kieran Mansley wrote:
> > > This patch set provides the standard net driver and accelerated network
> > > driver plug-ins for Solarflare range of 10G network cards.  The
> > > accelerated network plug-ins make use of the modifications contributed
> > > to netfront and netback last year.  
> > > 
> > > The four patches are as follows:
> > > 
> > > Patch 1: standard Linux net driver (provides module "sfc")
> > > Patch 2: resource manager driver (provides module "sfc_resource")
> > > Patch 3: netfront plug-in module (provides module "sfc_netfront" and
> > > "sfc_netutil")
> > > Patch 4: netback plug-in module (provides module "sfc_netback")
> 
> Hi Kieran,
> 
>    I attempted to build this on ia64 and ran into several issues:

Apologies for the problems you've seen.  You've clearly hit something of
a sweet spot (or should that be sour-spot?) for compiling in ways I
don't do as routinely as I should.

I've attached a set of patches that should resolve all the problems you
describe.  Unfortunately until I can test more thoroughly on our IA64
hardware I've marked the acceleration modules as requiring CONFIG_X86,
so this will prevent these modules from being used on IA64 systems for
the time being. 

The patches are as follows:

sfc_net_static_compile
 - fix the namespace clash problem from the module paramenters when
compiling the plugins statically into the kernel.

debugfs_create_u64
 - move the debugfs_create_u64 which had duplicate implementations in
each plugin into the standard Linux debugfs files.

sfc_debugfs_off_compile
 - Fix compile error when not selecting CONFIG_SFC_DEBUGFS

sfc_config_x86_dependency
 - Specify module "depends on" to more accurately describe what the real
dependencies are.
  
IA64_PRIx64
 - fix the definition of PRIx64 in the sfc modules to be properly
defined on IA64 architectures.

Thanks 

Kieran

[-- Attachment #2: debugfs_create_u64 --]
[-- Type: text/x-patch, Size: 4661 bytes --]

diff -r 7693635f3e97 drivers/xen/sfc_netback/accel_debugfs.c
--- a/drivers/xen/sfc_netback/accel_debugfs.c
+++ b/drivers/xen/sfc_netback/accel_debugfs.c
@@ -37,28 +37,6 @@ static struct netback_accel_global_dbfs 
 static struct netback_accel_global_dbfs  global_dbfs;
 #endif
 #endif
-
-/*
- * Extend debugfs helper functions to have a u64 version
- */ 
-static void debugfs_u64_set(void *data, u64 val)
-{
-  *(u64 *)data = val;
-}
-
-static u64 debugfs_u64_get(void *data)
-{
-  return *(u64 *)data;
-}
-
-DEFINE_SIMPLE_ATTRIBUTE(fops_u64, debugfs_u64_get, debugfs_u64_set, "%llu\n");
-
-struct dentry *debugfs_create_u64(const char *name, mode_t mode,
-                                  struct dentry *parent, u64 *value)
-{
-  return debugfs_create_file(name, mode, parent, value, &fops_u64);
-}
-
 
 void netback_accel_debugfs_init(void) 
 {
diff -r 7693635f3e97 drivers/xen/sfc_netfront/accel_debugfs.c
--- a/drivers/xen/sfc_netfront/accel_debugfs.c
+++ b/drivers/xen/sfc_netfront/accel_debugfs.c
@@ -30,29 +30,6 @@
 #if defined(CONFIG_DEBUG_FS)
 static struct dentry *sfc_debugfs_root = NULL;
 #endif
-
-
-/*
- * Extend debugfs helper functions to have a u64 version
- */
-static void debugfs_u64_set(void *data, u64 val)
-{
-  *(u64 *)data = val;
-}
-
-static u64 debugfs_u64_get(void *data)
-{
-  return *(u64 *)data;
-}
-
-DEFINE_SIMPLE_ATTRIBUTE(fops_u64, debugfs_u64_get, debugfs_u64_set, "%llu\n");
-
-struct dentry *debugfs_create_u64(const char *name, mode_t mode,
-				  struct dentry *parent, u64 *value)
-{
-  return debugfs_create_file(name, mode, parent, value, &fops_u64);
-}
-
 
 void netfront_accel_debugfs_init(void) 
 {
diff -r 7693635f3e97 fs/debugfs/file.c
--- a/fs/debugfs/file.c
+++ b/fs/debugfs/file.c
@@ -170,6 +170,51 @@ struct dentry *debugfs_create_u32(const 
 }
 EXPORT_SYMBOL_GPL(debugfs_create_u32);
 
+
+static void debugfs_u64_set(void *data, u64 val)
+{
+	*(u64 *)data = val;
+}
+
+static u64 debugfs_u64_get(void *data)
+{
+	return *(u64 *)data;
+}
+DEFINE_SIMPLE_ATTRIBUTE(fops_u64, debugfs_u64_get, debugfs_u64_set, "%llu\n");
+
+/**
+ * debugfs_create_u64 - create a file in the debugfs filesystem that is used to read and write an unsigned 64 bit value.
+ *
+ * @name: a pointer to a string containing the name of the file to create.
+ * @mode: the permission that the file should have
+ * @parent: a pointer to the parent dentry for this file.  This should be a
+ *          directory dentry if set.  If this paramater is NULL, then the
+ *          file will be created in the root of the debugfs filesystem.
+ * @value: a pointer to the variable that the file should read to and write
+ *         from.
+ *
+ * This function creates a file in debugfs with the given name that
+ * contains the value of the variable @value.  If the @mode variable is so
+ * set, it can be read from, and written to.
+ *
+ * This function will return a pointer to a dentry if it succeeds.  This
+ * pointer must be passed to the debugfs_remove() function when the file is
+ * to be removed (no automatic cleanup happens if your module is unloaded,
+ * you are responsible here.)  If an error occurs, NULL will be returned.
+ *
+ * If debugfs is not enabled in the kernel, the value -ENODEV will be
+ * returned.  It is not wise to check for this value, but rather, check for
+ * NULL or !NULL instead as to eliminate the need for #ifdef in the calling
+ * code.
+ */
+struct dentry *debugfs_create_u64(const char *name, mode_t mode,
+				  struct dentry *parent, u64 *value)
+{
+	return debugfs_create_file(name, mode, parent, value, &fops_u64);
+}
+EXPORT_SYMBOL_GPL(debugfs_create_u64);
+
+
 static ssize_t read_file_bool(struct file *file, char __user *user_buf,
 			      size_t count, loff_t *ppos)
 {
diff -r 7693635f3e97 include/linux/debugfs.h
--- a/include/linux/debugfs.h
+++ b/include/linux/debugfs.h
@@ -41,6 +41,8 @@ struct dentry *debugfs_create_u16(const 
 				  struct dentry *parent, u16 *value);
 struct dentry *debugfs_create_u32(const char *name, mode_t mode,
 				  struct dentry *parent, u32 *value);
+struct dentry *debugfs_create_u64(const char *name, mode_t mode,
+				  struct dentry *parent, u64 *value);
 struct dentry *debugfs_create_bool(const char *name, mode_t mode,
 				  struct dentry *parent, u32 *value);
 
@@ -94,6 +96,13 @@ static inline struct dentry *debugfs_cre
 	return ERR_PTR(-ENODEV);
 }
 
+static inline struct dentry *debugfs_create_u64(const char *name, mode_t mode,
+						struct dentry *parent,
+						u64 *value)
+{
+	return ERR_PTR(-ENODEV);
+}
+
 static inline struct dentry *debugfs_create_bool(const char *name, mode_t mode,
 						 struct dentry *parent,
 						 u32 *value)

[-- Attachment #3: IA64_PRIx64 --]
[-- Type: text/x-patch, Size: 451 bytes --]

diff -r d70b5b49d7a7 drivers/net/sfc/sfc_resource/ci/efhw/common_sysdep.h
--- a/drivers/net/sfc/sfc_resource/ci/efhw/common_sysdep.h
+++ b/drivers/net/sfc/sfc_resource/ci/efhw/common_sysdep.h
@@ -52,8 +52,12 @@
 
 /* Linux kernel also does not provide PRIx32... Sigh. */
 #define PRIx32 "x"
-#define PRIx64 "llx"
 
+#ifdef __ia64__
+# define PRIx64 "lx"
+#else
+# define PRIx64 "llx"
+#endif
 
 #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,19)
 enum {

[-- Attachment #4: sfc_config_x86_dependency --]
[-- Type: text/x-patch, Size: 1178 bytes --]

diff -r dcbff25269d2 drivers/net/sfc/Kconfig
--- a/drivers/net/sfc/Kconfig
+++ b/drivers/net/sfc/Kconfig
@@ -28,7 +28,7 @@ config SFC_MTD
 	  new boot ROM to the NIC.
 
 config SFC_RESOURCE
-	depends on SFC
+	depends on SFC && X86
 	tristate "Solarflare Solarstorm SFC4000 resource driver"
 	help
 	  This module provides the SFC resource manager driver.
diff -r dcbff25269d2 drivers/xen/Kconfig
--- a/drivers/xen/Kconfig
+++ b/drivers/xen/Kconfig
@@ -81,11 +81,12 @@ config XEN_NETDEV_PIPELINED_TRANSMITTER
 
 config XEN_NETDEV_ACCEL_SFC_UTIL
         tristate
+        depends on X86
         default n
 
 config XEN_NETDEV_ACCEL_SFC_BACKEND
 	tristate "Network-device backend driver acceleration for Solarflare NICs"
-        depends on XEN_NETDEV_BACKEND
+        depends on XEN_NETDEV_BACKEND && SFC && SFC_RESOURCE && X86
         select XEN_NETDEV_ACCEL_SFC_UTIL
         default m
 
@@ -194,7 +195,7 @@ config XEN_GRANT_DEV
 
 config XEN_NETDEV_ACCEL_SFC_FRONTEND
 	tristate "Network-device frontend driver acceleration for Solarflare NICs"
-	depends on XEN_NETDEV_FRONTEND
+	depends on XEN_NETDEV_FRONTEND && X86
         select XEN_NETDEV_ACCEL_SFC_UTIL
 	default m
 

[-- Attachment #5: sfc_debugfs_off_compile --]
[-- Type: text/x-patch, Size: 878 bytes --]

diff -r 480bb0e62d4c drivers/net/sfc/tenxpress.c
--- a/drivers/net/sfc/tenxpress.c
+++ b/drivers/net/sfc/tenxpress.c
@@ -391,9 +391,11 @@ static int tenxpress_phy_init(struct efx
 	if (rc < 0)
 		goto fail;
 
+#ifdef CONFIG_SFC_DEBUGFS
 	rc = tenxpress_debugfs_init(efx);
 	if (rc < 0)
 		goto fail;
+#endif
 
 	schedule_timeout_uninterruptible(HZ / 5); /* 200ms */
 
diff -r 480bb0e62d4c drivers/net/sfc/txc43128_phy.c
--- a/drivers/net/sfc/txc43128_phy.c
+++ b/drivers/net/sfc/txc43128_phy.c
@@ -644,8 +644,8 @@ static void txc43128_phy_fini(struct efx
  * don't use LASI interrupts. Also update the BER counters and poll the lm87 */
 static int txc43128_phy_check_hw(struct efx_nic *efx)
 {
+	struct txc43128_data *data = efx->phy_data;
 #ifdef CONFIG_SFC_DEBUGFS
-	struct txc43128_data *data = efx->phy_data;
 	int phy = efx->mii.phy_id;
 	int timer, count, i, mmd;
 #endif

[-- Attachment #6: sfc_net_static_compile --]
[-- Type: text/x-patch, Size: 4210 bytes --]

diff -r b851951738f4 drivers/xen/sfc_netback/accel.c
--- a/drivers/xen/sfc_netback/accel.c
+++ b/drivers/xen/sfc_netback/accel.c
@@ -71,8 +71,8 @@ static struct notifier_block netback_acc
 };
 
 
-unsigned max_pages = NETBACK_ACCEL_DEFAULT_MAX_BUF_PAGES;
-module_param(max_pages, int, 0666);
+unsigned sfc_netback_max_pages = NETBACK_ACCEL_DEFAULT_MAX_BUF_PAGES;
+module_param_named(max_pages, sfc_netback_max_pages, uint, 0666);
 MODULE_PARM_DESC(max_pages, 
 		 "The number of buffer pages to enforce on each guest");
 
diff -r b851951738f4 drivers/xen/sfc_netback/accel.h
--- a/drivers/xen/sfc_netback/accel.h
+++ b/drivers/xen/sfc_netback/accel.h
@@ -49,7 +49,7 @@
 #define NETBACK_ACCEL_DEFAULT_MAX_MCASTS (8)
 #define NETBACK_ACCEL_DEFAULT_MAX_BUF_PAGES (384)
 /* Variable to store module parameter for max_buf_pages */
-extern unsigned max_pages;
+extern unsigned sfc_netback_max_pages;
 
 #define NETBACK_ACCEL_STATS 1
 
diff -r b851951738f4 drivers/xen/sfc_netback/accel_xenbus.c
--- a/drivers/xen/sfc_netback/accel_xenbus.c
+++ b/drivers/xen/sfc_netback/accel_xenbus.c
@@ -113,7 +113,7 @@ void cfg_hw_quotas(struct xenbus_device 
 		 */
 		DPRINTK("Failed to read quotas from xenbus, using defaults\n");
 		bend->quotas.max_filters = NETBACK_ACCEL_DEFAULT_MAX_FILTERS;
-		bend->quotas.max_buf_pages = max_pages;
+		bend->quotas.max_buf_pages = sfc_netback_max_pages;
 		bend->quotas.max_mcasts = NETBACK_ACCEL_DEFAULT_MAX_MCASTS;
 	}
 
diff -r b851951738f4 drivers/xen/sfc_netfront/accel.h
--- a/drivers/xen/sfc_netfront/accel.h
+++ b/drivers/xen/sfc_netfront/accel.h
@@ -336,8 +336,8 @@ typedef struct netfront_accel_vnic {
 
 
 /* Module parameters */
-extern unsigned max_pages;
-extern unsigned buffer_split;
+extern unsigned sfc_netfront_max_pages;
+extern unsigned sfc_netfront_buffer_split;
 
 extern const char *frontend_name;
 extern struct netfront_accel_hooks accel_hooks;
diff -r b851951738f4 drivers/xen/sfc_netfront/accel_bufs.c
--- a/drivers/xen/sfc_netfront/accel_bufs.c
+++ b/drivers/xen/sfc_netfront/accel_bufs.c
@@ -72,12 +72,12 @@ int netfront_accel_alloc_buffer_mem(stru
 	int n, rc;
 
 	if ((rc = netfront_accel_alloc_buf_desc_blocks
-	     (rx_manager, pages - (pages / buffer_split))) < 0) {
+	     (rx_manager, pages - (pages / sfc_netfront_buffer_split))) < 0) {
 		goto rx_fail;
 	}
 
 	if ((rc = netfront_accel_alloc_buf_desc_blocks
-	     (tx_manager, pages / buffer_split)) < 0) {
+	     (tx_manager, pages / sfc_netfront_buffer_split)) < 0) {
 		goto tx_fail;
 	}
 
diff -r b851951738f4 drivers/xen/sfc_netfront/accel_msg.c
--- a/drivers/xen/sfc_netfront/accel_msg.c
+++ b/drivers/xen/sfc_netfront/accel_msg.c
@@ -157,7 +157,7 @@ static int vnic_add_bufs(netfront_accel_
 	offset = msg->u.mapbufs.reqid;
 
 	if (offset < vnic->bufpages.max_pages - 
-	    (vnic->bufpages.max_pages / buffer_split)) {
+	    (vnic->bufpages.max_pages / sfc_netfront_buffer_split)) {
 		bufinfo = vnic->rx_bufs;
 	} else
 		bufinfo = vnic->tx_bufs;
@@ -286,7 +286,7 @@ static int vnic_process_hello_msg(netfro
 				  struct net_accel_msg *msg)
 {
 	int err = 0;
-	unsigned pages = max_pages;
+	unsigned pages = sfc_netfront_max_pages;
 
 	if (vnic_check_hello_version(msg->u.hello.version) < 0) {
 		msg->id = NET_ACCEL_MSG_HELLO | NET_ACCEL_MSG_REPLY 
diff -r b851951738f4 drivers/xen/sfc_netfront/accel_netfront.c
--- a/drivers/xen/sfc_netfront/accel_netfront.c
+++ b/drivers/xen/sfc_netfront/accel_netfront.c
@@ -234,13 +234,14 @@ struct netfront_accel_hooks accel_hooks 
 };
 
 
-unsigned max_pages = NETFRONT_ACCEL_DEFAULT_BUF_PAGES;
-module_param (max_pages, int, 0666);
+unsigned sfc_netfront_max_pages = NETFRONT_ACCEL_DEFAULT_BUF_PAGES;
+module_param_named (max_pages, sfc_netfront_max_pages, uint, 0666);
 MODULE_PARM_DESC(max_pages, "Number of buffer pages to request");
 
-unsigned buffer_split = 2;
-module_param (buffer_split, int, 0666);
-MODULE_PARM_DESC(buffer_split, "Fraction of buffers to use for TX, rest for RX");
+unsigned sfc_netfront_buffer_split = 2;
+module_param_named (buffer_split, sfc_netfront_buffer_split, uint, 0666);
+MODULE_PARM_DESC(buffer_split, 
+		 "Fraction of buffers to use for TX, rest for RX");
 
 
 const char *frontend_name = "sfc_netfront";

[-- Attachment #7: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel