An imbalanced TX indirection table causes netvsc to have low
performance. This table is created and managed during runtime. To help
better diagnose performance issues caused by imbalanced tables, add
device attributes to show the content of TX indirection tables.
Signed-off-by: Chi Song <[email protected]>
---
v4: use a separated group to organize tx_indirection better, change
location of attributes init/exit to netvsc_drv_init/exit
v5: update variable orders
v6: update names to be more precise, remove useless assignment
Thank you all for comments, learned a lot.
drivers/net/hyperv/netvsc_drv.c | 48 +++++++++++++++++++++++++++++++++
1 file changed, 48 insertions(+)
diff --git a/drivers/net/hyperv/netvsc_drv.c b/drivers/net/hyperv/netvsc_drv.c
index 6267f706e8ee..f6ad13ed320f 100644
--- a/drivers/net/hyperv/netvsc_drv.c
+++ b/drivers/net/hyperv/netvsc_drv.c
@@ -2370,6 +2370,50 @@ static int netvsc_unregister_vf(struct net_device *vf_netdev)
return NOTIFY_OK;
}
+static struct device_attribute dev_attr_tx_indirection_attrs[VRSS_SEND_TAB_SIZE];
+static struct attribute *tx_indirection_attrs[VRSS_SEND_TAB_SIZE + 1];
+
+const struct attribute_group tx_indirection_group = {
+ .name = "tx_indirection",
+ .attrs = tx_indirection_attrs,
+};
+
+static ssize_t tx_indirection_show(struct device *dev,
+ struct device_attribute *dev_attr, char *buf)
+{
+ int index = dev_attr - dev_attr_tx_indirection_attrs;
+ struct net_device *ndev = to_net_dev(dev);
+ struct net_device_context *ndc = netdev_priv(ndev);
+
+ return sprintf(buf, "%u\n", ndc->tx_table[index]);
+}
+
+static void netvsc_attrs_init(void)
+{
+ char buffer[4];
+ int i;
+
+ for (i = 0; i < VRSS_SEND_TAB_SIZE; i++) {
+ sprintf(buffer, "%02u", i);
+ dev_attr_tx_indirection_attrs[i].attr.name =
+ kstrdup(buffer, GFP_KERNEL);
+ dev_attr_tx_indirection_attrs[i].attr.mode = 0444;
+ sysfs_attr_init(&dev_attr_tx_indirection_attrs[i].attr);
+
+ dev_attr_tx_indirection_attrs[i].show = tx_indirection_show;
+ tx_indirection_attrs[i] =
+ &dev_attr_tx_indirection_attrs[i].attr;
+ }
+}
+
+static void netvsc_attrs_exit(void)
+{
+ int i;
+
+ for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
+ kfree(dev_attr_tx_indirection_attrs[i].attr.name);
+}
+
static int netvsc_probe(struct hv_device *dev,
const struct hv_vmbus_device_id *dev_id)
{
@@ -2410,6 +2454,7 @@ static int netvsc_probe(struct hv_device *dev,
net->netdev_ops = &device_ops;
net->ethtool_ops = ðtool_ops;
+ net->sysfs_groups[0] = &tx_indirection_group;
SET_NETDEV_DEV(net, &dev->device);
/* We always need headroom for rndis header */
@@ -2665,6 +2710,7 @@ static void __exit netvsc_drv_exit(void)
{
unregister_netdevice_notifier(&netvsc_netdev_notifier);
vmbus_driver_unregister(&netvsc_drv);
+ netvsc_attrs_exit();
}
static int __init netvsc_drv_init(void)
@@ -2678,6 +2724,8 @@ static int __init netvsc_drv_init(void)
}
netvsc_ring_bytes = ring_size * PAGE_SIZE;
+ netvsc_attrs_init();
+
ret = vmbus_driver_register(&netvsc_drv);
if (ret)
return ret;
--
2.25.1
On Tue, 21 Jul 2020 04:58:59 +0000 Chi Song wrote:
> An imbalanced TX indirection table causes netvsc to have low
> performance. This table is created and managed during runtime. To help
> better diagnose performance issues caused by imbalanced tables, add
> device attributes to show the content of TX indirection tables.
>
> Signed-off-by: Chi Song <[email protected]>
Sorry for waiting until v6 but sysfs feel like a very strange place to
expose this. Especially under the netdev, not the bus device.
This looks like device specific state, perhaps ethtool -d is a more
appropriate place?
On 7/21/20 12:21 PM, Jakub Kicinski wrote:
> On Tue, 21 Jul 2020 04:58:59 +0000 Chi Song wrote:
>> An imbalanced TX indirection table causes netvsc to have low
>> performance. This table is created and managed during runtime. To help
>> better diagnose performance issues caused by imbalanced tables, add
>> device attributes to show the content of TX indirection tables.
>>
>> Signed-off-by: Chi Song <[email protected]>
>
> Sorry for waiting until v6 but sysfs feel like a very strange place to
> expose this. Especially under the netdev, not the bus device.
>
> This looks like device specific state, perhaps ethtool -d is a more
> appropriate place?
Agreed, or a devlink resource maybe?
--
Florian
> -----Original Message-----
> From: Chi Song <[email protected]>
> Sent: Tuesday, July 21, 2020 12:59 AM
> To: Stephen Hemminger <[email protected]>; David Miller
> <[email protected]>; Haiyang Zhang <[email protected]>; KY
> Srinivasan <[email protected]>; Stephen Hemminger
> <[email protected]>; [email protected]; [email protected]; linux-
> [email protected]
> Cc: [email protected]; [email protected]
> Subject: [PATCH v6 net-next] net: hyperv: Add attributes to show TX indirection
> table
>
> An imbalanced TX indirection table causes netvsc to have low performance.
> This table is created and managed during runtime. To help better diagnose
> performance issues caused by imbalanced tables, add device attributes to show
> the content of TX indirection tables.
>
> Signed-off-by: Chi Song <[email protected]>
Thanks you!
Reviewed-by: Haiyang Zhang <[email protected]>
On Tue, 21 Jul 2020, Florian Fainelli wrote:
> On 7/21/20 12:21 PM, Jakub Kicinski wrote:
> > On Tue, 21 Jul 2020 04:58:59 +0000 Chi Song wrote:
> >> An imbalanced TX indirection table causes netvsc to have low
> >> performance. This table is created and managed during runtime. To help
> >> better diagnose performance issues caused by imbalanced tables, add
> >> device attributes to show the content of TX indirection tables.
> >>
> >> Signed-off-by: Chi Song <[email protected]>
> >
> > Sorry for waiting until v6 but sysfs feel like a very strange place to
> > expose this. Especially under the netdev, not the bus device.
> >
> > This looks like device specific state, perhaps ethtool -d is a more
> > appropriate place?
>
> Agreed, or a devlink resource maybe?
Thank you for comments, I will move it to ethtool.