Hi Shamir,
On 27/03/2017 23:11, Shamir Rabinovitch wrote:
> On Mon, Mar 27, 2017 at 06:06:42PM +0300, Mark Bloch wrote:
>> Hi Shamir,
>>
>> Like I've said in v1 of this patch, I believe we are calling ipoib_delete_debug_files too many times.
>> We are calling in unconditionally in: ipoib_dev_cleanup
>> and also in ipoib_netdev_event when we get an NETDEV_UNREGISTER event.
>>
>> For example, I have a setup with ConnectX-4 dual port configured to be in IB mode.
>> So I have two ipoib interfaces (ib0, ib1)
>>
>> When I load and unload mlx5_ib (while ib_ipoib is loaded:
>>
>> root@dev-r-vrt-175 tools]# ./funccount.py 'ipoib_*_debug_files'
>> Tracing 2 functions for "ipoib_*_debug_files"... Hit Ctrl-C to end.
>> ^C
>> FUNC COUNT
>> ipoib_create_debug_files 2
>> ipoib_delete_debug_files 4
>> Detaching...
>>
>> Why not just remove the call in ipoib_dev_cleanup?
>>
>
> Hi Mark,
>
> v3 of this patch works fine on system that has CX3 with 2 ports and the
> below udev rules:
>
> # InfiniBand: Mellanox Technologies MT27500 Family [ConnectX-3]
> SUBSYSTEM=="net", ACTION=="add", DRIVERS=="mlx4_core", BUS=="pci",
> ID=="0002:01:00.0", ATTR{dev_id}=="0x0", KERNEL=="ib*", NAME="ib1"
> SUBSYSTEM=="net", ACTION=="add", DRIVERS=="mlx4_core", BUS=="pci",
> ID=="0002:01:00.0", ATTR{dev_id}=="0x1", KERNEL=="ib*", NAME="ib0"
>
> On this system, the udev rules rename ib0-ib1 & ib1->ib0 causing small
> chaos in the ipoib device names.
>
> The attached logs include the information collected when the openibd
> service was started and when it was stopped. You can have a look in the
> files and see what are the network events and how they are processed by
> the ipoib devices.
>
> I think it will answer your concerns.
>
> BR, Shamir
>
I'm not saying it doesn't work, I'm saying works != works correctly.
We are calling ipoib_delete_debug_file too many times, it works by luck/chance.
While testing the patch, I've encountered another issue, running:
modprobe ib_ipoib
echo "0x0043" > /sys/class/net/ib0/create_child
modprobe -r ib_ipoib
and then looking the at the debugfs dir:
[root@dev-r-vrt-175 ~]# ls /sys/kernel/debug/ipoib/
ib0.8043_mcg ib0.8043_pat1
As you can see the the debugfs entries for the ib0 child weren't removed.
Also notice that after that, I can't load ib_ipoib
[root@dev-r-vrt-175 ~]# modprobe ib_ipoib
modprobe: ERROR: could not insert 'ib_ipoib': Cannot allocate memory
The more interesting issue is, dmesg output has this:
[ 467.185609] ib0.8043: failed to create mcg debug file
[ 467.192551] ib0.8043: failed to create path debug file
so maybe this is a debugfs bug?
Sorry I can't look into it, I have some internal stuff I need to work on :/
Mark.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
