On Fri, 30 Nov 2007 10:52:29 +0300
Vasily Averin [off-list ref] wrote:

Hi Ayaz,


While testing new OpenVZ kernel based on RHEL5.1 kernel 2.6.18-53.el5 we have
observed that node loads forcedeth driver too long:

Nov 28 14:04:19 ts49 kjournald starting.  Commit interval 5 seconds
Nov 28 14:04:19 ts49 EXT3-fs: mounted filesystem with ordered data mode.
...
Nov 28 19:34:47 ts49 modprobe      D ffff8101ff772120     0  4101   4098
             (NOTLB)
Nov 28 19:34:47 ts49  ffff8101fe7e1c28 0000000000000086 0000000000000246
00000010000200d0
Nov 28 19:34:47 ts49  ffff8101ff772120 ffff8103ff93c930 000012229b7b3bd7
00002bcc55b0dda7
Nov 28 19:34:47 ts49  ffff8101ff772328 ffff8101ff93a000 ffffffff802de500
ffff8103ff93c930
Nov 28 19:34:47 ts49 Call Trace:
Nov 28 19:34:47 ts49  [<ffffffff80062cc2>] schedule_timeout+0x8a/0xad
Nov 28 19:34:47 ts49  [<ffffffff80091bf9>] process_timeout+0x0/0x5
Nov 28 19:34:47 ts49  [<ffffffff800925c0>] msleep+0x21/0x2c
Nov 28 19:34:47 ts49  [<ffffffff8819fc8f>] :forcedeth:nv_probe+0x873/0xd4d
Nov 28 19:34:47 ts49  [<ffffffff801427f8>] pci_device_probe+0x100/0x180
Nov 28 19:34:47 ts49  [<ffffffff8019ff2c>] driver_probe_device+0x52/0xaa
Nov 28 19:34:47 ts49  [<ffffffff801a005b>] __driver_attach+0x65/0xb6
Nov 28 19:34:47 ts49  [<ffffffff8019fff6>] __driver_attach+0x0/0xb6
Nov 28 19:34:47 ts49  [<ffffffff8019f96d>] bus_for_each_dev+0x43/0x6e
Nov 28 19:34:47 ts49  [<ffffffff8019f5b3>] bus_add_driver+0x7e/0x130
Nov 28 19:34:47 ts49  [<ffffffff801429dc>] __pci_register_driver+0x57/0x7e
Nov 28 19:34:47 ts49  [<ffffffff800aa815>] sys_init_module+0x16aa/0x185f
Nov 28 19:34:47 ts49  [<ffffffff8005e166>] system_call+0x7e/0x83
...
Nov 28 21:02:02 ts49 eth0: forcedeth.c: subsystem: 010f1:2912 bound to 0000:00:08.0
Nov 28 21:02:02 ts49 ACPI: PCI Interrupt Link [LMA2] enabled at IRQ 18
Nov 28 21:02:02 ts49 GSI 22 sharing vector 0x3A and IRQ 22
Nov 28 21:02:02 ts49 ACPI: PCI Interrupt 0000:00:09.0[A] -> Link [LMA2] -> GSI
18 (level, high) -> IRQ 58
Nov 28 21:02:02 ts49 PCI: Setting latency timer of device 0000:00:09.0 to 64
Nov 28 21:02:02 ts49
Nov 28 21:02:02 ts49 forcedeth: using HIGHDMA
Nov 28 21:02:02 ts49 eth1: forcedeth.c: subsystem: 010f1:2912 bound to 0000:00:09.0

nv_probe() was called under semaphore and blocks lots of other modprobe's from
loading other drivers.

linux-2.6-net-forcedeth-update-to-driver-version-0-60.patch patch from RHEL5.1
added the following piece of code to nv_probe():

Please don't complain about vendor kernel patches on the mainline kernel development
list. Complain to the distribution vendor.


-- 
Stephen Hemminger [off-list ref]