RE: PCI probing in the kernel

It specifically says *em* probing and then it calls the *em_register* function repeatedly which is the registration routine that the em driver defines as part of its initialization. This is whats confusing me  as to why is it repeatedly calling these functions for a specific module repeatedly

 

From: Warner Losh <imp@bsdimp.com> 
Sent: Saturday, June 26, 2021 10:40 AM
To: saimabaig11@gmail.com
Cc: FreeBSD Hackers <freebsd-hackers@freebsd.org>
Subject: Re: PCI probing in the kernel

 

 

 

On Fri, Jun 25, 2021 at 10:59 PM <saimabaig11@gmail.com <mailto:saimabaig11@gmail.com> > wrote:

My question relates to how does the kernel go about its business of
discovering PCI devices (after boot time) . I am loading the e1000(em)
driver by kldloading the `if_em` module. The kernel seems to be calling the
probe routine multiple times. So the log below includes messages coming from
the kernel and also includes my own debug logs.



```

Jun 25 11:16:45 FreeBSD12 kernel: em probing

Jun 25 11:16:45 FreeBSD12 syslogd: last message repeated 5 times

Jun 25 11:16:45 FreeBSD12 kernel: em0: +++ Calling em_register

Jun 25 11:16:45 FreeBSD12 kernel: em probing

Jun 25 11:16:45 FreeBSD12 syslogd: last message repeated 5 times

Jun 25 11:16:45 FreeBSD12 kernel: em0: +++ Calling em_register

Jun 25 11:16:45 FreeBSD12 kernel: em probing

Jun 25 11:16:45 FreeBSD12 kernel: em0: +++ Calling em_register

Jun 25 11:16:45 FreeBSD12 kernel: em0: <Intel(R) PRO/1000 Network
Connection> port 0x2000-0x203f mem
0xfd5c0000-0xfd5dffff,0xfdff0000-0xfdffffff irq 19 at device 1.0 on pci2

Jun 25 11:16:45 FreeBSD12 kernel: em attaching

Jun 25 11:16:45 FreeBSD12 kernel: em0: +++ Calling em_register

Jun 25 11:17:34 FreeBSD12 kernel: em0: Using 2048 TX descriptors and 2048 RX
descriptors

Jun 25 11:17:35 FreeBSD12 kernel: em probing

Jun 25 11:17:35 FreeBSD12 syslogd: last message repeated 5 times

Jun 25 11:17:35 FreeBSD12 kernel: em1: +++ Calling em_register

Jun 25 11:17:35 FreeBSD12 kernel: em probing

Jun 25 11:17:36 FreeBSD12 kernel: Base = 5ee7e000, Length = 8000

```

It seems that the `probe->register` sequence seems to be getting repeated
multiple times until the adapter is identified properly and its resources
allocated. Isn't multiple registration a cause of concern? Or may something
is wrong with the way I think. May be I am not getting it right? Could
someone elaborate?

 

When you load a driver, we call probe on all the devices on the bus that don't

yet have drivers. I believe you are seeing this. The probe routine tries a bunch

of different devices until finding the NIC that doesn't have a driver. It's completely

normal and ordinary.

 

Warner