Gotta start somewhere ... how many of us are really out there?

[Antony Mawer]

On 4/08/2006 4:58 AM, User Freebsd wrote:
> Getting a list of devices is actually pretty easy, and I've tried this 
> on my 4.x machines also, so it isn't something that will be a problem on 
> older versions:
> 
> # pciconf -l
> chip0 at pci0:0:0: class=0x060000 card=0x00000000 chip=0x700c1022 rev=0x20 
> hdr=0x00
...
> And, more specifically, we can get:
> 
> # pciconf -l -v
> asr0 at pci0:9:0:  class=0x010400 card=0xc0351044 chip=0xa5111044 rev=0x01 
> hdr=0x00
>     vendor   = 'Adaptec (Formerly: Distributed Processing Technology 
> (DPT))'
>     device   = 'Raptor SmartRAID Controller'
>     class    = mass storage
>     subclass = RAID

All of the expanded 'vendor', 'device', 'class' and 'subclass' 
information is present in the non -v version of the command output. The 
numbers shown earlier can be used to derive the text information:

     class=0x010400
       determines the class/subclass lines, using the table from here:
       http://fxr.watson.org/fxr/source/dev/pci/pci.c#L1340

     card=0xc0351044 chip=0xa5111044
       these make up the vendor and device lines, using the list in
       /usr/share/misc/pci_vendors (which is derived from the PCIDEVS.TXT
       listing).

       The last 4 hex digits of the card and chip lines are the vendor ID
       while the first 4 are the device ID. The card is often given by
       the vendor, while the chip identifies the actual part it uses to
       implement functionality. For instance, a Netcomm ethernet NIC may
       use a Realtek 8139 chip... so chip gives us the fact it's
       essentially a generic Realtek chipset, while the card tells us the
       vendor who manufactured the card & perhaps their name for it.

In short, there's no reason to have to transmit all the text names back 
to any server -- this can all be resolved at the server end,

> 
> So, with that one command, we can get a fair amount of hardware 
> information ... but, how to feed that into a proper HTTP request? 
> Storing all of that information would be cool, cause then we could build 
> reports based on device driver / vendor / device / class and subclass 
> ... but that might be a bit heavy to do in an HTTP request, no?  I take 
> it email isn't an option, in your case?

Email may be a viable alternative -- one concern with email is that 
various organisations SMTP servers blast their own disclaimer message 
and so on across the bottom of all out-going emails, which might 
complicate parsing of it on the server end.

If you're only encoding purely the numeric details, this would make the 
information far lighter to transmit than having the whole text blurb. 
Just the pciconf -l version as-is:

     ~$ pciconf -l|wc -c
     1545

So that's ~1500 bytes. Now strip out all the unnecessary text - the 
class=, card=, chip=, rev=, hdr=, extra spaces... something like:

     mpt0 at pci2:5:0: 010000 34358086 00301000 08 00
     mpt1 at pci2:5:1: 010000 34358086 00301000 08 00
     em0 at pci3:4:0: 020000 10798086 10798086 03 00
     em1 at pci3:4:1: 020000 10798086 10798086 03 00

     ~$ cat pciconf-stripped | wc -c
     899

We've nearly halved the size of the information. Now it's still in 
ASCII, so you could further shave bits off by converting that to binary 
if you wanted to...


With that amount of information, you'd probably be more inclined to want 
to use HTTP POST than HTTP GET. A quick glance suggests libfetch(3) 
doesn't support this; I haven't looked at the code enough to see if 
adding support for it would be trivial or not.

-Antony