Re: [Pcihpd-discuss] do I need ACPI to detect hot plugged cpci cards?

Re: [Pcihpd-discuss] do I need ACPI to detect hot plugged cpci cards?

[Florea Igor]

On Tuesday 17 January 2006 22:29, Scott Murray wrote:
> On Tue, 17 Jan 2006, user wrote:
> > Hi all
> >
> > do I need ACPI to detect hot plugged cpci cards? my problem is : I cannot
> > get IRQ after hot inserting the card.. it is 255 (I'm booting with
> > inserted cards , remove the card and insert it again so resources are
> > allocated..) I need some ideas where to start looking?
> > from my dmesg:
> > # dmesg |grep -i acpi
> > ACPI: Unable to locate RSDP
> > ACPI: Subsystem revision 20050902
> > ACPI: Interpreter disabled.
> >  pnp: PnP ACPI: disabled
> >  PCI quirk: region 4000-407f claimed by ICH4 ACPI/GPIO/TCO
>
> Sorry for the slow response to your queries, I was on vacation until last
> week, and it's taken me the past week to work through my work backlog to
> the point I can look at this.  As a first guess, I'd say it looks like
> your board has an IO-APIC of some form.  I've not tried hotplug on any
> CPCI boards with IO-APICs in 2.6 recently, but in 2.4 I had to do some
> hackery to get interrupt assignment to work on such boards.  Could you
> identify the board you're using and/or provide the output of "dmesg" and
> "lspci -vvv"?
>
> Thanks,
>
> Scott

The peripheral board in our project is developed by Topex ( I work here ) and 
the CPU is Kontron cp303.
We already have the driver for our CPCI card and it works fine when we boot 
the system with inserted cards (resources allocated by bios at boot).  We 
tried to use cpcihp_generic driver from kernel 2.6.14 for Compact PCI 
controller and it works fine with some modifications in cpci_hotplug_core.c 
and in cpci_hotplug_pci.c but we have problems with resources allocation : 
first of all when we boot the system without cards, load the cpcihp_generic 
driver and insert the card we get the card but without any resources ( IRQ, 
Memory) . If we boot with inserted cards , load drivers , remove a card and 
hot plug it back.. we get the memory [disabled] and we get IRQ%5 ( not 
valid ) and we cannot change it 
to some IRQ 0..15 .
the log messages i got :
lspci -vvv after i boot with inserted cards : http://favorites.go.ro/lspci1

lspci -vvv after i boot with inserted cards, load the hotplug controller 
drivers, remove the card, and hot plug it back: http://favorites.go.ro/lspci2
dmesg  (i've added some printk_s in kernel sources for debug): 
http://favorites.go.ro/dmesg
dmesg  (without my debug messages): http://favorites.go.ro/dmesg1

our cpci card is 02:0e.0 Co-processor: Xilinx Corporation Unknown device 0400
on 01:0c.0 PCI bridge: Intel Corporation 21154 PCI-to-PCI Bridge (prog-if 00 
[Normal decode])
The CPU board is running now a Gentoo Linux 2005 (kernel 2.6.14)



-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems?  Stop!  Download the new AJAX search engine that makes
searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
http://sel.as-us.falkag.net/sel?cmd_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel

Re: [Pcihpd-discuss] do I need ACPI to detect hot plugged cpci cards?

[Florea Igor]

Hi Scott

> It might be worth seeing if the apcihp driver works on the CP-306.
it works fine 
> >   	list_for_each_entry(slot, &slot_list, slot_list){
> > 		if (slot->extracting) {
> > 			disable_slot(slot->hotplug_slot);
> > 		}
> The intent of the driver is that a userspace component detects the
> extracting state, does any required software shutdown, and then drives
> the extract (i.e. disable_slot) by writing a 0 to the "power" node for
> the slot in sysfs.
I cannot get disable_slot called when i
echo 0 > /sys/bus/pci/devices/0000\:04\:0d.0/power/state
i don't know why.. so i let it with explicit call to disable_slot ..it does 
not bother me much.. but i would like to do it in the ritgh way.. the user 
program have to poll the enum bit of enum port? it is already done by 
cpcihp_generic driver..

> > I also have to expand the code  to support all my 3 hotplug controllers 
> > (one on another ..as is seen in my lspci ) if you have some warnings for
> > me..please let me know..
>
> If you have 3 separate hotplug controllers (separate ENUM# signalling),
> watching different buses, then the current driver won't handle that
> without some significant rework.  If there is only a single ENUM#
> interrupt or signal for 3 buses, then that can be made to work by
> calling cpci_hp_register_bus for the 3 buses.  If you can provide a bit
> more detail on what the hardware looks like, I can likely come up with
> an idea of how the driver needs to change to support it.
Good Idea!!
I did not notice that i can register only buses .. i implemented all the stuff 
to also register all the controllers .. and can
echo 0 > /sys/bus/pci/devices/0000\:04\:0d.0/power/state
work if i registered only one controller?
(anyway this part does not work now..so i could register only one..)

I have  3 hotplug controllers with one ENUM# signal (port 0x0286 bit 2)...and 
if i had 3 separate ENUM# sources i should only change the query_enum 
function..

tanks
Igor



-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems?  Stop!  Download the new AJAX search engine that makes
searching your log files as easy as surfing the  web.  DOWNLOAD SPLUNK!
http://sel.as-us.falkag.net/sel?cmd_______________________________________________
Linux-hotplug-devel mailing list  http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel