fault-injection-developer

RE: [Fault-injection-developer] [RFC][PATCH] PCI configuration interceptor patch (ficl, fsml)

[Wang, S. <sta...@in...>]

There is another way to capture the pci configuration access indeed.
We can capture the sequential access to 0xcf8 and 0xcfc instead of place
a probe in the kernel.
You can get the example @ http://easynews.dl.sourceforge.net/sourceforge/fault-injection/mock_e100.tgz

Best Regards,
Stan

> -----Original Message-----
> From: Fumitake ABE [mailto:ab...@jp...]
> Sent: 2003-04-10 17:28
> To: fau...@li...
> Subject: [Fault-injection-developer] [RFC][PATCH] PCI 
> configuration interceptor patch (ficl, fsml)
> 
> 
> Dear All:
> 
> This mail appends two files: a test faultset file (pcitest.fsml) and 
> a patch to ficl implementing a function to use a pci configuration 
> access interceptor (ficl_pci.patch).
> 
> Pci configuration access interceptor can be used by the following:
> 1. apply patches to kernel and ficl
> 2. install the kernel, modules and ficl
> 3. attach kprobe to a pci configuration access 
>    (ficl -p /boot/System.map,/usr/src/linux/vmlinux)
> 4. add a trigger (ficl -i pcitest.fsml)
> 5. access a pci configuration register (e.g. lspci -xx)
> 
> Ficl needs a System.map path and a unzipped kernel file path for 
> attaching kprobe to a pci configuration interceptor. I think that it 
> may be ugly, but I couldn't come up with a better idea than it.
> Please comment.
> 
> Sincerely,
> 
> Fumitake ABE
> ab...@jp...
> 

[Fault-injection-developer] IRQ interceptor

[arun <ar...@np...>]

Hi Stanley/Kevin,
The fith_utility/ficl_cmd.c doesnt seems to handle IRQ interceptor. It
only handle the "Trigger"
and for the rest errors are popped up.
I tried the below which was from the fith_manual from CVS directory
(might be old and obselete)

irq.fsml:
<?xml version="1.0" encoding="UTF-8" ?>
<!-- $Id: fith.texi,v 1.12 2002/06/03 07:45:22 stanley Exp $ -->
<!-- irq is the interrupt vector you want to intercept -->

<!-- devname is the device's name that be used in request_irq() -->

<!-- spurious_hertz="10" means one spurious IRQ is generated -->
<!-- per 10 timer clicks -->
<!-- delay_hertz="10" means delay one IRQ per 10 times -->
<!-- delay_time="3" means the IRQ will be delayed 3 timer clicks -->
<fake_irq irq="4"
modname=”fith_dummy”
devname="dummy"
spurious_hertz="10"
delay_hertz="10"
delay_time="3"
/>

[root@xcast1hclt-ntl test]# ficl -i irq1.fsml
Inserting irq1.fsml into fi_dm...
Document of the wrong type, FSML Namespace not found!
Can not parser out faultset from this file.


But this fsml file could be inserted using ficl command.

I guess the IRQ interceptor should be given inside the TRIGGER tag... I
tried
that, but in vain (The ficl_cmd.c handles only the parameters for the PF
interceptor)

I tried to "bk pull" to get the latest "fith tool", but there was no
updation.... I just
want to clarify whether there is a workaround for this or I should
update the ficl_cmd.c to
handle the IRQ interceptor parsing... Please pass on any document (if
available) about
this...

Thanks
-arun

Re: [Fault-injection-developer] [RFC][PATCH] PCI configurationinterceptor patch (ficl, fsml)

[Fumitake A. <ab...@jp...>]

Thank you, Stan.

"Wang, Stanley" wrote:
> You can get the example @ http://easynews.dl.sourceforge.net/sourceforge/fault-injection/mock_e100.tgz

I just thought that mock_e100.tgz supported only e100 driver because 
of its name. So I had missed it. I feel that it is simpler and smarter 
than my implementation.

BTW, 'PCI access mode' needs to be 'direct' (CONFIG_PCI_GODIRECT=y) to
use mock_e100.tgz, doesn't it? Actullay, I've tried to the way of 
trapping at 0xcf8 before I implement the way written in previous mail. 
But it didn't work. I think now pci configuration was accessed via BIOS. 
If pci access mode of kernel configuration is set up as "default", 
it becomes via BIOS.

If it's true, I think users have to be aware of this.
Do you think it is better to be written to README?

Sincerely,

Fumitake ABE
ab...@jp...