Re: 2.5.29: bug in ide and hd kernel option handling

Re: 2.5.29: bug in ide and hd kernel option handling

[Mikael Pettersson]

On Tue, 30 Jul 2002, I wrote:
> On my 486 test box (ISA/VLB only, CONFIG_PCI=n), passing any
> any ide or hd kernel option (like idebus=33) to 2.5.29 results
> in a kernel hang at boot: I get the initial "Uncompressing ..
> booting .." and then nothing.

Problem partially identified.

With CONFIG_PCI=n, include/asm-i386/ide.h:ide_init_default_hwifs()
is defined to ide_register_hw() the PC's standard IDE ports, but
with CONFIG_PCI=y, it's empty.

When drivers/ide/main.c:ide_setup() is called for some "ide..."
kernel option, it starts by calling init_global_data(), which
in turn calls ide_init_default_hwifs(). When CONFIG_PCI=n so
ide_init_default_hwifs() isn't empty, the kernel either hangs
or reboots at that point.

init_global_data() and ide_init_default_hwifs() can also be called
much later from 'module_init(init_ata)'. In that case there is no
hang or reboot -- so my guess is that the initialisation does something
which normally works but is illegal and causes a fault when done
at __setup()-time.

I tested every kernel from 2.5.29 and back, and the problem started
with 2.5.5.

As a workaround I applied the patch below to unconditionally
make ide_init_default_hwifs() do nothing. This solved my problem
and doesn't seem to have had any bad side-effects: the kernel still
finds all standard IDE ports on my 486.

/Mikael

--- linux-2.5.29/include/asm-i386/ide.h.~1~	Sat Jul 20 23:49:45 2002
+++ linux-2.5.29/include/asm-i386/ide.h	Thu Aug  1 02:20:31 2002
@@ -65,7 +65,7 @@
 
 static __inline__ void ide_init_default_hwifs(void)
 {
-#ifndef CONFIG_PCI
+#if 0 && !defined(CONFIG_PCI)
 	hw_regs_t hw;
 	int index;
 

Re: 2.5.29: bug in ide and hd kernel option handling

[Marcin Dalecki]

Mikael Pettersson wrote:
> On Tue, 30 Jul 2002, I wrote:
> 
>>On my 486 test box (ISA/VLB only, CONFIG_PCI=n), passing any
>>any ide or hd kernel option (like idebus=33) to 2.5.29 results
>>in a kernel hang at boot: I get the initial "Uncompressing ..
>>booting .." and then nothing.
> 
> 
> Problem partially identified.
> 
> With CONFIG_PCI=n, include/asm-i386/ide.h:ide_init_default_hwifs()
> is defined to ide_register_hw() the PC's standard IDE ports, but
> with CONFIG_PCI=y, it's empty.
> 
> When drivers/ide/main.c:ide_setup() is called for some "ide..."
> kernel option, it starts by calling init_global_data(), which
> in turn calls ide_init_default_hwifs(). When CONFIG_PCI=n so
> ide_init_default_hwifs() isn't empty, the kernel either hangs
> or reboots at that point.
> 
> init_global_data() and ide_init_default_hwifs() can also be called
> much later from 'module_init(init_ata)'. In that case there is no
> hang or reboot -- so my guess is that the initialisation does something
> which normally works but is illegal and causes a fault when done
> at __setup()-time.
> 
> I tested every kernel from 2.5.29 and back, and the problem started
> with 2.5.5.
> 
> As a workaround I applied the patch below to unconditionally
> make ide_init_default_hwifs() do nothing. This solved my problem
> and doesn't seem to have had any bad side-effects: the kernel still
> finds all standard IDE ports on my 486.

Ahh. Thats actually insightfull. OK some other archs even default to the
same if PCI bus support is enabled at all. I will have to test the
effects of this approach in bochs to see how it lives with "legacy"
host chips. Maybe this whole ide_init_default_hwifs can go completely
away then... (If anything it should be fallback only.)

Re: 2.5.29: bug in ide and hd kernel option handling

[Gerald Champagne]

Well, that would most likely be my fault.  I made some small changes to
the parameter parsing code recently.

> On my 486 test box (ISA/VLB only, CONFIG_PCI=n), passing any
> any ide or hd kernel option (like idebus=33) to 2.5.29 results
> in a kernel hang at boot: I get the initial "Uncompressing ..
> booting .." and then nothing.

The no display symptom makes sense if there's a problem in a setup
routine.  These routines are called before console_init.  Printk's
before console_init are queued up and only displayed after console_init
is called.  If the kernel hangs before this point, nothing is displayed.

> With 2.5.27, the kernel instantly rebooted itself instead.

The changes went in in 2.5.27.

Can you please send my your .config file?  I'll look at this tonight.

Thanks.

Gerald

2.5.29: bug in ide and hd kernel option handling

[Mikael Pettersson]

On my 486 test box (ISA/VLB only, CONFIG_PCI=n), passing any
any ide or hd kernel option (like idebus=33) to 2.5.29 results
in a kernel hang at boot: I get the initial "Uncompressing ..
booting .." and then nothing.
With 2.5.27, the kernel instantly rebooted itself instead.

I don't know when this problem appeared. I only recently starting
experimenting with ide kernel options in an attempt to get an
old qd6580 controller card going -- alas the second channel
never wants to show up :-(

/Mikael