Re: Patch?: linux-2.5.33/drivers/input/keyboard/atkbd.c allow SETLEDS to fail

Re: Patch?: linux-2.5.33/drivers/input/keyboard/atkbd.c allow SETLEDS to fail

[Adam J. Richter]

On Mon, 9 Sep 2002 at 16:07:23 +0200, Vojtech Pavlik wrote:
>On Mon, Sep 09, 2002 at 06:51:11AM -0700, Adam J. Richter wrote:

>> 	I have a computer with an Iwill VD133 motherboard doing
>> USB-to-PS/2 keyboard emulation (built into the chipset somewhere)
>> for a BTC 7932M USB keyboard.  Under this configuration, the
>> SETLEDS command in atkbd_probe fails (the first atkbd_sendbyte
>> in atkbd_command fails), but the keyboard otherwise works if
>> that failure is ignored.
[...]

>I cannot do that, since then the keyboard identification will give too
>many positives.

	What real problem does that cause?  There are other tests that
the keyboard still has to pass.  Does this change even cause a false
positive in any real case?  As I understand it, this is how it was
prior to 2.5.32.


>I assume you have an USB keyboard, and the BIOS emulates a PS/2 keyboard
>for non-USB capable OSes.

>Well, fix Linux irq router code, instead of damaging the keyboard code.

	How is the Linux IRQ routing code broken?  As far as I know,
it only reads the information given to it and cannot reprogram the
hardware.

	My understanding is that the BIOS is responsible for setting
up the IRQ routes of each device and writing them into each device's
PCI configuration register dword 15 byte 0 (INTERRUPT_LINE), which is
normally just 8 bits of RAM.  The mechanism for *making* those routes
is vendor specific and handled by the BIOS at boot time.  In the
particular case of my VD133, it apparently uses a Via VT82C598 PCI
bridge ("Apollo MVP3 AGP"), which I think is the component that would
have the secret bits for setting up routing from the main PCI bus to
the interrupt controllers.  And those bits do appear to be secret.
While I have found programming information for a number of Via chips
on my VD133 motherboard, I have only been able to find technical
marketing information on the VT82C598, and the form for requesting
additional information from the Via web site says they normally only
give more detailed documentation to high-volume partners under
non-disclosure agreements.

	By the way, there is a note in drivers/pci/quirks.c that
mentioned that a different Via chipset did IRQ configuration for the
USB device by having the write to the INTERRUPT_LINE register actually
do the configuration. I tried it ("setpci -v -s 00:01.0 INTERRUPT_LINE=7"),
but that didn't update the kernel's idea of the IRQ routing (although
I could read it back from hardware with "lspci -H 1 -vv -s 00:01.0").

	However, even if some further development of this kludge works
(I would welcome that), I suspect that the bug that I am reporting is
common to many USB emulated keyboards, and that my fix does not cause
a real problem, and applications that do not need USB for any other
purpose to avoid including or configuring Linux USB.

	On the other hand, if there is real hardware is broken by my
change (and kernels prior to 2.5.32 I assume), that would change things,
and I'd be interested in knowing what that hardware is and how it breaks.

Adam J. Richter     __     ______________   575 Oroville Road
adam@yggdrasil.com     \ /                  Milpitas, California 95035
+1 408 309-6081         | g g d r a s i l   United States of America
                         "Free Software For The Rest Of Us."

Re: Patch?: linux-2.5.33/drivers/input/keyboard/atkbd.c allow SETLEDS to fail

[Vojtech Pavlik]

On Mon, Sep 09, 2002 at 08:04:11AM -0700, Adam J. Richter wrote:
> On Mon, 9 Sep 2002 at 16:07:23 +0200, Vojtech Pavlik wrote:
> >On Mon, Sep 09, 2002 at 06:51:11AM -0700, Adam J. Richter wrote:
> 
> >> 	I have a computer with an Iwill VD133 motherboard doing
> >> USB-to-PS/2 keyboard emulation (built into the chipset somewhere)
> >> for a BTC 7932M USB keyboard.  Under this configuration, the
> >> SETLEDS command in atkbd_probe fails (the first atkbd_sendbyte
> >> in atkbd_command fails), but the keyboard otherwise works if
> >> that failure is ignored.
> [...]
> 
> >I cannot do that, since then the keyboard identification will give too
> >many positives.
> 
> 	What real problem does that cause?  There are other tests that
> the keyboard still has to pass. 

There are not. If SETLEDS fails, then there is the GETID command, which
is also allowed to fail (fails on AT (non-PS/2) keyboards, which are
still common), and it most likely also fails on your keyboard. That's
all.

If you kill the SETLEDS command, then even without a keyboard, the
keyboard will be detected, and sent commands, which will have to time
out every time.

> Does this change even cause a false
> positive in any real case?  As I understand it, this is how it was
> prior to 2.5.32.

Yes. The keyboard was assumed to be always present.

> >I assume you have an USB keyboard, and the BIOS emulates a PS/2 keyboard
> >for non-USB capable OSes.
> 
> >Well, fix Linux irq router code, instead of damaging the keyboard code.
> 
> 	How is the Linux IRQ routing code broken?  As far as I know,
> it only reads the information given to it and cannot reprogram the
> hardware.

It can.

> 	My understanding is that the BIOS is responsible for setting
> up the IRQ routes of each device and writing them into each device's
> PCI configuration register dword 15 byte 0 (INTERRUPT_LINE), which is
> normally just 8 bits of RAM.

Yes, but because the BIOS often writes nonsense to these config
registers, Linux knows several southbridges and can fix the routing by
hand.

> The mechanism for *making* those routes
> is vendor specific and handled by the BIOS at boot time.  In the
> particular case of my VD133, it apparently uses a Via VT82C598 PCI
> bridge ("Apollo MVP3 AGP"), which I think is the component that would
> have the secret bits for setting up routing from the main PCI bus to
> the interrupt controllers.

Most likely the interrupt routing is controlled by the southbridge (the
ISA bridge device). See arch/i386/pci/irq.c. Maybe it's as easy as an
entry for your southbridge is missing.

> And those bits do appear to be secret.

Not really. Most of them are actually documented. Namely for built-in
devices like USB controllers, because those cannot be wired differently
by different board manufacturers.

> While I have found programming information for a number of Via chips
> on my VD133 motherboard, I have only been able to find technical
> marketing information on the VT82C598, and the form for requesting
> additional information from the Via web site says they normally only
> give more detailed documentation to high-volume partners under
> non-disclosure agreements.

Try the southbridge instead of the northbridge. I have several of the
documents, and can send it to you if you tell me the right chip number.

> 	By the way, there is a note in drivers/pci/quirks.c that
> mentioned that a different Via chipset did IRQ configuration for the
> USB device by having the write to the INTERRUPT_LINE register actually
> do the configuration. I tried it ("setpci -v -s 00:01.0 INTERRUPT_LINE=7"),
> but that didn't update the kernel's idea of the IRQ routing (although
> I could read it back from hardware with "lspci -H 1 -vv -s 00:01.0").
> 
> 	However, even if some further development of this kludge works
> (I would welcome that), I suspect that the bug that I am reporting is
> common to many USB emulated keyboards, and that my fix does not cause
> a real problem, and applications that do not need USB for any other
> purpose to avoid including or configuring Linux USB.

Well, the emulated keyboard is a very very ugly kludge. It uses SMM and
SMI to handle the keyboard, and then pushes keystrokes into the keyboard
controller. It disables any extra keys on the keyboard, and the keyboard
LEDs. The Linux USB stack isn't too big to enable the keyboard the sane
way on those systems.

> 	On the other hand, if there is real hardware is broken by my
> change (and kernels prior to 2.5.32 I assume), that would change things,
> and I'd be interested in knowing what that hardware is and how it breaks.

Any machine with an USB keyboard enabled in the normal way will try to
set the LEDs on a non-existing PS/2 keyboard. Which will take time.

-- 
Vojtech Pavlik
SuSE Labs

Re: Patch?: linux-2.5.33/drivers/input/keyboard/atkbd.c allow SETLEDS to fail

[Adam J. Richter]

[-- Attachment #1: Type: text/plain, Size: 8784 bytes --]

On Mon, Sep 09, 2002 at 05:20:04PM +0200, Vojtech Pavlik wrote:
> There are not. If SETLEDS fails, then there is the GETID command, which
> is also allowed to fail (fails on AT (non-PS/2) keyboards, which are
> still common), and it most likely also fails on your keyboard. That's
> all.
> 
> If you kill the SETLEDS command, then even without a keyboard, the
> keyboard will be detected, and sent commands, which will have to time
> out every time.

	100 milliseconds, which normally only happens when the user
hits CapsLock, NumLock or ScrollLock, which generally implies that a
keyboard is present.  I suspect that you could change it to use mdelay
instead of udelay, so that other processes could do useful work during
the wait, which might be relevant even with a keyboard plugged in that
just happens to be slow if one is running a screen saver that cycles
the keyboard LED's.  If it's important, you could also set a flag
when SETLEDS fails and not attempt further SETLEDS until another
sign of life from the keyboard is received (which would also address
the case of someone unplugging the keyboard after initialization).

	I also noticed a problem with plugging in a PS/2 keyboard
after initialization where it would not work, at least when I also had
the USB-to-PS/2 emulation enabled (if I booted with PS/2 keyboard
plugged in, both the USB and the PS/2 keyboards would work).  I have
not checked the purely non-USB case for whether your code works for
loading the module without the keyboard present and then plugging the
keyboard in later.

	Without the USB-to-PS/2 emulation on my VD133 motherboard, I
can't type into lilo.  Perhaps newer BIOSes have USB keyboard input
without need for hardware emulation, but I imagine that there are
quite a few computers with this problem.  That would be OK if there
were some way for Linux to turn off USB-to-PS/2 keyboard emulation
once the Linux USB code is loaded.  Currently, I get an infinite loop
of carriage returns when I do "modprobe hid" if I am using USB-to-PS/2
keyboard emulation.

	I agree that USB-to-PS/2 seems to be an ugly kludge, but I
think that the small cost due to orphaning certain oddball hardware
combinations is greater than the even smaller benefit of faster
SETLEDs (which I think could be achieved in other ways).

	The rest of this email is just about the PCI interrupt routing.

> > >I assume you have an USB keyboard, and the BIOS emulates a PS/2 keyboard
> > >for non-USB capable OSes.
> > 
> > >Well, fix Linux irq router code, instead of damaging the keyboard code.
> > 
> > 	How is the Linux IRQ routing code broken?  As far as I know,
> > it only reads the information given to it and cannot reprogram the
> > hardware.
> 
> It can.

	Thanks for the pointers to arch/i386/pci/irq.c and
drivers/pci/quirks.c.  This code programs interrupt routing in a few
special cases and arch/i386/pci/irq.c has somewhat general underlying
facilities that could be put to more general use, but I do not see any
existing programming interface by which a user level program could
somehow get the kernel to try, for example, to reprogram pci device
00:07.2 to IRQ 10.


> > 	My understanding is that the BIOS is responsible for setting
> > up the IRQ routes of each device and writing them into each device's
> > PCI configuration register dword 15 byte 0 (INTERRUPT_LINE), which is
> > normally just 8 bits of RAM.
> 
> Yes, but because the BIOS often writes nonsense to these config
> registers, Linux knows several southbridges and can fix the routing by
> hand.
> 
> > The mechanism for *making* those routes
> > is vendor specific and handled by the BIOS at boot time.  In the
> > particular case of my VD133, it apparently uses a Via VT82C598 PCI
> > bridge ("Apollo MVP3 AGP"), which I think is the component that would
> > have the secret bits for setting up routing from the main PCI bus to
> > the interrupt controllers.
> 
> Most likely the interrupt routing is controlled by the southbridge (the
> ISA bridge device). See arch/i386/pci/irq.c. Maybe it's as easy as an
> entry for your southbridge is missing.
> 
> > And those bits do appear to be secret.
> 
> Not really. Most of them are actually documented. Namely for built-in
> devices like USB controllers, because those cannot be wired differently
> by different board manufacturers.

	There appears to be more black magic involved than is covered
by drivers/pci/quirks.c and arch/i386/pci/irq.c.  Thanks for the
pointers.  From the files that you mentioned and a lot of
experimentation of setpci and a minor kernel change, I have determined
that when I have my natsemi ethernet card plugged in, then, and *only
then*, I can get the USB controller to use the same IRQ as the
ethernet card, and, with the attached modification to
arch/i386/pci/irq.c, I can get the kernel to use that IRQ even though
there is no pirq entry for it.  By the way, this isn't just the USB
controller being polled whenever an ethernet packet arrives.  I've
verified that it works with the ethernet cable unplugged, as long as
the card is present.  However, without the ethernet card installed,
the only way I have gotten the USB keyboard to work is with the
USB-to-PS/2 emulation.  Right now, I have the ethernet card installed
and am composing this email on that USB keyboard via the Linux USB
drivers.  Thanks again particularly for the pointer to
arch/i386/pci/irq.c.

	Here are the details of what I found, as the information may
be useful for other purposes in the future.

	pirq_via_{get,set} in arch/i386/pci/irq.c show a general
mechanism for assigning IRQ's though the Via south bridge chip.
dev_perq_info->irq[pin].link is an index into a table of 4 bit
nibbles starting at PCI configuration register 0x55 of the south
bridge.  You write the IRQ that you want to route to into the
appropriate nibble.  Here is an example from my system, with some
comments in brackets:

	setpci -v -s 00:07.0 55 56 57 58
	00:07.0:55 = c0					[ pirq#1 --> IRQ 12 ]
	00:07.0:56 = 0b					[ pirq#2 --> IRQ 11 ]
	00:07.0:57 = 00
	00:07.0:58 = 00

	I can change the ethernet's IRQ routing from 11 to, say, 9,
and watch the ethernet immediately stop, by doing something like
"setpci -s 00:07.0 56=09".

	However, the built-in USB on the VT82C586B does not seem to use
this mechanism.  There is no PIRQ entry for the pin that the USB controller
uses, and I cannot, for example, get the USB interrupts to show up on IRQ
ten by filling all of the southbridge configuration registers from 0x55
through 0x58 with 0xaa.

	It seems that the built-in USB on the VT82C586B has the same
feature as is described for the Via 686A/B in quirk_via_irqpic in
drivers/pci/quirks.c:

 * Via 686A/B:  The PCI_INTERRUPT_LINE register for the on-chip
 * devices, USB0/1, AC97, MC97, and ACPI, has an unusual feature:
 * when written, it makes an internal connection to the PIC.

	This seems to work on my VD133 motherboard also, but it seems
only to work if I route it to the interrupt being used by the ethernet
card.  It seems that there is some additional initialization needed
(and yes, I've checked that my choice of IRQ conforms that various IRQ
masks in arch/i386/pci/irq.c, and I've tried setting the south
bridge's configuration registers at 0x55 in conjunction with this.
So, it appears that there is some further interrupt routing
preparation that the kernel does not know about.

> Try the southbridge instead of the northbridge. I have several of the
> documents, and can send it to you if you tell me the right chip number.

	Great!  Here is output of lspci on my system.  I would be
interested in documents for any of the relevant chips.  Although I've
kludged around the problem enough for my own use, I'd prefer to
develop a more general and correct fix for other VD133 users at least
and perhaps other boards if they have a similar BIOS problem.

00:00.0 Host bridge: VIA Technologies, Inc. VT82C691 [Apollo PRO] (rev 44)
00:01.0 PCI bridge: VIA Technologies, Inc. VT82C598 [Apollo MVP3 AGP]
00:07.0 ISA bridge: VIA Technologies, Inc. VT82C596 ISA [Apollo PRO] (rev 12)
00:07.1 IDE interface: VIA Technologies, Inc. VT82C586 IDE [Apollo] (rev 06)
00:07.2 USB Controller: VIA Technologies, Inc. VT82C586B USB (rev 08)
00:07.3 Host bridge: VIA Technologies, Inc.: Unknown device 3050 (rev 20)
00:10.0 Ethernet controller: National Semiconductor Corporation DP83815 10/100Mbps Ethernet with Wake on LAN
01:00.0 VGA compatible controller: ATI Technologies Inc 3D Rage Pro AGP AGP 1X/2X (rev 5c)


	Thanks for your help.

-- 
Adam J. Richter     __     ______________   575 Oroville Road
adam@yggdrasil.com     \ /                  Milpitas, California 95035
+1 408 309-6081         | g g d r a s i l   United States of America
                         "Free Software For The Rest Of Us."

[-- Attachment #2: irq.route --]
[-- Type: text/plain, Size: 546 bytes --]

--- linux-2.5.33/arch/i386/pci/irq.c	2002-08-31 15:05:37.000000000 -0700
+++ linux/arch/i386/pci/irq.c	2002-09-10 05:29:09.000000000 -0700
@@ -591,8 +591,13 @@
 	pirq = info->irq[pin].link;
 	mask = info->irq[pin].bitmap;
 	if (!pirq) {
-		DBG(" -> not routed\n");
-		return 0;
+		pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &dev->irq);
+		if (dev->irq)
+			return 1;
+		else {
+			DBG(" -> not routed\n");
+			return 0;
+		}
 	}
 	DBG(" -> PIRQ %02x, mask %04x, excl %04x", pirq, mask, pirq_table->exclusive_irqs);
 	mask &= pcibios_irq_mask;

Re: Patch?: linux-2.5.33/drivers/input/keyboard/atkbd.c allow SETLEDS to fail

[Vojtech Pavlik]

On Tue, Sep 10, 2002 at 09:05:20AM -0700, Adam J. Richter wrote:
> On Mon, Sep 09, 2002 at 05:20:04PM +0200, Vojtech Pavlik wrote:
> > There are not. If SETLEDS fails, then there is the GETID command, which
> > is also allowed to fail (fails on AT (non-PS/2) keyboards, which are
> > still common), and it most likely also fails on your keyboard. That's
> > all.
> > 
> > If you kill the SETLEDS command, then even without a keyboard, the
> > keyboard will be detected, and sent commands, which will have to time
> > out every time.
> 
> 	100 milliseconds, which normally only happens when the user
> hits CapsLock, NumLock or ScrollLock, which generally implies that a
> keyboard is present.

100 milliseconds is a long time. When a keyboard is present, it replies
much faster than that, usually around 1-4 ms.

> I suspect that you could change it to use mdelay
> instead of udelay, so that other processes could do useful work during
> the wait,

Actually, mdelay just calls udelay several times, processes are stopped.
I could do a sleep, but that'd imply quite complex code involving a
kernel thread most likely - keypresses come from interrupt context.

> which might be relevant even with a keyboard plugged in that
> just happens to be slow if one is running a screen saver that cycles
> the keyboard LED's.  If it's important, you could also set a flag
> when SETLEDS fails and not attempt further SETLEDS until another
> sign of life from the keyboard is received (which would also address
> the case of someone unplugging the keyboard after initialization).

We do this for USB keyboards, it might make sense for AT keyboards as
well, but only in case the LED thread is implemented.

> 	I also noticed a problem with plugging in a PS/2 keyboard
> after initialization where it would not work, at least when I also had
> the USB-to-PS/2 emulation enabled (if I booted with PS/2 keyboard
> plugged in, both the USB and the PS/2 keyboards would work).  I have
> not checked the purely non-USB case for whether your code works for
> loading the module without the keyboard present and then plugging the
> keyboard in later.

Autoloading the modules doesn't work yet, but if the modules are loaded
(or better compiled in, as keyboard support used to be), then a keyboard
is autodetected after plugging it in, even if you connect it to the
mouse port.

> 	Without the USB-to-PS/2 emulation on my VD133 motherboard, I
> can't type into lilo. 

Then leave it enabled, it's not a problem.

> Perhaps newer BIOSes have USB keyboard input
> without need for hardware emulation, but I imagine that there are
> quite a few computers with this problem.

Actually the USB code in the Linux kernel does a thing called "USB
emulation takeover", where it takes the USB controller over from BIOS
and disables the PS/2 emulation. This way you can have your keyboard
both in LILO and in Linux using a proper USB driver.

> That would be OK if there
> were some way for Linux to turn off USB-to-PS/2 keyboard emulation
> once the Linux USB code is loaded. 

This is what happens.

> Currently, I get an infinite loop
> of carriage returns when I do "modprobe hid" if I am using USB-to-PS/2
> keyboard emulation.

This is interesting. I never had a report like that. Care to supply more
detail? Also this would mean you have the irqs assigned correctly now?

> 	I agree that USB-to-PS/2 seems to be an ugly kludge, but I
> think that the small cost due to orphaning certain oddball hardware
> combinations is greater than the even smaller benefit of faster
> SETLEDs (which I think could be achieved in other ways).

I'll be adding a switch to not communicate with the keyboard and only
accept keystrokes. This will solve your problem. But I still think it's
better to fix the other problems you have and have the hardware operate
in the intended way.

> 	The rest of this email is just about the PCI interrupt routing.
> 
> > > >I assume you have an USB keyboard, and the BIOS emulates a PS/2 keyboard
> > > >for non-USB capable OSes.
> > > 
> > > >Well, fix Linux irq router code, instead of damaging the keyboard code.
> > > 
> > > 	How is the Linux IRQ routing code broken?  As far as I know,
> > > it only reads the information given to it and cannot reprogram the
> > > hardware.
> > 
> > It can.
> 
> 	Thanks for the pointers to arch/i386/pci/irq.c and
> drivers/pci/quirks.c.  This code programs interrupt routing in a few
> special cases and arch/i386/pci/irq.c has somewhat general underlying
> facilities that could be put to more general use, but I do not see any
> existing programming interface by which a user level program could
> somehow get the kernel to try, for example, to reprogram pci device
> 00:07.2 to IRQ 10.

It's done at boot time, so it's not designed to be programmed by
userspace. However, the usb driver does pci_enable_device(), where the
PCI code will assign an IRQ to it if the BIOS didn't.

> > > 	My understanding is that the BIOS is responsible for setting
> > > up the IRQ routes of each device and writing them into each device's
> > > PCI configuration register dword 15 byte 0 (INTERRUPT_LINE), which is
> > > normally just 8 bits of RAM.
> > 
> > Yes, but because the BIOS often writes nonsense to these config
> > registers, Linux knows several southbridges and can fix the routing by
> > hand.
> > 
> > > The mechanism for *making* those routes
> > > is vendor specific and handled by the BIOS at boot time.  In the
> > > particular case of my VD133, it apparently uses a Via VT82C598 PCI
> > > bridge ("Apollo MVP3 AGP"), which I think is the component that would
> > > have the secret bits for setting up routing from the main PCI bus to
> > > the interrupt controllers.
> > 
> > Most likely the interrupt routing is controlled by the southbridge (the
> > ISA bridge device). See arch/i386/pci/irq.c. Maybe it's as easy as an
> > entry for your southbridge is missing.
> > 
> > > And those bits do appear to be secret.
> > 
> > Not really. Most of them are actually documented. Namely for built-in
> > devices like USB controllers, because those cannot be wired differently
> > by different board manufacturers.
> 
> 	There appears to be more black magic involved than is covered
> by drivers/pci/quirks.c and arch/i386/pci/irq.c.  Thanks for the
> pointers.  From the files that you mentioned and a lot of
> experimentation of setpci and a minor kernel change, I have determined
> that when I have my natsemi ethernet card plugged in, then, and *only
> then*, I can get the USB controller to use the same IRQ as the
> ethernet card, and, with the attached modification to
> arch/i386/pci/irq.c, I can get the kernel to use that IRQ even though
> there is no pirq entry for it.  By the way, this isn't just the USB
> controller being polled whenever an ethernet packet arrives.  I've
> verified that it works with the ethernet cable unplugged, as long as
> the card is present.  However, without the ethernet card installed,
> the only way I have gotten the USB keyboard to work is with the
> USB-to-PS/2 emulation.  Right now, I have the ethernet card installed
> and am composing this email on that USB keyboard via the Linux USB
> drivers.  Thanks again particularly for the pointer to
> arch/i386/pci/irq.c.

Glad to hear you got it to work. More comments below.

> 	Here are the details of what I found, as the information may
> be useful for other purposes in the future.
> 
> 	pirq_via_{get,set} in arch/i386/pci/irq.c show a general
> mechanism for assigning IRQ's though the Via south bridge chip.
> dev_perq_info->irq[pin].link is an index into a table of 4 bit
> nibbles starting at PCI configuration register 0x55 of the south
> bridge.  You write the IRQ that you want to route to into the
> appropriate nibble.  Here is an example from my system, with some
> comments in brackets:
> 
> 	setpci -v -s 00:07.0 55 56 57 58
> 	00:07.0:55 = c0					[ pirq#1 --> IRQ 12 ]
> 	00:07.0:56 = 0b					[ pirq#2 --> IRQ 11 ]
> 	00:07.0:57 = 00
> 	00:07.0:58 = 00
> 
> 	I can change the ethernet's IRQ routing from 11 to, say, 9,
> and watch the ethernet immediately stop, by doing something like
> "setpci -s 00:07.0 56=09".
> 
> 	However, the built-in USB on the VT82C586B does not seem to use
> this mechanism.  There is no PIRQ entry for the pin that the USB controller
> uses, and I cannot, for example, get the USB interrupts to show up on IRQ
> ten by filling all of the southbridge configuration registers from 0x55
> through 0x58 with 0xaa.
> 
> 	It seems that the built-in USB on the VT82C586B has the same
> feature as is described for the Via 686A/B in quirk_via_irqpic in
> drivers/pci/quirks.c:
> 
>  * Via 686A/B:  The PCI_INTERRUPT_LINE register for the on-chip
>  * devices, USB0/1, AC97, MC97, and ACPI, has an unusual feature:
>  * when written, it makes an internal connection to the PIC.

Indeed, the 586b specification says that writing ofset 0x3c of the USB
controller with the irq value causes USB irq to be routed to the irq
written.

> 	This seems to work on my VD133 motherboard also, but it seems
> only to work if I route it to the interrupt being used by the ethernet
> card.

It most likely will work when shared with any PCI interrupt.

> It seems that there is some additional initialization needed

Couldn't find anything like that in the docs. Maybe the other irqs are
used for something else, non-shareable. Maybe you can set them as ISA in
BIOS setup and then the BIOS won't assign anything to them.

> (and yes, I've checked that my choice of IRQ conforms that various IRQ
> masks in arch/i386/pci/irq.c, and I've tried setting the south
> bridge's configuration registers at 0x55 in conjunction with this.
> So, it appears that there is some further interrupt routing
> preparation that the kernel does not know about.

The 586b register description doesn't mention anything else.

> > Try the southbridge instead of the northbridge. I have several of the
> > documents, and can send it to you if you tell me the right chip number.
> 
> 	Great!  Here is output of lspci on my system.  I would be
> interested in documents for any of the relevant chips.  Although I've
> kludged around the problem enough for my own use, I'd prefer to
> develop a more general and correct fix for other VD133 users at least
> and perhaps other boards if they have a similar BIOS problem.
> 
> 00:00.0 Host bridge: VIA Technologies, Inc. VT82C691 [Apollo PRO] (rev 44)
> 00:01.0 PCI bridge: VIA Technologies, Inc. VT82C598 [Apollo MVP3 AGP]
> 00:07.0 ISA bridge: VIA Technologies, Inc. VT82C596 ISA [Apollo PRO] (rev 12)
> 00:07.1 IDE interface: VIA Technologies, Inc. VT82C586 IDE [Apollo] (rev 06)
> 00:07.2 USB Controller: VIA Technologies, Inc. VT82C586B USB (rev 08)
> 00:07.3 Host bridge: VIA Technologies, Inc.: Unknown device 3050 (rev 20)
> 00:10.0 Ethernet controller: National Semiconductor Corporation DP83815 10/100Mbps Ethernet with Wake on LAN
> 01:00.0 VGA compatible controller: ATI Technologies Inc 3D Rage Pro AGP AGP 1X/2X (rev 5c)
> 
> 
> 	Thanks for your help.

586b. Quite old. I have that document. You can get it from
http://twilight.ucw.cz/vt82c586b.pdf

> --- linux-2.5.33/arch/i386/pci/irq.c	2002-08-31 15:05:37.000000000 -0700
> +++ linux/arch/i386/pci/irq.c	2002-09-10 05:29:09.000000000 -0700
> @@ -591,8 +591,13 @@
>  	pirq = info->irq[pin].link;
>  	mask = info->irq[pin].bitmap;
>  	if (!pirq) {
> -		DBG(" -> not routed\n");
> -		return 0;
> +		pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &dev->irq);
> +		if (dev->irq)
> +			return 1;
> +		else {
> +			DBG(" -> not routed\n");
> +			return 0;
> +		}
>  	}
>  	DBG(" -> PIRQ %02x, mask %04x, excl %04x", pirq, mask, pirq_table->exclusive_irqs);
>  	mask &= pcibios_irq_mask;


-- 
Vojtech Pavlik
SuSE Labs

Re: Patch?: linux-2.5.33/drivers/input/keyboard/atkbd.c allow SETLEDS to fail

[Horst von Brand]

"Adam J. Richter" <adam@yggdrasil.com> said:

[...]

> 	100 milliseconds, which normally only happens when the user
> hits CapsLock, NumLock or ScrollLock, which generally implies that a
> keyboard is present.  I suspect that you could change it to use mdelay
> instead of udelay, so that other processes could do useful work during
> the wait,

udelay() is busy wait, mdelay() is just calling udelay(1000) over and over.
-- 
Dr. Horst H. von Brand                   User #22616 counter.li.org
Departamento de Informatica                     Fono: +56 32 654431
Universidad Tecnica Federico Santa Maria              +56 32 654239
Casilla 110-V, Valparaiso, Chile                Fax:  +56 32 797513

Re: Patch?: linux-2.5.33/drivers/input/keyboard/atkbd.c allow SETLEDS to fail

[Vojtech Pavlik]

On Mon, Sep 09, 2002 at 06:51:11AM -0700, Adam J. Richter wrote:

> 	I have a computer with an Iwill VD133 motherboard doing
> USB-to-PS/2 keyboard emulation (built into the chipset somewhere)
> for a BTC 7932M USB keyboard.  Under this configuration, the
> SETLEDS command in atkbd_probe fails (the first atkbd_sendbyte
> in atkbd_command fails), but the keyboard otherwise works if
> that failure is ignored.
> 
> 	I noticed this when my keyboard stopped working in 2.5.32.
> I have verified that 2.5.31 (and probably all kernels before it)
> also do not set the LEDs on my USB-emulating-PS/2 keyboard.
> 
> 	I cannot say whether the problem occurs with other USB
> keyboards because I don't have one handy.
> 
> 	One cannot run the USB keyboard "natively" via Linux USB on this
> motherboard because the BIOS does not seem to set an IRQ line for the USB
> controller.  I've tried booting with "pci=biosirq", and I think I've tried
> all the relevant BIOS menu options, and I'm running the latest BIOS,
> and the VD133 product has been terminated.  So the patch does seem to be
> "necessary" to support some configurations, at least for now.

I cannot do that, since then the keyboard identification will give too
many positives.

I assume you have an USB keyboard, and the BIOS emulates a PS/2 keyboard
for non-USB capable OSes.

Well, fix Linux irq router code, instead of damaging the keyboard code.

The mainboard has VIA Apollo Pro133 chipset - it should be fairly easily
possible to change the IRQ routes in the pci-irq.c code for the USB
controller.

> --- linux-2.5.33/drivers/input/keyboard/atkbd.c	2002-08-31 15:05:31.000000000 -0700
> +++ linux/drivers/input/keyboard/atkbd.c	2002-09-09 06:29:04.000000000 -0700
> @@ -359,13 +359,13 @@
>  #endif
>  
>  /*
> - * Next we check we can set LEDs on the keyboard. This should work on every
> - * keyboard out there. It also turns the LEDs off, which we want anyway.
> + * Turn off LEDs.  This command fails on at least a BTC 7932M USB keyboard
> + * connected to an Iwill VD133 motherboard that is configured to emulate
> + * a PS/2 keyboard via USB.
>   */
>  
>  	param[0] = 0;
> -	if (atkbd_command(atkbd, param, ATKBD_CMD_SETLEDS))
> -		return -1;
> +	atkbd_command(atkbd, param, ATKBD_CMD_SETLEDS);
>  
>  /*
>   * Then we check the keyboard ID. We should get 0xab83 under normal conditions.


-- 
Vojtech Pavlik
SuSE Labs

Patch?: linux-2.5.33/drivers/input/keyboard/atkbd.c allow SETLEDS to fail

[Adam J. Richter]

[-- Attachment #1: Type: text/plain, Size: 1306 bytes --]

	I have a computer with an Iwill VD133 motherboard doing
USB-to-PS/2 keyboard emulation (built into the chipset somewhere)
for a BTC 7932M USB keyboard.  Under this configuration, the
SETLEDS command in atkbd_probe fails (the first atkbd_sendbyte
in atkbd_command fails), but the keyboard otherwise works if
that failure is ignored.

	I noticed this when my keyboard stopped working in 2.5.32.
I have verified that 2.5.31 (and probably all kernels before it)
also do not set the LEDs on my USB-emulating-PS/2 keyboard.

	I cannot say whether the problem occurs with other USB
keyboards because I don't have one handy.

	One cannot run the USB keyboard "natively" via Linux USB on this
motherboard because the BIOS does not seem to set an IRQ line for the USB
controller.  I've tried booting with "pci=biosirq", and I think I've tried
all the relevant BIOS menu options, and I'm running the latest BIOS,
and the VD133 product has been terminated.  So the patch does seem to be
"necessary" to support some configurations, at least for now.

-- 
Adam J. Richter     __     ______________   575 Oroville Road
adam@yggdrasil.com     \ /                  Milpitas, California 95035
+1 408 309-6081         | g g d r a s i l   United States of America
                         "Free Software For The Rest Of Us."

[-- Attachment #2: kbd.diff --]
[-- Type: text/plain, Size: 755 bytes --]

--- linux-2.5.33/drivers/input/keyboard/atkbd.c	2002-08-31 15:05:31.000000000 -0700
+++ linux/drivers/input/keyboard/atkbd.c	2002-09-09 06:29:04.000000000 -0700
@@ -359,13 +359,13 @@
 #endif
 
 /*
- * Next we check we can set LEDs on the keyboard. This should work on every
- * keyboard out there. It also turns the LEDs off, which we want anyway.
+ * Turn off LEDs.  This command fails on at least a BTC 7932M USB keyboard
+ * connected to an Iwill VD133 motherboard that is configured to emulate
+ * a PS/2 keyboard via USB.
  */
 
 	param[0] = 0;
-	if (atkbd_command(atkbd, param, ATKBD_CMD_SETLEDS))
-		return -1;
+	atkbd_command(atkbd, param, ATKBD_CMD_SETLEDS);
 
 /*
  * Then we check the keyboard ID. We should get 0xab83 under normal conditions.