Hang in yenta socket driver in 2.4.x w/ PCI->PCMCIA adapter

Hang in yenta socket driver in 2.4.x w/ PCI->PCMCIA adapter

[David Moore]

Hi, I'm using kernel 2.4.9, but have observed this problem with all 2.4.x
kernels I've tried.  I have a PCI card with a single PCMCIA slot in it.
It's has a TI Cardbus controller which is detected successfully and
assigned IRQs without trouble.  I have also used it successfully with the
pcmcia-cs modules and PCMCIA support not enabled in the main kernel
sources.

I get the following messages upon loading the pcmcia_core, yenta_socket,
and ds modules:

PCI: Enabling device 01:09.1 (0000 -> 0002)
PCI: Enabling device 01:09.0 (0000 -> 0002)
Yenta IRQ list 0000, PCI irq5
Socket status: 30000010
Yenta IRQ list 0000, PCI irq5
Socket status: 30000006
cs: socket c79bb800 timed out during reset.  Try increasing setup_delay.

Then, immediately after I start the cardmgr daemon, the system hangs
solidly (cannot switch consoles or scroll up).  Upon closer inspection,
socket 0 is not attached to a physical PCMCIA slot, and is the one timing
out above.  Socket 1 is the only real slot.  As a workaround, I modified
cardmgr so that it only initializes socket 1, and avoids socket 0
completely.  When I do this, everything works perfectly and I can use
cards in the socket.  (It should be noted that the hang occurs even
without any cards in the socket).  Also, I doubt this is an IRQ problem
because irq 5 (assigned above) is not used by another device as reported
by lspci.

Since the pcmcia-cs modules work (even with kernel 2.4.x) I know that it's
theoretically possible to have socket 0 fail gracefully rather than hang
the system when it's initialized, even though it doesn't really exist.

Any ideas how to fix this?

I have attached the output when debugging is enabled in yenta_socket (this
is with cardmgr loading the modules itself):

cb_readl: c88c6740 0008 30001818
exca_readb: c88c6740 0001 4c
exca_readb: c88c6740 0003 50
^-- these three lines repeat for a while....
cb_readl: c88c6740 0008 30001818
exca_readb: c88c6740 0001 4c
exca_readb: c88c6740 0003 50
cs: socket c79a3800 timed out during reset.  Try increasing setup_delay.
cb_readl: c88c67c8 0008 30000086
exca_readb: c88c67c8 0001 00
exca_readb: c88c67c8 0003 50
exca_writeb: c88c6740 0040 a0
exca_writew: c88c6740 0010 0000
exca_writew: c88c6740 0012 8000
exca_writew: c88c6740 0014 4000
exca_writeb: c88c6740 0006 01 <------------------- hangs here

The last few lines look like they are from the mem_map function, but I'm
not sure if the hang occurs immediately at this point, or is caused by
code somewhere else.

Thanks for the help.

Best Regards,

David Moore

Re: Hang in yenta socket driver in 2.4.x w/ PCI->PCMCIA adapter

[Jim Roland]

Does it work (not lock-up) with your PC card removed?  Have you verified 
(beyond a shadow of a doubt) that nothing else is using IRQ5?

David Moore wrote:

>Hi, I'm using kernel 2.4.9, but have observed this problem with all 2.4.x
>kernels I've tried.  I have a PCI card with a single PCMCIA slot in it.
>It's has a TI Cardbus controller which is detected successfully and
>assigned IRQs without trouble.  I have also used it successfully with the
>pcmcia-cs modules and PCMCIA support not enabled in the main kernel
>sources.
>
>I get the following messages upon loading the pcmcia_core, yenta_socket,
>and ds modules:
>
>PCI: Enabling device 01:09.1 (0000 -> 0002)
>PCI: Enabling device 01:09.0 (0000 -> 0002)
>Yenta IRQ list 0000, PCI irq5
>Socket status: 30000010
>Yenta IRQ list 0000, PCI irq5
>Socket status: 30000006
>cs: socket c79bb800 timed out during reset.  Try increasing setup_delay.
>
>Then, immediately after I start the cardmgr daemon, the system hangs
>solidly (cannot switch consoles or scroll up).  Upon closer inspection,
>socket 0 is not attached to a physical PCMCIA slot, and is the one timing
>out above.  Socket 1 is the only real slot.  As a workaround, I modified
>cardmgr so that it only initializes socket 1, and avoids socket 0
>completely.  When I do this, everything works perfectly and I can use
>cards in the socket.  (It should be noted that the hang occurs even
>without any cards in the socket).  Also, I doubt this is an IRQ problem
>because irq 5 (assigned above) is not used by another device as reported
>by lspci.
>
>Since the pcmcia-cs modules work (even with kernel 2.4.x) I know that it's
>theoretically possible to have socket 0 fail gracefully rather than hang
>the system when it's initialized, even though it doesn't really exist.
>
>Any ideas how to fix this?
>
>I have attached the output when debugging is enabled in yenta_socket (this
>is with cardmgr loading the modules itself):
>
>cb_readl: c88c6740 0008 30001818
>exca_readb: c88c6740 0001 4c
>exca_readb: c88c6740 0003 50
>^-- these three lines repeat for a while....
>cb_readl: c88c6740 0008 30001818
>exca_readb: c88c6740 0001 4c
>exca_readb: c88c6740 0003 50
>cs: socket c79a3800 timed out during reset.  Try increasing setup_delay.
>cb_readl: c88c67c8 0008 30000086
>exca_readb: c88c67c8 0001 00
>exca_readb: c88c67c8 0003 50
>exca_writeb: c88c6740 0040 a0
>exca_writew: c88c6740 0010 0000
>exca_writew: c88c6740 0012 8000
>exca_writew: c88c6740 0014 4000
>exca_writeb: c88c6740 0006 01 <------------------- hangs here
>
>The last few lines look like they are from the mem_map function, but I'm
>not sure if the hang occurs immediately at this point, or is caused by
>code somewhere else.
>
>Thanks for the help.
>
>Best Regards,
>
>David Moore
>
>-
>To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>the body of a message to majordomo@vger.kernel.org
>More majordomo info at  http://vger.kernel.org/majordomo-info.html
>Please read the FAQ at  http://www.tux.org/lkml/
>

Re: Hang in yenta socket driver in 2.4.x w/ PCI->PCMCIA adapter

[David Moore]

The lock-up occurs even if there is no PC card in the slot.  In its most
verbose mode, lspci reports no other PCI device using IRQ5, so I'm fairly
confident it's not an IRQ problem.  I'm further convinced of this because
the PCMCIA slot works perfectly if I hack cardmgr to ignore socket 0
(which has no physical slot attached) and only initialize socket 1 (which
is the one card slot on the PCI card).

Regards,

David

On Thu, 30 Aug 2001, Jim Roland wrote:

> Does it work (not lock-up) with your PC card removed?  Have you verified
> (beyond a shadow of a doubt) that nothing else is using IRQ5?
>
> David Moore wrote:
>
> >Hi, I'm using kernel 2.4.9, but have observed this problem with all 2.4.x
> >kernels I've tried.  I have a PCI card with a single PCMCIA slot in it.
> >It's has a TI Cardbus controller which is detected successfully and
> >assigned IRQs without trouble.  I have also used it successfully with the
> >pcmcia-cs modules and PCMCIA support not enabled in the main kernel
> >sources.
> >
> >I get the following messages upon loading the pcmcia_core, yenta_socket,
> >and ds modules:
> >
> >PCI: Enabling device 01:09.1 (0000 -> 0002)
> >PCI: Enabling device 01:09.0 (0000 -> 0002)
> >Yenta IRQ list 0000, PCI irq5
> >Socket status: 30000010
> >Yenta IRQ list 0000, PCI irq5
> >Socket status: 30000006
> >cs: socket c79bb800 timed out during reset.  Try increasing setup_delay.
> >
> >Then, immediately after I start the cardmgr daemon, the system hangs
> >solidly (cannot switch consoles or scroll up).  Upon closer inspection,
> >socket 0 is not attached to a physical PCMCIA slot, and is the one timing
> >out above.  Socket 1 is the only real slot.  As a workaround, I modified
> >cardmgr so that it only initializes socket 1, and avoids socket 0
> >completely.  When I do this, everything works perfectly and I can use
> >cards in the socket.  (It should be noted that the hang occurs even
> >without any cards in the socket).  Also, I doubt this is an IRQ problem
> >because irq 5 (assigned above) is not used by another device as reported
> >by lspci.
> >
> >Since the pcmcia-cs modules work (even with kernel 2.4.x) I know that it's
> >theoretically possible to have socket 0 fail gracefully rather than hang
> >the system when it's initialized, even though it doesn't really exist.
> >
> >Any ideas how to fix this?
> >
> >I have attached the output when debugging is enabled in yenta_socket (this
> >is with cardmgr loading the modules itself):
> >
> >cb_readl: c88c6740 0008 30001818
> >exca_readb: c88c6740 0001 4c
> >exca_readb: c88c6740 0003 50
> >^-- these three lines repeat for a while....
> >cb_readl: c88c6740 0008 30001818
> >exca_readb: c88c6740 0001 4c
> >exca_readb: c88c6740 0003 50
> >cs: socket c79a3800 timed out during reset.  Try increasing setup_delay.
> >cb_readl: c88c67c8 0008 30000086
> >exca_readb: c88c67c8 0001 00
> >exca_readb: c88c67c8 0003 50
> >exca_writeb: c88c6740 0040 a0
> >exca_writew: c88c6740 0010 0000
> >exca_writew: c88c6740 0012 8000
> >exca_writew: c88c6740 0014 4000
> >exca_writeb: c88c6740 0006 01 <------------------- hangs here
> >
> >The last few lines look like they are from the mem_map function, but I'm
> >not sure if the hang occurs immediately at this point, or is caused by
> >code somewhere else.
> >
> >Thanks for the help.
> >
> >Best Regards,
> >
> >David Moore
> >
> >-
> >To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> >the body of a message to majordomo@vger.kernel.org
> >More majordomo info at  http://vger.kernel.org/majordomo-info.html
> >Please read the FAQ at  http://www.tux.org/lkml/
> >
>
>
>