lkml.org 
[lkml]   [2004]   [Aug]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
Subject[PATCH] Cardbus card memory assignment on x86_64 (if base==maxbase)
Hi,
I've sent the patch below to Ivan Kokshaysky (bk login: ink) and got
a mail from him back that I should forward it and telling that it looks
correct. According to bitkeeper logs for 2.4 he initially did the pci_size
function this patch is about.

I've debugged a memory assignment problem which happens even with 2.6.8.1
on Amilo Laptops from Siemens Fujitsu with a RT2500 CardBus card.

The problem was that the PCI scan didn't report a memory range for the
card and thus the driver finally complained about not beind able to remap
the memory.

After plaching a dump_stack() into bus_add_device() to know thich
functions were really involved, I worked my way forward into the PCI
code.

Enabling the debug statement in the PCI code, I've found pci_setup_device().
From there I added more printk's. Ultimatively, I've found that pci_size()
returned 0 for the memory size at this place in drivers/pci/probe.c:

sz = pci_size(l, sz, PCI_BASE_ADDRESS_MEM_MASK);
if (!sz)
continue;

Here is the pci_size():

/*
* Find the extent of a PCI decode..
*/
static u32 pci_size(u32 base, u32 maxbase, unsigned long mask)
{
u32 size = mask & maxbase; /* Find the significant bits */
if (!size)
return 0;

/* Get the lowest of them to find the decode size, and
from that the extent. */
size = (size & ~(size-1)) - 1;

/* base == maxbase can be valid only if the BAR has
already been programmed with all 1s. */
if (base == maxbase && ((base | size) & mask) != mask)
return 0;

return size;
}

Adding the results of my debug, it's called like this - for my card:
pci_size(u32 base = ffffe000, maxbase = ffffe000, mask = fffffffffffffff0)

Finally the CPU reaches this line:
if (base == maxbase && ((base | size) & mask) != mask)
return 0;

With the values from my debug printks, it looks like this:

if (ffffe000 == ffffe000
&& ((ffffe000 | 1fff) & fffffffffffffff0) != fffffffffffffff0)

Evaluation of the if:

1): ffffe000 == ffffe000 is true

2): (ffffe000 | 1fff) & fffffffffffffff0) results in fffffff0

but this means 2) will always != fffffffffffffff0.
(it's a large #defined mask)

It can never be equal, because "(base | size) & mask)" (base and size 32bit)
- the result of ((32-bit | 32-bit) & 64-bit)) can never match a large 64-bit
value such as fffffffffffffff0.

So, on 64-bit, calls to

static u32 pci_size(u32 base, u32 maxbase, unsigned long mask)

with base == maxbase and mask = ~0x0fUL will always result in
return 0.

I think that's the bug here, this function is intented to work
on 32-bit PCI values.

PCI64 is handled in a separate way, in pci_read_bases(), just a few
lines after pci_size() is called:

#if BITS_PER_LONG == 64
res->start |= ((unsigned long) l) << 32;
res->end = res->start + sz;
pci_write_config_dword(dev, reg+4, ~0);
pci_read_config_dword(dev, reg+4, &sz);
pci_write_config_dword(dev, reg+4, l);
if (~sz)
res->end = res->start + 0xffffffff +
(((unsigned long) ~sz) << 32);

I believe, pci_size() should work with 32-bit vaules
only and I've tested that the diff of

-static u32 pci_size(u32 base, u32 maxbase, unsigned long mask)
+static u32 pci_size(u32 base, u32 maxbase, u32 mask)

makes my card's memory appear as it does in a 32-bit kernel where mask
is a 32-bit value too.

It seems to be really a corner case, I don't know if many other
Cards provide such PCI config space where base == maxbase,
but there could be some others out there as well.

For these cards, the above diff fixes the memory assignment problem.

I hope this mail helps some of you, the patch is attached, I'll forward
this to the kernel maintainers.

Bernhard--- drivers/pci/probe.c 2004/08/23 14:36:04 1.1
+++ drivers/pci/probe.c 2004/08/23 14:36:21
@@ -82,7 +82,7 @@
/*
* Find the extent of a PCI decode..
*/
-static u32 pci_size(u32 base, u32 maxbase, unsigned long mask)
+static u32 pci_size(u32 base, u32 maxbase, u32 mask)
{
u32 size = mask & maxbase; /* Find the significant bits */
if (!size)
\
 
 \ /
  Last update: 2005-03-22 14:05    [W:0.025 / U:0.204 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site