On Mon, Apr 11, 2011 at 6:18 AM, Daniel Hellstrom <daniel@xxxxxxxxxxx> wrote:
>In commit 13583b16592a ("PCI: refactor io size calculation code"):
>>
>> refactor code that calculates the io size in pbus_size_io()
>> and pbus_mem_io() into separate functions.
>>
>> Signed-off-by: Ram Pai <linuxram@xxxxxxxxxx>
>> ---
>> drivers/pci/setup-bus.c | 66
>> ++++++++++++++++++++++++++++++-----------------
>> 1 files changed, 42 insertions(+), 24 deletions(-)
>>
>> diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
>> index 66cb8f4..2121215 100644
>> --- a/drivers/pci/setup-bus.c
>> +++ b/drivers/pci/setup-bus.c
>> @@ -404,6 +404,43 @@ static struct resource *find_free_bus_resource(struct
>> pci_bus *bus, unsigned lon
>> return NULL;
>> }
>>
> ...
>
>> +
>> +static resource_size_t calculate_memsize(resource_size_t size,
>> + resource_size_t min_size,
>> + resource_size_t size1,
>> + resource_size_t old_size,
>> + resource_size_t align)
>> +{
>> + if (size < min_size)
>> + size = min_size;
>> + if (old_size == 1 )
>> + old_size = 0;
>> + if (size < old_size)
>> + size = old_size;
>> + size = ALIGN(size + size1, align);
>> + return size;
>> +}
>> +
>> /* Sizing the IO windows of the PCI-PCI bridge is trivial,
>> since these windows have 4K granularity and the IO ranges
>> of non-bridge PCI devices are limited to 256 bytes.
>>
>>
>
> ...
>
>> @@ -516,14 +542,6 @@ static int pbus_size_mem(struct pci_bus *bus,
>> unsigned long mask,
>> mem64_mask &= r->flags & IORESOURCE_MEM_64;
>> }
>> }
>> - if (size < min_size)
>> - size = min_size;
>> - old_size = resource_size(b_res);
>> - if (old_size == 1)
>> - old_size = 0;
>> - if (size < old_size)
>> - size = old_size;
>> -
>> align = 0;
>> min_align = 0;
>> for (order = 0; order <= max_order; order++) {
>> @@ -537,7 +555,7 @@ static int pbus_size_mem(struct pci_bus *bus, unsigned
>> long mask,
>> min_align = align1 >> 1;
>> align += aligns[order];
>> }
>> - size = ALIGN(size, min_align);
>> + size = calculate_memsize(size, min_size, 0, resource_size(b_res),
>> align);
>>
>
> On my SPARC32/LEON4 PCI system I get overlapped areas, double mapped
> resources. Some BARs on PCIBUS0 are in the same non-prefetchable memory
> range as the secondary bus PCIBUS1. Changing align to min_align in the above
> call to calculate_memsize() fixes the problem, and the memory allocation is
> the same as with 2.6.36.4 kernel.
>
> I belive this is just a typo.
It does seem that way. The original code used 'min_align' and 'align'
itself is meaningless in that place.
Can you confirm that the patch you talk about as fixing things is the attached?
Linus
drivers/pci/setup-bus.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
index 89d0a6a88df7..ebf51ad1b714 100644
--- a/drivers/pci/setup-bus.c
+++ b/drivers/pci/setup-bus.c
@@ -676,10 +676,10 @@ static int pbus_size_mem(struct pci_bus *bus, unsigned long mask,
min_align = align1 >> 1;
align += aligns[order];
}
- size0 = calculate_memsize(size, min_size, 0, resource_size(b_res), align);
+ size0 = calculate_memsize(size, min_size, 0, resource_size(b_res), min_align);
size1 = !add_size ? size :
calculate_memsize(size, min_size+add_size, 0,
- resource_size(b_res), align);
+ resource_size(b_res), min_align);
if (!size0 && !size1) {
if (b_res->start || b_res->end)
dev_info(&bus->self->dev, "disabling bridge window "
