On Fri, Feb 10, 2012 at 1:00 PM, Jesse Barnes <jbarnes@xxxxxxxxxxxxxxxx> wrote:
> On Sat, 4 Feb 2012 22:55:08 -0800
> Yinghai Lu <yinghai@xxxxxxxxxx> wrote:
>
>> If bios does not assign those BAR or wrong address, then kernel will
>> try to do pci realloc.
>>
>> in that case, user still can use pci=realloc=off to override it.
>>
>> Signed-off-by: Yinghai Lu <yinghai@xxxxxxxxxx>
>> ---
>> drivers/pci/setup-bus.c | 28 ++++++++++++++++++++++++++++
>> 1 files changed, 28 insertions(+), 0 deletions(-)
>>
>> diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
>> index 9526038..520f256 100644
>> --- a/drivers/pci/setup-bus.c
>> +++ b/drivers/pci/setup-bus.c
>> @@ -1293,6 +1293,31 @@ static bool __init pci_realloc_enabled(void)
>> return pci_realloc_enable >= enable_yes_user;
>> }
>>
>> +static void __init pci_realloc_detect(void)
>> +{
>> + struct pci_dev *dev = NULL;
>> +
>> + if (pci_realloc_enable != enable_not_set)
>> + return;
>> +
>> +#ifdef CONFIG_PCI_IOV
>> + for_each_pci_dev(dev) {
>> + int i;
>> +
>> + for (i = PCI_IOV_RESOURCES; i <= PCI_IOV_RESOURCE_END; i++) {
>> + struct resource *r = &dev->resource[i];
>> +
>> + /* Not assigned, or rejected by kernel */
>> + if (r->flags && !r->start) {
>> + pci_realloc_enable = enable_yes_detected;
>> +
>> + return;
>> + }
>> + }
>> + }
>> +#endif
>> +}
>> +
>> /*
>> * first try will not touch pci bridge res
>> * second and later try will clear small leaf bridge res
>> @@ -1314,6 +1339,7 @@ pci_assign_unassigned_resources(void)
>> int pci_try_num = 1;
>>
>> /* don't realloc if asked to do so */
>> + pci_realloc_detect();
>> if (pci_realloc_enabled()) {
>> int max_depth = pci_get_max_depth();
>>
>> @@ -1348,6 +1374,8 @@ again:
>> if (tried_times >= pci_try_num) {
>> if (pci_realloc_enable == enable_not_set)
>> printk(KERN_INFO "Some pci devices resources are not assigned, please try to boot with pci=realloc\n");
>> + else if (pci_realloc_enable == enable_yes_detected)
>> + printk(KERN_INFO "Automatically enabled pci realloc, if you have problem, please try to boot with pci=realloc=off\n");
>>
>> free_list(&fail_head);
>> goto enable_and_dump;
>
> Not sure I like this one either. Distros will want to enable IOV by
> default, but in most configs users still won't use it, and many BIOSes
> won't bother to allocate IOV space. So defaulting to trying to do so
> will likely cause a lot of unnecessary re-allocation.
>
> So distro guidance would be appreciated here for the default behavior.
Or do you prefer to:
if all SRIOV bar are not assigned, We will not enable SRIOV feature even?
Yinghai
--
To unsubscribe from this list: send the line "unsubscribe linux-pci" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
