Re: [PATCH v2 1/2] PCI ACPI: Avoid panic when PCI IO resource's size is not page aligned

[Yisheng Xie]

Hi Rafael,

On 2018/5/9 5:20, Rafael J. Wysocki wrote:
> On Mon, May 7, 2018 at 10:06 AM, Yisheng Xie <xieyisheng1@xxxxxxxxxx> wrote:
>>
>>
>> On 2018/5/1 17:28, Rafael J. Wysocki wrote:
>>>>  drivers/acpi/pci_root.c | 2 +-
>>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/drivers/acpi/pci_root.c b/drivers/acpi/pci_root.c
>>>>> index 6fc204a..b758ca3 100644
>>>>> --- a/drivers/acpi/pci_root.c
>>>>> +++ b/drivers/acpi/pci_root.c
>>>>> @@ -746,7 +746,7 @@ static void acpi_pci_root_remap_iospace(struct resource_entry *entry)
>>>>>            goto err;
>>>>>
>>>>>    res->start = port;
>>>>> -  res->end = port + length - 1;
>>>>> +  res->end = PAGE_ALIGN(port + length) - 1;
>>> Shouldn't pci_remap_iospace() sanitize its arguments instead?
>>
>> Yeah, I thought that pci_remap_iospace() will be called at many place, and presently I
>> had not seen any problem at other place except acpi_pci_root_remap_iospace(). Anyway,
>> sanitize arguments in pci_remap_iospace() can resolve the problem more thoroughly, but
>> should more common, right?
>>
>> Therefore, is the follow change ok from your point of view?
>>
>> diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
>> index e597655..8607298 100644
>> --- a/drivers/pci/pci.c
>> +++ b/drivers/pci/pci.c
>> @@ -3527,6 +3527,9 @@ int pci_remap_iospace(const struct resource *res, phys_addr_t phys_addr)
>>         if (res->end > IO_SPACE_LIMIT)
>>                 return -EINVAL;
>>
>> +       if (!PAGE_ALIGNED(vaddr) || !PAGE_ALIGNED(resource_size(res)))
>> +               return -EINVAL;
>> +
>>         return ioremap_page_range(vaddr, vaddr + resource_size(res), phys_addr,
>>                                   pgprot_device(PAGE_KERNEL));
>>  #else
> 
> I'd rather apply PAGE_ALIGN() to the arguments of ioremap_page_range()
> and call it anyway.  It will fail if the mapping cannot be created.
> 
Hmm, here is a corner case which take 64k page size as an example:
 step 1. someone call pci_remap_iospace() with res:
		res->start 0x1000, resource_size(res) 0x1000
	after PAGE_ALIGN(), the arguments of ioremap_page_range() will be
		addr: (PCI_IOBASE + PAGE_SIZE), end: (PCI_IOBASE + 2*PAGE_SIZE)
	and ioremap_page_range will be ok.

 step 2. another one call pci_remap_iospace() with res:
		res->start 0x2000, resource_size(res) 0x1000
	after PAGE_ALIGN(), the arguments of ioremap_page_range() also will be:
		addr: (PCI_IOBASE + PAGE_SIZE), end: (PCI_IOBASE + 2*PAGE_SIZE)
	then ioremap_page_range() will also trigger BUG_ON(!pte_none(*pte));
	for the pte is not none after step 1, right?

 ps, if let addr = vaddr && PAGE_MASK, above case seems also will trigger BUG_ON.

Actually, I am not sure whether the above case is exist in real system.

Thanks
Yisheng

> .
>