在 2023/6/6 10:27, Huacai Chen 写道:
> On Mon, Jun 5, 2023 at 10:51 PM Will Deacon <will@xxxxxxxxxx> wrote:
>>
>> On Mon, Jun 05, 2023 at 11:44:23AM +0800, Bibo Mao wrote:
>>> Some PCI devices has only 4K memory space size, it is normal in general
>>> machines and aligned with page size. However some architectures which
>>> support different page size, default page size on LoongArch is 16K, and
>>> ARM64 supports page size varying from 4K to 64K. On machines where larger
>>> page size is use, memory space region of two different pci devices may be
>>> in one page. It is not safe with mmu protection, also VFIO pci device
>>> driver requires base address of pci memory space page aligned, so that it
>>> can be memory mapped to qemu user space when it is pass-through to vm.
>>>
>>> It consumes more pci memory resource with page size alignment requirement,
>>> on 64 bit system it should not be a problem. And UEFI bios set pci memory
>>> base address with 4K fixed-size aligned, the safer solution is to align
>>> with larger size on UEFI BIOS stage on these architectures, linux kernel
>>> can reuse resource from UEFI bios. For new devices such hotplug pci
>>> devices and sriov devices, pci resource is assigned in Linux kernel side.
>>>
>>> Signed-off-by: Bibo Mao <maobibo@xxxxxxxxxxx>
>>> ---
>>> arch/arm64/kernel/pci.c | 13 +++++++++++++
>>> arch/loongarch/pci/pci.c | 18 ++++++++++++++++++
>>> 2 files changed, 31 insertions(+)
>>>
>>> diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c
>>> index 2276689b5411..e2f7b176b156 100644
>>> --- a/arch/arm64/kernel/pci.c
>>> +++ b/arch/arm64/kernel/pci.c
>>> @@ -232,4 +232,17 @@ void pcibios_remove_bus(struct pci_bus *bus)
>>> acpi_pci_remove_bus(bus);
>>> }
>>>
>>> +resource_size_t pcibios_align_resource(void *data, const struct resource *res,
>>> + resource_size_t size, resource_size_t align)
>>> +{
>>> + resource_size_t start = res->start;
>>> +
>>> + /*
>>> + * align base address of pci memory resource with page size
>>> + */
>>> + if ((res->flags & IORESOURCE_MEM) && (align < PAGE_SIZE))
>>> + start = ALIGN(start, PAGE_SIZE);
>>> +
>>> + return start;
>>> +}
>>> #endif
>>
>> If this is needed by all architectures with !4k page size, why don't we
>> instead handle the alignment in the core code, given that it is aware
>> of the page size?
> Agree, modifying the common weak version of pcibios_align_resource()
> seems better. Maybe we can use CONFIG_VFIO_PCI to guard the align
> operation.
will do. only that I am not sure it is ok for all architectures, after all it consumes
more pci memory resource if forced to page size aligned.
Regards
Bibo, Mao
>
> Huacai
>>
>> Will
