On 2017/9/23 15:24, Zhou Wang wrote:
> When double checking 64bit prefetch range, we will change the prefetch range
> in a time slot by writing 0xffffffff to PCI_PREF_BASE_UPPER32. This may break
> transfers through related bridge at that time.
>
> E.g. if we have below PCIe topology:
>
> -[0000:00]-+-00.0-[01-02]--+-00.0 Device 8086:10fb
> | \-00.1 Device 8086:10fb
> \-08.0-[03]----00.0 Device 8086:0953
>
> When rescan 00:08.0, it will call:
>
> pci_rescan_bus
> -> pci_assign_unassigned_bus_resources
> -> __pci_bus_size_bridges
> -> pci_bridge_check_ranges
> -> pci_write_config_dword(bridge, PCI_PREF_BASE_UPPER32, 0xffffffff)
>
> This will change the prefetch range of 00:00.0 in a time slot, so traffic of
> 01:00.0 or 01:00.1 may be broken.
>
> In fact, we can get if one bridge supports 64bit range by the bottom 4bits of
> prefetchable memory base/limit. Honestly speaking, I don't know why 1f82de10d6b1
> ("PCI/86: don't assume prefetchable ranges are 64bit") has added the double
> check code.
>
> So Can we remove the double checking of prefetchable range to avoid this problem?
>
> Signed-off-by: Zhou Wang <wangzhou1@xxxxxxxxxxxxx>
> ---
> drivers/pci/setup-bus.c | 14 --------------
> 1 file changed, 14 deletions(-)
>
> diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
> index 958da7d..23010a9 100644
> --- a/drivers/pci/setup-bus.c
> +++ b/drivers/pci/setup-bus.c
> @@ -778,20 +778,6 @@ static void pci_bridge_check_ranges(struct pci_bus *bus)
> b_res[2].flags |= PCI_PREF_RANGE_TYPE_64;
> }
> }
> -
> - /* double check if bridge does support 64 bit pref */
> - if (b_res[2].flags & IORESOURCE_MEM_64) {
> - u32 mem_base_hi, tmp;
> - pci_read_config_dword(bridge, PCI_PREF_BASE_UPPER32,
> - &mem_base_hi);
> - pci_write_config_dword(bridge, PCI_PREF_BASE_UPPER32,
> - 0xffffffff);
> - pci_read_config_dword(bridge, PCI_PREF_BASE_UPPER32, &tmp);
> - if (!tmp)
> - b_res[2].flags &= ~IORESOURCE_MEM_64;
> - pci_write_config_dword(bridge, PCI_PREF_BASE_UPPER32,
> - mem_base_hi);
> - }
> }
>
> /* Helper function for sizing routines: find first available
>
Any idea about this problem?
Best Regards，
Zhou
