On Fri, Jun 10, 2022 at 11:25:20AM +0300, Serge Semin wrote:
> It was wrong to use the region size parameter in order to determine
> whether the INCREASE_REGION_SIZE flag needs to be set for the outbound
> iATU entry because in general there are cases when combining a region base
> address and size together produces the out of bounds upper range limit
> while upper_32_bits(size) still returns zero. So having a region size
> within the permitted values doesn't mean the region limit address will fit
> to the corresponding CSR. Here is the way iATU calculates the in- and
> outbound untranslated regions if the INCREASE_REGION_SIZE flag is cleared
> [1]:
>
> Start address: End address:
> 63 31 0 63 31 0
> +---------------+---------------+ +---------------+---------------+
> | | | 0s | | | | Fs |
> +---------------+---------------+ +---------------+---------------+
> upper base | lower base !upper! base | limit address
> address address address
>
> So the region start address is determined by the iATU lower and upper base
> address registers, while the region upper boundary is calculated based on
> the 32-bits limit address register and the upper part of the base address.
> In accordance with that logic for instance the range
> 0xf0000000 @ 0x20000000 does have the size smaller than 4GB, but the
> actual limit address turns to be invalid forming the untranslated address
> map as [0xf0000000; 0x1000FFFF], which isn't what the original range was.
> In order to fix that we need to check whether the size after being added
> to the lower part of the base address causes the 4GB range overflow. If it
> does then we need to set the INCREASE_REGION_SIZE flag thus activating the
> extended limit address by means of an additional iATU CSR (upper limit
> address register) [2]:
>
> Start address: End address:
> 63 31 0 63 x 31 0
> +---------------+---------------+ +---------------+---------------+
> | | | 0s | | | | | Fs |
> +---------------+---------------+ +---------------+---------------+
> upper base | lower base upper | upper | limit address
> address address base | limit |
> address|address|
>
> Otherwise there is enough room in the 32-bits wide limit address register,
> and the flag can be left unset.
>
> Note the case when the size-based flag setting approach is correct implies
> requiring to have the size-aligned base addresses only. But that
> constraint isn't relevant to the PCIe ranges accepted by the kernel.
> There is also no point in implementing it either seeing the problem can be
> easily fixed by checking the whole limit address instead of the region
> size.
>
> [1] DesignWare Cores PCI Express Controller Databook - DWC PCIe Root Port,
> v5.40a, March 2019, fig.3-36, p.175
> [2] DesignWare Cores PCI Express Controller Databook - DWC PCIe Root Port,
> v5.40a, March 2019, fig.3-37, p.176
>
> Fixes: 5b4cf0f65324 ("PCI: dwc: Add upper limit address for outbound iATU")
> Signed-off-by: Serge Semin <Sergey.Semin@xxxxxxxxxxxxxxxxxxxx>
> Reviewed-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
> Tested-by: Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>
>
> ---
>
> Changelog v2:
> - Fix the end address in the example of the patch log. It should be
> 0x1000FFFF and not 0x0000FFFF (@Manivannan).
> ---
> drivers/pci/controller/dwc/pcie-designware.c | 16 ++++++++++------
> 1 file changed, 10 insertions(+), 6 deletions(-)
Reviewed-by: Rob Herring <robh@xxxxxxxxxx>
