[PATCH v4 1/5] PCI: designware: add memory barrier after enabling region

[Stanimir Varbanov]

On 12/09/2015 11:52 AM, Arnd Bergmann wrote:
> On Wednesday 09 December 2015 10:10:05 Pratyush Anand wrote:
>> On Tue, Dec 8, 2015 at 2:31 PM, Stanimir Varbanov
>>>> Signed-off-by: Stanimir Varbanov <stanimir.varbanov at linaro.org>
>>>> ---
>>>>  drivers/pci/host/pcie-designware.c |    5 +++++
>>>>  1 file changed, 5 insertions(+)
>>>>
>>>> diff --git a/drivers/pci/host/pcie-designware.c b/drivers/pci/host/pcie-designware.c
>>>> index 02a7452bdf23..ed4dc2e2553b 100644
>>>> --- a/drivers/pci/host/pcie-designware.c
>>>> +++ b/drivers/pci/host/pcie-designware.c
>>>> @@ -164,6 +164,11 @@ static void dw_pcie_prog_outbound_atu(struct pcie_port *pp, int index,
>>>>       dw_pcie_writel_rc(pp, upper_32_bits(pci_addr), PCIE_ATU_UPPER_TARGET);
>>>>       dw_pcie_writel_rc(pp, type, PCIE_ATU_CR1);
>>>>       dw_pcie_writel_rc(pp, PCIE_ATU_ENABLE, PCIE_ATU_CR2);
>>>> +     /*
>>>> +      * ensure that the ATU enable has been happaned before accessing
>>>> +      * pci configuration/io spaces through dw_pcie_cfg_[read|write].
>>>> +      */
>>>> +     wmb();
>>>>  }
>>>>
>>
>>
>> My understnading is that since writel() of dw_pcie_writel_rc() in
>> above code and readl(), writel() of dw_pcie_cfg_[read|write]() (which
>> will follow) goes through same device (ie PCIe host here). So, it is
>> guaranteed that 1st writel() will be executed before later
>> readl()/writel(). If that is true then we do not need any explicit
>> barrier here.
>>
>> Arnd, Russel: whats your opinion here.
> 
> I think the ordering is only enforced if the two register accesses are
> on the same device as seen from the bus, and it's possible that the
> RC registers and the config space registers are not considered the
> same thing here.
> 
> For config write, this is not a problem, because the config space write
> has a wmb() that enforces ordering, but it's possible that the config
> space read may hit the device in parallel with the PCIE_ATU_ENABLE
> write.

Hmm, just a matter of fact - as I described in the patch description
this wmb() fixed an issue with pcie device enumeration (I came down to
pci_bus_read_dev_vendor_id() returns zero) i.e. exactly a pci
configuration space read.

-- 
regards,
Stan