[V3] PCI: fu740: Drop to 2.5GT/s to fix initial device probing on some boards

[Alexandre Ghiti]

On 3/21/22 14:49, Alexandre Ghiti wrote:
> Hi Ben,
>
> On 3/18/22 16:24, Ben Dooks wrote:
>> The fu740 PCIe core does not probe any devices on the SiFive Unmatched
>> board without this fix (or having U-Boot explicitly start the PCIe via
>> either boot-script or user command). The fix is to start the link at
>> 2.5GT/s speeds and once the link is up then change the maximum speed 
>> back
>> to the default.
>>
>> The U-Boot driver claims to set the link-speed to 2.5GT/s to get the 
>> probe
>> to work (and U-Boot does print link up at 2.5GT/s) in the following 
>> code:
>> https://source.denx.de/u-boot/u-boot/-/blob/master/drivers/pci/pcie_dw_sifive.c?id=v2022.01#L271 
>>
>>
>> Signed-off-by: Ben Dooks <ben.dooks at codethink.co.uk>
>> -- 
>> Note, this patch has had significant re-work since the previous 4
>> sets, including trying to fix style, message, reliance on the U-Boot
>> fix and the comments about usage of LINK_CAP and reserved fields.
>>
>> v2:
>> - fix issues with Gen1/2.5GTs
>> - updated comment on the initial probe
>> - run tests with both uninitialised and initialsed pcie from uboot
>> ---
>>   drivers/pci/controller/dwc/pcie-fu740.c | 52 ++++++++++++++++++++++++-
>>   1 file changed, 51 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/pci/controller/dwc/pcie-fu740.c 
>> b/drivers/pci/controller/dwc/pcie-fu740.c
>> index 842b7202b96e..ecac0364178a 100644
>> --- a/drivers/pci/controller/dwc/pcie-fu740.c
>> +++ b/drivers/pci/controller/dwc/pcie-fu740.c
>> @@ -181,10 +181,60 @@ static int fu740_pcie_start_link(struct dw_pcie 
>> *pci)
>>   {
>>       struct device *dev = pci->dev;
>>       struct fu740_pcie *afp = dev_get_drvdata(dev);
>> +    u8 cap_exp = dw_pcie_find_capability(pci, PCI_CAP_ID_EXP);
>> +    int ret;
>> +    u32 orig, tmp;
>> +
>> +    /*
>> +     * Force 2.5GT/s when starting the link, due to some devices not
>> +     * probing at higher speeds. This happens with the PCIe switch
>> +     * on the Unmatched board when U-Boot has not initialised the PCIe.
>> +     * The fix in U-Boot is to force 2.5GT/s, which then gets cleared
>> +     * by the soft reset does by this driver.
>> +     */
>> +
>> +    dev_dbg(dev, "cap_exp at %x\n", cap_exp);
>> +    dw_pcie_dbi_ro_wr_en(pci);
>> +
>> +    tmp = dw_pcie_readl_dbi(pci, cap_exp + PCI_EXP_LNKCAP);
>> +    orig = tmp & PCI_EXP_LNKCAP_SLS;
>> +    tmp &= ~PCI_EXP_LNKCAP_SLS;
>> +    tmp |= PCI_EXP_LNKCAP_SLS_2_5GB;
>> +    dw_pcie_writel_dbi(pci, cap_exp + PCI_EXP_LNKCAP, tmp);
>>         /* Enable LTSSM */
>>       writel_relaxed(0x1, afp->mgmt_base + PCIEX8MGMT_APP_LTSSM_ENABLE);
>> -    return 0;
>> +
>> +    ret = dw_pcie_wait_for_link(pci);
>> +    if (ret) {
>> +        dev_err(dev, "error: link did not start\n");
>> +        goto err;
>> +    }
>> +
>> +    tmp = dw_pcie_readl_dbi(pci, cap_exp + PCI_EXP_LNKCAP);
>> +    if ((tmp & PCI_EXP_LNKCAP_SLS) != orig) {
>> +        dev_dbg(dev, "changing speed back to original\n");
>> +
>> +        tmp &= ~PCI_EXP_LNKCAP_SLS;
>> +        tmp |= orig;
>> +        dw_pcie_writel_dbi(pci, cap_exp + PCI_EXP_LNKCAP, tmp);
>> +
>> +        tmp = dw_pcie_readl_dbi(pci, PCIE_LINK_WIDTH_SPEED_CONTROL);
>> +        tmp |= PORT_LOGIC_SPEED_CHANGE;
>> +        dw_pcie_writel_dbi(pci, PCIE_LINK_WIDTH_SPEED_CONTROL, tmp);
>> +
>> +        ret = dw_pcie_wait_for_link(pci);
>> +        if (ret) {
>> +            dev_err(dev, "error: link did not start at new speed\n");
>> +            goto err;
>> +        }
>> +    }
>> +
>> +    ret = 0;
>> +err:
>> +    WARN_ON(ret);    /* we assume that errors will be very rare */
>> +    dw_pcie_dbi_ro_wr_dis(pci);
>> +    return ret;
>>   }
>>     static int fu740_pcie_host_init(struct pcie_port *pp)
>
> +cc Maciej and David as there is this other fix that seems to do the 
> same but differently, it's been under review for some time now: 
> https://lore.kernel.org/all/20220302000043.GA662523@bhelgaas/t/
>
> I fell onto this issue recently, I'll give your patch and the above a 
> try soon.


FWIW, I have tested this and it solved my issue with nvme not being 
probed, so:

Tested-by: Alexandre Ghiti <alexandre.ghiti at canonical.com>

Thanks,

Alex


>
> Thanks
>
> Alex
>
>
> _______________________________________________
> linux-riscv mailing list
> linux-riscv at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-riscv