[PATCH 1/2] PCI: dra7xx: Fix link removal on probe error

[Luca Ceresoli]

Hi Saravana,

On 14/05/22 05:46, Saravana Kannan wrote:
> On Thu, May 12, 2022 at 7:07 AM Luca Ceresoli <luca at lucaceresoli.net> wrote:
>>
>> Hi Lorenzo,
>>
>> On 11/05/22 18:41, Lorenzo Pieralisi wrote:
>>> On Sat, Jan 15, 2022 at 10:02:00AM -0600, Rob Herring wrote:
>>>> +Saravana
>>>>
>>>> On Tue, Jan 11, 2022 at 4:35 AM Luca Ceresoli <luca at lucaceresoli.net> wrote:
>>>>>
>>>>> Hi Rob,
>>>>>
>>>>> On 16/12/21 10:08, Luca Ceresoli wrote:
>>>>>> Hi Rob,
>>>>>>
>>>>>> thanks for the quick feedback!
>>>>>>
>>>>>> On 14/12/21 23:42, Rob Herring wrote:
>>>>>>> On Tue, Dec 14, 2021 at 4:15 PM Luca Ceresoli <luca at lucaceresoli.net> wrote:
>>>>>>>>
>>>>>>>> If a devm_phy_get() calls fails with phy_count==N (N > 0), then N links
>>>>>>>> have already been added by device_link_add() and won't be deleted by
>>>>>>>> device_link_del() because the code calls 'return' and not 'goto err_link'.
>>>>>>>>
>>>>>>>> Fix in a very simple way by doing all the devm_phy_get() calls before all
>>>>>>>> the device_link_add() calls.
>>>>>>>>
>>>>>>>> Fixes: 7a4db656a635 ("PCI: dra7xx: Create functional dependency between PCIe and PHY")
>>>>>>>> Signed-off-by: Luca Ceresoli <luca at lucaceresoli.net>
>>>>>>>> ---
>>>>>>>>  drivers/pci/controller/dwc/pci-dra7xx.c | 2 ++
>>>>>>>>  1 file changed, 2 insertions(+)
>>>>>>>>
>>>>>>>> diff --git a/drivers/pci/controller/dwc/pci-dra7xx.c b/drivers/pci/controller/dwc/pci-dra7xx.c
>>>>>>>> index f7f1490e7beb..2ccc53869e13 100644
>>>>>>>> --- a/drivers/pci/controller/dwc/pci-dra7xx.c
>>>>>>>> +++ b/drivers/pci/controller/dwc/pci-dra7xx.c
>>>>>>>> @@ -757,7 +757,9 @@ static int dra7xx_pcie_probe(struct platform_device *pdev)
>>>>>>>>                 phy[i] = devm_phy_get(dev, name);
>>>>>>>>                 if (IS_ERR(phy[i]))
>>>>>>>>                         return PTR_ERR(phy[i]);
>>>>>>>> +       }
>>>>>>>>
>>>>>>>> +       for (i = 0; i < phy_count; i++) {
>>>>>>>>                 link[i] = device_link_add(dev, &phy[i]->dev, DL_FLAG_STATELESS);
>>>>>>>
>>>>>>> I think this should happen automatically now with fw_devlink being
>>>>>>> enabled by default. Can you try?
>>>>>>
>>>>>> Do you mean removal should be done automatically? I think they are not
>>>>>> due to the DL_FLAG_STATELESS flag.
>>>>>
>>>>> I would love to have feedback because, as said, I think my patch is
>>>>> correct, but if I'm wrong (which might well be) I have to drop patch 1
>>>>> and rewrite patch 2 in a slightly more complex form.
>>>>
>>>> I mean that why do you need explicit dependency tracking here when
>>>> dependencies on a PHY should happen automatically now. IOW, what is
>>>> special about this driver and dependency?
>>>
>>> Any update on this patch ? I think patch 2 can be merged, please
>>> let me know if this one can be dropped.
>>
>> Thanks for the feedback! You would say yes, you can merge patch 2,
>> except it probably does not even apply as it is written in a way that is
>> based on the changes in patch 1.
>>
>> I could rewrite patch 2 to not depend on patch 1 of course, but it
>> wouldn't make code simpler, perhaps more complex. And moreover the
>> hardware that I used to have access to has phy_count==1 so I could never
>> test the failing case, and sadly now I have no access to that hardware.
> 
> Hi Luca,
> 
> The fw_devlink code to create device links from consumers to "phys"
> suppliers is pretty well exercised. Most/all Android devices running
> 5.10+ kernels (including Pixel 6) use fw_devlink=on to be able to boot
> properly.
> 
> So I'd be pretty confident in deleting the device_link_add/del() code
> in drivers/pci/controller/dwc/pci-dra7xx.c. The device links should
> already be there before the probe is even called.
> 
> Also, if you want to check if the device links (even the 1 phy one you
> have) are being created, you can look at /sys/class/devlink to see the
> list of all device links that are currently present. You can delete
> the code and then use this to check too.

Thank you for your feedback. Unfortunately as I said I have no access to
the hardware, and won't have anymore. I don't think it is a good idea to
send a patch that I cannot test on real hardware, especially since it is
for a generic hardware that thus might affect others. But I would be
glad to review any such patch that might be sent, FWIW.

-- 
Luca