[PATCH] PCI: qcom: Fix the fixup of PCI_VENDOR_ID_QCOM

[PATCH] PCI: qcom: Fix the fixup of PCI_VENDOR_ID_QCOM

[Bjorn Andersson]

There exists non-bridge PCIe devices with PCI_VENDOR_ID_QCOM, so limit
the fixup to only affect the PCIe 2.0 (0x106) and PCIe 3.0 (0x107)
bridges.

Cc: stable@vger.kernel.org
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
---
 drivers/pci/controller/dwc/pcie-qcom.c | 3 ++-
 include/linux/pci_ids.h                | 2 ++
 2 files changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/pci/controller/dwc/pcie-qcom.c b/drivers/pci/controller/dwc/pcie-qcom.c
index 35f4980480bb..b91abf4d4905 100644
--- a/drivers/pci/controller/dwc/pcie-qcom.c
+++ b/drivers/pci/controller/dwc/pcie-qcom.c
@@ -1441,7 +1441,8 @@ static void qcom_fixup_class(struct pci_dev *dev)
 {
 	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
 }
-DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCI_ANY_ID, qcom_fixup_class);
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE20, qcom_fixup_class);
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE30, qcom_fixup_class);
 
 static struct platform_driver qcom_pcie_driver = {
 	.probe = qcom_pcie_probe,
diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
index 21a572469a4e..3d0724ee4d2f 100644
--- a/include/linux/pci_ids.h
+++ b/include/linux/pci_ids.h
@@ -2413,6 +2413,8 @@
 #define PCI_VENDOR_ID_LENOVO		0x17aa
 
 #define PCI_VENDOR_ID_QCOM		0x17cb
+#define PCIE_DEVICE_ID_QCOM_PCIE20	0x0106
+#define PCIE_DEVICE_ID_QCOM_PCIE30	0x0107
 
 #define PCI_VENDOR_ID_CDNS		0x17cd
 
-- 
2.23.0

Re: [PATCH] PCI: qcom: Fix the fixup of PCI_VENDOR_ID_QCOM

[Vinod Koul]

On 01-11-19, 17:24, Bjorn Andersson wrote:
> There exists non-bridge PCIe devices with PCI_VENDOR_ID_QCOM, so limit
> the fixup to only affect the PCIe 2.0 (0x106) and PCIe 3.0 (0x107)
> bridges.

Reviewed-by: Vinod Koul <vkoul@kernel.org>

-- 
~Vinod

Re: [PATCH] PCI: qcom: Fix the fixup of PCI_VENDOR_ID_QCOM

[Stanimir Varbanov]

Hi Bjorn,

Thanks for the fix!

On 11/2/19 2:24 AM, Bjorn Andersson wrote:
> There exists non-bridge PCIe devices with PCI_VENDOR_ID_QCOM, so limit
> the fixup to only affect the PCIe 2.0 (0x106) and PCIe 3.0 (0x107)
> bridges.

Are you sure that this will not break ops_1_0_0 (Qcom IP rev.: 1.0.0
Synopsys IP rev.: 4.11a) i.e. apq8084 ?

> 
> Cc: stable@vger.kernel.org
> Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
> ---
>  drivers/pci/controller/dwc/pcie-qcom.c | 3 ++-
>  include/linux/pci_ids.h                | 2 ++
>  2 files changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/controller/dwc/pcie-qcom.c b/drivers/pci/controller/dwc/pcie-qcom.c
> index 35f4980480bb..b91abf4d4905 100644
> --- a/drivers/pci/controller/dwc/pcie-qcom.c
> +++ b/drivers/pci/controller/dwc/pcie-qcom.c
> @@ -1441,7 +1441,8 @@ static void qcom_fixup_class(struct pci_dev *dev)
>  {
>  	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
>  }
> -DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCI_ANY_ID, qcom_fixup_class);
> +DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE20, qcom_fixup_class);
> +DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE30, qcom_fixup_class);
>  
>  static struct platform_driver qcom_pcie_driver = {
>  	.probe = qcom_pcie_probe,
> diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
> index 21a572469a4e..3d0724ee4d2f 100644
> --- a/include/linux/pci_ids.h
> +++ b/include/linux/pci_ids.h
> @@ -2413,6 +2413,8 @@
>  #define PCI_VENDOR_ID_LENOVO		0x17aa
>  
>  #define PCI_VENDOR_ID_QCOM		0x17cb
> +#define PCIE_DEVICE_ID_QCOM_PCIE20	0x0106
> +#define PCIE_DEVICE_ID_QCOM_PCIE30	0x0107
>  
>  #define PCI_VENDOR_ID_CDNS		0x17cd
>  
> 

-- 
regards,
Stan

Re: [PATCH] PCI: qcom: Fix the fixup of PCI_VENDOR_ID_QCOM

[Bjorn Andersson]

On Mon 04 Nov 05:00 PST 2019, Stanimir Varbanov wrote:

> Hi Bjorn,
> 
> Thanks for the fix!
> 
> On 11/2/19 2:24 AM, Bjorn Andersson wrote:
> > There exists non-bridge PCIe devices with PCI_VENDOR_ID_QCOM, so limit
> > the fixup to only affect the PCIe 2.0 (0x106) and PCIe 3.0 (0x107)
> > bridges.
> 
> Are you sure that this will not break ops_1_0_0 (Qcom IP rev.: 1.0.0
> Synopsys IP rev.: 4.11a) i.e. apq8084 ?
> 

I am not, I've only tested this on db820c and db845c. Unfortunately I'm
unable to find a branch with the necessary dts snippets to test it. Do
you perhaps have a branch somewhere?

Regards,
Bjorn

> > 
> > Cc: stable@vger.kernel.org
> > Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
> > ---
> >  drivers/pci/controller/dwc/pcie-qcom.c | 3 ++-
> >  include/linux/pci_ids.h                | 2 ++
> >  2 files changed, 4 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/pci/controller/dwc/pcie-qcom.c b/drivers/pci/controller/dwc/pcie-qcom.c
> > index 35f4980480bb..b91abf4d4905 100644
> > --- a/drivers/pci/controller/dwc/pcie-qcom.c
> > +++ b/drivers/pci/controller/dwc/pcie-qcom.c
> > @@ -1441,7 +1441,8 @@ static void qcom_fixup_class(struct pci_dev *dev)
> >  {
> >  	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
> >  }
> > -DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCI_ANY_ID, qcom_fixup_class);
> > +DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE20, qcom_fixup_class);
> > +DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE30, qcom_fixup_class);
> >  
> >  static struct platform_driver qcom_pcie_driver = {
> >  	.probe = qcom_pcie_probe,
> > diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
> > index 21a572469a4e..3d0724ee4d2f 100644
> > --- a/include/linux/pci_ids.h
> > +++ b/include/linux/pci_ids.h
> > @@ -2413,6 +2413,8 @@
> >  #define PCI_VENDOR_ID_LENOVO		0x17aa
> >  
> >  #define PCI_VENDOR_ID_QCOM		0x17cb
> > +#define PCIE_DEVICE_ID_QCOM_PCIE20	0x0106
> > +#define PCIE_DEVICE_ID_QCOM_PCIE30	0x0107
> >  
> >  #define PCI_VENDOR_ID_CDNS		0x17cd
> >  
> > 
> 
> -- 
> regards,
> Stan

Re: [PATCH] PCI: qcom: Fix the fixup of PCI_VENDOR_ID_QCOM

[Stanimir Varbanov]

Hi Bjorn,

On 11/6/19 10:23 PM, Bjorn Andersson wrote:
> On Mon 04 Nov 05:00 PST 2019, Stanimir Varbanov wrote:
> 
>> Hi Bjorn,
>>
>> Thanks for the fix!
>>
>> On 11/2/19 2:24 AM, Bjorn Andersson wrote:
>>> There exists non-bridge PCIe devices with PCI_VENDOR_ID_QCOM, so limit
>>> the fixup to only affect the PCIe 2.0 (0x106) and PCIe 3.0 (0x107)
>>> bridges.
>>
>> Are you sure that this will not break ops_1_0_0 (Qcom IP rev.: 1.0.0
>> Synopsys IP rev.: 4.11a) i.e. apq8084 ?

OK, I've checked DEVICE IDs for all supported SoCs in pcie-qcom driver:

apq8084 0x101
ipq8064 0x101
apq8064 0x101
msm8996 0x104
ipq8074 0x302
ipq4019 0x1001
qcs404  ??? (can you check for this SoC)

As you can see they are various, so I don't think we have to expose them
in pci_ids.h. See my proposal below.

Of course the last word will be from PCI maintainers.

>>
> 
> I am not, I've only tested this on db820c and db845c. Unfortunately I'm
> unable to find a branch with the necessary dts snippets to test it. Do
> you perhaps have a branch somewhere?
> 
> Regards,
> Bjorn
> 
>>>
>>> Cc: stable@vger.kernel.org
>>> Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>> ---
>>>  drivers/pci/controller/dwc/pcie-qcom.c | 3 ++-
>>>  include/linux/pci_ids.h                | 2 ++
>>>  2 files changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/pci/controller/dwc/pcie-qcom.c b/drivers/pci/controller/dwc/pcie-qcom.c
>>> index 35f4980480bb..b91abf4d4905 100644
>>> --- a/drivers/pci/controller/dwc/pcie-qcom.c
>>> +++ b/drivers/pci/controller/dwc/pcie-qcom.c
>>> @@ -1441,7 +1441,8 @@ static void qcom_fixup_class(struct pci_dev *dev)
>>>  {
>>>  	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
>>>  }
>>> -DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCI_ANY_ID, qcom_fixup_class);
>>> +DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE20, qcom_fixup_class);
>>> +DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE30, qcom_fixup_class);

DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x0101, qcom_fixup_class);
DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x0104, qcom_fixup_class);
DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x0106, qcom_fixup_class);
DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x0107, qcom_fixup_class);
DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x0302, qcom_fixup_class);
DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x1001, qcom_fixup_class);

>>>  
>>>  static struct platform_driver qcom_pcie_driver = {
>>>  	.probe = qcom_pcie_probe,
>>> diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
>>> index 21a572469a4e..3d0724ee4d2f 100644
>>> --- a/include/linux/pci_ids.h
>>> +++ b/include/linux/pci_ids.h
>>> @@ -2413,6 +2413,8 @@
>>>  #define PCI_VENDOR_ID_LENOVO		0x17aa
>>>  
>>>  #define PCI_VENDOR_ID_QCOM		0x17cb
>>> +#define PCIE_DEVICE_ID_QCOM_PCIE20	0x0106
>>> +#define PCIE_DEVICE_ID_QCOM_PCIE30	0x0107

... and drop these additions.

>>>  
>>>  #define PCI_VENDOR_ID_CDNS		0x17cd
>>>  
>>>
>>
>> -- 
>> regards,
>> Stan

-- 
regards,
Stan

Re: [PATCH] PCI: qcom: Fix the fixup of PCI_VENDOR_ID_QCOM

[Stanimir Varbanov]

Hi,

On 11/7/19 11:32 AM, Stanimir Varbanov wrote:
> Hi Bjorn,
> 
> On 11/6/19 10:23 PM, Bjorn Andersson wrote:
>> On Mon 04 Nov 05:00 PST 2019, Stanimir Varbanov wrote:
>>
>>> Hi Bjorn,
>>>
>>> Thanks for the fix!
>>>
>>> On 11/2/19 2:24 AM, Bjorn Andersson wrote:
>>>> There exists non-bridge PCIe devices with PCI_VENDOR_ID_QCOM, so limit
>>>> the fixup to only affect the PCIe 2.0 (0x106) and PCIe 3.0 (0x107)
>>>> bridges.
>>>
>>> Are you sure that this will not break ops_1_0_0 (Qcom IP rev.: 1.0.0
>>> Synopsys IP rev.: 4.11a) i.e. apq8084 ?
> 
> OK, I've checked DEVICE IDs for all supported SoCs in pcie-qcom driver:
> 
> apq8084 0x101
> ipq8064 0x101
> apq8064 0x101
> msm8996 0x104
> ipq8074 0x302
> ipq4019 0x1001
> qcs404  ??? (can you check for this SoC)
> 
> As you can see they are various, so I don't think we have to expose them
> in pci_ids.h. See my proposal below.
> 
> Of course the last word will be from PCI maintainers.
> 
>>>
>>
>> I am not, I've only tested this on db820c and db845c. Unfortunately I'm
>> unable to find a branch with the necessary dts snippets to test it. Do
>> you perhaps have a branch somewhere?
>>
>> Regards,
>> Bjorn
>>
>>>>
>>>> Cc: stable@vger.kernel.org
>>>> Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
>>>> ---
>>>>  drivers/pci/controller/dwc/pcie-qcom.c | 3 ++-
>>>>  include/linux/pci_ids.h                | 2 ++
>>>>  2 files changed, 4 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/pci/controller/dwc/pcie-qcom.c b/drivers/pci/controller/dwc/pcie-qcom.c
>>>> index 35f4980480bb..b91abf4d4905 100644
>>>> --- a/drivers/pci/controller/dwc/pcie-qcom.c
>>>> +++ b/drivers/pci/controller/dwc/pcie-qcom.c
>>>> @@ -1441,7 +1441,8 @@ static void qcom_fixup_class(struct pci_dev *dev)
>>>>  {
>>>>  	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
>>>>  }
>>>> -DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCI_ANY_ID, qcom_fixup_class);
>>>> +DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE20, qcom_fixup_class);
>>>> +DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE30, qcom_fixup_class);
> 
> DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x0101, qcom_fixup_class);
> DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x0104, qcom_fixup_class);
> DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x0106, qcom_fixup_class);
> DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x0107, qcom_fixup_class);
> DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x0302, qcom_fixup_class);
> DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x1001, qcom_fixup_class);
> 
>>>>  
>>>>  static struct platform_driver qcom_pcie_driver = {
>>>>  	.probe = qcom_pcie_probe,
>>>> diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
>>>> index 21a572469a4e..3d0724ee4d2f 100644
>>>> --- a/include/linux/pci_ids.h
>>>> +++ b/include/linux/pci_ids.h
>>>> @@ -2413,6 +2413,8 @@
>>>>  #define PCI_VENDOR_ID_LENOVO		0x17aa
>>>>  
>>>>  #define PCI_VENDOR_ID_QCOM		0x17cb
>>>> +#define PCIE_DEVICE_ID_QCOM_PCIE20	0x0106
>>>> +#define PCIE_DEVICE_ID_QCOM_PCIE30	0x0107

I looked in pci_ids.h, and it looks like this is the right place for
DEVICE_IDs so instead of pcie gen we could list the SoCs names?

#define PCIE_DEVICE_ID_QCOM_APQ8084	0x0101
#define PCIE_DEVICE_ID_QCOM_APQ8064	0x0101
....

-- 
regards,
Stan

Re: [PATCH] PCI: qcom: Fix the fixup of PCI_VENDOR_ID_QCOM

[Marc Gonzalez]

On 02/11/2019 01:24, Bjorn Andersson wrote:

> There exists non-bridge PCIe devices with PCI_VENDOR_ID_QCOM, so limit
> the fixup to only affect the PCIe 2.0 (0x106) and PCIe 3.0 (0x107)
> bridges.

Hey, git blames me! Why didn't you CC me? :-)

Commit 322f03436692481993d389f539c016d20bb0fa1d
PCI: qcom: Use default config space read function

The patch's history is of interest:
https://lkml.org/lkml/2019/3/11/614
https://www.spinics.net/lists/linux-arm-msm/msg49090.html

> Cc: stable@vger.kernel.org
> Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
> ---
>  drivers/pci/controller/dwc/pcie-qcom.c | 3 ++-
>  include/linux/pci_ids.h                | 2 ++
>  2 files changed, 4 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/pci/controller/dwc/pcie-qcom.c b/drivers/pci/controller/dwc/pcie-qcom.c
> index 35f4980480bb..b91abf4d4905 100644
> --- a/drivers/pci/controller/dwc/pcie-qcom.c
> +++ b/drivers/pci/controller/dwc/pcie-qcom.c
> @@ -1441,7 +1441,8 @@ static void qcom_fixup_class(struct pci_dev *dev)
>  {
>  	dev->class = PCI_CLASS_BRIDGE_PCI << 8;
>  }
> -DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCI_ANY_ID, qcom_fixup_class);
> +DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE20, qcom_fixup_class);
> +DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, PCIE_DEVICE_ID_QCOM_PCIE30, qcom_fixup_class);
>  
>  static struct platform_driver qcom_pcie_driver = {
>  	.probe = qcom_pcie_probe,
> diff --git a/include/linux/pci_ids.h b/include/linux/pci_ids.h
> index 21a572469a4e..3d0724ee4d2f 100644
> --- a/include/linux/pci_ids.h
> +++ b/include/linux/pci_ids.h
> @@ -2413,6 +2413,8 @@
>  #define PCI_VENDOR_ID_LENOVO		0x17aa
>  
>  #define PCI_VENDOR_ID_QCOM		0x17cb
> +#define PCIE_DEVICE_ID_QCOM_PCIE20	0x0106
> +#define PCIE_DEVICE_ID_QCOM_PCIE30	0x0107

I don't think the fixup is required for 0x106 and 0x107...

In v1, I wrote:
FWIW, this quirk is no longer required on recent chips:
msm8996 (tested by Stanimir), msm8998 (tested by me), sdm845 (untested) are unaffected
apq/ipq8064 is affected => what is the device ID for these chips?
others?

IIRC, 0x0101 requires the fixup because
dw_pcie_wr_own_conf(pp, PCI_CLASS_DEVICE, 2, PCI_CLASS_BRIDGE_PCI);
is broken on that platform (grrr, HW devs)
(See my v3, tested by Srinivas)

Stan wrote:

Yes it is good but to avoid breaking another SoCs could you add fixups
for the following SoCs:

SoC		device ID
ipq4019 	0x1001
ipq8064		0x101
ipq8074		0x108

ipq8064 has the same device ID as apq8064, but I'm not sure do we need
defines per SoC or just rename DEV_ID_8064 ? I'm fine with both ways.


In conclusion, my analysis in v5 was wrong
"Changes from v4 to v5: Apply fixup to all qcom chips, the same way it was before
(thus the code remains functionally equivalent)"
=> The fixup was applied *more widely* than before, so not functionally equivalent.

Regards.