[1/1] PCI: qcom: Add early fixup to set the max payload size for IPQ9574

Message ID 20230724124711.2346886-2-quic_ipkumar@quicinc.com
State New
Headers
Series Set may payload size for IPQ9574 |

Commit Message

Praveenkumar I July 24, 2023, 12:47 p.m. UTC
  Set 256 bytes as payload size for IPQ9574 via early fixup. This allows
PCIe RC to use the max payload size when a capable link partner is
connected.

Signed-off-by: Praveenkumar I <quic_ipkumar@quicinc.com>
---
This patch depends on the below series which adds support for PCIe
controllers in IPQ9574
https://lore.kernel.org/linux-arm-msm/20230519090219.15925-1-quic_devipriy@quicinc.com/

 drivers/pci/controller/dwc/pcie-qcom.c | 6 ++++++
 1 file changed, 6 insertions(+)
  

Comments

Praveenkumar I July 24, 2023, 1:51 p.m. UTC | #1
On 7/24/2023 6:23 PM, Konrad Dybcio wrote:
> On 24.07.2023 14:47, Praveenkumar I wrote:
>> Set 256 bytes as payload size for IPQ9574 via early fixup. This allows
>> PCIe RC to use the max payload size when a capable link partner is
>> connected.
>>
>> Signed-off-by: Praveenkumar I <quic_ipkumar@quicinc.com>
>> ---
> [...]
>
>> +static void qcom_fixup_mps_256(struct pci_dev *dev)
>> +{
>> +	pcie_set_mps(dev, 256);
> Looks like setting "dev->pcie_mpss = 1" here would make the PCIe generic
> code take care of this.
Yes, but that is not helping as the generic code pci_configure_mps() has 
a check for,
/         if (!bridge || !pci_is_pcie(bridge))/
/            return;/
/Here it is returning and mps is not set to new 256 bytes./
>
> Konrad
--
Thanks,
Praveenkumar
  
Manivannan Sadhasivam July 24, 2023, 2:09 p.m. UTC | #2
On Mon, Jul 24, 2023 at 06:38:55PM +0530, Manivannan Sadhasivam wrote:
> On Mon, Jul 24, 2023 at 02:53:37PM +0200, Konrad Dybcio wrote:
> > On 24.07.2023 14:47, Praveenkumar I wrote:
> > > Set 256 bytes as payload size for IPQ9574 via early fixup. This allows
> > > PCIe RC to use the max payload size when a capable link partner is
> > > connected.
> > > 
> > > Signed-off-by: Praveenkumar I <quic_ipkumar@quicinc.com>
> > > ---
> > [...]
> > 
> > > 
> > > +static void qcom_fixup_mps_256(struct pci_dev *dev)
> > > +{
> > > +	pcie_set_mps(dev, 256);
> > Looks like setting "dev->pcie_mpss = 1" here would make the PCIe generic
> > code take care of this.
> > 
> 
> Right, also this setting should not be PCI-PCI bridge specific but rather
> controller specific.
> 

Wait, have you tested this patch with PCIe devices having MPS < 256 i.e.,
default 128?

Take a look at this discussion: https://lore.kernel.org/all/20230608093652.1409485-1-vidyas@nvidia.com/

- Mani

> - Mani
> 
> > Konrad
> 
> -- 
> மணிவண்ணன் சதாசிவம்
  
Praveenkumar I July 25, 2023, 4:46 a.m. UTC | #3
On 7/24/2023 7:39 PM, Manivannan Sadhasivam wrote:
> On Mon, Jul 24, 2023 at 06:38:55PM +0530, Manivannan Sadhasivam wrote:
>> On Mon, Jul 24, 2023 at 02:53:37PM +0200, Konrad Dybcio wrote:
>>> On 24.07.2023 14:47, Praveenkumar I wrote:
>>>> Set 256 bytes as payload size for IPQ9574 via early fixup. This allows
>>>> PCIe RC to use the max payload size when a capable link partner is
>>>> connected.
>>>>
>>>> Signed-off-by: Praveenkumar I <quic_ipkumar@quicinc.com>
>>>> ---
>>> [...]
>>>
>>>> +static void qcom_fixup_mps_256(struct pci_dev *dev)
>>>> +{
>>>> +	pcie_set_mps(dev, 256);
>>> Looks like setting "dev->pcie_mpss = 1" here would make the PCIe generic
>>> code take care of this.
>>>
>> Right, also this setting should not be PCI-PCI bridge specific but rather
>> controller specific.
>>
> Wait, have you tested this patch with PCIe devices having MPS < 256 i.e.,
> default 128?
>
> Take a look at this discussion: https://lore.kernel.org/all/20230608093652.1409485-1-vidyas@nvidia.com/
>
> - Mani
Yes, tested this patch with PCIe devices having default 128 and RC is 
falling back to 128 when pci device is added.
This is handled inside pci_configure_mps().
/        mpss = 128 << dev->pcie_mpss;/
/        if (mpss < p_mps && pci_pcie_type(bridge) == 
PCI_EXP_TYPE_ROOT_PORT) {/
/                pcie_set_mps(bridge, mpss);/
/                pci_info(dev, "Upstream bridge's Max Payload Size set 
to %d (was %d, max %d)\n",/
/                         mpss, p_mps, 128 << bridge->pcie_mpss);/
/                p_mps = pcie_get_mps(bridge);/
/        }/
//
Also getting the below print,
/[    2.011963] pci 0003:01:00.0: Upstream bridge's Max Payload Size set 
to 128 (was 256, max 256)/
>> - Mani
>>
>>> Konrad
>> -- 
>> மணிவண்ணன் சதாசிவம்
--
Thanks,
Praveenkumar
  
Manivannan Sadhasivam July 25, 2023, 6:06 a.m. UTC | #4
On Tue, Jul 25, 2023 at 10:16:04AM +0530, Praveenkumar I wrote:
> 
> On 7/24/2023 7:39 PM, Manivannan Sadhasivam wrote:
> > On Mon, Jul 24, 2023 at 06:38:55PM +0530, Manivannan Sadhasivam wrote:
> > > On Mon, Jul 24, 2023 at 02:53:37PM +0200, Konrad Dybcio wrote:
> > > > On 24.07.2023 14:47, Praveenkumar I wrote:
> > > > > Set 256 bytes as payload size for IPQ9574 via early fixup. This allows
> > > > > PCIe RC to use the max payload size when a capable link partner is
> > > > > connected.
> > > > > 
> > > > > Signed-off-by: Praveenkumar I <quic_ipkumar@quicinc.com>
> > > > > ---
> > > > [...]
> > > > 
> > > > > +static void qcom_fixup_mps_256(struct pci_dev *dev)
> > > > > +{
> > > > > +	pcie_set_mps(dev, 256);
> > > > Looks like setting "dev->pcie_mpss = 1" here would make the PCIe generic
> > > > code take care of this.
> > > > 
> > > Right, also this setting should not be PCI-PCI bridge specific but rather
> > > controller specific.
> > > 
> > Wait, have you tested this patch with PCIe devices having MPS < 256 i.e.,
> > default 128?
> > 
> > Take a look at this discussion: https://lore.kernel.org/all/20230608093652.1409485-1-vidyas@nvidia.com/
> > 
> > - Mani
> Yes, tested this patch with PCIe devices having default 128 and RC is
> falling back to 128 when pci device is added.
> This is handled inside pci_configure_mps().
> /        mpss = 128 << dev->pcie_mpss;/
> /        if (mpss < p_mps && pci_pcie_type(bridge) ==
> PCI_EXP_TYPE_ROOT_PORT) {/
> /                pcie_set_mps(bridge, mpss);/
> /                pci_info(dev, "Upstream bridge's Max Payload Size set to %d
> (was %d, max %d)\n",/
> /                         mpss, p_mps, 128 << bridge->pcie_mpss);/
> /                p_mps = pcie_get_mps(bridge);/
> /        }/
> //
> Also getting the below print,
> /[    2.011963] pci 0003:01:00.0: Upstream bridge's Max Payload Size set to
> 128 (was 256, max 256)/

Ok. But for setting MPS, you need to change the DEVCTL register in post_init
sequence for IPQ9574. It is not a quirk, so you cannot use fixups.

- Mani

> > > - Mani
> > > 
> > > > Konrad
> > > -- 
> > > மணிவண்ணன் சதாசிவம்
> --
> Thanks,
> Praveenkumar
  
Praveenkumar I July 26, 2023, 7:28 a.m. UTC | #5
On 7/25/2023 11:36 AM, Manivannan Sadhasivam wrote:
> On Tue, Jul 25, 2023 at 10:16:04AM +0530, Praveenkumar I wrote:
>> On 7/24/2023 7:39 PM, Manivannan Sadhasivam wrote:
>>> On Mon, Jul 24, 2023 at 06:38:55PM +0530, Manivannan Sadhasivam wrote:
>>>> On Mon, Jul 24, 2023 at 02:53:37PM +0200, Konrad Dybcio wrote:
>>>>> On 24.07.2023 14:47, Praveenkumar I wrote:
>>>>>> Set 256 bytes as payload size for IPQ9574 via early fixup. This allows
>>>>>> PCIe RC to use the max payload size when a capable link partner is
>>>>>> connected.
>>>>>>
>>>>>> Signed-off-by: Praveenkumar I <quic_ipkumar@quicinc.com>
>>>>>> ---
>>>>> [...]
>>>>>
>>>>>> +static void qcom_fixup_mps_256(struct pci_dev *dev)
>>>>>> +{
>>>>>> +	pcie_set_mps(dev, 256);
>>>>> Looks like setting "dev->pcie_mpss = 1" here would make the PCIe generic
>>>>> code take care of this.
>>>>>
>>>> Right, also this setting should not be PCI-PCI bridge specific but rather
>>>> controller specific.
>>>>
>>> Wait, have you tested this patch with PCIe devices having MPS < 256 i.e.,
>>> default 128?
>>>
>>> Take a look at this discussion: https://lore.kernel.org/all/20230608093652.1409485-1-vidyas@nvidia.com/
>>>
>>> - Mani
>> Yes, tested this patch with PCIe devices having default 128 and RC is
>> falling back to 128 when pci device is added.
>> This is handled inside pci_configure_mps().
>> /        mpss = 128 << dev->pcie_mpss;/
>> /        if (mpss < p_mps && pci_pcie_type(bridge) ==
>> PCI_EXP_TYPE_ROOT_PORT) {/
>> /                pcie_set_mps(bridge, mpss);/
>> /                pci_info(dev, "Upstream bridge's Max Payload Size set to %d
>> (was %d, max %d)\n",/
>> /                         mpss, p_mps, 128 << bridge->pcie_mpss);/
>> /                p_mps = pcie_get_mps(bridge);/
>> /        }/
>> //
>> Also getting the below print,
>> /[    2.011963] pci 0003:01:00.0: Upstream bridge's Max Payload Size set to
>> 128 (was 256, max 256)/
> Ok. But for setting MPS, you need to change the DEVCTL register in post_init
> sequence for IPQ9574. It is not a quirk, so you cannot use fixups.
Sure, will add in post_init of IPQ9574.

--
Thanks,
Praveenkumar
>
> - Mani
>
>>>> - Mani
>>>>
>>>>> Konrad
>>>> -- 
>>>> மணிவண்ணன் சதாசிவம்
>> --
>> Thanks,
>> Praveenkumar
  
Praveenkumar I July 26, 2023, 9:22 a.m. UTC | #6
On 7/25/2023 11:36 AM, Manivannan Sadhasivam wrote:
> On Tue, Jul 25, 2023 at 10:16:04AM +0530, Praveenkumar I wrote:
>> On 7/24/2023 7:39 PM, Manivannan Sadhasivam wrote:
>>> On Mon, Jul 24, 2023 at 06:38:55PM +0530, Manivannan Sadhasivam wrote:
>>>> On Mon, Jul 24, 2023 at 02:53:37PM +0200, Konrad Dybcio wrote:
>>>>> On 24.07.2023 14:47, Praveenkumar I wrote:
>>>>>> Set 256 bytes as payload size for IPQ9574 via early fixup. This allows
>>>>>> PCIe RC to use the max payload size when a capable link partner is
>>>>>> connected.
>>>>>>
>>>>>> Signed-off-by: Praveenkumar I <quic_ipkumar@quicinc.com>
>>>>>> ---
>>>>> [...]
>>>>>
>>>>>> +static void qcom_fixup_mps_256(struct pci_dev *dev)
>>>>>> +{
>>>>>> +	pcie_set_mps(dev, 256);
>>>>> Looks like setting "dev->pcie_mpss = 1" here would make the PCIe generic
>>>>> code take care of this.
>>>>>
>>>> Right, also this setting should not be PCI-PCI bridge specific but rather
>>>> controller specific.
>>>>
>>> Wait, have you tested this patch with PCIe devices having MPS < 256 i.e.,
>>> default 128?
>>>
>>> Take a look at this discussion: https://lore.kernel.org/all/20230608093652.1409485-1-vidyas@nvidia.com/
>>>
>>> - Mani
>> Yes, tested this patch with PCIe devices having default 128 and RC is
>> falling back to 128 when pci device is added.
>> This is handled inside pci_configure_mps().
>> /        mpss = 128 << dev->pcie_mpss;/
>> /        if (mpss < p_mps && pci_pcie_type(bridge) ==
>> PCI_EXP_TYPE_ROOT_PORT) {/
>> /                pcie_set_mps(bridge, mpss);/
>> /                pci_info(dev, "Upstream bridge's Max Payload Size set to %d
>> (was %d, max %d)\n",/
>> /                         mpss, p_mps, 128 << bridge->pcie_mpss);/
>> /                p_mps = pcie_get_mps(bridge);/
>> /        }/
>> //
>> Also getting the below print,
>> /[    2.011963] pci 0003:01:00.0: Upstream bridge's Max Payload Size set to
>> 128 (was 256, max 256)/
> Ok. But for setting MPS, you need to change the DEVCTL register in post_init
> sequence for IPQ9574. It is not a quirk, so you cannot use fixups.
Sorry, if I do so, then the above mentioned issue will come here as well 
right?

This one: https://lore.kernel.org/all/20230608093652.1409485-1-vidyas@nvidia.com/

> - Mani
>
>>>> - Mani
>>>>
>>>>> Konrad
>>>> -- 
>>>> மணிவண்ணன் சதாசிவம்
>> --
>> Thanks,
>> Praveenkumar
  
Praveenkumar I July 26, 2023, 10:49 a.m. UTC | #7
On 7/26/2023 2:52 PM, Praveenkumar I wrote:
>
> On 7/25/2023 11:36 AM, Manivannan Sadhasivam wrote:
>> On Tue, Jul 25, 2023 at 10:16:04AM +0530, Praveenkumar I wrote:
>>> On 7/24/2023 7:39 PM, Manivannan Sadhasivam wrote:
>>>> On Mon, Jul 24, 2023 at 06:38:55PM +0530, Manivannan Sadhasivam wrote:
>>>>> On Mon, Jul 24, 2023 at 02:53:37PM +0200, Konrad Dybcio wrote:
>>>>>> On 24.07.2023 14:47, Praveenkumar I wrote:
>>>>>>> Set 256 bytes as payload size for IPQ9574 via early fixup. This 
>>>>>>> allows
>>>>>>> PCIe RC to use the max payload size when a capable link partner is
>>>>>>> connected.
>>>>>>>
>>>>>>> Signed-off-by: Praveenkumar I <quic_ipkumar@quicinc.com>
>>>>>>> ---
>>>>>> [...]
>>>>>>
>>>>>>> +static void qcom_fixup_mps_256(struct pci_dev *dev)
>>>>>>> +{
>>>>>>> +    pcie_set_mps(dev, 256);
>>>>>> Looks like setting "dev->pcie_mpss = 1" here would make the PCIe 
>>>>>> generic
>>>>>> code take care of this.
>>>>>>
>>>>> Right, also this setting should not be PCI-PCI bridge specific but 
>>>>> rather
>>>>> controller specific.
>>>>>
>>>> Wait, have you tested this patch with PCIe devices having MPS < 256 
>>>> i.e.,
>>>> default 128?
>>>>
>>>> Take a look at this discussion: 
>>>> https://lore.kernel.org/all/20230608093652.1409485-1-vidyas@nvidia.com/ 
>>>>
>>>>
>>>> - Mani
>>> Yes, tested this patch with PCIe devices having default 128 and RC is
>>> falling back to 128 when pci device is added.
>>> This is handled inside pci_configure_mps().
>>> /        mpss = 128 << dev->pcie_mpss;/
>>> /        if (mpss < p_mps && pci_pcie_type(bridge) ==
>>> PCI_EXP_TYPE_ROOT_PORT) {/
>>> /                pcie_set_mps(bridge, mpss);/
>>> /                pci_info(dev, "Upstream bridge's Max Payload Size 
>>> set to %d
>>> (was %d, max %d)\n",/
>>> /                         mpss, p_mps, 128 << bridge->pcie_mpss);/
>>> /                p_mps = pcie_get_mps(bridge);/
>>> /        }/
>>> //
>>> Also getting the below print,
>>> /[    2.011963] pci 0003:01:00.0: Upstream bridge's Max Payload Size 
>>> set to
>>> 128 (was 256, max 256)/
>> Ok. But for setting MPS, you need to change the DEVCTL register in 
>> post_init
>> sequence for IPQ9574. It is not a quirk, so you cannot use fixups.
> Sorry, if I do so, then the above mentioned issue will come here as 
> well right?
>
> This one: 
> https://lore.kernel.org/all/20230608093652.1409485-1-vidyas@nvidia.com/
Sorry, confused a bit here. After moving the MPS setting to post_init, 
pci_configure_mps() is taking care if 128 byte PCIe device is connected.
Posted a updated patch.

- Praveenkumar
>
>> - Mani
>>
>>>>> - Mani
>>>>>
>>>>>> Konrad
>>>>> -- 
>>>>> மணிவண்ணன் சதாசிவம்
>>> -- 
>>> Thanks,
>>> Praveenkumar
  

Patch

diff --git a/drivers/pci/controller/dwc/pcie-qcom.c b/drivers/pci/controller/dwc/pcie-qcom.c
index cee4e400a695..6695bc3b122f 100644
--- a/drivers/pci/controller/dwc/pcie-qcom.c
+++ b/drivers/pci/controller/dwc/pcie-qcom.c
@@ -1631,6 +1631,11 @@  static void qcom_fixup_class(struct pci_dev *dev)
 {
 	dev->class = PCI_CLASS_BRIDGE_PCI_NORMAL;
 }
+
+static void qcom_fixup_mps_256(struct pci_dev *dev)
+{
+	pcie_set_mps(dev, 256);
+}
 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);
@@ -1638,6 +1643,7 @@  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, 0x1000, qcom_fixup_class);
 DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x1001, qcom_fixup_class);
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_QCOM, 0x1108, qcom_fixup_mps_256);
 
 static const struct dev_pm_ops qcom_pcie_pm_ops = {
 	NOIRQ_SYSTEM_SLEEP_PM_OPS(qcom_pcie_suspend_noirq, qcom_pcie_resume_noirq)