Re: [PATCH] pcie: ti: Provide patch to force GEN1 PCIe operation

[Kishon Vijay Abraham I <kishon@ti.com>]

+ Joao, Jingoo

Hi,

On Monday 16 January 2017 03:01 PM, Lukasz Majewski wrote:
> Hi Kishon,
> 
>> Hi Łukasz,
>>
>> On Monday 16 January 2017 12:19 PM, Lukasz Majewski wrote:
>>> Hi Kishon,
>>>
>>>> Hi,
>>>>
>>>> On Sunday 15 January 2017 06:49 PM, Lukasz Majewski wrote:
>>>>> Some devices (due to e.g. bad PCIe signal integrity) require to
>>>>> run with forced GEN1 speed on PCIe bus.
>>>>>
>>>>> This patch changes the speed explicitly on dra7 based devices when
>>>>> proper device tree attribute is defined for the PCIe controller.
>>>>>
>>>>> Signed-off-by: Lukasz Majewski <lukma@denx.de>
>>>>
>>>> Bjorn has already queued a patch to do the same thing
>>>> https://git.kernel.org/cgit/linux/kernel/git/helgaas/pci.git/log/?h=pci/host-dra7xx
>>>
>>> It seems like Bjorn only modifies CAP registers.
>>
>> The patch also modifies the LNKCTL2 register.
>>>
>>> He also needs to change register with 0x080C offset to actually
>>> ( PCIECTRL_PL_WIDTH_SPEED_CTL )
>>
>> This bit is used to initiate speed change (after the link is
>> initialized in GEN1). Resetting the bit (like what you have done
>> here) prevents speed change.
> 
> This is strange, but e2e advised me to do things as I did in the patch
> to _force_ GEN1 operation on PCIe2 port [1] (AM5728)
> 
> Link:
> [1] https://e2e.ti.com/support/arm/sitara_arm/f/791/t/566421
> 
> Both patches modify 0x5180 007C register to set GEN1 capability
> (PCI_EXP_LNKCAP_SLS_2_5GB)
> 
> The problem is with second register (in your patch):
> 
> From SPRUHZ6G TRM:
> 
> PCIECTRL_EP_DBICS_LNK_CAS_2 (0x5180 00A0)
> - TRGT_LINK_SPEED (Reset 0x1) - "Target Link Speed" - no more
>   description in TRM
> 
> It is set to PCI_EXP_LNKCAP_SLS_2_5GB = 0x1, which is the same as
> default /reset value.

The default value is 0x2 (or else none of the cards would have enumerated in GEN2)
> 
> 
> Could you clarify which way to _force_ PCIe GEN1 operation is correct?
> Mine shows differences in lspci output (as posted in [1]).

You'll see the difference even with the patch in Bjorn's tree ;-)

I think these are 2 different approaches to keep the link at GEN1. Joao or
Jingoo, do you have any suggestion here?

> 
>>
>> IMO the better way is to set the LNKCTL2 to GEN1 instead of hacking
>> the IP register.
> 
> From the original patch description:
> 
> "Add support to force Root Complex to work in GEN1 mode if so desired,
> but don't force GEN1 mode on any board just yet."
> 
> Are there any (floating around) patches allowing forcing GEN1 operation
> on any board (I would like to reuse/port them to my current solution)?

For setting to GEN1 mode, "max-link-speed" should be set to 1 in dt with the
patch in Bjorn's tree.

Thanks
Kishon