[U-Boot] [PATCH v3 25/35] dm: tegra: pci: Convert to livetree

[Stephen Warren]

On 07/06/2017 09:57 PM, Simon Glass wrote:
> Hi Stephen,
> 
> On 6 July 2017 at 12:55, Stephen Warren <swarren at wwwdotorg.org> wrote:
>>
>> On 07/06/2017 11:36 AM, sjg at google.com wrote:
>>>
>>> Update the tegra pci driver to support a live device tree. Fix the check
>>> for nvidia,num-lanes so that an error will actually be detected.
>>
>>
>> Simon,
>>
>> Something in the latest u-boot-dm/master branch breaks PCI or Ethernet on Jeton TK1 and TX1. I don't know which patch it is. test/py's ping test simply fails due to lack of any Ethernet adapter. I don't see any errors during either boot or the network test setup scripts, which include "pci enum":
>>
>> TK1 (built-in RTL8169 PCI Ethernet):
>> Tegra124 (Jetson TK1) # ping $serverip
>> No ethernet found.
>> ping failed; host 192.168.100.1 is not alive
>> Tegra124 (Jetson TK1) #
>>
>> TX1 (add-on RTL8169 PCI Ethernet):
>> Tegra210 (P2371-2180) # pci enum
>> ERROR: tegra-pcie: failed to power on PHY: -110
>> at /var/lib/jenkins/workspace/u-boot-denx_uboot_dm-master-build/src/u-boot/drivers/pci/pci_tegra.c:774/tegra_pcie_enable_controller()
>> ERROR: tegra-pcie: failed to enable controller
>> at /var/lib/jenkins/workspace/u-boot-denx_uboot_dm-master-build/src/u-boot/drivers/pci/pci_tegra.c:1154/pci_tegra_probe()
>> Tegra210 (P2371-2180) # setenv autoload no
>> Tegra210 (P2371-2180) # dhcp
>> No ethernet found.
>> Tegra210 (P2371-2180) #
>>
>> I also retested u-boot/master to make sure there's no test system infrastructure issue (that branch passes), and re-ran the failing test 3 times on TK1 with identical results. Other Tegra boards that I test all seem fine, at least as far as passing test/py.
>>
>> P.S. I'll be away next week and the week after. Tom Warren may be able to monitor the test system emails, but I don't think I've set him up with access to e.g. retrigger tests etc.; perhaps I should look into that...
> 
> As mentioned elsewhere that error (failing to power on PHY) happens
> for me always on my TK1 (and has for about a year) because it seems to
> have a broken PMIC. I do have a TX1 somewhere but cannot find it
> despite much searching.

I'm very puzzled that you claim your TK1 always has this issue; this 
issue is clearly not present in mainline U-Boot (unless it's very 
intermittent and my test system has got lucky, but it's very 
reproducible so I don't think so), and I don't believe there's anything 
in the the PMIC that can be altered permanently; the boot configuration 
is in OTP (One Time Programmable) memory (fuses?) and can't be changed. 
Are you absolutely 100% sure that you cleanly replaced your development 
version of U-Boot correctly and tested with a 100% unmodified mainline 
U-Boot? Also, please try (a) flashing unmodified mainline U-Boot (b) 
disconnecting EVERY cable from the TK1 and leaving it that way for 10 
minutes to drain any capacitors (or perhaps even overnight) (c) powering 
up to see if any problems still exist.

> Obviously I chose Tegra for the livetree work based on Nyan which I
> have and which is quick to develop on. I am beginning to regret this
> because:
> 
> - No one at Nvidia appears to be maintaining the code at present -
> e.g. DM enablement, Kconfig versions

Well, the existing code works perfectly. There's no need to change it. 
If you want to change it, that's fine, but you must take on the job of 
validating your changes and fixing any issues they introduce. It's not 
reasonable to introduce changes that break things and then try and push 
forward with them even in the face of known breakage caused by those 
changes. It's not reasonable to create patches that introduce problems 
and then expect others to do the work of debugging and fixing them.

> - The test report you provides is a signal, but it is not actionable for me

Why is it not actionable? You possess hardware that will reproduce the 
change (your TX1 board even if not the TK1 system). The test system 
source is checked into the U-Boot source tree and available to you, 
although I believe that simply running "pci enum; dhcp Image" should 
reproduce the problem so it's even simpler than running test/py.

> - As above due to broken/lost boards I am not able to test as well as
> I originally thought

Perhaps you should only modify the boards you can test? You mentioned 
that Nyan works for you. If so, limit the modifications to the 
boards/drivers/features you have available.

> Is there any way someone at Nvidia could jump into this and work on
> bringing mainline up to scratch?

If anyone can, it'd be Tom Warren.

P.S. I'm on vacation for the next 2 weeks, so won't be able to respond 
further for a while.