Hi Stephen,
On 14 June 2017 at 10:48, Stephen Warren swarren@wwwdotorg.org wrote:
On 06/14/2017 10:32 AM, Stephen Warren wrote:
On 06/14/2017 05:30 AM, Simon Glass wrote: ...
At this point I'm thinking of applying the patches except for the one that actually enables livetree. Once we figure out what is happening with jetson-tk1, or you can confirm that it works on beaver, we can look at the final patch.
Beaver PCIe works OK with or without this series. Note that while my automated test system contains a Beaver, it actually uses a USB Ethernet adapter for network tests rather than the built-in PCIe Ethernet, since the tests co-exist with another kernel testing system that only supports USB Ethernet on this board.
This series (I didn't check which patch) breaks PCIe on both p2371-2180 (Jetson TX1) and Jetson TK1. PCIe Ethernet works fine without this series on both these boards, and is tested by my automated system.
Note that I didn't see the boot-time MMC timeout errors (that were mentioned elsewhere) on any of these 3 boards with or without this series, with the baseline commit being 8cb3ce64f936 Merge git://git.denx.de/u-boot-dm.
I should also mention that I'm not convinced that applying this series, even with the new feature only enabled on Beaver, is a good idea.
If we were confident that there was some purely logic bug that only affects Jetson TK1/TX1, then applying the series only for Beaver would be fine. However, we haven't root-caused the issue yet. If the problem is in fact some kind of memory corruption, stack overflow, or uninitialized data issue, then it's just a complete fluke that we happen not to see the issue on Beaver. If that's the case, then the issue could randomly appear on Beaver later due to unrelated code/data layout changes. I'd prefer to avoid that possibility. Let's root-cause the bug first and fix it, then apply the changes. Also, as I mentioned I don't test Beaver PCIe in my automated system, so any problems there run the risk of being undetected for longer.
Thanks for testing. I suspect the bisect would just point you to the last patch which enables things.
Unfortunately my Jetson-TK1 does not work with PCI before or after this series as I mentioned before. I have a TX1 somewhere but cannot find it at present. Is there anyone there that can take a look at what is different on Jetson-TK1? Or can I send my board somewhere to be re-imaged (to fix the PMIC)?
Since you don't test beaver, one fallback would be apply it just for nyan-big, as per my original series. The problem is likely in the device-tree decoding for PCI, and without a working board I doubt it is possible to resolve it.
Regards, Simon