open-power/snap

CAPI2.0 SNAP actions hang on certain linux kernels

luyong6 opened this issue · 3 comments

luyong6 commented

I created several SNAP actions like hls_helloworld, hls_memcopy, hls_mm_test on FX609 (and Semptian card), and I found on following Ubuntu kernel:

  • 4.14.49

  • 4.18.0
    After running 3-5 times successfully, the application will "hang". No response.

  • 4.15.0 and 4.17.0 don't have the phenomenon.

Some hints:

  1. Vaibhav has build the cxl module from 4.15.0 into 4.18.0, but the "hang" still happen.
  2. I built afp2 for FX609 and run on 4.18.0, there is no "hang".
  3. "hang" happens at the last MMIO polling.
jsvogt commented

Please describe the test procedure and "hang" situation.
I think I heard the problem is that the action doesn't finish the job, i.e. not "done", or the mmio reads polling for completion are failing.

jsvogt commented

Is this still occurring?

bmesnet commented

No description + this wasn't faced since a while