Comment 10 for bug 1913395

Revision history for this message
bugproxy (bugproxy) wrote : Comment bridged from LTC Bugzilla

------- Comment From <email address hidden> 2021-02-01 11:04 EDT-------
Thanks Christian.

I ran tests on both focal and groovy using the linked qemu PPA builds on s390x with a vfio-pci passthrough mlx adapter. Since the kernel fix hasn't hit focal yet, for both scenarios I used the patched groovy kernel that Frank provided in 1907421 to test the full fix (kernel advertisement of the DMA limit, which the modified QEMU can then exploit to ensure the limit isn't overrun).

For testing, I forced data-transfer workloads that could, prior to this fix, quickly and reliably overrun the DMA space - I left these workloads run for about 30 minutes each, verifying that were no transfer stalls / no errors in the QEMU log. Looks good!