etnaviv: perf fixes
This MR should improve/fix perfmon counter implementation. It makes use of the generic accumulated HW queries foundation.
To test the correctness of perfmon queries the following was done:
- add two sim queries to the kernel. The first one returns alternating values of 50 and 100 and the seconds one returns rising values.
- in mesa I added valuation for both simulated queries to ensure the values are as expected.
I can provide (and even try to upstream) the simulated counters too and ensure it does not get broken in the future.
Tests where on with a GC3000.
No regressions regarding occlusion queries.