Fix riscv_machine_timer mtime read

[Dolu1990]

This pullrequest fix #10788

[codecov-io]

Codecov Report

Merging #10859 into master will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master   #10859   +/-   ##
=======================================
  Coverage   53.27%   53.27%           
=======================================
  Files         215      215           
  Lines       25851    25851           
  Branches     5695     5695           
=======================================
  Hits        13773    13773           
  Misses       9762     9762           
  Partials     2316     2316

Impacted Files	Coverage Δ
subsys/bluetooth/host/hci_core.c	2.46% <0%> (ø)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 40bf45a...f25db14. Read the comment docs.

[Dolu1990]

This issue could lead to some timer freeze.
For instance on a system which take (let's say) 5 cycles between reading the mtime low and the mtime high, a timer freez would occure each 5/2^32 timer interrupts.

If that system is running at 100 Mhz, the freez would be about 40 seconds, and if the timer is running at 1Khz it would happen each 10 days.

[Dolu1990]

@pgielda @kgugala Ready for review

[nategraff-sifive]

Thanks for your contribution @Dolu1990!

Right now there's also a rewrite of a whole bunch of timer drivers in PR #10556, which includes a similar change for the RISC-V machine timer driver along with a number of other improvements.

[brouhaha]

@Dolu1990 The code looks good. I think your analysis of the freeze is wrong. Unless some nested higher-priority interrupt(s) block the timer handler for near 2^32 mtime cycles, the loop will never iterate more than twice before it succeeds to get two consecutive identical reads of mtimeh.

[Dolu1990]

Oulala, look like there is some serious business around the timer ^^
thanks :)

[nategraff-sifive]

Closing, #10556 included this same change. Thanks again for your contribution @Dolu1990!

[Dolu1990]

<3