1. Tech-Unixplatformspec
  2. Messages

Re: MTIME update frequency

Greg Favor
 

On Wed, Nov 17, 2021 at 2:30 PM Darius Rad <darius@...> wrote:
> CSRs need to have the consistent time at the observation point. The fastest way in most system to "observe" value in CSRs is through the cache or through memory. So the difference between the two CSRs should not be observable to the extent that the following test fails:
>
> Hart-X (sends a message to Hart-Y):
>     Read time
>     Write timestamp message to memory
> Hart-Y:
>     Read timestamp message from memory (gets the data from cache or memory)
>     Read time
>     compare timestamp to time (the timestamp should not be in future)
>
> If this test is passes then the fact that the CSR in Hart-Y is 2 ticks behind or ahead of Hart-X is not observable.

Literally, such a "loosely synchronized" implementation would not be compliant with RISC-V architecture and its tight synchronization requirement.  As Darius noted (below), that is just one way to observe the clocks and, most importantly, the arch spec requirement is not defined in terms of an observation litmus test like above.  The literal arch spec requirement is the actual requirement for arch compliance.

That is perhaps a useful test, but not the only way to observe the clocks,
nor is it how the accuracy is specified in the architecture.  The
architecture does not say the accuracy is with respect to what is
observable in another hart.  The time could, for example, be observed by
something external to both harts and with higher precision.

> Setting the granularity to 1 ns provides the path for RISC-V implementations
> that want the fine resolution to acheive the goal without penalizing
> implementations that do not.

As Anup noted early in this thread:  One could require a fixed 1ns resolution (like ARM SBSA), but for RISC-V one must also honor the architectural requirement that all time CSRs are synchronized to within 1 tick.  And for multi-socket (or multi-die) systems it becomes even more challenging (arguably impossible) to synchronize time to within 1ns resolution across an SMP domain.

Greg
Join tech-unixplatformspec@lists.riscv.org to automatically receive all group messages.