Support run-time configuration of the PIPE_MINDIRECT threshold.

[rwatson]

Summary

PIPE_MINDIRECT determines at what (blocking) write size one-copy
optimizations are applied in pipe(2) I/O. That threshold hasn't
been tuned since the 1990s when this code was originally
committed, and allowing run-time reconfiguration will make it
easier to assess whether contemporary microarchitectures would
prefer a different threshold.

(On our local RPi4 baords, the 8k default would ideally be at least
32k, but it's not clear how generalizable that observation is.)

[jrtc27]

pipe.h has:

/*
 * PIPE_MINDIRECT MUST be smaller than PIPE_SIZE and MUST be bigger
 * than PIPE_BUF.
 */

Do we need to do any validation of the value?

[rwatson]

In D29819#669007, @jrtc27 wrote:

pipe.h has:

/*
 * PIPE_MINDIRECT MUST be smaller than PIPE_SIZE and MUST be bigger
 * than PIPE_BUF.
 */

Do we need to do any validation of the value?

I wasn't imagining we keep this around indefinitely or that it was a regularly frobbed setting. But adding validation is not difficult, and some other similar-ish sysctls do have such checking (e.g., SB_MAX). The value of PIPE_SIZE might also be interesting to revisit -- or at least its implied use here. MAXPIPESIZE has a more sensible value, and if one's using MAXPIPESIZE, one might as well not bother doing a check as other limits will prevent overflowing that [I believe].

[rwatson]

Use a sysctl handler to limit possible values of kern.ipc.pipe_mindirect.

[rwatson]

Whitespace fix.

[rwatson]

Type consistency.

[jrtc27]

Style nit otherwise looks sensible

sys/kern/sys_pipe.c

274

style(9) is to not rely on implicit boolean conversions, ie != 0 and == NULL

[rwatson]

• Style fix: Differentiate arbitrary integer values from booleans.