Re: git: 73f9251c0b52 - main - Compute 'maxproc'/'maxfiles' from memory amount; Expand/fix comments

On Thu, May 15, 2025 at 02:23:29PM +0000, Olivier Certner wrote:
> The branch main has been updated by olce:
> 
> URL: https://cgit.FreeBSD.org/src/commit/?id=73f9251c0b52ac0bfb2f2baff27493d6bae216f6
> 
> commit 73f9251c0b52ac0bfb2f2baff27493d6bae216f6
> Author:     Olivier Certner <olce@FreeBSD.org>
> AuthorDate: 2025-05-02 12:47:32 +0000
> Commit:     Olivier Certner <olce@FreeBSD.org>
> CommitDate: 2025-05-15 14:22:53 +0000
> 
>     Compute 'maxproc'/'maxfiles' from memory amount; Expand/fix comments
>     
>     Change the formulae so that 'maxproc' and 'maxfiles' are computed based
>     on the physical memory amount, and not the corresponding number of
>     pages.  Make sure that the practical results are unchanged for all
>     architectures (which is possible since all have 4096 as their
>     PAGE_SIZE).  Despite this, we do this change to make it clearer how many
>     units of these values are computed per MB, which is easier on readers.
Not all architectures have PAGE_SIZE of 4K.  Not going back to the historic
arches, and out of tree patches for x86, I believe that arm64 has in-tree
support for 16K pages.

VM system is fine with the page sizes from 4K to 32K.

>     
>     Change the comments accordingly and expand them to indicate which parts
>     of the computations are actually not used by default.  In passing, fix the
>     comments about default values and limits for 'maxfiles' (they were off).
>     
>     No functional change (intended).
>     
>     MFC after:      3 days
>     Sponsored by:   The FreeBSD Foundation
>     Differential Revision:  https://reviews.freebsd.org/D50124
> ---
>  sys/kern/subr_param.c | 44 +++++++++++++++++++++++++++-----------------
>  1 file changed, 27 insertions(+), 17 deletions(-)
> 
> diff --git a/sys/kern/subr_param.c b/sys/kern/subr_param.c
> index f4359efec466..471640c290a7 100644
> --- a/sys/kern/subr_param.c
> +++ b/sys/kern/subr_param.c
> @@ -69,13 +69,14 @@
>  #    define	HZ_VM HZ
>  #  endif
>  #endif
> +/* See the comments in init_param2() for these. */
>  #define	NPROC (20 + 16 * maxusers)
> -#ifndef NBUF
> -#define NBUF 0
> -#endif
>  #ifndef MAXFILES
>  #define	MAXFILES (40 + 32 * maxusers)
>  #endif
> +#ifndef NBUF
> +#define NBUF 0
> +#endif
>  
>  static int sysctl_kern_vm_guest(SYSCTL_HANDLER_ARGS);
>  
> @@ -275,13 +276,14 @@ init_param1(void)
>  void
>  init_param2(long physpages)
>  {
> +	long maxproc_clamp, maxfiles_clamp;
>  
>  	TSENTER();
>  	/* Base parameters */
>  	maxusers = MAXUSERS;
>  	TUNABLE_INT_FETCH("kern.maxusers", &maxusers);
>  	if (maxusers == 0) {
> -		maxusers = physpages / (2 * 1024 * 1024 / PAGE_SIZE);
> +		maxusers = pgtok(physpages) / (2 * 1024);
>  		if (maxusers < 32)
>  			maxusers = 32;
>  #ifdef VM_MAX_AUTOTUNE_MAXUSERS
> @@ -290,35 +292,43 @@ init_param2(long physpages)
>  #endif
>                  /*
>                   * Scales down the function in which maxusers grows once
> -                 * we hit 384.
> +                 * we hit 384 (16MB to get a new "user").
>                   */
>                  if (maxusers > 384)
>                          maxusers = 384 + ((maxusers - 384) / 8);
>          }
>  
>  	/*
> -	 * The following can be overridden after boot via sysctl.  Note:
> -	 * unless overridden, these macros are ultimately based on maxusers.
> -	 * Limit maxproc so that kmap entries cannot be exhausted by
> -	 * processes.
> +	 * The following can be overridden after boot via sysctl.  Note: unless
> +	 * overridden, these macros are ultimately based on 'maxusers'.  Limit
> +	 * maxproc so that kmap entries cannot be exhausted by processes.  The
> +	 * default for 'maxproc' linearly scales as 16 times 'maxusers' (so,
> +	 * linearly with 8 processes per MB up to 768MB, then 1 process per MB;
> +	 * overridable by a tunable), and is then clamped at 21 + 1/3 processes
> +	 * per MB (which never happens by default as long as physical memory is
> +	 * > ~1.5MB).
>  	 */
>  	maxproc = NPROC;
>  	TUNABLE_INT_FETCH("kern.maxproc", &maxproc);
> -	if (maxproc > (physpages / 12))
> -		maxproc = physpages / 12;
> +	maxproc_clamp = pgtok(physpages) / (3 * 1024 / 64);
> +	if (maxproc > maxproc_clamp)
> +		maxproc = maxproc_clamp;
>  	if (maxproc > pid_max)
>  		maxproc = pid_max;
>  	maxprocperuid = (maxproc * 9) / 10;
>  
>  	/*
> -	 * The default limit for maxfiles is 1/12 of the number of
> -	 * physical page but not less than 16 times maxusers.
> -	 * At most it can be 1/6 the number of physical pages.
> +	 * 'maxfiles' by default is set to 32 files per MB (overridable by
> +	 * a tunable), and is then clamped at 64 files per MB (which thus never
> +	 * happens by default).  (The default MAXFILES is for all practical
> +	 * purposes not used, as it gives a lower value than 32 files per MB as
> +	 * soon as there is more than ~2.5MB of memory.)
>  	 */
> -	maxfiles = imax(MAXFILES, physpages / 8);
> +	maxfiles = imax(MAXFILES, pgtok(physpages) / (1024 / 32));
>  	TUNABLE_INT_FETCH("kern.maxfiles", &maxfiles);
> -	if (maxfiles > (physpages / 4))
> -		maxfiles = physpages / 4;
> +	maxfiles_clamp = pgtok(physpages) / (1024 / 64);
> +	if (maxfiles > maxfiles_clamp)
> +		maxfiles = maxfiles_clamp;
>  	maxfilesperproc = (maxfiles / 10) * 9;
>  	TUNABLE_INT_FETCH("kern.maxfilesperproc", &maxfilesperproc);
>