svn commit: r309658 - head/sys/vm

[Mark Johnston]

On Wed, Dec 07, 2016 at 01:26:47PM -0800, Gleb Smirnoff wrote:
> On Tue, Dec 06, 2016 at 10:52:45PM +0000, Mark Johnston wrote:
> M> Author: markj
> M> Date: Tue Dec  6 22:52:45 2016
> M> New Revision: 309658
> M> URL: https://svnweb.freebsd.org/changeset/base/309658
> M> 
> M> Log:
> M>   Provide dummy sysctls for v_cache_count and v_tcached.
> M>   
> M>   Some utilities (notably top(1)) exit if any of their input sysctls don't
> M>   exist, and the removal of the above-mentioned PG_CACHE-related sysctls
> M>   makes it difficult to run such utilities on different versions of the
> M>   kernel without recompiling.
> M>   
> M>   Requested by:	bde
> M> 
> M> Modified:
> M>   head/sys/vm/vm_meter.c
> M> 
> M> Modified: head/sys/vm/vm_meter.c
> M> ==============================================================================
> M> --- head/sys/vm/vm_meter.c	Tue Dec  6 22:48:28 2016	(r309657)
> M> +++ head/sys/vm/vm_meter.c	Tue Dec  6 22:52:45 2016	(r309658)
> M> @@ -314,3 +314,14 @@ VM_STATS_VM(v_forkpages, "VM pages affec
> M>  VM_STATS_VM(v_vforkpages, "VM pages affected by vfork()");
> M>  VM_STATS_VM(v_rforkpages, "VM pages affected by rfork()");
> M>  VM_STATS_VM(v_kthreadpages, "VM pages affected by fork() by kernel");
> M> +
> M> +#ifndef BURN_BRIDGES
> M> +/*
> M> + * Provide compatibility sysctls for the benefit of old utilities which exit
> M> + * with an error if they cannot be found.
> M> + */
> M> +SYSCTL_UINT(_vm_stats_vm, OID_AUTO, v_cache_count, CTLFLAG_RD,
> M> +    (u_int *)NULL, 0, "Dummy for compatibility");
> M> +SYSCTL_UINT(_vm_stats_vm, OID_AUTO, v_tcached, CTLFLAG_RD,
> M> +    (u_int *)NULL, 0, "Dummy for compatibility");
> M> +#endif
> 
> IMHO, there should be some garbage collecting timeout for them. I'd suggest to
> delete them from head after stable/12 branch is forked.

The intent was to improve compatibility with old userlands, not just to
help bridge the transition from 11.0 to 12.0. In general I'm all for
removing cruft, but these sysctls are unobtrusive and I don't see any
reason to remove them in a specific time frame provided that nothing
else breaks compatibility for top(1) (and vmstat(8)). Please feel free
to add a timeout if you still disagree, though.