From: Theo Buehler <tb@theobuehler.org>
Subject: Re: Towards killing uvm_km_kmemalloc_pla()
To: tech@openbsd.org
Date: Mon, 10 Nov 2025 08:13:08 +0100

On Sun, Nov 09, 2025 at 10:44:51PM +0000, Martin Pieuchot wrote:
> 14 years later, the conversion to km_alloc(9) is almost finished.  Diff
> below removes all dead code from uvm_km_kmemalloc_pla() to ease reviews
> and make it simpler to discuss the final conversion.  It includes:
> 
> - No behavior change
> - Remove unused arguments
> - Use `kmem_map' directly
> - Kill UVM_KMF_* flags and use UVM_FLAG_TRYLOCK where necessary
> 
> Once applied we can easily see the remaining differences with
> km_alloc(9) internal which will help finding a consensus for the next
> step.

ok tb

One tiny comment below.

> Index: kern/kern_malloc.c
> ===================================================================
> RCS file: /cvs/src/sys/kern/kern_malloc.c,v
> diff -u -p -r1.155 kern_malloc.c
> --- kern/kern_malloc.c	12 Jun 2025 20:37:58 -0000	1.155
> +++ kern/kern_malloc.c	9 Nov 2025 22:27:57 -0000
> @@ -217,12 +217,9 @@ malloc(size_t size, int type, int flags)
>  		mtx_leave(&malloc_mtx);
>  		npg = atop(round_page(allocsize));
>  		s = splvm();
> -		va = (caddr_t)uvm_km_kmemalloc_pla(kmem_map, NULL,
> -		    (vsize_t)ptoa(npg), 0,
> -		    ((flags & M_NOWAIT) ? UVM_KMF_NOWAIT : 0) |
> -		    ((flags & M_CANFAIL) ? UVM_KMF_CANFAIL : 0),
> -		    no_constraint.ucr_low, no_constraint.ucr_high,
> -		    0, 0, 0);
> +		va = (caddr_t)uvm_km_kmemalloc_pla((vsize_t)ptoa(npg),
> +		    (flags & (M_NOWAIT|M_CANFAIL)),

Not fussed, but I don't think the outer parentheses here help.

> +		    no_constraint.ucr_low, no_constraint.ucr_high);
>  		splx(s);
>  		if (va == NULL) {
>  			/*