From: Miod Vallat <miod@online.fr>
Subject: Re: uvm_fault_unwire() & truncation
To: Martin Pieuchot <mpi@grenadille.net>
Cc: tech@openbsd.org
Date: Sat, 2 Nov 2024 08:02:55 +0000

> Index: uvm/uvm_fault.c
> ===================================================================
> RCS file: /cvs/src/sys/uvm/uvm_fault.c,v
> diff -u -p -r1.135 uvm_fault.c
> --- uvm/uvm_fault.c	5 Sep 2023 05:08:26 -0000	1.135
> +++ uvm/uvm_fault.c	21 Oct 2024 11:02:33 -0000
> @@ -552,7 +552,7 @@ struct uvm_faultctx {
>  
>  int		uvm_fault_check(
>  		    struct uvm_faultinfo *, struct uvm_faultctx *,
> -		    struct vm_anon ***);
> +		    struct vm_anon ***, vm_fault_t);
>  
>  int		uvm_fault_upper(
>  		    struct uvm_faultinfo *, struct uvm_faultctx *,
> @@ -585,11 +585,6 @@ uvm_fault(vm_map_t orig_map, vaddr_t vad
>  	ufi.orig_map = orig_map;
>  	ufi.orig_rvaddr = trunc_page(vaddr);
>  	ufi.orig_size = PAGE_SIZE;	/* can't get any smaller than this */
> -	if (fault_type == VM_FAULT_WIRE)
> -		flt.narrow = TRUE;	/* don't look for neighborhood
> -					 * pages on wire */
> -	else
> -		flt.narrow = FALSE;	/* normal fault */

You should keep a default initialization of flt.narrow to FALSE, or at
least make sure uvm_fault_check() initializes it to FALSE in the
non-narrow code paths (e.g. around line 700 where all the fields are
initialized).

Apart from this, the diff is ok.