re: kern/57683: uvm_pglistalloc_s_ps() et al integer overflow

[matthew green <mrg%netbsd.org@localhost>]

The following reply was made to PR kern/57683; it has been noted by GNATS.

From: matthew green <mrg%netbsd.org@localhost>
To: gnats-bugs%netbsd.org@localhost, tnn%nygren.pp.se@localhost
Cc: kern-bug-people%netbsd.org@localhost, gnats-admin%netbsd.org@localhost,
    netbsd-bugs%netbsd.org@localhost
Subject: re: kern/57683: uvm_pglistalloc_s_ps() et al integer overflow
Date: Mon, 06 Nov 2023 13:25:18 +1100

 > But problems will I think only occur here on systems with more than 16 T=
 B of RAM.
 
 we have all sorts of issues with anything beyond 8TB (signed 32 bit
 4K pages overflow) because we use "int" all over uvm to count pages.
 
 this is one of them, and if it's enough to fix your problem, then it
 is worth doing but i expect to see *all sorts of issues* in UVM when
 there's actually huge ram.
 
 this is not a small problem AFAICT.  getting to 16T by using unsigned
 page count would help, but it's not nearly enough for the future.
 
 
 .mrg.