>Number:         36019
>Category:       kern
>Synopsis:       uvm_readahead.c:ra_startio() assumes (MAXPHYS & (MAXPHYS - 1)) == 0
>Confidential:   no
>Severity:       non-critical
>Priority:       low
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Mar 16 19:45:00 +0000 2007
>Originator:     Izumi Tsutsui
>Release:        NetBSD 4.99.15
>Organization:
>Environment:
System: NetBSD mirage 4.99.15
Architecture: m68000
Machine: sun2
maybe sun3 is also affected and
4.0_BETA2 has the same problem

>Description:
There is the following KASSERT() in sys/uvm/uvm_readahead.c:ra_startoi():
---
#define	RA_IOCHUNK	MAXPHYS			/* read-ahead i/o chunk size */
 :
		const size_t chunksize = RA_IOCHUNK;
 :
		KASSERT((chunksize & (chunksize - 1)) == 0);
		KASSERT((off & PAGE_MASK) == 0);
		bytelen = ((off + chunksize) & -(off_t)chunksize) - off;
		KASSERT((bytelen & PAGE_MASK) == 0);
		npages = orignpages = bytelen >> PAGE_SHIFT;
---
but (MAXPHYS & (MAXPHYS - 1)) could be != 0 on some ports
so "KASSERT((chunksize & (chunksize - 1)) == 0)" can fail.

On sun2 (and sun3), MAXPHYS is 0xe000.
This will cause a panic "genfs_getpages: too many pages" later
on a kernel without DIAGNOSTIC due to wrong bytelen and npages.

>How-To-Repeat:
Boot sun2 GENERIC kernel (on TME) and copy a file into filesystem.

>Fix:
Choose proper RA_IOCHUNK in any case?