NetBSD Problem Report #25639

Received: (qmail 14201 invoked by uid 605); 19 May 2004 13:46:04 -0000
Message-Id: <1084974359.582750.8991.nullmailer@yamt.dyndns.org>
Date: Wed, 19 May 2004 22:45:59 +0900
From: yamt@mwd.biglobe.ne.jp
Sender: gnats-bugs-owner@NetBSD.org
Reply-To: yamt@mwd.biglobe.ne.jp
To: gnats-bugs@gnats.netbsd.org
Subject: uvm_vslock / uvm_vsunlock problems
X-Send-Pr-Version: 3.95

>Number:         25639
>Category:       kern
>Synopsis:       uvm_vslock / uvm_vsunlock problems
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    kern-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Wed May 19 13:47:00 +0000 2004
>Closed-Date:    
>Last-Modified:  Thu May 20 16:25:00 +0000 2004
>Originator:     YAMAMOTO Takashi <yamt@mwd.biglobe.ne.jp>
>Release:        NetBSD 2.0E
>Organization:

>Environment:


System: NetBSD kaeru 2.0E NetBSD 2.0E (build.kaeru) #1310: Wed May 19 21:29:48 JST 2004 takashi@kaeru:/usr/home/takashi/work/kernel/build.kaeru i386
Architecture: i386
Machine: i386
>Description:
	from Stephan Uphoff.
	(http://mail-index.NetBSD.org/tech-kern/2003/07/10/0013.html)

> Problem 1:
> -----------
> uvm_vslock assumes that once a page is wired - it will stay wired until
> uvm_vsunlock (potentially) unwires it.
> This is not true for mmaped file pages when the file is truncated.
> (All managed mappings are released - page is freed)
> 
> Problem 2:
> ----------
> The uvm_vslock() / uvm_vsunlock() pair assumes only one thread 
> of control per address space.
> Because of clone(2) or scheduler activations (lwps) this is no longer always
> true.
> 
> Problem 3:
> ----------
> uvm_vslock indirectly calls uvm_fault.
> If uvm_fault is called without an VM_PROT_WRITE 
> bit in accesstype to wire a loaned page it will not
> break the loan.
> This can cause the buffer created by physio to eventual map to the
> wrong pages and could break a KASSERT in uvm_pagefree.

>How-To-Repeat:
>Fix:

> Solutions:
> ----------
> sys___sysctl can probably use a kernel buffer for the (hopefully few) 
> occasions
> where copying is not allowed to block. (And copyout from the kernel buffer
> when it is save to do so)
> ( I assume this is the reason for the wiring but was too lazy to verify ) 
> 
> The uvm_fault problem (3) could be fixed by always breaking loans on wire 
> faults.
> 
> I believe due to the first two problems physio can not be made to work 
> correctly without resorting to mechanisms closely related to page loaning. 
> (or the use of bounce buffers :-)
> Unfortunately this will need a major effort.
>Release-Note:
>Audit-Trail:

From: "Darrin B. Jewell" <dbj@netbsd.org>
To: yamt@mwd.biglobe.ne.jp
Cc: gnats-bugs@gnats.netbsd.org
Subject: Re: kern/25639: uvm_vslock / uvm_vsunlock problems
Date: 20 May 2004 12:21:32 -0400

 --=-=-=


 I noticed this problem as well and had boiled it down to this
 reproducible test case.  Compile the following program and run it on
 on a raw character disk device.

 On i386, it will quickly start printing out pmap unwiring complaints.


 --=-=-=
 Content-Disposition: attachment; filename=physiotest.c
 Content-Description: test program to find physio bug


 #include <sys/param.h>
 #include <sys/types.h>
 #include <sys/uio.h>
 #include <unistd.h>
 #include <fcntl.h>

 #include <pthread.h>

 #define NTHREADS 2

 pthread_t threads[NTHREADS];

 char iobuf[DEV_BSIZE];
 int fd;

 void *
 doio(void *p)
 {
 	int index = ((pthread_t *)p)-(&threads[0]);
 	int cnt = 0;
 	while (1) {
 		pwrite(fd, iobuf, sizeof(iobuf),
 		  10*MAXBSIZE*((cnt++)*NTHREADS+index));
 	}
 }

 int
 main(int argc, char *argv[])
 {
 	int i;

 	fd = open(argv[1], O_WRONLY, 0);

 	for (i = 0; i < NTHREADS; i++)
 		pthread_create(&threads[i], NULL, &doio, &threads[i]);

 	pthread_join(threads[0], NULL);

 	return 0;
 }

 --=-=-=--
>Unformatted:

Home	PR Database Search