From mboxrd@z Thu Jan  1 00:00:00 1970
Date: Mon,  2 Jun 2014 23:00:15 -0700
From: Anthony Martin <ality@pbrane.org>
To: Fans of the OS Plan 9 from Bell Labs <9fans@9fans.net>
Message-ID: <20140603060015.GA28564@dinah>
References: <C29EB1DC-4A22-4149-9510-3054933D77C5@fb.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <C29EB1DC-4A22-4149-9510-3054933D77C5@fb.com>
Subject: Re: [9fans] kernel bug
Topicbox-Message-UUID: f4c20e18-ead8-11e9-9d60-3106f5b1d025

Yoann Padioleau <pad@fb.com> once said:
> in the newseg() function there is:
>
> [...]
>
> I think it should be
> if(mapsize > (SEGMAPSIZE))
>             mapsize = SEGMAPSIZE;

Yes, you're correct.

The code allows the creation of a segment VM map with more
than SEGMAPSIZE Ptes.

Personally, I'd remove the check entirely along with the
optimization of doubling the segment size, perhaps moving
it to ibrk. I think it's more likely that the segment will
grow if brk is called at least once.

  Anthony

P.S. It looks like this is a 15 year old bug:

1998/0916/sys/src/9/port/segment.c:62,67 - 1998/0919/sys/src/9/port/segment.c:62,70

    mapsize = ROUND(size, PTEPERTAB)/PTEPERTAB;
    if(mapsize > nelem(s->ssegmap)){
+       mapsize *= 2;
+       if(mapsize > (SEGMAPSIZE*PTEPERTAB))
+           mapsize = (SEGMAPSIZE*PTEPERTAB);
        s->map = smalloc(mapsize*sizeof(Pte*));
        s->mapsize = mapsize;
    }