From mboxrd@z Thu Jan  1 00:00:00 1970
Received: from ar.aichi-u.ac.jp ([202.250.160.40]) by ttr; Sun May  4 08:40:28 EDT 2014
Received: from [192.168.1.105] ([125.193.25.135]) by ar; Sun May  4 21:40:17 JST 2014
Content-Type: text/plain; charset=us-ascii
Mime-Version: 1.0 (Mac OS X Mail 6.6 \(1510\))
Subject: Re: [9front] cwfs fsmempercent
From: arisawa <arisawa@ar.aichi-u.ac.jp>
In-Reply-To: <edd12afe17b1d29d9dfa30cc39d59228@felloff.net>
Date: Sun, 4 May 2014 21:40:16 +0900
Content-Transfer-Encoding: 7bit
Message-Id: <DB91CCB8-7CF7-4DAE-A291-4F0DD77E6133@ar.aichi-u.ac.jp>
List-ID: <9front.9front.org>
X-Glyph: ➈
X-Bullshit: cloud-oriented grid backend
References: <edd12afe17b1d29d9dfa30cc39d59228@felloff.net>
To: 9front@9front.org
X-Mailer: Apple Mail (2.1510)


Now we can allocate large memory allocation for cwfs on 9pc64.
it seems the change below works for me.

change of malloc.c

static ulong
memsize(void)
{
	...
//		pgmax = (1024*1024*1024)/pgsize;	/* 1GB max for 386 */
		pgmax = (3*1024*1024*1024)/pgsize;	/* 3GB max for x64 */
	...
}

void*
sbrkz(ulong n)
{
	char *p;
	p = sbrk(n);
	memset(p,0,n);
	return p;
}

/*
 * allocate rest of mem
 * for io buffers.
 */
void
iobufinit(void)
{
	...
//	hiob = ialloc(nhiob * sizeof(Hiob), 0);
	hiob = sbrkz(nhiob * sizeof(Hiob));
	...
//	p = ialloc(niob * sizeof(Iobuf), 0);
	p = sbrkz(niob * sizeof(Iobuf));
	...
//	xiop = ialloc(niob * RBUFSIZE, 0);
	xiop = sbrkz(niob * RBUFSIZE);
	...
}


term% ps
...
arisawa        2649    0:00   0:00  3160928K Semacqui 6.cwfs64x
arisawa        2650    0:00   0:00  3160928K Pread    6.cwfs64x
arisawa        2651    0:00   0:00  3160928K Pread    6.cwfs64x
arisawa        2652    0:00   0:00  3160928K Pread    6.cwfs64x
arisawa        2653    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2654    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2655    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2656    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2657    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2658    0:00   0:00  3160928K Semacqui 6.cwfs64x
arisawa        2659    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2660    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2661    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2662    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2663    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2664    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2665    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2666    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2667    0:00   0:00  3160928K Rendez   6.cwfs64x
arisawa        2668    0:00   0:00  3160928K Sleep    6.cwfs64x
arisawa        2669    0:00   0:00  3160928K Sleep    6.cwfs64x
...
term%

don't remove
 #define malloc(n)	ialloc(n, 0)
in all.h

thank you cinap.

Kenji Arisawa


On 2014/04/30, at 5:11, cinap_lenrek@felloff.net wrote:

> that might work. i havnt tried it. you need to be carefull
> with overflows as 2GB becomes negative with signed 32 bit ints.
> 
> 	if(dsize >= 0x80000000UL){	/* for sanity, overflow */
> 		werrstr("invalid allocation size");
> 		return nil;
> 	}
> 
> the only routine that does substantial allocation is iobufinit()
> in cwfs/malloc.c.
> 
> as far as i can see, this memory is never freed, so we might
> just remove the define:
> 
> all.h:13: #define malloc(n)	ialloc(n, 0)
> 
> and provide our own ialloc() that just uses brk().
> 
> --
> cinap