From mboxrd@z Thu Jan  1 00:00:00 1970
From: Alessandro Rubini <rubini-list@gnudd.com>
Date: Thu, 29 Jan 2009 11:37:11 +0100
Subject: [U-Boot] IP_t should be a "packed" struct
In-Reply-To: <20090129102022.A71D3832E416@gemini.denx.de>
References: <20090129102022.A71D3832E416@gemini.denx.de>
	<200901271932.11060.luigi.mantellini.ml@gmail.com>
	<49809C21.6070004@gmail.com> <4980A5AA.2090306@ge.com>
	<4980AAC3.7020601@gmail.com>
	<b73e93990901281240l71c60e96y837802e434e5a681@mail.gmail.com>
	<4980CC59.1070302@gmail.com>
	<20090128213828.13423832E416@gemini.denx.de>
	<b73e93990901281416k766b2bf3qc6429f77545f9191@mail.gmail.com>
	<20090128222747.CEBFD832E416@gemini.denx.de>
	<b73e93990901281504k3940b356nbff298e137c0e707@mail.gmail.com>
Message-ID: <20090129103711.GA32504@mail.gnudd.com>
List-Id: <u-boot.lists.denx.de>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: u-boot@lists.denx.de

> Now study the data structures used  for  netwrking  -  they  are  all
> carefully  crafted that the natural aligment "just fits", i. e. there
> are no gaps between variables if they are aligned naturally.

Actually, a packed structure not only doesn't have gaps, but can live
at any address. So the compiler must access it byte-by-byte
regardless.  This means that a naturally-aligned structure performs
much worse if declared as packed. And single-word loads and stores are
not atomic any more.

I tried with eldk-4.2 for arm and an older gcc for mips, I have
no ppc or other archs handy right now:

   #include <stdint.h>
   #include <stdio.h>

   #ifdef PACKEDTEST
   #  define P __attribute__ ((packed))
   #else
   #  define P
   #endif

   struct {uint32_t i; uint16_t s; uint8_t c, d;} P datum;

   uint32_t i;
   uint16_t s;
   uint8_t c;

   int main(int argc, char **argv)
   {
   	printf("ih\n");
   	i = datum.i;
   	printf("oh\n");
   	s = datum.s;
   	printf("ah\n");
   	c = datum.c;
   	return 0;
   }  

Compilation: ARM non-packed (only interesting lines):

    8378:       e5942000        ldr     r2, [r4]
    8384:       e5832000        str     r2, [r3]

    838c:       e1d420b4        ldrh    r2, [r4, #4]
    8398:       e1c320b0        strh    r2, [r3]

    83a0:       e5d42006        ldrb    r2, [r4, #6]
    83ac:       e5c32000        strb    r2, [r3]

Compilation: ARM packed (only interesting lines):

    8378:       e5d42001        ldrb    r2, [r4, #1]
    837c:       e5d43000        ldrb    r3, [r4]
    8380:       e5d40002        ldrb    r0, [r4, #2]
    8384:       e5d41003        ldrb    r1, [r4, #3]
    8388:       e1833402        orr     r3, r3, r2, lsl #8
    838c:       e1833800        orr     r3, r3, r0, lsl #16
    8394:       e1833c01        orr     r3, r3, r1, lsl #24
    8398:       e5823000        str     r3, [r2]

    83a4:       e5d43005        ldrb    r3, [r4, #5]
    83a8:       e5d42004        ldrb    r2, [r4, #4]
    83b0:       e1822403        orr     r2, r2, r3, lsl #8
    83b8:       e1c320b0        strh    r2, [r3]

    83c0:       e5d42006        ldrb    r2, [r4, #6]
    83cc:       e5c32000        strb    r2, [r3]

Compilation: MIPS non-packed (only interesting lines):

  44:   8e030000        lw      v1,0(s0)
  5c:   ac230000        sw      v1,0(at)

  74:   96030004        lhu     v1,4(s0)
  8c:   a4230000        sh      v1,0(at)

  a4:   92030006        lbu     v1,6(s0)
  bc:   a0230000        sb      v1,0(at)


Compilation: MIPS packed (only interesting lines):

  44:   8a030003        lwl     v1,3(s0)
  48:   9a030000        lwr     v1,0(s0)
  60:   ac230000        sw      v1,0(at)

  78:   92030005        lbu     v1,5(s0)
  7c:   92020004        lbu     v0,4(s0)
  80:   00031a00        sll     v1,v1,0x8
  84:   00621825        or      v1,v1,v0
  9c:   a4230000        sh      v1,0(at)

  b4:   92030006        lbu     v1,6(s0)
  cc:   a0230000        sb      v1,0(at)


/alessandro