patch for vmlinux.lds.S

This fixes building the kernel for me; possibly a recent binutils thing? (Otherwise, I'm getting dangling references to the text.exit code in .IA_64.unwind.text.exit sections.)

Bill

@@ -12,6 +12,8 @@
   /DISCARD/ : {
 	*(.text.exit)
 	*(.data.exit)
+	*(*.text.exit)
+	*(*.data.exit)
 	*(.exitcall.exit)
 	}
 

Which binutils?

What does "objdump -h init/main.o vmlinux" report?

2.11.92.0.12 + patches.

init/main.o: file format elf64-ia64-little

Sections:

Idx Name          Size      VMA               LMA               File off  Algn
  0 .text         00000740  0000000000000000  0000000000000000  00000040  2**5

CONTENTS, ALLOC, LOAD, RELOC, READONLY, CODE
1 .data 00000108 0000000000000000 0000000000000000 00000780 2**3
CONTENTS, ALLOC, LOAD, RELOC, DATA
2 .bss 00000000 0000000000000000 0000000000000000 00000888 2**0
ALLOC
3 .rodata.str1.8 00000708 0000000000000000 0000000000000000 00000888 2**3
CONTENTS, ALLOC, LOAD, READONLY, DATA
4 .sdata 00000020 0000000000000000 0000000000000000 00000f90 2**3
CONTENTS, ALLOC, LOAD, DATA, SMALL_DATA
5 .text.init 000013c0 0000000000000000 0000000000000000 00000fc0 2**5
CONTENTS, ALLOC, LOAD, RELOC, READONLY, CODE
6 .IA_64.unwind_info.text.init 00000130 0000000000000000 0000000000000000 00002380 2**3
CONTENTS, ALLOC, LOAD, READONLY, DATA
7 .IA_64.unwind.text.init 000000f0 0000000000000000 0000000000000000 000024b0 2**3
CONTENTS, ALLOC, LOAD, RELOC, READONLY, DATA
8 .data.init 00000708 0000000000000000 0000000000000000 000025a0 2**3
CONTENTS, ALLOC, LOAD, RELOC, DATA
9 .setup.init 00000060 0000000000000000 0000000000000000 00002ca8 2**3
CONTENTS, ALLOC, LOAD, RELOC, DATA
10 .IA_64.unwind_info 00000078 0000000000000000 0000000000000000 00002d08 2**3
CONTENTS, ALLOC, LOAD, READONLY, DATA
11 .IA_64.unwind 00000060 0000000000000000 0000000000000000 00002d80 2**3
CONTENTS, ALLOC, LOAD, RELOC, READONLY, DATA
12 .comment 00000039 0000000000000000 0000000000000000 00002de0 2**0
CONTENTS, READONLY

vmlinux: file format elf64-ia64-little

Sections:

Idx Name          Size      VMA               LMA               File off  Algn
  0 .text         00411d00  e000000004400000  0000000004400000  00010000  2**15

CONTENTS, ALLOC, LOAD, READONLY, CODE
1 .text.lock 00003480 e000000004811d00 0000000004811d00 00421d00 2**0
CONTENTS, ALLOC, LOAD, READONLY, CODE
2 __ex_table 00001d98 e000000004815180 0000000004815180 00425180 2**2
CONTENTS, ALLOC, LOAD, READONLY, DATA
3 __ksymtab 00005070 e000000004816f18 0000000004816f18 00426f18 2**3
CONTENTS, ALLOC, LOAD, READONLY, DATA
4 .IA_64.unwind_info 00023c18 e00000000481bf88 000000000481bf88 0042bf88 2**3
CONTENTS, ALLOC, LOAD, READONLY, DATA
5 .IA_64.unwind 0001b9f0 e00000000483fba0 000000000483fba0 0044fba0 2**3
CONTENTS, ALLOC, LOAD, READONLY, DATA
6 .rodata 000311f5 e00000000485b590 000000000485b590 0046b590 2**3
CONTENTS, ALLOC, LOAD, READONLY, DATA
7 .kstrtab 0000a5a0 e00000000488c788 000000000488c788 0049c788 2**3
CONTENTS, ALLOC, LOAD, READONLY, DATA
8 .opd 000097c0 e000000004896d30 0000000004896d30 004a6d30 2**4
CONTENTS, ALLOC, LOAD, READONLY, DATA
9 .text.init 00029ce0 e0000000048a4000 00000000048a4000 004b4000 2**5
CONTENTS, ALLOC, LOAD, READONLY, CODE
10 .data.init 000307c8 e0000000048cdce0 00000000048cdce0 004ddce0 2**3
CONTENTS, ALLOC, LOAD, DATA
11 .setup.init 00000270 e0000000048fe4b0 00000000048fe4b0 0050e4b0 2**3
CONTENTS, ALLOC, LOAD, DATA
12 .initcall.init 00000150 e0000000048fe720 00000000048fe720 0050e720 2**3
CONTENTS, ALLOC, LOAD, DATA
13 init_task 00008000 e000000004900000 0000000004900000 00510000 2**4
CONTENTS, ALLOC, LOAD, DATA
14 .data.cacheline_aligned 00032dc0 e000000004908000 0000000004908000 00518000 2**6
CONTENTS, ALLOC, LOAD, DATA
15 .data 000dd3b8 e00000000493adc0 000000000493adc0 0054adc0 2**6
CONTENTS, ALLOC, LOAD, DATA
16 .got 0000d1c8 e000000004a18178 0000000004a18178 00628178 2**3
CONTENTS, ALLOC, LOAD, DATA, SMALL_DATA
17 .sdata 00000b44 e000000004a25340 0000000004a25340 00635340 2**3
CONTENTS, ALLOC, LOAD, DATA, SMALL_DATA
18 .sbss 00000748 e000000004a25e88 0000000004a25e88 00635e88 2**3
ALLOC, SMALL_DATA
19 .bss 0006df30 e000000004a265d0 0000000004a25e88 00635e88 2**4
ALLOC

Bill

Your patch looks OK, I just want to confirm that the binutils change to generate multiple unwind sections has not broken the requirement that unwind entries appear in ascending order. Run this Perl script against your vmlinux, it should say 0 out of order entries detected.

#!/usr/bin/perl -w
# Verify that ia64 unwind data is valid.
# Needs readelf command from recent binutils.

use strict;
use integer;
die($0 . " takes exactly one argument, the vmlinux to be explored\n") if($#ARGV);

my ($data, $section, $start, $size);
my @f;

open(VMLINUX, $ARGV[0]) || die("cannot open $ARGV[0] for reading");

$data = readelf -S $ARGV[0] | fgrep -A 1 ' .IA_64.unwind '; chomp($data);
$data =~ s/\n/ /m;
die("$0 could not find .IA_64.unwind section in $ARGV[0]\n") if($data eq ""); print("Unwind section ", $data, "\n");
$data =~ s/[][]//g;
(@f) = split(' ', $data);
$section = $f[0];
$start = hex($f[4]);
$size = hex($f[5]);

sysseek(VMLINUX, $start, 0) || die "cannot seek to offset $start in $ARGV[0]"; sysread(VMLINUX, $data, $size) || die "cannot read $size bytes from offset $start in $ARGV[0]"; close(VMLINUX);

my @unwind;
my @previous = (0, 0, 0, 0, 0, 0);
my $i;
my $error = 0;

for ($i = 0; $i < $size; $i += 24) {
	# unpack Q is optional, don't rely on it
	@unwind = unpack("LLLLLL", substr($data, $i, 24));
	if ($unwind[0] < $previous[0] ||
	    $unwind[0] = $previous[0] && $unwind[1] < $previous[1]) {
		printf("Out of order\n");
		printf("start 0x%08x%08x end 0x%08x%08x info 0x%08x%08x\n",
			$previous[1], $previous[0], $previous[3], $previous[2], $previous[5], $previous[4]);
		printf("      0x%08x%08x     0x%08x%08x      0x%08x%08x\n",
			$unwind[1], $unwind[0], $unwind[3], $unwind[2], $unwind[5], $unwind[4]);
		++$error;
	}
	@previous = @unwind;
}

printf("%d out of order entries detected\n", $error); exit($error);

Oops, wrong version of the script in previous mail.

Your patch looks OK, I just want to confirm that the binutils change to generate multiple unwind sections has not broken the requirement that unwind entries appear in ascending order. Run this Perl script against your vmlinux, it should say 0 out of order entries detected.

#!/usr/bin/perl -w
# Verify that ia64 unwind data is valid.
# Needs readelf command from recent binutils.

use strict;
use integer;
die($0 . " takes exactly one argument, the vmlinux to be explored\n") if($#ARGV);

my ($data, $start, $size);
my @f;

open(VMLINUX, $ARGV[0]) || die("cannot open $ARGV[0] for reading");

$data = readelf -S $ARGV[0] | fgrep -A 1 ' .IA_64.unwind '; chomp($data);
$data =~ s/\n/ /m;
die("$0 could not find .IA_64.unwind section in $ARGV[0]\n") if($data eq ""); print("Unwind section ", $data, "\n");
$data =~ s/[][]//g;
(@f) = split(' ', $data);
$start = hex($f[4]);
$size = hex($f[5]);

sysseek(VMLINUX, $start, 0) || die "cannot seek to offset $start in $ARGV[0]"; sysread(VMLINUX, $data, $size) || die "cannot read $size bytes from offset $start in $ARGV[0]"; close(VMLINUX);

my @unwind;
my @previous = (0, 0, 0, 0, 0, 0);
my $i;
my $error = 0;

for ($i = 0; $i < $size; $i += 24) {
	# unpack Q is optional, don't rely on it
	@unwind = unpack("LLLLLL", substr($data, $i, 24));
	if ($unwind[1] < $previous[1] ||
	    $unwind[1] = $previous[1] && $unwind[0] < $previous[0]) {
		printf("Out of order\n");
		printf("start 0x%08x%08x end 0x%08x%08x info 0x%08x%08x\n",
			$previous[1], $previous[0], $previous[3], $previous[2], $previous[5], $previous[4]);
		printf("      0x%08x%08x     0x%08x%08x      0x%08x%08x\n",
			$unwind[1], $unwind[0], $unwind[3], $unwind[2], $unwind[5], $unwind[4]);
		++$error;
	}
	@previous = @unwind;
}

printf("%d out of order entries detected\n", $error); exit($error);

It appears to be OK:
Unwind section [ 6] .IA_64.unwind IA_64_UNWIND e00000000483fba0 0044fba0 000000000001b9f0 0000000000000000 AL 1 1 8
0 out of order entries detected

Bill

Bill> This fixes building the kernel for me; possibly a recent Bill> binutils thing? (Otherwise, I'm getting dangling references to Bill> the text.exit code in .IA_64.unwind.text.exit sections.)

Yes, I added a similar fix. It will be in the next patch (hopefully next week, but with LinuxWorld happening, it may end up being the week after).

	--david