PATCH: Fix 2.6 kernel for the new ia64 assembler
The new ia64 assembler uses slot 1 for the offset of a long (2-slot) instruction and the old assembler uses slot 2. The 2.6 kernel assumes slot 2 and won't boot when the new assembler is used:
http://sources.redhat.com/bugzilla/show_bug.cgi?id=1433
This patch will work with either slot 1 or 2.
H.J.
arch/ia64/kernel/patch.c renamed from arch/ia64/kernel/patch.c.slot +12 -4
@@ -64,22 +64,30 @@ ia64_patch (u64 insn_addr, u64 mask, u64void ia64_patch_imm64 (u64 insn_addr,u64 val) { -ia64_patch (insn_addr, +/* The assembler may generate offset pointing to either slot 1 +or slot 2 for a long (2-slot) instruction, occupying slots 1 +and 2. */ + insn_addr &=-16UL ; +ia64_patch (insn_addr +2 ,0x01fffefe000UL , ( ((val &0x8000000000000000UL ) >>27 )/* bit 63 -> 36 */ | ((val &0x0000000000200000UL ) <<0 )/* bit 21 -> 21 */ | ((val &0x00000000001f0000UL ) <<6 )/* bit 16 -> 22 */ | ((val &0x000000000000ff80UL ) <<20 )/* bit 7 -> 27 */ | ((val &0x000000000000007fUL ) <<13 )/* bit 0 -> 13 */ )); -ia64_patch (insn_addr -1 ,0x1ffffffffffUL , val >>22 ); +ia64_patch (insn_addr +1 ,0x1ffffffffffUL , val >>22 ); }void ia64_patch_imm60 (u64 insn_addr,u64 val) { -ia64_patch (insn_addr, +/* The assembler may generate offset pointing to either slot 1 +or slot 2 for a long (2-slot) instruction, occupying slots 1 +and 2. */ + insn_addr &=-16UL ; +ia64_patch (insn_addr +2 ,0x011ffffe000UL , ( ((val &0x0800000000000000UL ) >>23 )/* bit 59 -> 36 */ | ((val &0x00000000000fffffUL ) <<13 )/* bit 0 -> 13 */ )); -ia64_patch (insn_addr -1 ,0x1fffffffffcUL , val >>18 ); +ia64_patch (insn_addr +1 ,0x1fffffffffcUL , val >>18 ); }/*
Looks fine to me.
--david
Quoted 4 lines · H. J. Lu
The new ia64 assembler uses slot 1 for the offset of a long (2-slot) instruction and the old assembler uses slot 2. The 2.6 kernel assumes slot 2 and won't boot when the new assembler is used:
Quoted 9 lines
http://sources.redhat.com/bugzilla/show_bug.cgi?id33
This patch will work with either slot 1 or 2.
H.J.
--
Mosberger Consulting LLC, voice/fax: 510-744-9372,
http://www.mosberger-consulting.com/
35706 Runckel Lane, Fremont, CA 94536
Quoted 3 lines · H. J. Lu
The new ia64 assembler uses slot 1 for the offset of a long (2-slot) instruction and the old assembler uses slot 2. The 2.6 kernel assumes slot 2 and won't boot when the new assembler is used:
Specifically, this is in reference to tags occurring immediately before a long instruction.
Given something like
.text
nop
[L1:]
movl
The value of L1 should be .text+1 but it was incorrectly .text+2 until about 2 weeks ago. Unfortunately, there is code in the kernel that relies on the incorrect value, and needs to be fixed. H.J.'s patch fixes this.
Will the fixed code work on both old and new assemblers? If not, isn't there a potential gcc/kernel version mismatch problem?
Quoted 34 lines
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of James E Wilson
Sent: Friday, October 07, 2005 12:20 PM
Cc: linux ia64 kernel
Subject: Re: PATCH: Fix 2.6 kernel for the new ia64 assemblerQuoted 2 lines · H. J. Lu
The new ia64 assembler uses slot 1 for the offset of a long (2-slot) instruction and the old assembler uses slot 2. The 2.6
kernel assumes
Quoted 1 line
slot 2 and won't boot when the new assembler is used:
Specifically, this is in reference to tags occurring immediately before
a long instruction.Given something like
.text
nop
[L1:]
movl
The value of L1 should be .text+1 but it was incorrectly .text+2 until about 2 weeks ago. Unfortunately, there is code in the kernel that relies on the incorrect value, and needs to be fixed. H.J.'s patch fixes this.-
To unsubscribe from this list: send the line "unsubscribe linux-ia64" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Quoted 1 line · Dan (HP Labs Fort Collins)
Will the fixed code work on both old and new assemblers?
Yes. It is independent of the assembler, at the expense of being slightly slower than necessary. The code does an &=-16 to get a bundle address, then adds 1 or 2 as appropriate to reach each slot. Ideally, the &=-16 should not be needed, but it is there to work around the assembler problem.