arch: Remove Itanium (IA-64) architecture

The Itanium architecture is obsolete, and an informal survey [0] reveals
that any residual use of Itanium hardware in production is mostly HP-UX
or OpenVMS based. The use of Linux on Itanium appears to be limited to
enthusiasts that occasionally boot a fresh Linux kernel to see whether
things are still working as intended, and perhaps to churn out some
distro packages that are rarely used in practice.

None of the original companies behind Itanium still produce or support
any hardware or software for the architecture, and it is listed as
'Orphaned' in the MAINTAINERS file, as apparently, none of the engineers
that contributed on behalf of those companies (nor anyone else, for that
matter) have been willing to support or maintain the architecture
upstream or even be responsible for applying the odd fix. The Intel
firmware team removed all IA-64 support from the Tianocore/EDK2
reference implementation of EFI in 2018. (Itanium is the original
architecture for which EFI was developed, and the way Linux supports it
deviates significantly from other architectures.) Some distros, such as
Debian and Gentoo, still maintain [unofficial] ia64 ports, but many have
dropped support years ago.

While the argument is being made [1] that there is a 'for the common
good' angle to being able to build and run existing projects such as the
Grid Community Toolkit [2] on Itanium for interoperability testing, the
fact remains that none of those projects are known to be deployed on
Linux/ia64, and very few people actually have access to such a system in
the first place. Even if there were ways imaginable in which Linux/ia64
could be put to good use today, what matters is whether anyone is
actually doing that, and this does not appear to be the case.

There are no emulators widely available, and so boot testing Itanium is
generally infeasible for ordinary contributors. GCC still supports IA-64
but its compile farm [3] no longer has any IA-64 machines. GLIBC would
like to get rid of IA-64 [4] too because it would permit some overdue
code cleanups. In summary, the benefits to the ecosystem of having IA-64
be part of it are mostly theoretical, whereas the maintenance overhead
of keeping it supported is real.

So let's rip off the band aid, and remove the IA-64 arch code entirely.
This follows the timeline proposed by the Debian/ia64 maintainer [5],
which removes support in a controlled manner, leaving IA-64 in a known
good state in the most recent LTS release. Other projects will follow
once the kernel support is removed.

[0] https://lore.kernel.org/all/CAMj1kXFCMh_578jniKpUtx_j8ByHnt=s7S+yQ+vGbKt9ud7+kQ@mail.gmail.com/
[1] https://lore.kernel.org/all/0075883c-7c51-00f5-2c2d-5119c1820410@web.de/
[2] https://gridcf.org/gct-docs/latest/index.html
[3] https://cfarm.tetaneutral.net/machines/list/
[4] https://lore.kernel.org/all/87bkiilpc4.fsf@mid.deneb.enyo.de/
[5] https://lore.kernel.org/all/ff58a3e76e5102c94bb5946d99187b358def688a.camel@physik.fu-berlin.de/

Acked-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>

1 files changed, 0 insertions, 195 deletions

diff --git a/arch/ia64/lib/strlen.S b/arch/ia64/lib/strlen.S
deleted file mode 100644
index 1f4a46c15127..000000000000
--- a/arch/ia64/lib/strlen.S
+++ /dev/null
@@ -1,195 +0,0 @@
-/* SPDX-License-Identifier: GPL-2.0 */
-/*
- *
- * Optimized version of the standard strlen() function
- *
- *
- * Inputs:
- *	in0	address of string
- *
- * Outputs:
- *	ret0	the number of characters in the string (0 if empty string)
- *	does not count the \0
- *
- * Copyright (C) 1999, 2001 Hewlett-Packard Co
- *	Stephane Eranian <eranian@hpl.hp.com>
- *
- * 09/24/99 S.Eranian add speculation recovery code
- */
-
-#include <linux/export.h>
-#include <asm/asmmacro.h>
-
-//
-//
-// This is an enhanced version of the basic strlen. it includes a combination
-// of compute zero index (czx), parallel comparisons, speculative loads and
-// loop unroll using rotating registers.
-//
-// General Ideas about the algorithm:
-//	  The goal is to look at the string in chunks of 8 bytes.
-//	  so we need to do a few extra checks at the beginning because the
-//	  string may not be 8-byte aligned. In this case we load the 8byte
-//	  quantity which includes the start of the string and mask the unused
-//	  bytes with 0xff to avoid confusing czx.
-//	  We use speculative loads and software pipelining to hide memory
-//	  latency and do read ahead safely. This way we defer any exception.
-//
-//	  Because we don't want the kernel to be relying on particular
-//	  settings of the DCR register, we provide recovery code in case
-//	  speculation fails. The recovery code is going to "redo" the work using
-//	  only normal loads. If we still get a fault then we generate a
-//	  kernel panic. Otherwise we return the strlen as usual.
-//
-//	  The fact that speculation may fail can be caused, for instance, by
-//	  the DCR.dm bit being set. In this case TLB misses are deferred, i.e.,
-//	  a NaT bit will be set if the translation is not present. The normal
-//	  load, on the other hand, will cause the translation to be inserted
-//	  if the mapping exists.
-//
-//	  It should be noted that we execute recovery code only when we need
-//	  to use the data that has been speculatively loaded: we don't execute
-//	  recovery code on pure read ahead data.
-//
-// Remarks:
-//	- the cmp r0,r0 is used as a fast way to initialize a predicate
-//	  register to 1. This is required to make sure that we get the parallel
-//	  compare correct.
-//
-//	- we don't use the epilogue counter to exit the loop but we need to set
-//	  it to zero beforehand.
-//
-//	- after the loop we must test for Nat values because neither the
-//	  czx nor cmp instruction raise a NaT consumption fault. We must be
-//	  careful not to look too far for a Nat for which we don't care.
-//	  For instance we don't need to look at a NaT in val2 if the zero byte
-//	  was in val1.
-//
-//	- Clearly performance tuning is required.
-//
-//
-//
-#define saved_pfs	r11
-#define	tmp		r10
-#define base		r16
-#define orig		r17
-#define saved_pr	r18
-#define src		r19
-#define mask		r20
-#define val		r21
-#define val1		r22
-#define val2		r23
-
-GLOBAL_ENTRY(strlen)
-	.prologue
-	.save ar.pfs, saved_pfs
-	alloc saved_pfs=ar.pfs,11,0,0,8 // rotating must be multiple of 8
-
-	.rotr v[2], w[2]	// declares our 4 aliases
-
-	extr.u tmp=in0,0,3	// tmp=least significant 3 bits
-	mov orig=in0		// keep trackof initial byte address
-	dep src=0,in0,0,3	// src=8byte-aligned in0 address
-	.save pr, saved_pr
-	mov saved_pr=pr		// preserve predicates (rotation)
-	;;
-
-	.body
-
-	ld8 v[1]=[src],8	// must not speculate: can fail here
-	shl tmp=tmp,3		// multiply by 8bits/byte
-	mov mask=-1		// our mask
-	;;
-	ld8.s w[1]=[src],8	// speculatively load next
-	cmp.eq p6,p0=r0,r0	// sets p6 to true for cmp.and
-	sub tmp=64,tmp		// how many bits to shift our mask on the right
-	;;
-	shr.u	mask=mask,tmp	// zero enough bits to hold v[1] valuable part
-	mov ar.ec=r0		// clear epilogue counter (saved in ar.pfs)
-	;;
-	add base=-16,src	// keep track of aligned base
-	or v[1]=v[1],mask	// now we have a safe initial byte pattern
-	;;
-1:
-	ld8.s v[0]=[src],8	// speculatively load next
-	czx1.r val1=v[1]	// search 0 byte from right
-	czx1.r val2=w[1]	// search 0 byte from right following 8bytes
-	;;
-	ld8.s w[0]=[src],8	// speculatively load next to next
-	cmp.eq.and p6,p0=8,val1	// p6 = p6 and val1==8
-	cmp.eq.and p6,p0=8,val2	// p6 = p6 and mask==8
-(p6)	br.wtop.dptk 1b		// loop until p6 == 0
-	;;
-	//
-	// We must return try the recovery code iff
-	// val1_is_nat || (val1==8 && val2_is_nat)
-	//
-	// XXX Fixme
-	//	- there must be a better way of doing the test
-	//
-	cmp.eq  p8,p9=8,val1	// p6 = val1 had zero (disambiguate)
-	tnat.nz p6,p7=val1	// test NaT on val1
-(p6)	br.cond.spnt .recover	// jump to recovery if val1 is NaT
-	;;
-	//
-	// if we come here p7 is true, i.e., initialized for // cmp
-	//
-	cmp.eq.and  p7,p0=8,val1// val1==8?
-	tnat.nz.and p7,p0=val2	// test NaT if val2
-(p7)	br.cond.spnt .recover	// jump to recovery if val2 is NaT
-	;;
-(p8)	mov val1=val2		// the other test got us out of the loop
-(p8)	adds src=-16,src	// correct position when 3 ahead
-(p9)	adds src=-24,src	// correct position when 4 ahead
-	;;
-	sub ret0=src,orig	// distance from base
-	sub tmp=8,val1		// which byte in word
-	mov pr=saved_pr,0xffffffffffff0000
-	;;
-	sub ret0=ret0,tmp	// adjust
-	mov ar.pfs=saved_pfs	// because of ar.ec, restore no matter what
-	br.ret.sptk.many rp	// end of normal execution
-
-	//
-	// Outlined recovery code when speculation failed
-	//
-	// This time we don't use speculation and rely on the normal exception
-	// mechanism. that's why the loop is not as good as the previous one
-	// because read ahead is not possible
-	//
-	// IMPORTANT:
-	// Please note that in the case of strlen() as opposed to strlen_user()
-	// we don't use the exception mechanism, as this function is not
-	// supposed to fail. If that happens it means we have a bug and the
-	// code will cause of kernel fault.
-	//
-	// XXX Fixme
-	//	- today we restart from the beginning of the string instead
-	//	  of trying to continue where we left off.
-	//
-.recover:
-	ld8 val=[base],8	// will fail if unrecoverable fault
-	;;
-	or val=val,mask		// remask first bytes
-	cmp.eq p0,p6=r0,r0	// nullify first ld8 in loop
-	;;
-	//
-	// ar.ec is still zero here
-	//
-2:
-(p6)	ld8 val=[base],8	// will fail if unrecoverable fault
-	;;
-	czx1.r val1=val		// search 0 byte from right
-	;;
-	cmp.eq p6,p0=8,val1	// val1==8 ?
-(p6)	br.wtop.dptk 2b		// loop until p6 == 0
-	;;			// (avoid WAW on p63)
-	sub ret0=base,orig	// distance from base
-	sub tmp=8,val1
-	mov pr=saved_pr,0xffffffffffff0000
-	;;
-	sub ret0=ret0,tmp	// length=now - back -1
-	mov ar.pfs=saved_pfs	// because of ar.ec, restore no matter what
-	br.ret.sptk.many rp	// end of successful recovery code
-END(strlen)
-EXPORT_SYMBOL(strlen)