arch: remove tile port

The Tile architecture port was added by Chris Metcalf in 2010, and
maintained until early 2018 when he orphaned it due to his departure
from Mellanox, and nobody else stepped up to maintain it. The product
line is still around in the form of the BlueField SoC, but no longer
uses the Tile architecture.

There are also still products for sale with Tile-GX SoCs, notably the
Mikrotik CCR router family. The products all use old (linux-3.3) kernels
with lots of patches and won't be upgraded by their manufacturers. There
have been efforts to port both OpenWRT and Debian to these, but both
projects have stalled and are very unlikely to be continued in the future.

Given that we are reasonably sure that nobody is still using the port
with an upstream kernel any more, it seems better to remove it now while
the port is in a good shape than to let it bitrot for a few years first.

Cc: Chris Metcalf <chris.d.metcalf@gmail.com>
Cc: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Link: http://www.mellanox.com/page/npu_multicore_overview
Link: https://jenkins.debian.net/view/rebootstrap/job/rebootstrap_tilegx_gcc7/
Signed-off-by: Arnd Bergmann <arnd@arndb.de>

1 files changed, 0 insertions, 205 deletions

diff --git a/arch/tile/lib/atomic_asm_32.S b/arch/tile/lib/atomic_asm_32.S
deleted file mode 100644
index 94709ab..0000000
--- a/arch/tile/lib/atomic_asm_32.S
+++ /dev/null
@@ -1,205 +0,0 @@
-/*
- * Copyright 2010 Tilera Corporation. All Rights Reserved.
- *
- *   This program is free software; you can redistribute it and/or
- *   modify it under the terms of the GNU General Public License
- *   as published by the Free Software Foundation, version 2.
- *
- *   This program is distributed in the hope that it will be useful, but
- *   WITHOUT ANY WARRANTY; without even the implied warranty of
- *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
- *   NON INFRINGEMENT.  See the GNU General Public License for
- *   more details.
- *
- * Support routines for atomic operations.  Each function takes:
- *
- * r0: address to manipulate
- * r1: pointer to atomic lock guarding this operation (for ATOMIC_LOCK_REG)
- * r2: new value to write, or for cmpxchg/add_unless, value to compare against
- * r3: (cmpxchg/xchg_add_unless) new value to write or add;
- *     (atomic64 ops) high word of value to write
- * r4/r5: (cmpxchg64/add_unless64) new value to write or add
- *
- * The 32-bit routines return a "struct __get_user" so that the futex code
- * has an opportunity to return -EFAULT to the user if needed.
- * The 64-bit routines just return a "long long" with the value,
- * since they are only used from kernel space and don't expect to fault.
- * Support for 16-bit ops is included in the framework but we don't provide any.
- *
- * Note that the caller is advised to issue a suitable L1 or L2
- * prefetch on the address being manipulated to avoid extra stalls.
- * In addition, the hot path is on two icache lines, and we start with
- * a jump to the second line to make sure they are both in cache so
- * that we never stall waiting on icache fill while holding the lock.
- * (This doesn't work out with most 64-bit ops, since they consume
- * too many bundles, so may take an extra i-cache stall.)
- *
- * These routines set the INTERRUPT_CRITICAL_SECTION bit, just
- * like sys_cmpxchg(), so that NMIs like PERF_COUNT will not interrupt
- * the code, just page faults.
- *
- * If the load or store faults in a way that can be directly fixed in
- * the do_page_fault_ics() handler (e.g. a vmalloc reference) we fix it
- * directly, return to the instruction that faulted, and retry it.
- *
- * If the load or store faults in a way that potentially requires us
- * to release the atomic lock, then retry (e.g. a migrating PTE), we
- * reset the PC in do_page_fault_ics() to the "tns" instruction so
- * that on return we will reacquire the lock and restart the op.  We
- * are somewhat overloading the exception_table_entry notion by doing
- * this, since those entries are not normally used for migrating PTEs.
- *
- * If the main page fault handler discovers a bad address, it will see
- * the PC pointing to the "tns" instruction (due to the earlier
- * exception_table_entry processing in do_page_fault_ics), and
- * re-reset the PC to the fault handler, atomic_bad_address(), which
- * effectively takes over from the atomic op and can either return a
- * bad "struct __get_user" (for user addresses) or can just panic (for
- * bad kernel addresses).
- *
- * Note that if the value we would store is the same as what we
- * loaded, we bypass the store.  Other platforms with true atomics can
- * make the guarantee that a non-atomic __clear_bit(), for example,
- * can safely race with an atomic test_and_set_bit(); this example is
- * from bit_spinlock.h in slub_lock() / slub_unlock().  We can't do
- * that on Tile since the "atomic" op is really just a
- * read/modify/write, and can race with the non-atomic
- * read/modify/write.  However, if we can short-circuit the write when
- * it is not needed, in the atomic case, we avoid the race.
- */
-
-#include <linux/linkage.h>
-#include <asm/atomic_32.h>
-#include <asm/page.h>
-#include <asm/processor.h>
-
-	.section .text.atomic,"ax"
-ENTRY(__start_atomic_asm_code)
-
-	.macro  atomic_op, name, bitwidth, body
-	.align  64
-STD_ENTRY_SECTION(__atomic\name, .text.atomic)
-	{
-	 movei  r24, 1
-	 j      4f		/* branch to second cache line */
-	}
-1:	{
-	 .ifc \bitwidth,16
-	 lh     r22, r0
-	 .else
-	 lw     r22, r0
-	 addi   r28, r0, 4
-	 .endif
-	}
-	.ifc \bitwidth,64
-	lw      r23, r28
-	.endif
-	\body /* set r24, and r25 if 64-bit */
-	{
-	 seq    r26, r22, r24
-	 seq    r27, r23, r25
-	}
-	.ifc \bitwidth,64
-	bbnst   r27, 2f
-	.endif
-	bbs     r26, 3f		/* skip write-back if it's the same value */
-2:	{
-	 .ifc \bitwidth,16
-	 sh     r0, r24
-	 .else
-	 sw     r0, r24
-	 .endif
-	}
-	.ifc \bitwidth,64
-	sw      r28, r25
-	.endif
-	mf
-3:	{
-	 move   r0, r22
-	 .ifc \bitwidth,64
-	 move   r1, r23
-	 .else
-	 move   r1, zero
-	 .endif
-	 sw     ATOMIC_LOCK_REG_NAME, zero
-	}
-	mtspr   INTERRUPT_CRITICAL_SECTION, zero
-	jrp     lr
-4:	{
-	 move   ATOMIC_LOCK_REG_NAME, r1
-	 mtspr  INTERRUPT_CRITICAL_SECTION, r24
-	}
-#ifndef CONFIG_SMP
-	j       1b		/* no atomic locks */
-#else
-	{
-	 tns    r21, ATOMIC_LOCK_REG_NAME
-	 moveli r23, 2048       /* maximum backoff time in cycles */
-	}
-	{
-	 bzt    r21, 1b		/* branch if lock acquired */
-	 moveli r25, 32         /* starting backoff time in cycles */
-	}
-5:	mtspr   INTERRUPT_CRITICAL_SECTION, zero
-	mfspr   r26, CYCLE_LOW  /* get start point for this backoff */
-6:	mfspr   r22, CYCLE_LOW  /* test to see if we've backed off enough */
-	sub     r22, r22, r26
-	slt     r22, r22, r25
-	bbst    r22, 6b
-	{
-	 mtspr  INTERRUPT_CRITICAL_SECTION, r24
-	 shli   r25, r25, 1     /* double the backoff; retry the tns */
-	}
-	{
-	 tns    r21, ATOMIC_LOCK_REG_NAME
-	 slt    r26, r23, r25   /* is the proposed backoff too big? */
-	}
-	{
-	 bzt    r21, 1b		/* branch if lock acquired */
-	 mvnz   r25, r26, r23
-	}
-	j       5b
-#endif
-	STD_ENDPROC(__atomic\name)
-	.ifc \bitwidth,32
-	.pushsection __ex_table,"a"
-	.align  4
-	.word   1b, __atomic\name
-	.word   2b, __atomic\name
-	.word   __atomic\name, __atomic_bad_address
-	.popsection
-	.endif
-	.endm
-
-
-/*
- * Use __atomic32 prefix to avoid collisions with GCC builtin __atomic functions.
- */
-
-atomic_op 32_cmpxchg, 32, "seq r26, r22, r2; { bbns r26, 3f; move r24, r3 }"
-atomic_op 32_xchg, 32, "move r24, r2"
-atomic_op 32_xchg_add, 32, "add r24, r22, r2"
-atomic_op 32_xchg_add_unless, 32, \
-	"sne r26, r22, r2; { bbns r26, 3f; add r24, r22, r3 }"
-atomic_op 32_fetch_or, 32, "or r24, r22, r2"
-atomic_op 32_fetch_and, 32, "and r24, r22, r2"
-atomic_op 32_fetch_andn, 32, "nor r2, r2, zero; and r24, r22, r2"
-atomic_op 32_fetch_xor, 32, "xor r24, r22, r2"
-
-atomic_op 64_cmpxchg, 64, "{ seq r26, r22, r2; seq r27, r23, r3 }; \
-	{ bbns r26, 3f; move r24, r4 }; { bbns r27, 3f; move r25, r5 }"
-atomic_op 64_xchg, 64, "{ move r24, r2; move r25, r3 }"
-atomic_op 64_xchg_add, 64, "{ add r24, r22, r2; add r25, r23, r3 }; \
-	slt_u r26, r24, r22; add r25, r25, r26"
-atomic_op 64_xchg_add_unless, 64, \
-	"{ sne r26, r22, r2; sne r27, r23, r3 }; \
-	{ bbns r26, 3f; add r24, r22, r4 }; \
-	{ bbns r27, 3f; add r25, r23, r5 }; \
-	slt_u r26, r24, r22; add r25, r25, r26"
-atomic_op 64_fetch_or, 64, "{ or r24, r22, r2; or r25, r23, r3 }"
-atomic_op 64_fetch_and, 64, "{ and r24, r22, r2; and r25, r23, r3 }"
-atomic_op 64_fetch_xor, 64, "{ xor r24, r22, r2; xor r25, r23, r3 }"
-
-	jrp     lr              /* happy backtracer */
-
-ENTRY(__end_atomic_asm_code)