arch/tile: finish enabling support for TILE-Gx 64-bit chip

This support was partially present in the existing code (look for
"__tilegx__" ifdefs) but with this change you can build a working
kernel using the TILE-Gx toolchain and ARCH=tilegx.

Most of these files are new, generally adding a foo_64.c file
where previously there was just a foo_32.c file.

The ARCH=tilegx directive redirects to arch/tile, not arch/tilegx,
using the existing SRCARCH mechanism in the top-level Makefile.

Changes to existing files:

- <asm/bitops.h> and <asm/bitops_32.h> changed to factor the
  include of <asm-generic/bitops/non-atomic.h> in the common header.

- <asm/compat.h> and arch/tile/kernel/compat.c changed to remove
  the "const" markers I had put on compat_sys_execve() when trying
  to match some recent similar changes to the non-compat execve.
  It turns out the compat version wasn't "upgraded" to use const.

- <asm/opcode-tile_64.h> and <asm/opcode_constants_64.h> were
  previously included accidentally, with the 32-bit contents.  Now
  they have the proper 64-bit contents.

Finally, I had to hack the existing hacky drivers/input/input-compat.h
to add yet another "#ifdef" for INPUT_COMPAT_TEST (same as x86_64).

Signed-off-by: Chris Metcalf <cmetcalf@tilera.com>
Acked-by: Dmitry Torokhov <dmitry.torokhov@gmail.com> [drivers/input]

12 files changed, 2527 insertions, 1346 deletions

diff --git a/arch/tile/include/arch/chip_tilegx.h b/arch/tile/include/arch/chip_tilegx.h
new file mode 100644
index 00000000000..ea8e4f2c948
--- /dev/null
+++ b/arch/tile/include/arch/chip_tilegx.h
@@ -0,0 +1,258 @@
+/*
+ * Copyright 2011 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.
+ */
+
+/*
+ * @file
+ * Global header file.
+ * This header file specifies defines for TILE-Gx.
+ */
+
+#ifndef __ARCH_CHIP_H__
+#define __ARCH_CHIP_H__
+
+/** Specify chip version.
+ * When possible, prefer the CHIP_xxx symbols below for future-proofing.
+ * This is intended for cross-compiling; native compilation should
+ * use the predefined __tile_chip__ symbol.
+ */
+#define TILE_CHIP 10
+
+/** Specify chip revision.
+ * This provides for the case of a respin of a particular chip type;
+ * the normal value for this symbol is "0".
+ * This is intended for cross-compiling; native compilation should
+ * use the predefined __tile_chip_rev__ symbol.
+ */
+#define TILE_CHIP_REV 0
+
+/** The name of this architecture. */
+#define CHIP_ARCH_NAME "tilegx"
+
+/** The ELF e_machine type for binaries for this chip. */
+#define CHIP_ELF_TYPE() EM_TILEGX
+
+/** The alternate ELF e_machine type for binaries for this chip. */
+#define CHIP_COMPAT_ELF_TYPE() 0x2597
+
+/** What is the native word size of the machine? */
+#define CHIP_WORD_SIZE() 64
+
+/** How many bits of a virtual address are used. Extra bits must be
+ * the sign extension of the low bits.
+ */
+#define CHIP_VA_WIDTH() 42
+
+/** How many bits are in a physical address? */
+#define CHIP_PA_WIDTH() 40
+
+/** Size of the L2 cache, in bytes. */
+#define CHIP_L2_CACHE_SIZE() 262144
+
+/** Log size of an L2 cache line in bytes. */
+#define CHIP_L2_LOG_LINE_SIZE() 6
+
+/** Size of an L2 cache line, in bytes. */
+#define CHIP_L2_LINE_SIZE() (1 << CHIP_L2_LOG_LINE_SIZE())
+
+/** Associativity of the L2 cache. */
+#define CHIP_L2_ASSOC() 8
+
+/** Size of the L1 data cache, in bytes. */
+#define CHIP_L1D_CACHE_SIZE() 32768
+
+/** Log size of an L1 data cache line in bytes. */
+#define CHIP_L1D_LOG_LINE_SIZE() 6
+
+/** Size of an L1 data cache line, in bytes. */
+#define CHIP_L1D_LINE_SIZE() (1 << CHIP_L1D_LOG_LINE_SIZE())
+
+/** Associativity of the L1 data cache. */
+#define CHIP_L1D_ASSOC() 2
+
+/** Size of the L1 instruction cache, in bytes. */
+#define CHIP_L1I_CACHE_SIZE() 32768
+
+/** Log size of an L1 instruction cache line in bytes. */
+#define CHIP_L1I_LOG_LINE_SIZE() 6
+
+/** Size of an L1 instruction cache line, in bytes. */
+#define CHIP_L1I_LINE_SIZE() (1 << CHIP_L1I_LOG_LINE_SIZE())
+
+/** Associativity of the L1 instruction cache. */
+#define CHIP_L1I_ASSOC() 2
+
+/** Stride with which flush instructions must be issued. */
+#define CHIP_FLUSH_STRIDE() CHIP_L2_LINE_SIZE()
+
+/** Stride with which inv instructions must be issued. */
+#define CHIP_INV_STRIDE() CHIP_L2_LINE_SIZE()
+
+/** Stride with which finv instructions must be issued. */
+#define CHIP_FINV_STRIDE() CHIP_L2_LINE_SIZE()
+
+/** Can the local cache coherently cache data that is homed elsewhere? */
+#define CHIP_HAS_COHERENT_LOCAL_CACHE() 1
+
+/** How many simultaneous outstanding victims can the L2 cache have? */
+#define CHIP_MAX_OUTSTANDING_VICTIMS() 128
+
+/** Does the TLB support the NC and NOALLOC bits? */
+#define CHIP_HAS_NC_AND_NOALLOC_BITS() 1
+
+/** Does the chip support hash-for-home caching? */
+#define CHIP_HAS_CBOX_HOME_MAP() 1
+
+/** Number of entries in the chip's home map tables. */
+#define CHIP_CBOX_HOME_MAP_SIZE() 128
+
+/** Do uncacheable requests miss in the cache regardless of whether
+ * there is matching data? */
+#define CHIP_HAS_ENFORCED_UNCACHEABLE_REQUESTS() 1
+
+/** Does the mf instruction wait for victims? */
+#define CHIP_HAS_MF_WAITS_FOR_VICTIMS() 0
+
+/** Does the chip have an "inv" instruction that doesn't also flush? */
+#define CHIP_HAS_INV() 1
+
+/** Does the chip have a "wh64" instruction? */
+#define CHIP_HAS_WH64() 1
+
+/** Does this chip have a 'dword_align' instruction? */
+#define CHIP_HAS_DWORD_ALIGN() 0
+
+/** Number of performance counters. */
+#define CHIP_PERFORMANCE_COUNTERS() 4
+
+/** Does this chip have auxiliary performance counters? */
+#define CHIP_HAS_AUX_PERF_COUNTERS() 1
+
+/** Is the CBOX_MSR1 SPR supported? */
+#define CHIP_HAS_CBOX_MSR1() 0
+
+/** Is the TILE_RTF_HWM SPR supported? */
+#define CHIP_HAS_TILE_RTF_HWM() 1
+
+/** Is the TILE_WRITE_PENDING SPR supported? */
+#define CHIP_HAS_TILE_WRITE_PENDING() 0
+
+/** Is the PROC_STATUS SPR supported? */
+#define CHIP_HAS_PROC_STATUS_SPR() 1
+
+/** Is the DSTREAM_PF SPR supported? */
+#define CHIP_HAS_DSTREAM_PF() 1
+
+/** Log of the number of mshims we have. */
+#define CHIP_LOG_NUM_MSHIMS() 2
+
+/** Are the bases of the interrupt vector areas fixed? */
+#define CHIP_HAS_FIXED_INTVEC_BASE() 0
+
+/** Are the interrupt masks split up into 2 SPRs? */
+#define CHIP_HAS_SPLIT_INTR_MASK() 0
+
+/** Is the cycle count split up into 2 SPRs? */
+#define CHIP_HAS_SPLIT_CYCLE() 0
+
+/** Does the chip have a static network? */
+#define CHIP_HAS_SN() 0
+
+/** Does the chip have a static network processor? */
+#define CHIP_HAS_SN_PROC() 0
+
+/** Size of the L1 static network processor instruction cache, in bytes. */
+/* #define CHIP_L1SNI_CACHE_SIZE() -- does not apply to chip 10 */
+
+/** Does the chip have DMA support in each tile? */
+#define CHIP_HAS_TILE_DMA() 0
+
+/** Does the chip have the second revision of the directly accessible
+ *  dynamic networks?  This encapsulates a number of characteristics,
+ *  including the absence of the catch-all, the absence of inline message
+ *  tags, the absence of support for network context-switching, and so on.
+ */
+#define CHIP_HAS_REV1_XDN() 1
+
+/** Does the chip have cmpexch and similar (fetchadd, exch, etc.)? */
+#define CHIP_HAS_CMPEXCH() 1
+
+/** Does the chip have memory-mapped I/O support? */
+#define CHIP_HAS_MMIO() 1
+
+/** Does the chip have post-completion interrupts? */
+#define CHIP_HAS_POST_COMPLETION_INTERRUPTS() 1
+
+/** Does the chip have native single step support? */
+#define CHIP_HAS_SINGLE_STEP() 1
+
+#ifndef __OPEN_SOURCE__  /* features only relevant to hypervisor-level code */
+
+/** How many entries are present in the instruction TLB? */
+#define CHIP_ITLB_ENTRIES() 16
+
+/** How many entries are present in the data TLB? */
+#define CHIP_DTLB_ENTRIES() 32
+
+/** How many MAF entries does the XAUI shim have? */
+#define CHIP_XAUI_MAF_ENTRIES() 32
+
+/** Does the memory shim have a source-id table? */
+#define CHIP_HAS_MSHIM_SRCID_TABLE() 0
+
+/** Does the L1 instruction cache clear on reset? */
+#define CHIP_HAS_L1I_CLEAR_ON_RESET() 1
+
+/** Does the chip come out of reset with valid coordinates on all tiles?
+ * Note that if defined, this also implies that the upper left is 1,1.
+ */
+#define CHIP_HAS_VALID_TILE_COORD_RESET() 1
+
+/** Does the chip have unified packet formats? */
+#define CHIP_HAS_UNIFIED_PACKET_FORMATS() 1
+
+/** Does the chip support write reordering? */
+#define CHIP_HAS_WRITE_REORDERING() 1
+
+/** Does the chip support Y-X routing as well as X-Y? */
+#define CHIP_HAS_Y_X_ROUTING() 1
+
+/** Is INTCTRL_3 managed with the correct MPL? */
+#define CHIP_HAS_INTCTRL_3_STATUS_FIX() 1
+
+/** Is it possible to configure the chip to be big-endian? */
+#define CHIP_HAS_BIG_ENDIAN_CONFIG() 1
+
+/** Is the CACHE_RED_WAY_OVERRIDDEN SPR supported? */
+#define CHIP_HAS_CACHE_RED_WAY_OVERRIDDEN() 0
+
+/** Is the DIAG_TRACE_WAY SPR supported? */
+#define CHIP_HAS_DIAG_TRACE_WAY() 0
+
+/** Is the MEM_STRIPE_CONFIG SPR supported? */
+#define CHIP_HAS_MEM_STRIPE_CONFIG() 1
+
+/** Are the TLB_PERF SPRs supported? */
+#define CHIP_HAS_TLB_PERF() 1
+
+/** Is the VDN_SNOOP_SHIM_CTL SPR supported? */
+#define CHIP_HAS_VDN_SNOOP_SHIM_CTL() 0
+
+/** Does the chip support rev1 DMA packets? */
+#define CHIP_HAS_REV1_DMA_PACKETS() 1
+
+/** Does the chip have an IPI shim? */
+#define CHIP_HAS_IPI() 1
+
+#endif /* !__OPEN_SOURCE__ */
+#endif /* __ARCH_CHIP_H__ */
diff --git a/arch/tile/include/arch/interrupts_64.h b/arch/tile/include/arch/interrupts_64.h
new file mode 100644
index 00000000000..5bb58b2e4e6
--- /dev/null
+++ b/arch/tile/include/arch/interrupts_64.h
@@ -0,0 +1,276 @@
+/*
+ * Copyright 2011 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.
+ */
+
+#ifndef __ARCH_INTERRUPTS_H__
+#define __ARCH_INTERRUPTS_H__
+
+/** Mask for an interrupt. */
+#ifdef __ASSEMBLER__
+/* Note: must handle breaking interrupts into high and low words manually. */
+#define INT_MASK(intno) (1 << (intno))
+#else
+#define INT_MASK(intno) (1ULL << (intno))
+#endif
+
+
+/** Where a given interrupt executes */
+#define INTERRUPT_VECTOR(i, pl) (0xFC000000 + ((pl) << 24) + ((i) << 8))
+
+/** Where to store a vector for a given interrupt. */
+#define USER_INTERRUPT_VECTOR(i) INTERRUPT_VECTOR(i, 0)
+
+/** The base address of user-level interrupts. */
+#define USER_INTERRUPT_VECTOR_BASE INTERRUPT_VECTOR(0, 0)
+
+
+/** Additional synthetic interrupt. */
+#define INT_BREAKPOINT (63)
+
+#define INT_MEM_ERROR    0
+#define INT_SINGLE_STEP_3    1
+#define INT_SINGLE_STEP_2    2
+#define INT_SINGLE_STEP_1    3
+#define INT_SINGLE_STEP_0    4
+#define INT_IDN_COMPLETE    5
+#define INT_UDN_COMPLETE    6
+#define INT_ITLB_MISS    7
+#define INT_ILL    8
+#define INT_GPV    9
+#define INT_IDN_ACCESS   10
+#define INT_UDN_ACCESS   11
+#define INT_SWINT_3   12
+#define INT_SWINT_2   13
+#define INT_SWINT_1   14
+#define INT_SWINT_0   15
+#define INT_ILL_TRANS   16
+#define INT_UNALIGN_DATA   17
+#define INT_DTLB_MISS   18
+#define INT_DTLB_ACCESS   19
+#define INT_IDN_FIREWALL   20
+#define INT_UDN_FIREWALL   21
+#define INT_TILE_TIMER   22
+#define INT_AUX_TILE_TIMER   23
+#define INT_IDN_TIMER   24
+#define INT_UDN_TIMER   25
+#define INT_IDN_AVAIL   26
+#define INT_UDN_AVAIL   27
+#define INT_IPI_3   28
+#define INT_IPI_2   29
+#define INT_IPI_1   30
+#define INT_IPI_0   31
+#define INT_PERF_COUNT   32
+#define INT_AUX_PERF_COUNT   33
+#define INT_INTCTRL_3   34
+#define INT_INTCTRL_2   35
+#define INT_INTCTRL_1   36
+#define INT_INTCTRL_0   37
+#define INT_BOOT_ACCESS   38
+#define INT_WORLD_ACCESS   39
+#define INT_I_ASID   40
+#define INT_D_ASID   41
+#define INT_DOUBLE_FAULT   42
+
+#define NUM_INTERRUPTS 43
+
+#ifndef __ASSEMBLER__
+#define QUEUED_INTERRUPTS ( \
+    INT_MASK(INT_MEM_ERROR) | \
+    INT_MASK(INT_IDN_COMPLETE) | \
+    INT_MASK(INT_UDN_COMPLETE) | \
+    INT_MASK(INT_IDN_FIREWALL) | \
+    INT_MASK(INT_UDN_FIREWALL) | \
+    INT_MASK(INT_TILE_TIMER) | \
+    INT_MASK(INT_AUX_TILE_TIMER) | \
+    INT_MASK(INT_IDN_TIMER) | \
+    INT_MASK(INT_UDN_TIMER) | \
+    INT_MASK(INT_IDN_AVAIL) | \
+    INT_MASK(INT_UDN_AVAIL) | \
+    INT_MASK(INT_IPI_3) | \
+    INT_MASK(INT_IPI_2) | \
+    INT_MASK(INT_IPI_1) | \
+    INT_MASK(INT_IPI_0) | \
+    INT_MASK(INT_PERF_COUNT) | \
+    INT_MASK(INT_AUX_PERF_COUNT) | \
+    INT_MASK(INT_INTCTRL_3) | \
+    INT_MASK(INT_INTCTRL_2) | \
+    INT_MASK(INT_INTCTRL_1) | \
+    INT_MASK(INT_INTCTRL_0) | \
+    INT_MASK(INT_BOOT_ACCESS) | \
+    INT_MASK(INT_WORLD_ACCESS) | \
+    INT_MASK(INT_I_ASID) | \
+    INT_MASK(INT_D_ASID) | \
+    INT_MASK(INT_DOUBLE_FAULT) | \
+    0)
+#define NONQUEUED_INTERRUPTS ( \
+    INT_MASK(INT_SINGLE_STEP_3) | \
+    INT_MASK(INT_SINGLE_STEP_2) | \
+    INT_MASK(INT_SINGLE_STEP_1) | \
+    INT_MASK(INT_SINGLE_STEP_0) | \
+    INT_MASK(INT_ITLB_MISS) | \
+    INT_MASK(INT_ILL) | \
+    INT_MASK(INT_GPV) | \
+    INT_MASK(INT_IDN_ACCESS) | \
+    INT_MASK(INT_UDN_ACCESS) | \
+    INT_MASK(INT_SWINT_3) | \
+    INT_MASK(INT_SWINT_2) | \
+    INT_MASK(INT_SWINT_1) | \
+    INT_MASK(INT_SWINT_0) | \
+    INT_MASK(INT_ILL_TRANS) | \
+    INT_MASK(INT_UNALIGN_DATA) | \
+    INT_MASK(INT_DTLB_MISS) | \
+    INT_MASK(INT_DTLB_ACCESS) | \
+    0)
+#define CRITICAL_MASKED_INTERRUPTS ( \
+    INT_MASK(INT_MEM_ERROR) | \
+    INT_MASK(INT_SINGLE_STEP_3) | \
+    INT_MASK(INT_SINGLE_STEP_2) | \
+    INT_MASK(INT_SINGLE_STEP_1) | \
+    INT_MASK(INT_SINGLE_STEP_0) | \
+    INT_MASK(INT_IDN_COMPLETE) | \
+    INT_MASK(INT_UDN_COMPLETE) | \
+    INT_MASK(INT_IDN_FIREWALL) | \
+    INT_MASK(INT_UDN_FIREWALL) | \
+    INT_MASK(INT_TILE_TIMER) | \
+    INT_MASK(INT_AUX_TILE_TIMER) | \
+    INT_MASK(INT_IDN_TIMER) | \
+    INT_MASK(INT_UDN_TIMER) | \
+    INT_MASK(INT_IDN_AVAIL) | \
+    INT_MASK(INT_UDN_AVAIL) | \
+    INT_MASK(INT_IPI_3) | \
+    INT_MASK(INT_IPI_2) | \
+    INT_MASK(INT_IPI_1) | \
+    INT_MASK(INT_IPI_0) | \
+    INT_MASK(INT_PERF_COUNT) | \
+    INT_MASK(INT_AUX_PERF_COUNT) | \
+    INT_MASK(INT_INTCTRL_3) | \
+    INT_MASK(INT_INTCTRL_2) | \
+    INT_MASK(INT_INTCTRL_1) | \
+    INT_MASK(INT_INTCTRL_0) | \
+    0)
+#define CRITICAL_UNMASKED_INTERRUPTS ( \
+    INT_MASK(INT_ITLB_MISS) | \
+    INT_MASK(INT_ILL) | \
+    INT_MASK(INT_GPV) | \
+    INT_MASK(INT_IDN_ACCESS) | \
+    INT_MASK(INT_UDN_ACCESS) | \
+    INT_MASK(INT_SWINT_3) | \
+    INT_MASK(INT_SWINT_2) | \
+    INT_MASK(INT_SWINT_1) | \
+    INT_MASK(INT_SWINT_0) | \
+    INT_MASK(INT_ILL_TRANS) | \
+    INT_MASK(INT_UNALIGN_DATA) | \
+    INT_MASK(INT_DTLB_MISS) | \
+    INT_MASK(INT_DTLB_ACCESS) | \
+    INT_MASK(INT_BOOT_ACCESS) | \
+    INT_MASK(INT_WORLD_ACCESS) | \
+    INT_MASK(INT_I_ASID) | \
+    INT_MASK(INT_D_ASID) | \
+    INT_MASK(INT_DOUBLE_FAULT) | \
+    0)
+#define MASKABLE_INTERRUPTS ( \
+    INT_MASK(INT_MEM_ERROR) | \
+    INT_MASK(INT_SINGLE_STEP_3) | \
+    INT_MASK(INT_SINGLE_STEP_2) | \
+    INT_MASK(INT_SINGLE_STEP_1) | \
+    INT_MASK(INT_SINGLE_STEP_0) | \
+    INT_MASK(INT_IDN_COMPLETE) | \
+    INT_MASK(INT_UDN_COMPLETE) | \
+    INT_MASK(INT_IDN_FIREWALL) | \
+    INT_MASK(INT_UDN_FIREWALL) | \
+    INT_MASK(INT_TILE_TIMER) | \
+    INT_MASK(INT_AUX_TILE_TIMER) | \
+    INT_MASK(INT_IDN_TIMER) | \
+    INT_MASK(INT_UDN_TIMER) | \
+    INT_MASK(INT_IDN_AVAIL) | \
+    INT_MASK(INT_UDN_AVAIL) | \
+    INT_MASK(INT_IPI_3) | \
+    INT_MASK(INT_IPI_2) | \
+    INT_MASK(INT_IPI_1) | \
+    INT_MASK(INT_IPI_0) | \
+    INT_MASK(INT_PERF_COUNT) | \
+    INT_MASK(INT_AUX_PERF_COUNT) | \
+    INT_MASK(INT_INTCTRL_3) | \
+    INT_MASK(INT_INTCTRL_2) | \
+    INT_MASK(INT_INTCTRL_1) | \
+    INT_MASK(INT_INTCTRL_0) | \
+    0)
+#define UNMASKABLE_INTERRUPTS ( \
+    INT_MASK(INT_ITLB_MISS) | \
+    INT_MASK(INT_ILL) | \
+    INT_MASK(INT_GPV) | \
+    INT_MASK(INT_IDN_ACCESS) | \
+    INT_MASK(INT_UDN_ACCESS) | \
+    INT_MASK(INT_SWINT_3) | \
+    INT_MASK(INT_SWINT_2) | \
+    INT_MASK(INT_SWINT_1) | \
+    INT_MASK(INT_SWINT_0) | \
+    INT_MASK(INT_ILL_TRANS) | \
+    INT_MASK(INT_UNALIGN_DATA) | \
+    INT_MASK(INT_DTLB_MISS) | \
+    INT_MASK(INT_DTLB_ACCESS) | \
+    INT_MASK(INT_BOOT_ACCESS) | \
+    INT_MASK(INT_WORLD_ACCESS) | \
+    INT_MASK(INT_I_ASID) | \
+    INT_MASK(INT_D_ASID) | \
+    INT_MASK(INT_DOUBLE_FAULT) | \
+    0)
+#define SYNC_INTERRUPTS ( \
+    INT_MASK(INT_SINGLE_STEP_3) | \
+    INT_MASK(INT_SINGLE_STEP_2) | \
+    INT_MASK(INT_SINGLE_STEP_1) | \
+    INT_MASK(INT_SINGLE_STEP_0) | \
+    INT_MASK(INT_IDN_COMPLETE) | \
+    INT_MASK(INT_UDN_COMPLETE) | \
+    INT_MASK(INT_ITLB_MISS) | \
+    INT_MASK(INT_ILL) | \
+    INT_MASK(INT_GPV) | \
+    INT_MASK(INT_IDN_ACCESS) | \
+    INT_MASK(INT_UDN_ACCESS) | \
+    INT_MASK(INT_SWINT_3) | \
+    INT_MASK(INT_SWINT_2) | \
+    INT_MASK(INT_SWINT_1) | \
+    INT_MASK(INT_SWINT_0) | \
+    INT_MASK(INT_ILL_TRANS) | \
+    INT_MASK(INT_UNALIGN_DATA) | \
+    INT_MASK(INT_DTLB_MISS) | \
+    INT_MASK(INT_DTLB_ACCESS) | \
+    0)
+#define NON_SYNC_INTERRUPTS ( \
+    INT_MASK(INT_MEM_ERROR) | \
+    INT_MASK(INT_IDN_FIREWALL) | \
+    INT_MASK(INT_UDN_FIREWALL) | \
+    INT_MASK(INT_TILE_TIMER) | \
+    INT_MASK(INT_AUX_TILE_TIMER) | \
+    INT_MASK(INT_IDN_TIMER) | \
+    INT_MASK(INT_UDN_TIMER) | \
+    INT_MASK(INT_IDN_AVAIL) | \
+    INT_MASK(INT_UDN_AVAIL) | \
+    INT_MASK(INT_IPI_3) | \
+    INT_MASK(INT_IPI_2) | \
+    INT_MASK(INT_IPI_1) | \
+    INT_MASK(INT_IPI_0) | \
+    INT_MASK(INT_PERF_COUNT) | \
+    INT_MASK(INT_AUX_PERF_COUNT) | \
+    INT_MASK(INT_INTCTRL_3) | \
+    INT_MASK(INT_INTCTRL_2) | \
+    INT_MASK(INT_INTCTRL_1) | \
+    INT_MASK(INT_INTCTRL_0) | \
+    INT_MASK(INT_BOOT_ACCESS) | \
+    INT_MASK(INT_WORLD_ACCESS) | \
+    INT_MASK(INT_I_ASID) | \
+    INT_MASK(INT_D_ASID) | \
+    INT_MASK(INT_DOUBLE_FAULT) | \
+    0)
+#endif /* !__ASSEMBLER__ */
+#endif /* !__ARCH_INTERRUPTS_H__ */
diff --git a/arch/tile/include/arch/spr_def_64.h b/arch/tile/include/arch/spr_def_64.h
new file mode 100644
index 00000000000..cd3e5f95d5f
--- /dev/null
+++ b/arch/tile/include/arch/spr_def_64.h
@@ -0,0 +1,173 @@
+/*
+ * Copyright 2011 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.
+ */
+
+#ifndef __DOXYGEN__
+
+#ifndef __ARCH_SPR_DEF_H__
+#define __ARCH_SPR_DEF_H__
+
+#define SPR_AUX_PERF_COUNT_0 0x2105
+#define SPR_AUX_PERF_COUNT_1 0x2106
+#define SPR_AUX_PERF_COUNT_CTL 0x2107
+#define SPR_AUX_PERF_COUNT_STS 0x2108
+#define SPR_CMPEXCH_VALUE 0x2780
+#define SPR_CYCLE 0x2781
+#define SPR_DONE 0x2705
+#define SPR_DSTREAM_PF 0x2706
+#define SPR_EVENT_BEGIN 0x2782
+#define SPR_EVENT_END 0x2783
+#define SPR_EX_CONTEXT_0_0 0x2580
+#define SPR_EX_CONTEXT_0_1 0x2581
+#define SPR_EX_CONTEXT_0_1__PL_SHIFT 0
+#define SPR_EX_CONTEXT_0_1__PL_RMASK 0x3
+#define SPR_EX_CONTEXT_0_1__PL_MASK  0x3
+#define SPR_EX_CONTEXT_0_1__ICS_SHIFT 2
+#define SPR_EX_CONTEXT_0_1__ICS_RMASK 0x1
+#define SPR_EX_CONTEXT_0_1__ICS_MASK  0x4
+#define SPR_EX_CONTEXT_1_0 0x2480
+#define SPR_EX_CONTEXT_1_1 0x2481
+#define SPR_EX_CONTEXT_1_1__PL_SHIFT 0
+#define SPR_EX_CONTEXT_1_1__PL_RMASK 0x3
+#define SPR_EX_CONTEXT_1_1__PL_MASK  0x3
+#define SPR_EX_CONTEXT_1_1__ICS_SHIFT 2
+#define SPR_EX_CONTEXT_1_1__ICS_RMASK 0x1
+#define SPR_EX_CONTEXT_1_1__ICS_MASK  0x4
+#define SPR_EX_CONTEXT_2_0 0x2380
+#define SPR_EX_CONTEXT_2_1 0x2381
+#define SPR_EX_CONTEXT_2_1__PL_SHIFT 0
+#define SPR_EX_CONTEXT_2_1__PL_RMASK 0x3
+#define SPR_EX_CONTEXT_2_1__PL_MASK  0x3
+#define SPR_EX_CONTEXT_2_1__ICS_SHIFT 2
+#define SPR_EX_CONTEXT_2_1__ICS_RMASK 0x1
+#define SPR_EX_CONTEXT_2_1__ICS_MASK  0x4
+#define SPR_FAIL 0x2707
+#define SPR_ILL_TRANS_REASON__I_STREAM_VA_RMASK 0x1
+#define SPR_INTCTRL_0_STATUS 0x2505
+#define SPR_INTCTRL_1_STATUS 0x2405
+#define SPR_INTCTRL_2_STATUS 0x2305
+#define SPR_INTERRUPT_CRITICAL_SECTION 0x2708
+#define SPR_INTERRUPT_MASK_0 0x2506
+#define SPR_INTERRUPT_MASK_1 0x2406
+#define SPR_INTERRUPT_MASK_2 0x2306
+#define SPR_INTERRUPT_MASK_RESET_0 0x2507
+#define SPR_INTERRUPT_MASK_RESET_1 0x2407
+#define SPR_INTERRUPT_MASK_RESET_2 0x2307
+#define SPR_INTERRUPT_MASK_SET_0 0x2508
+#define SPR_INTERRUPT_MASK_SET_1 0x2408
+#define SPR_INTERRUPT_MASK_SET_2 0x2308
+#define SPR_INTERRUPT_VECTOR_BASE_0 0x2509
+#define SPR_INTERRUPT_VECTOR_BASE_1 0x2409
+#define SPR_INTERRUPT_VECTOR_BASE_2 0x2309
+#define SPR_INTERRUPT_VECTOR_BASE_3 0x2209
+#define SPR_IPI_EVENT_0 0x1f05
+#define SPR_IPI_EVENT_1 0x1e05
+#define SPR_IPI_EVENT_2 0x1d05
+#define SPR_IPI_EVENT_RESET_0 0x1f06
+#define SPR_IPI_EVENT_RESET_1 0x1e06
+#define SPR_IPI_EVENT_RESET_2 0x1d06
+#define SPR_IPI_EVENT_SET_0 0x1f07
+#define SPR_IPI_EVENT_SET_1 0x1e07
+#define SPR_IPI_EVENT_SET_2 0x1d07
+#define SPR_IPI_MASK_0 0x1f08
+#define SPR_IPI_MASK_1 0x1e08
+#define SPR_IPI_MASK_2 0x1d08
+#define SPR_IPI_MASK_RESET_0 0x1f09
+#define SPR_IPI_MASK_RESET_1 0x1e09
+#define SPR_IPI_MASK_RESET_2 0x1d09
+#define SPR_IPI_MASK_SET_0 0x1f0a
+#define SPR_IPI_MASK_SET_1 0x1e0a
+#define SPR_IPI_MASK_SET_2 0x1d0a
+#define SPR_MPL_AUX_TILE_TIMER_SET_0 0x1700
+#define SPR_MPL_AUX_TILE_TIMER_SET_1 0x1701
+#define SPR_MPL_AUX_TILE_TIMER_SET_2 0x1702
+#define SPR_MPL_INTCTRL_0_SET_0 0x2500
+#define SPR_MPL_INTCTRL_0_SET_1 0x2501
+#define SPR_MPL_INTCTRL_0_SET_2 0x2502
+#define SPR_MPL_INTCTRL_1_SET_0 0x2400
+#define SPR_MPL_INTCTRL_1_SET_1 0x2401
+#define SPR_MPL_INTCTRL_1_SET_2 0x2402
+#define SPR_MPL_INTCTRL_2_SET_0 0x2300
+#define SPR_MPL_INTCTRL_2_SET_1 0x2301
+#define SPR_MPL_INTCTRL_2_SET_2 0x2302
+#define SPR_MPL_UDN_ACCESS_SET_0 0x0b00
+#define SPR_MPL_UDN_ACCESS_SET_1 0x0b01
+#define SPR_MPL_UDN_ACCESS_SET_2 0x0b02
+#define SPR_MPL_UDN_AVAIL_SET_0 0x1b00
+#define SPR_MPL_UDN_AVAIL_SET_1 0x1b01
+#define SPR_MPL_UDN_AVAIL_SET_2 0x1b02
+#define SPR_MPL_UDN_COMPLETE_SET_0 0x0600
+#define SPR_MPL_UDN_COMPLETE_SET_1 0x0601
+#define SPR_MPL_UDN_COMPLETE_SET_2 0x0602
+#define SPR_MPL_UDN_FIREWALL_SET_0 0x1500
+#define SPR_MPL_UDN_FIREWALL_SET_1 0x1501
+#define SPR_MPL_UDN_FIREWALL_SET_2 0x1502
+#define SPR_MPL_UDN_TIMER_SET_0 0x1900
+#define SPR_MPL_UDN_TIMER_SET_1 0x1901
+#define SPR_MPL_UDN_TIMER_SET_2 0x1902
+#define SPR_MPL_WORLD_ACCESS_SET_0 0x2700
+#define SPR_MPL_WORLD_ACCESS_SET_1 0x2701
+#define SPR_MPL_WORLD_ACCESS_SET_2 0x2702
+#define SPR_PASS 0x2709
+#define SPR_PERF_COUNT_0 0x2005
+#define SPR_PERF_COUNT_1 0x2006
+#define SPR_PERF_COUNT_CTL 0x2007
+#define SPR_PERF_COUNT_DN_CTL 0x2008
+#define SPR_PERF_COUNT_STS 0x2009
+#define SPR_PROC_STATUS 0x2784
+#define SPR_SIM_CONTROL 0x2785
+#define SPR_SINGLE_STEP_CONTROL_0 0x0405
+#define SPR_SINGLE_STEP_CONTROL_0__CANCELED_MASK  0x1
+#define SPR_SINGLE_STEP_CONTROL_0__INHIBIT_MASK  0x2
+#define SPR_SINGLE_STEP_CONTROL_1 0x0305
+#define SPR_SINGLE_STEP_CONTROL_1__CANCELED_MASK  0x1
+#define SPR_SINGLE_STEP_CONTROL_1__INHIBIT_MASK  0x2
+#define SPR_SINGLE_STEP_CONTROL_2 0x0205
+#define SPR_SINGLE_STEP_CONTROL_2__CANCELED_MASK  0x1
+#define SPR_SINGLE_STEP_CONTROL_2__INHIBIT_MASK  0x2
+#define SPR_SINGLE_STEP_EN_0_0 0x250a
+#define SPR_SINGLE_STEP_EN_0_1 0x240a
+#define SPR_SINGLE_STEP_EN_0_2 0x230a
+#define SPR_SINGLE_STEP_EN_1_0 0x250b
+#define SPR_SINGLE_STEP_EN_1_1 0x240b
+#define SPR_SINGLE_STEP_EN_1_2 0x230b
+#define SPR_SINGLE_STEP_EN_2_0 0x250c
+#define SPR_SINGLE_STEP_EN_2_1 0x240c
+#define SPR_SINGLE_STEP_EN_2_2 0x230c
+#define SPR_SYSTEM_SAVE_0_0 0x2582
+#define SPR_SYSTEM_SAVE_0_1 0x2583
+#define SPR_SYSTEM_SAVE_0_2 0x2584
+#define SPR_SYSTEM_SAVE_0_3 0x2585
+#define SPR_SYSTEM_SAVE_1_0 0x2482
+#define SPR_SYSTEM_SAVE_1_1 0x2483
+#define SPR_SYSTEM_SAVE_1_2 0x2484
+#define SPR_SYSTEM_SAVE_1_3 0x2485
+#define SPR_SYSTEM_SAVE_2_0 0x2382
+#define SPR_SYSTEM_SAVE_2_1 0x2383
+#define SPR_SYSTEM_SAVE_2_2 0x2384
+#define SPR_SYSTEM_SAVE_2_3 0x2385
+#define SPR_TILE_COORD 0x270b
+#define SPR_TILE_RTF_HWM 0x270c
+#define SPR_TILE_TIMER_CONTROL 0x1605
+#define SPR_UDN_AVAIL_EN 0x1b05
+#define SPR_UDN_DATA_AVAIL 0x0b80
+#define SPR_UDN_DEADLOCK_TIMEOUT 0x1906
+#define SPR_UDN_DEMUX_COUNT_0 0x0b05
+#define SPR_UDN_DEMUX_COUNT_1 0x0b06
+#define SPR_UDN_DEMUX_COUNT_2 0x0b07
+#define SPR_UDN_DEMUX_COUNT_3 0x0b08
+#define SPR_UDN_DIRECTION_PROTECT 0x1505
+
+#endif /* !defined(__ARCH_SPR_DEF_H__) */
+
+#endif /* !defined(__DOXYGEN__) */
diff --git a/arch/tile/include/asm/atomic_64.h b/arch/tile/include/asm/atomic_64.h
new file mode 100644
index 00000000000..32170529480
--- /dev/null
+++ b/arch/tile/include/asm/atomic_64.h
@@ -0,0 +1,169 @@
+/*
+ * Copyright 2011 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.
+ *
+ * Do not include directly; use <asm/atomic.h>.
+ */
+
+#ifndef _ASM_TILE_ATOMIC_64_H
+#define _ASM_TILE_ATOMIC_64_H
+
+#ifndef __ASSEMBLY__
+
+#include <arch/spr_def.h>
+
+/* First, the 32-bit atomic ops that are "real" on our 64-bit platform. */
+
+#define atomic_set(v, i) ((v)->counter = (i))
+
+/*
+ * The smp_mb() operations throughout are to support the fact that
+ * Linux requires memory barriers before and after the operation,
+ * on any routine which updates memory and returns a value.
+ */
+
+static inline int atomic_cmpxchg(atomic_t *v, int o, int n)
+{
+	int val;
+	__insn_mtspr(SPR_CMPEXCH_VALUE, o);
+	smp_mb();  /* barrier for proper semantics */
+	val = __insn_cmpexch4((void *)&v->counter, n);
+	smp_mb();  /* barrier for proper semantics */
+	return val;
+}
+
+static inline int atomic_xchg(atomic_t *v, int n)
+{
+	int val;
+	smp_mb();  /* barrier for proper semantics */
+	val = __insn_exch4((void *)&v->counter, n);
+	smp_mb();  /* barrier for proper semantics */
+	return val;
+}
+
+static inline void atomic_add(int i, atomic_t *v)
+{
+	__insn_fetchadd4((void *)&v->counter, i);
+}
+
+static inline int atomic_add_return(int i, atomic_t *v)
+{
+	int val;
+	smp_mb();  /* barrier for proper semantics */
+	val = __insn_fetchadd4((void *)&v->counter, i) + i;
+	barrier();  /* the "+ i" above will wait on memory */
+	return val;
+}
+
+static inline int atomic_add_unless(atomic_t *v, int a, int u)
+{
+	int guess, oldval = v->counter;
+	do {
+		if (oldval == u)
+			break;
+		guess = oldval;
+		oldval = atomic_cmpxchg(v, guess, guess + a);
+	} while (guess != oldval);
+	return oldval != u;
+}
+
+/* Now the true 64-bit operations. */
+
+#define ATOMIC64_INIT(i)	{ (i) }
+
+#define atomic64_read(v)		((v)->counter)
+#define atomic64_set(v, i) ((v)->counter = (i))
+
+static inline long atomic64_cmpxchg(atomic64_t *v, long o, long n)
+{
+	long val;
+	smp_mb();  /* barrier for proper semantics */
+	__insn_mtspr(SPR_CMPEXCH_VALUE, o);
+	val = __insn_cmpexch((void *)&v->counter, n);
+	smp_mb();  /* barrier for proper semantics */
+	return val;
+}
+
+static inline long atomic64_xchg(atomic64_t *v, long n)
+{
+	long val;
+	smp_mb();  /* barrier for proper semantics */
+	val = __insn_exch((void *)&v->counter, n);
+	smp_mb();  /* barrier for proper semantics */
+	return val;
+}
+
+static inline void atomic64_add(long i, atomic64_t *v)
+{
+	__insn_fetchadd((void *)&v->counter, i);
+}
+
+static inline long atomic64_add_return(long i, atomic64_t *v)
+{
+	int val;
+	smp_mb();  /* barrier for proper semantics */
+	val = __insn_fetchadd((void *)&v->counter, i) + i;
+	barrier();  /* the "+ i" above will wait on memory */
+	return val;
+}
+
+static inline long atomic64_add_unless(atomic64_t *v, long a, long u)
+{
+	long guess, oldval = v->counter;
+	do {
+		if (oldval == u)
+			break;
+		guess = oldval;
+		oldval = atomic64_cmpxchg(v, guess, guess + a);
+	} while (guess != oldval);
+	return oldval != u;
+}
+
+#define atomic64_sub_return(i, v)	atomic64_add_return(-(i), (v))
+#define atomic64_sub(i, v)		atomic64_add(-(i), (v))
+#define atomic64_inc_return(v)		atomic64_add_return(1, (v))
+#define atomic64_dec_return(v)		atomic64_sub_return(1, (v))
+#define atomic64_inc(v)			atomic64_add(1, (v))
+#define atomic64_dec(v)			atomic64_sub(1, (v))
+
+#define atomic64_inc_and_test(v)	(atomic64_inc_return(v) == 0)
+#define atomic64_dec_and_test(v)	(atomic64_dec_return(v) == 0)
+#define atomic64_sub_and_test(i, v)	(atomic64_sub_return((i), (v)) == 0)
+#define atomic64_add_negative(i, v)	(atomic64_add_return((i), (v)) < 0)
+
+#define atomic64_inc_not_zero(v)	atomic64_add_unless((v), 1, 0)
+
+/* Atomic dec and inc don't implement barrier, so provide them if needed. */
+#define smp_mb__before_atomic_dec()	smp_mb()
+#define smp_mb__after_atomic_dec()	smp_mb()
+#define smp_mb__before_atomic_inc()	smp_mb()
+#define smp_mb__after_atomic_inc()	smp_mb()
+
+#define xchg(ptr, x)							\
+	((typeof(*(ptr)))						\
+	 ((sizeof(*(ptr)) == sizeof(atomic_t)) ?			\
+	  atomic_xchg((atomic_t *)(ptr), (long)(x)) :			\
+	  (sizeof(*(ptr)) == sizeof(atomic_long_t)) ?			\
+	  atomic_long_xchg((atomic_long_t *)(ptr), (long)(x)) :		\
+	  __xchg_called_with_bad_pointer()))
+
+#define cmpxchg(ptr, o, n)						\
+	((typeof(*(ptr)))						\
+	 ((sizeof(*(ptr)) == sizeof(atomic_t)) ?			\
+	  atomic_cmpxchg((atomic_t *)(ptr), (long)(o), (long)(n)) :	\
+	  (sizeof(*(ptr)) == sizeof(atomic_long_t)) ?			\
+	  atomic_long_cmpxchg((atomic_long_t *)(ptr), (long)(o), (long)(n)) : \
+	  __cmpxchg_called_with_bad_pointer()))
+
+#endif /* !__ASSEMBLY__ */
+
+#endif /* _ASM_TILE_ATOMIC_64_H */
diff --git a/arch/tile/include/asm/bitops.h b/arch/tile/include/asm/bitops.h
index 132e6bbd07e..16f1fa51fea 100644
--- a/arch/tile/include/asm/bitops.h
+++ b/arch/tile/include/asm/bitops.h
@@ -122,6 +122,7 @@ static inline unsigned long __arch_hweight64(__u64 w)
 #include <asm-generic/bitops/lock.h>
 #include <asm-generic/bitops/find.h>
 #include <asm-generic/bitops/sched.h>
+#include <asm-generic/bitops/non-atomic.h>
 #include <asm-generic/bitops/le.h>
 
 #endif /* _ASM_TILE_BITOPS_H */
diff --git a/arch/tile/include/asm/bitops_32.h b/arch/tile/include/asm/bitops_32.h
index 2638be51a16..d31ab905cfa 100644
--- a/arch/tile/include/asm/bitops_32.h
+++ b/arch/tile/include/asm/bitops_32.h
@@ -126,7 +126,6 @@ static inline int test_and_change_bit(unsigned nr,
 #define smp_mb__before_clear_bit()	smp_mb()
 #define smp_mb__after_clear_bit()	do {} while (0)
 
-#include <asm-generic/bitops/non-atomic.h>
 #include <asm-generic/bitops/ext2-atomic.h>
 
 #endif /* _ASM_TILE_BITOPS_32_H */
diff --git a/arch/tile/include/asm/bitops_64.h b/arch/tile/include/asm/bitops_64.h
new file mode 100644
index 00000000000..99615e8d2d8
--- /dev/null
+++ b/arch/tile/include/asm/bitops_64.h
@@ -0,0 +1,105 @@
+/*
+ * Copyright 2011 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<