Kernel/linux
1
0
Fork 0
mirror of https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git synced 2025-01-04 12:13:43 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Xtensa updates for v5.16

Browse Source 
- add support for xtensa cores witout windowed registers option
 -----BEGIN PGP SIGNATURE-----
 
 iQJHBAABCgAxFiEEK2eFS5jlMn3N6xfYUfnMkfg/oEQFAmGFgpUTHGpjbXZia2Jj
 QGdtYWlsLmNvbQAKCRBR+cyR+D+gRGA9D/9aADtHeDIynLdlnMuvF38vM1m3N+4H
 bQU6e69JHMdLoQQ67Bg1mpReFOyR/GEwi86p9pZvnMpiKS5oGqzgJY5IoVcaUlcS
 yUB+SJ527WeVLv0NuLzIsoQdHrhsEjenT24dUew6MCS5poFs8BRoJE9xrJzBvnpJ
 Vcb5XzBWvHSB7+9dwwaHhT6ZyNdQUykDN+ST8EkZOEZDfOWZ1lP37FwpuRHdhltS
 Gbzk3uG0REaZUyhpbzABVo9wyE5xRvzjttTaczqXZgyUw6jT42bURgmQeEQDKDfQ
 ta4ro/mup2JSVAF1NSMtn8qwscfNNfUNKBoVr1EsjSnvakHvtBwudLICFY+fnJOo
 8D/HMhHakRPH43DMde4+XIb/c3JQyLsqScKCjNdXpw/BbDf0oDifq6VQPXtZ7bn3
 JFMu0f6dF93dglO+bobEayY07+MdeuTZShYr6KGSMphMYaU1IgtvDSYvDCiPAuey
 0JJRLHwTJNKmvX5tyHWCCQ/U6jAxaEHQtbaQyMIDX0Z045puHT4H/PwN0jpvE/aL
 SRVkzSBH56eMIrxJstqy6CeaAeAS+J5A9desED4vfUFIAJjpuXiErfbS1ul5MwNy
 106UxWrCtbY0wwrcIVaOd0S4ste1fiN1PDRjzd6v19d4j9PL6TXT3aS9XwOHf8hG
 C1jRdoRMyZkFGg==
 =JWIj
 -----END PGP SIGNATURE-----

Merge tag 'xtensa-20211105' of git://github.com/jcmvbkbc/linux-xtensa

Pull xtensa updates from Max Filippov:

 - add support for xtensa cores without windowed registers option

* tag 'xtensa-20211105' of git://github.com/jcmvbkbc/linux-xtensa:
  xtensa: move section symbols to asm/sections.h
  xtensa: remove unused variable wmask
  xtensa: only build windowed register support code when needed
  xtensa: use register window specific opcodes only when present
  xtensa: implement call0 ABI support in assembly
  xtensa: definitions for call0 ABI
  xtensa: don't use a12 in __xtensa_copy_user in call0 ABI
  xtensa: don't use a12 in strncpy_user
  xtensa: use a14 instead of a15 in inline assembly
  xtensa: move _SimulateUserKernelVectorException out of WindowVectors
This commit is contained in:
Linus Torvalds 2021-11-06 14:54:03 -07:00
commit 00f178e150
21 changed files with 541 additions and 265 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
arch/xtensa/boot/boot-elf/bootstrap.S
View File

@ -42,12 +42,14 @@ _bootparam:
.align 4 .align 4
_SetupMMU: _SetupMMU:
#if XCHAL_HAVE_WINDOWED
movi a0, 0 movi a0, 0
wsr a0, windowbase wsr a0, windowbase
rsync rsync
movi a0, 1 movi a0, 1
wsr a0, windowstart wsr a0, windowstart
rsync rsync
#endif
movi a0, 0x1F movi a0, 0x1F
wsr a0, ps wsr a0, ps
rsync rsync

72
arch/xtensa/boot/boot-redboot/bootstrap.S
View File

@ -3,6 +3,7 @@
#include <asm/regs.h> #include <asm/regs.h>
#include <asm/asmmacro.h> #include <asm/asmmacro.h>
#include <asm/cacheasm.h> #include <asm/cacheasm.h>
#include <asm/processor.h>
/* /*
* RB-Data: RedBoot data/bss * RB-Data: RedBoot data/bss
* P: Boot-Parameters * P: Boot-Parameters
@ -36,7 +37,7 @@
.globl __start .globl __start
/* this must be the first byte of the loader! */ /* this must be the first byte of the loader! */
__start: __start:
entry sp, 32 # we do not intend to return abi_entry(32) # we do not intend to return
_call0 _start _call0 _start
__start_a0: __start_a0:
.align 4 .align 4
@ -55,17 +56,19 @@ _start:
movi a4, 1 movi a4, 1
wsr a4, ps wsr a4, ps
rsync rsync
#if XCHAL_HAVE_WINDOWED
rsr a5, windowbase rsr a5, windowbase
ssl a5 ssl a5
sll a4, a4 sll a4, a4
wsr a4, windowstart wsr a4, windowstart
rsync rsync
#endif
movi a4, 0x00040000 movi a4, KERNEL_PS_WOE_MASK
wsr a4, ps wsr a4, ps
rsync rsync
KABI_C0 mov abi_saved0, abi_arg0
/* copy the loader to its address /* copy the loader to its address
* Note: The loader itself is a very small piece, so we assume we * Note: The loader itself is a very small piece, so we assume we
* don't partially overlap. We also assume (even more important) * don't partially overlap. We also assume (even more important)
@ -168,52 +171,52 @@ _reloc:
movi a3, __image_load movi a3, __image_load
sub a4, a3, a4 sub a4, a3, a4
add a8, a0, a4 add abi_arg2, a0, a4
# a1 Stack # a1 Stack
# a8(a4) Load address of the image # a8(a4) Load address of the image
movi a6, _image_start movi abi_arg0, _image_start
movi a10, _image_end movi abi_arg4, _image_end
movi a7, 0x1000000 movi abi_arg1, 0x1000000
sub a11, a10, a6 sub abi_tmp0, abi_arg4, abi_arg0
movi a9, complen movi abi_arg3, complen
s32i a11, a9, 0 s32i abi_tmp0, abi_arg3, 0
movi a0, 0 movi a0, 0
# a6 destination # abi_arg0 destination
# a7 maximum size of destination # abi_arg1 maximum size of destination
# a8 source # abi_arg2 source
# a9 ptr to length # abi_arg3 ptr to length
.extern gunzip .extern gunzip
movi a4, gunzip movi abi_tmp0, gunzip
beqz a4, 1f beqz abi_tmp0, 1f
callx4 a4 abi_callx abi_tmp0
j 2f j 2f
# a6 destination start # abi_arg0 destination start
# a7 maximum size of destination # abi_arg1 maximum size of destination
# a8 source start # abi_arg2 source start
# a9 ptr to length # abi_arg3 ptr to length
# a10 destination end # abi_arg4 destination end
1: 1:
l32i a9, a8, 0 l32i abi_tmp0, abi_arg2, 0
l32i a11, a8, 4 l32i abi_tmp1, abi_arg2, 4
s32i a9, a6, 0 s32i abi_tmp0, abi_arg0, 0
s32i a11, a6, 4 s32i abi_tmp1, abi_arg0, 4
l32i a9, a8, 8 l32i abi_tmp0, abi_arg2, 8
l32i a11, a8, 12 l32i abi_tmp1, abi_arg2, 12
s32i a9, a6, 8 s32i abi_tmp0, abi_arg0, 8
s32i a11, a6, 12 s32i abi_tmp1, abi_arg0, 12
addi a6, a6, 16 addi abi_arg0, abi_arg0, 16
addi a8, a8, 16 addi abi_arg2, abi_arg2, 16
blt a6, a10, 1b blt abi_arg0, abi_arg4, 1b
/* jump to the kernel */ /* jump to the kernel */
@ -230,6 +233,7 @@ _reloc:
# a2 Boot parameter list # a2 Boot parameter list
KABI_C0 mov abi_arg0, abi_saved0
movi a0, _image_start movi a0, _image_start
jx a0 jx a0

65
arch/xtensa/include/asm/asmmacro.h
View File

@ -194,6 +194,12 @@
#define XTENSA_STACK_ALIGNMENT 16 #define XTENSA_STACK_ALIGNMENT 16
#if defined(__XTENSA_WINDOWED_ABI__) #if defined(__XTENSA_WINDOWED_ABI__)
/* Assembly instructions for windowed kernel ABI. */
#define KABI_W
/* Assembly instructions for call0 kernel ABI (will be ignored). */
#define KABI_C0 #
#define XTENSA_FRAME_SIZE_RESERVE 16 #define XTENSA_FRAME_SIZE_RESERVE 16
#define XTENSA_SPILL_STACK_RESERVE 32 #define XTENSA_SPILL_STACK_RESERVE 32
@ -206,8 +212,34 @@
#define abi_ret(frame_size) retw #define abi_ret(frame_size) retw
#define abi_ret_default retw #define abi_ret_default retw
/* direct call */
#define abi_call call4
/* indirect call */
#define abi_callx callx4
/* outgoing call argument registers */
#define abi_arg0 a6
#define abi_arg1 a7
#define abi_arg2 a8
#define abi_arg3 a9
#define abi_arg4 a10
#define abi_arg5 a11
/* return value */
#define abi_rv a6
/* registers preserved across call */
#define abi_saved0 a2
#define abi_saved1 a3
/* none of the above */
#define abi_tmp0 a4
#define abi_tmp1 a5
#elif defined(__XTENSA_CALL0_ABI__) #elif defined(__XTENSA_CALL0_ABI__)
/* Assembly instructions for windowed kernel ABI (will be ignored). */
#define KABI_W #
/* Assembly instructions for call0 kernel ABI. */
#define KABI_C0
#define XTENSA_SPILL_STACK_RESERVE 0 #define XTENSA_SPILL_STACK_RESERVE 0
#define abi_entry(frame_size) __abi_entry (frame_size) #define abi_entry(frame_size) __abi_entry (frame_size)
@ -233,10 +265,43 @@
#define abi_ret_default ret #define abi_ret_default ret
/* direct call */
#define abi_call call0
/* indirect call */
#define abi_callx callx0
/* outgoing call argument registers */
#define abi_arg0 a2
#define abi_arg1 a3
#define abi_arg2 a4
#define abi_arg3 a5
#define abi_arg4 a6
#define abi_arg5 a7
/* return value */
#define abi_rv a2
/* registers preserved across call */
#define abi_saved0 a12
#define abi_saved1 a13
/* none of the above */
#define abi_tmp0 a8
#define abi_tmp1 a9
#else #else
#error Unsupported Xtensa ABI #error Unsupported Xtensa ABI
#endif #endif
#if defined(USER_SUPPORT_WINDOWED)
/* Assembly instructions for windowed user ABI. */
#define UABI_W
/* Assembly instructions for call0 user ABI (will be ignored). */
#define UABI_C0 #
#else
/* Assembly instructions for windowed user ABI (will be ignored). */
#define UABI_W #
/* Assembly instructions for call0 user ABI. */
#define UABI_C0
#endif
#define __XTENSA_HANDLER .section ".exception.text", "ax" #define __XTENSA_HANDLER .section ".exception.text", "ax"
#endif /* _XTENSA_ASMMACRO_H */ #endif /* _XTENSA_ASMMACRO_H */

26
arch/xtensa/include/asm/atomic.h
View File

@ -25,15 +25,15 @@
* *
* Locking interrupts looks like this: * Locking interrupts looks like this:
* *
* rsil a15, TOPLEVEL * rsil a14, TOPLEVEL
* <code> * <code>
* wsr a15, PS * wsr a14, PS
* rsync * rsync
* *
* Note that a15 is used here because the register allocation * Note that a14 is used here because the register allocation
* done by the compiler is not guaranteed and a window overflow * done by the compiler is not guaranteed and a window overflow
* may not occur between the rsil and wsr instructions. By using * may not occur between the rsil and wsr instructions. By using
* a15 in the rsil, the machine is guaranteed to be in a state * a14 in the rsil, the machine is guaranteed to be in a state
* where no register reference will cause an overflow. * where no register reference will cause an overflow.
*/ */
@ -185,15 +185,15 @@ static inline void arch_atomic_##op(int i, atomic_t * v) \
unsigned int vval; \ unsigned int vval; \
\ \
__asm__ __volatile__( \ __asm__ __volatile__( \
" rsil a15, "__stringify(TOPLEVEL)"\n" \ " rsil a14, "__stringify(TOPLEVEL)"\n" \
" l32i %[result], %[mem]\n" \ " l32i %[result], %[mem]\n" \
" " #op " %[result], %[result], %[i]\n" \ " " #op " %[result], %[result], %[i]\n" \
" s32i %[result], %[mem]\n" \ " s32i %[result], %[mem]\n" \
" wsr a15, ps\n" \ " wsr a14, ps\n" \
" rsync\n" \ " rsync\n" \
: [result] "=&a" (vval), [mem] "+m" (*v) \ : [result] "=&a" (vval), [mem] "+m" (*v) \
: [i] "a" (i) \ : [i] "a" (i) \
: "a15", "memory" \ : "a14", "memory" \
); \ ); \
} \ } \
@ -203,15 +203,15 @@ static inline int arch_atomic_##op##_return(int i, atomic_t * v) \
unsigned int vval; \ unsigned int vval; \
\ \
__asm__ __volatile__( \ __asm__ __volatile__( \
" rsil a15,"__stringify(TOPLEVEL)"\n" \ " rsil a14,"__stringify(TOPLEVEL)"\n" \
" l32i %[result], %[mem]\n" \ " l32i %[result], %[mem]\n" \
" " #op " %[result], %[result], %[i]\n" \ " " #op " %[result], %[result], %[i]\n" \
" s32i %[result], %[mem]\n" \ " s32i %[result], %[mem]\n" \
" wsr a15, ps\n" \ " wsr a14, ps\n" \
" rsync\n" \ " rsync\n" \
: [result] "=&a" (vval), [mem] "+m" (*v) \ : [result] "=&a" (vval), [mem] "+m" (*v) \
: [i] "a" (i) \ : [i] "a" (i) \
: "a15", "memory" \ : "a14", "memory" \
); \ ); \
\ \
return vval; \ return vval; \
@ -223,16 +223,16 @@ static inline int arch_atomic_fetch_##op(int i, atomic_t * v) \
unsigned int tmp, vval; \ unsigned int tmp, vval; \
\ \
__asm__ __volatile__( \ __asm__ __volatile__( \
" rsil a15,"__stringify(TOPLEVEL)"\n" \ " rsil a14,"__stringify(TOPLEVEL)"\n" \
" l32i %[result], %[mem]\n" \ " l32i %[result], %[mem]\n" \
" " #op " %[tmp], %[result], %[i]\n" \ " " #op " %[tmp], %[result], %[i]\n" \
" s32i %[tmp], %[mem]\n" \ " s32i %[tmp], %[mem]\n" \
" wsr a15, ps\n" \ " wsr a14, ps\n" \
" rsync\n" \ " rsync\n" \
: [result] "=&a" (vval), [tmp] "=&a" (tmp), \ : [result] "=&a" (vval), [tmp] "=&a" (tmp), \
[mem] "+m" (*v) \ [mem] "+m" (*v) \
: [i] "a" (i) \ : [i] "a" (i) \
: "a15", "memory" \ : "a14", "memory" \
); \ ); \
\ \
return vval; \ return vval; \

16
arch/xtensa/include/asm/cmpxchg.h
View File

@ -52,16 +52,16 @@ __cmpxchg_u32(volatile int *p, int old, int new)
return new; return new;
#else #else
__asm__ __volatile__( __asm__ __volatile__(
" rsil a15, "__stringify(TOPLEVEL)"\n" " rsil a14, "__stringify(TOPLEVEL)"\n"
" l32i %[old], %[mem]\n" " l32i %[old], %[mem]\n"
" bne %[old], %[cmp], 1f\n" " bne %[old], %[cmp], 1f\n"
" s32i %[new], %[mem]\n" " s32i %[new], %[mem]\n"
"1:\n" "1:\n"
" wsr a15, ps\n" " wsr a14, ps\n"
" rsync\n" " rsync\n"
: [old] "=&a" (old), [mem] "+m" (*p) : [old] "=&a" (old), [mem] "+m" (*p)
: [cmp] "a" (old), [new] "r" (new) : [cmp] "a" (old), [new] "r" (new)
: "a15", "memory"); : "a14", "memory");
return old; return old;
#endif #endif
} }
@ -116,10 +116,10 @@ static inline unsigned long __cmpxchg_local(volatile void *ptr,
/* /*
* xchg_u32 * xchg_u32
* *
* Note that a15 is used here because the register allocation * Note that a14 is used here because the register allocation
* done by the compiler is not guaranteed and a window overflow * done by the compiler is not guaranteed and a window overflow
* may not occur between the rsil and wsr instructions. By using * may not occur between the rsil and wsr instructions. By using
* a15 in the rsil, the machine is guaranteed to be in a state * a14 in the rsil, the machine is guaranteed to be in a state
* where no register reference will cause an overflow. * where no register reference will cause an overflow.
*/ */
@ -157,14 +157,14 @@ static inline unsigned long xchg_u32(volatile int * m, unsigned long val)
#else #else
unsigned long tmp; unsigned long tmp;
__asm__ __volatile__( __asm__ __volatile__(
" rsil a15, "__stringify(TOPLEVEL)"\n" " rsil a14, "__stringify(TOPLEVEL)"\n"
" l32i %[tmp], %[mem]\n" " l32i %[tmp], %[mem]\n"
" s32i %[val], %[mem]\n" " s32i %[val], %[mem]\n"
" wsr a15, ps\n" " wsr a14, ps\n"
" rsync\n" " rsync\n"
: [tmp] "=&a" (tmp), [mem] "+m" (*m) : [tmp] "=&a" (tmp), [mem] "+m" (*m)
: [val] "a" (val) : [val] "a" (val)
: "a15", "memory"); : "a14", "memory");
return tmp; return tmp;
#endif #endif
} }

11
arch/xtensa/include/asm/core.h
View File

@ -26,4 +26,15 @@
#define XCHAL_SPANNING_WAY 0 #define XCHAL_SPANNING_WAY 0
#endif #endif
#if XCHAL_HAVE_WINDOWED
#if defined(CONFIG_USER_ABI_DEFAULT) || defined(CONFIG_USER_ABI_CALL0_PROBE)
/* Whether windowed ABI is supported in userspace. */
#define USER_SUPPORT_WINDOWED
#endif
#if defined(__XTENSA_WINDOWED_ABI__) || defined(USER_SUPPORT_WINDOWED)
/* Whether windowed ABI is supported either in userspace or in the kernel. */
#define SUPPORT_WINDOWED
#endif
#endif
#endif #endif

32
arch/xtensa/include/asm/processor.h
View File

@ -18,12 +18,6 @@
#include <asm/types.h> #include <asm/types.h>
#include <asm/regs.h> #include <asm/regs.h>
/* Assertions. */
#if (XCHAL_HAVE_WINDOWED != 1)
# error Linux requires the Xtensa Windowed Registers Option.
#endif
/* Xtensa ABI requires stack alignment to be at least 16 */ /* Xtensa ABI requires stack alignment to be at least 16 */
#define STACK_ALIGN (XCHAL_DATA_WIDTH > 16 ? XCHAL_DATA_WIDTH : 16) #define STACK_ALIGN (XCHAL_DATA_WIDTH > 16 ? XCHAL_DATA_WIDTH : 16)
@ -105,8 +99,18 @@
#define WSBITS (XCHAL_NUM_AREGS / 4) /* width of WINDOWSTART in bits */ #define WSBITS (XCHAL_NUM_AREGS / 4) /* width of WINDOWSTART in bits */
#define WBBITS (XCHAL_NUM_AREGS_LOG2 - 2) /* width of WINDOWBASE in bits */ #define WBBITS (XCHAL_NUM_AREGS_LOG2 - 2) /* width of WINDOWBASE in bits */
#if defined(__XTENSA_WINDOWED_ABI__)
#define KERNEL_PS_WOE_MASK PS_WOE_MASK
#elif defined(__XTENSA_CALL0_ABI__)
#define KERNEL_PS_WOE_MASK 0
#else
#error Unsupported xtensa ABI
#endif
#ifndef __ASSEMBLY__ #ifndef __ASSEMBLY__
#if defined(__XTENSA_WINDOWED_ABI__)
/* Build a valid return address for the specified call winsize. /* Build a valid return address for the specified call winsize.
* winsize must be 1 (call4), 2 (call8), or 3 (call12) * winsize must be 1 (call4), 2 (call8), or 3 (call12)
*/ */
@ -117,6 +121,22 @@
*/ */
#define MAKE_PC_FROM_RA(ra,sp) (((ra) & 0x3fffffff) | ((sp) & 0xc0000000)) #define MAKE_PC_FROM_RA(ra,sp) (((ra) & 0x3fffffff) | ((sp) & 0xc0000000))
#elif defined(__XTENSA_CALL0_ABI__)
/* Build a valid return address for the specified call winsize.
* winsize must be 1 (call4), 2 (call8), or 3 (call12)
*/
#define MAKE_RA_FOR_CALL(ra, ws) (ra)
/* Convert return address to a valid pc
* Note: We assume that the stack pointer is in the same 1GB ranges as the ra
*/
#define MAKE_PC_FROM_RA(ra, sp) (ra)
#else
#error Unsupported Xtensa ABI
#endif
/* Spill slot location for the register reg in the spill area under the stack /* Spill slot location for the register reg in the spill area under the stack
* pointer sp. reg must be in the range [0..4). * pointer sp. reg must be in the range [0..4).
*/ */

41
arch/xtensa/include/asm/sections.h Normal file
View File

@ -0,0 +1,41 @@
/* SPDX-License-Identifier: GPL-2.0 */
#ifndef _XTENSA_SECTIONS_H
#define _XTENSA_SECTIONS_H
#include <asm-generic/sections.h>
#ifdef CONFIG_VECTORS_ADDR
extern char _WindowVectors_text_start[];
extern char _WindowVectors_text_end[];
extern char _DebugInterruptVector_text_start[];
extern char _DebugInterruptVector_text_end[];
extern char _KernelExceptionVector_text_start[];
extern char _KernelExceptionVector_text_end[];
extern char _UserExceptionVector_text_start[];
extern char _UserExceptionVector_text_end[];
extern char _DoubleExceptionVector_text_start[];
extern char _DoubleExceptionVector_text_end[];
extern char _exception_text_start[];
extern char _exception_text_end[];
extern char _Level2InterruptVector_text_start[];
extern char _Level2InterruptVector_text_end[];
extern char _Level3InterruptVector_text_start[];
extern char _Level3InterruptVector_text_end[];
extern char _Level4InterruptVector_text_start[];
extern char _Level4InterruptVector_text_end[];
extern char _Level5InterruptVector_text_start[];
extern char _Level5InterruptVector_text_end[];
extern char _Level6InterruptVector_text_start[];
extern char _Level6InterruptVector_text_end[];
#endif
#ifdef CONFIG_SMP
extern char _SecondaryResetVector_text_start[];
extern char _SecondaryResetVector_text_end[];
#endif
#ifdef CONFIG_XIP_KERNEL
extern char _xip_start[];
extern char _xip_end[];
#endif
#endif

2
arch/xtensa/include/asm/traps.h
View File

@ -56,6 +56,7 @@ void secondary_trap_init(void);
static inline void spill_registers(void) static inline void spill_registers(void)
{ {
#if defined(__XTENSA_WINDOWED_ABI__)
#if XCHAL_NUM_AREGS > 16 #if XCHAL_NUM_AREGS > 16
__asm__ __volatile__ ( __asm__ __volatile__ (
" call8 1f\n" " call8 1f\n"
@ -96,6 +97,7 @@ static inline void spill_registers(void)
" mov a12, a12\n" " mov a12, a12\n"
: : : "memory"); : : : "memory");
#endif #endif
#endif
} }
struct debug_table { struct debug_table {

2
arch/xtensa/kernel/align.S
View File

@ -58,7 +58,9 @@
* BE shift left / mask 0 0 X X * BE shift left / mask 0 0 X X
*/ */
#if XCHAL_HAVE_WINDOWED
#define UNALIGNED_USER_EXCEPTION #define UNALIGNED_USER_EXCEPTION
#endif
#if XCHAL_HAVE_BE #if XCHAL_HAVE_BE

216
arch/xtensa/kernel/entry.S
View File

@ -158,6 +158,7 @@ _user_exception:
/* Rotate ws so that the current windowbase is at bit0. */ /* Rotate ws so that the current windowbase is at bit0. */
/* Assume ws = xxwww1yyyy. Rotate ws right, so that a2 = yyyyxxwww1 */ /* Assume ws = xxwww1yyyy. Rotate ws right, so that a2 = yyyyxxwww1 */
#if defined(USER_SUPPORT_WINDOWED)
rsr a2, windowbase rsr a2, windowbase
rsr a3, windowstart rsr a3, windowstart
ssr a2 ssr a2
@ -167,24 +168,33 @@ _user_exception:
src a2, a3, a2 src a2, a3, a2
srli a2, a2, 32-WSBITS srli a2, a2, 32-WSBITS
s32i a2, a1, PT_WMASK # needed for restoring registers s32i a2, a1, PT_WMASK # needed for restoring registers
#else
movi a2, 0
movi a3, 1
s32i a2, a1, PT_WINDOWBASE
s32i a3, a1, PT_WINDOWSTART
s32i a3, a1, PT_WMASK
#endif
/* Save only live registers. */ /* Save only live registers. */
_bbsi.l a2, 1, 1f UABI_W _bbsi.l a2, 1, 1f
s32i a4, a1, PT_AREG4 s32i a4, a1, PT_AREG4
s32i a5, a1, PT_AREG5 s32i a5, a1, PT_AREG5
s32i a6, a1, PT_AREG6 s32i a6, a1, PT_AREG6
s32i a7, a1, PT_AREG7 s32i a7, a1, PT_AREG7
_bbsi.l a2, 2, 1f UABI_W _bbsi.l a2, 2, 1f
s32i a8, a1, PT_AREG8 s32i a8, a1, PT_AREG8
s32i a9, a1, PT_AREG9 s32i a9, a1, PT_AREG9
s32i a10, a1, PT_AREG10 s32i a10, a1, PT_AREG10
s32i a11, a1, PT_AREG11 s32i a11, a1, PT_AREG11
_bbsi.l a2, 3, 1f UABI_W _bbsi.l a2, 3, 1f
s32i a12, a1, PT_AREG12 s32i a12, a1, PT_AREG12
s32i a13, a1, PT_AREG13 s32i a13, a1, PT_AREG13
s32i a14, a1, PT_AREG14 s32i a14, a1, PT_AREG14
s32i a15, a1, PT_AREG15 s32i a15, a1, PT_AREG15
#if defined(USER_SUPPORT_WINDOWED)
_bnei a2, 1, 1f # only one valid frame? _bnei a2, 1, 1f # only one valid frame?
/* Only one valid frame, skip saving regs. */ /* Only one valid frame, skip saving regs. */
@ -239,7 +249,7 @@ _user_exception:
rsync rsync
/* We are back to the original stack pointer (a1) */ /* We are back to the original stack pointer (a1) */
#endif
2: /* Now, jump to the common exception handler. */ 2: /* Now, jump to the common exception handler. */
j common_exception j common_exception
@ -295,6 +305,7 @@ _kernel_exception:
s32i a3, a1, PT_SAR s32i a3, a1, PT_SAR
s32i a2, a1, PT_ICOUNTLEVEL s32i a2, a1, PT_ICOUNTLEVEL
#if defined(__XTENSA_WINDOWED_ABI__)
/* Rotate ws so that the current windowbase is at bit0. */ /* Rotate ws so that the current windowbase is at bit0. */
/* Assume ws = xxwww1yyyy. Rotate ws right, so that a2 = yyyyxxwww1 */ /* Assume ws = xxwww1yyyy. Rotate ws right, so that a2 = yyyyxxwww1 */
@ -305,27 +316,28 @@ _kernel_exception:
src a2, a3, a2 src a2, a3, a2
srli a2, a2, 32-WSBITS srli a2, a2, 32-WSBITS
s32i a2, a1, PT_WMASK # needed for kernel_exception_exit s32i a2, a1, PT_WMASK # needed for kernel_exception_exit
#endif
/* Save only the live window-frame */ /* Save only the live window-frame */
_bbsi.l a2, 1, 1f KABI_W _bbsi.l a2, 1, 1f
s32i a4, a1, PT_AREG4 s32i a4, a1, PT_AREG4
s32i a5, a1, PT_AREG5 s32i a5, a1, PT_AREG5
s32i a6, a1, PT_AREG6 s32i a6, a1, PT_AREG6
s32i a7, a1, PT_AREG7 s32i a7, a1, PT_AREG7
_bbsi.l a2, 2, 1f KABI_W _bbsi.l a2, 2, 1f
s32i a8, a1, PT_AREG8 s32i a8, a1, PT_AREG8
s32i a9, a1, PT_AREG9 s32i a9, a1, PT_AREG9
s32i a10, a1, PT_AREG10 s32i a10, a1, PT_AREG10
s32i a11, a1, PT_AREG11 s32i a11, a1, PT_AREG11
_bbsi.l a2, 3, 1f KABI_W _bbsi.l a2, 3, 1f
s32i a12, a1, PT_AREG12 s32i a12, a1, PT_AREG12
s32i a13, a1, PT_AREG13 s32i a13, a1, PT_AREG13
s32i a14, a1, PT_AREG14 s32i a14, a1, PT_AREG14
s32i a15, a1, PT_AREG15 s32i a15, a1, PT_AREG15
#ifdef __XTENSA_WINDOWED_ABI__
_bnei a2, 1, 1f _bnei a2, 1, 1f
/* Copy spill slots of a0 and a1 to imitate movsp /* Copy spill slots of a0 and a1 to imitate movsp
* in order to keep exception stack continuous * in order to keep exception stack continuous
*/ */
@ -333,6 +345,7 @@ _kernel_exception:
l32i a0, a1, PT_SIZE + 4 l32i a0, a1, PT_SIZE + 4
s32e a3, a1, -16 s32e a3, a1, -16
s32e a0, a1, -12 s32e a0, a1, -12
#endif
1: 1:
l32i a0, a1, PT_AREG0 # restore saved a0 l32i a0, a1, PT_AREG0 # restore saved a0
wsr a0, depc wsr a0, depc
@ -419,16 +432,16 @@ common_exception:
movi a3, LOCKLEVEL movi a3, LOCKLEVEL
.Lexception: .Lexception:
movi a0, PS_WOE_MASK KABI_W movi a0, PS_WOE_MASK
or a3, a3, a0 KABI_W or a3, a3, a0
#else #else
addi a2, a2, -EXCCAUSE_LEVEL1_INTERRUPT addi a2, a2, -EXCCAUSE_LEVEL1_INTERRUPT
movi a0, LOCKLEVEL movi a0, LOCKLEVEL
extui a3, a3, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH extui a3, a3, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH
# a3 = PS.INTLEVEL # a3 = PS.INTLEVEL
moveqz a3, a0, a2 # a3 = LOCKLEVEL iff interrupt moveqz a3, a0, a2 # a3 = LOCKLEVEL iff interrupt
movi a2, PS_WOE_MASK KABI_W movi a2, PS_WOE_MASK
or a3, a3, a2 KABI_W or a3, a3, a2
rsr a2, exccause rsr a2, exccause
#endif #endif
@ -461,14 +474,14 @@ common_exception:
*/ */
rsr a4, excsave1 rsr a4, excsave1
mov a6, a1 # pass stack frame
mov a7, a2 # pass EXCCAUSE
addx4 a4, a2, a4 addx4 a4, a2, a4
l32i a4, a4, EXC_TABLE_DEFAULT # load handler l32i a4, a4, EXC_TABLE_DEFAULT # load handler
mov abi_arg1, a2 # pass EXCCAUSE
mov abi_arg0, a1 # pass stack frame
/* Call the second-level handler */ /* Call the second-level handler */
callx4 a4 abi_callx a4
/* Jump here for exception exit */ /* Jump here for exception exit */
.global common_exception_return .global common_exception_return
@ -482,15 +495,15 @@ common_exception_return:
1: 1:
irq_save a2, a3 irq_save a2, a3
#ifdef CONFIG_TRACE_IRQFLAGS #ifdef CONFIG_TRACE_IRQFLAGS
call4 trace_hardirqs_off abi_call trace_hardirqs_off
#endif #endif
/* Jump if we are returning from kernel exceptions. */ /* Jump if we are returning from kernel exceptions. */
l32i a3, a1, PT_PS l32i abi_saved1, a1, PT_PS
GET_THREAD_INFO(a2, a1) GET_THREAD_INFO(a2, a1)
l32i a4, a2, TI_FLAGS l32i a4, a2, TI_FLAGS
_bbci.l a3, PS_UM_BIT, 6f _bbci.l abi_saved1, PS_UM_BIT, 6f
/* Specific to a user exception exit: /* Specific to a user exception exit:
* We need to check some flags for signal handling and rescheduling, * We need to check some flags for signal handling and rescheduling,
@ -509,20 +522,20 @@ common_exception_return:
/* Call do_signal() */ /* Call do_signal() */
#ifdef CONFIG_TRACE_IRQFLAGS #ifdef CONFIG_TRACE_IRQFLAGS
call4 trace_hardirqs_on abi_call trace_hardirqs_on
#endif #endif
rsil a2, 0 rsil a2, 0
mov a6, a1 mov abi_arg0, a1
call4 do_notify_resume # int do_notify_resume(struct pt_regs*) abi_call do_notify_resume # int do_notify_resume(struct pt_regs*)
j 1b j 1b
3: /* Reschedule */ 3: /* Reschedule */
#ifdef CONFIG_TRACE_IRQFLAGS #ifdef CONFIG_TRACE_IRQFLAGS
call4 trace_hardirqs_on abi_call trace_hardirqs_on
#endif #endif
rsil a2, 0 rsil a2, 0
call4 schedule # void schedule (void) abi_call schedule # void schedule (void)
j 1b j 1b
#ifdef CONFIG_PREEMPTION #ifdef CONFIG_PREEMPTION
@ -533,33 +546,33 @@ common_exception_return:
l32i a4, a2, TI_PRE_COUNT l32i a4, a2, TI_PRE_COUNT
bnez a4, 4f bnez a4, 4f
call4 preempt_schedule_irq abi_call preempt_schedule_irq
j 4f j 4f
#endif #endif
#if XTENSA_FAKE_NMI #if XTENSA_FAKE_NMI
.LNMIexit: .LNMIexit:
l32i a3, a1, PT_PS l32i abi_saved1, a1, PT_PS
_bbci.l a3, PS_UM_BIT, 4f _bbci.l abi_saved1, PS_UM_BIT, 4f
#endif #endif
5: 5:
#ifdef CONFIG_HAVE_HW_BREAKPOINT #ifdef CONFIG_HAVE_HW_BREAKPOINT
_bbci.l a4, TIF_DB_DISABLED, 7f _bbci.l a4, TIF_DB_DISABLED, 7f
call4 restore_dbreak abi_call restore_dbreak
7: 7:
#endif #endif
#ifdef CONFIG_DEBUG_TLB_SANITY #ifdef CONFIG_DEBUG_TLB_SANITY
l32i a4, a1, PT_DEPC l32i a4, a1, PT_DEPC
bgeui a4, VALID_DOUBLE_EXCEPTION_ADDRESS, 4f bgeui a4, VALID_DOUBLE_EXCEPTION_ADDRESS, 4f
call4 check_tlb_sanity abi_call check_tlb_sanity
#endif #endif
6: 6:
4: 4:
#ifdef CONFIG_TRACE_IRQFLAGS #ifdef CONFIG_TRACE_IRQFLAGS
extui a4, a3, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH extui a4, abi_saved1, PS_INTLEVEL_SHIFT, PS_INTLEVEL_WIDTH
bgei a4, LOCKLEVEL, 1f bgei a4, LOCKLEVEL, 1f
call4 trace_hardirqs_on abi_call trace_hardirqs_on
1: 1:
#endif #endif
/* Restore optional registers. */ /* Restore optional registers. */
@ -572,14 +585,15 @@ common_exception_return:
l32i a2, a1, PT_SCOMPARE1 l32i a2, a1, PT_SCOMPARE1
wsr a2, scompare1 wsr a2, scompare1
#endif #endif
wsr a3, ps /* disable interrupts */ wsr abi_saved1, ps /* disable interrupts */
_bbci.l a3, PS_UM_BIT, kernel_exception_exit _bbci.l abi_saved1, PS_UM_BIT, kernel_exception_exit
user_exception_exit: user_exception_exit:
/* Restore the state of the task and return from the exception. */ /* Restore the state of the task and return from the exception. */
#if defined(USER_SUPPORT_WINDOWED)
/* Switch to the user thread WINDOWBASE. Save SP temporarily in DEPC */ /* Switch to the user thread WINDOWBASE. Save SP temporarily in DEPC */
l32i a2, a1, PT_WINDOWBASE l32i a2, a1, PT_WINDOWBASE
@ -634,8 +648,10 @@ user_exception_exit:
* frame where we had loaded a2), or at least the lower 4 bits * frame where we had loaded a2), or at least the lower 4 bits
* (if we have restored WSBITS-1 frames). * (if we have restored WSBITS-1 frames).
*/ */
2: 2:
#else
movi a2, 1
#endif
#if XCHAL_HAVE_THREADPTR #if XCHAL_HAVE_THREADPTR
l32i a3, a1, PT_THREADPTR l32i a3, a1, PT_THREADPTR
wur a3, threadptr wur a3, threadptr
@ -650,6 +666,7 @@ user_exception_exit:
kernel_exception_exit: kernel_exception_exit:
#if defined(__XTENSA_WINDOWED_ABI__)
/* Check if we have to do a movsp. /* Check if we have to do a movsp.
* *
* We only have to do a movsp if the previous window-frame has * We only have to do a movsp if the previous window-frame has
@ -702,6 +719,9 @@ kernel_exception_exit:
* *
* Note: We expect a2 to hold PT_WMASK * Note: We expect a2 to hold PT_WMASK
*/ */
#else
movi a2, 1
#endif
common_exception_exit: common_exception_exit:
@ -920,14 +940,16 @@ unrecoverable_text:
ENTRY(unrecoverable_exception) ENTRY(unrecoverable_exception)
#if XCHAL_HAVE_WINDOWED
movi a0, 1 movi a0, 1
movi a1, 0 movi a1, 0
wsr a0, windowstart wsr a0, windowstart
wsr a1, windowbase wsr a1, windowbase
rsync rsync
#endif
movi a1, PS_WOE_MASK | LOCKLEVEL movi a1, KERNEL_PS_WOE_MASK | LOCKLEVEL
wsr a1, ps wsr a1, ps
rsync rsync
@ -935,8 +957,8 @@ ENTRY(unrecoverable_exception)
movi a0, 0 movi a0, 0
addi a1, a1, PT_REGS_OFFSET addi a1, a1, PT_REGS_OFFSET
movi a6, unrecoverable_text movi abi_arg0, unrecoverable_text
call4 panic abi_call panic
1: j 1b 1: j 1b
@ -947,6 +969,7 @@ ENDPROC(unrecoverable_exception)
__XTENSA_HANDLER __XTENSA_HANDLER
.literal_position .literal_position
#ifdef SUPPORT_WINDOWED
/* /*
* Fast-handler for alloca exceptions * Fast-handler for alloca exceptions
* *
@ -1010,6 +1033,7 @@ ENTRY(fast_alloca)
8: j _WindowUnderflow8 8: j _WindowUnderflow8
4: j _WindowUnderflow4 4: j _WindowUnderflow4
ENDPROC(fast_alloca) ENDPROC(fast_alloca)
#endif
#ifdef CONFIG_USER_ABI_CALL0_PROBE #ifdef CONFIG_USER_ABI_CALL0_PROBE
/* /*
@ -1206,7 +1230,8 @@ ENDPROC(fast_syscall_xtensa)
* Note: We assume the stack pointer is EXC_TABLE_KSTK in the fixup handler. * Note: We assume the stack pointer is EXC_TABLE_KSTK in the fixup handler.
*/ */
#ifdef CONFIG_FAST_SYSCALL_SPILL_REGISTERS #if defined(CONFIG_FAST_SYSCALL_SPILL_REGISTERS) && \
defined(USER_SUPPORT_WINDOWED)
ENTRY(fast_syscall_spill_registers) ENTRY(fast_syscall_spill_registers)
@ -1403,12 +1428,12 @@ ENTRY(fast_syscall_spill_registers)
rsr a3, excsave1 rsr a3, excsave1
l32i a1, a3, EXC_TABLE_KSTK l32i a1, a3, EXC_TABLE_KSTK
movi a4, PS_WOE_MASK | LOCKLEVEL movi a4, KERNEL_PS_WOE_MASK | LOCKLEVEL
wsr a4, ps wsr a4, ps
rsync rsync
movi a6, SIGSEGV movi abi_arg0, SIGSEGV
call4 do_exit abi_call do_exit
/* shouldn't return, so panic */ /* shouldn't return, so panic */
@ -1887,57 +1912,77 @@ ENDPROC(fast_store_prohibited)
ENTRY(system_call) ENTRY(system_call)
#if defined(__XTENSA_WINDOWED_ABI__)
abi_entry_default abi_entry_default
#elif defined(__XTENSA_CALL0_ABI__)
abi_entry(12)
s32i a0, sp, 0
s32i abi_saved0, sp, 4
s32i abi_saved1, sp, 8
mov abi_saved0, a2
#else
#error Unsupported Xtensa ABI
#endif
/* regs->syscall = regs->areg[2] */ /* regs->syscall = regs->areg[2] */
l32i a7, a2, PT_AREG2 l32i a7, abi_saved0, PT_AREG2
s32i a7, a2, PT_SYSCALL s32i a7, abi_saved0, PT_SYSCALL
GET_THREAD_INFO(a4, a1) GET_THREAD_INFO(a4, a1)
l32i a3, a4, TI_FLAGS l32i abi_saved1, a4, TI_FLAGS
movi a4, _TIF_WORK_MASK movi a4, _TIF_WORK_MASK
and a3, a3, a4 and abi_saved1, abi_saved1, a4
beqz a3, 1f beqz abi_saved1, 1f
mov a6, a2 mov abi_arg0, abi_saved0
call4 do_syscall_trace_enter abi_call do_syscall_trace_enter
beqz a6, .Lsyscall_exit beqz abi_rv, .Lsyscall_exit
l32i a7, a2, PT_SYSCALL l32i a7, abi_saved0, PT_SYSCALL
1: 1:
/* syscall = sys_call_table[syscall_nr] */ /* syscall = sys_call_table[syscall_nr] */
movi a4, sys_call_table movi a4, sys_call_table
movi a5, __NR_syscalls movi a5, __NR_syscalls
movi a6, -ENOSYS movi abi_rv, -ENOSYS
bgeu a7, a5, 1f bgeu a7, a5, 1f
addx4 a4, a7, a4 addx4 a4, a7, a4
l32i a4, a4, 0 l32i abi_tmp0, a4, 0
/* Load args: arg0 - arg5 are passed via regs. */ /* Load args: arg0 - arg5 are passed via regs. */
l32i a6, a2, PT_AREG6 l32i abi_arg0, abi_saved0, PT_AREG6
l32i a7, a2, PT_AREG3 l32i abi_arg1, abi_saved0, PT_AREG3
l32i a8, a2, PT_AREG4 l32i abi_arg2, abi_saved0, PT_AREG4
l32i a9, a2, PT_AREG5 l32i abi_arg3, abi_saved0, PT_AREG5
l32i a10, a2, PT_AREG8 l32i abi_arg4, abi_saved0, PT_AREG8
l32i a11, a2, PT_AREG9 l32i abi_arg5, abi_saved0, PT_AREG9
callx4 a4 abi_callx abi_tmp0
1: /* regs->areg[2] = return_value */ 1: /* regs->areg[2] = return_value */
s32i a6, a2, PT_AREG2 s32i abi_rv, abi_saved0, PT_AREG2
bnez a3, 1f bnez abi_saved1, 1f
.Lsyscall_exit: .Lsyscall_exit:
#if defined(__XTENSA_WINDOWED_ABI__)
abi_ret_default abi_ret_default
#elif defined(__XTENSA_CALL0_ABI__)
l32i a0, sp, 0
l32i abi_saved0, sp, 4
l32i abi_saved1, sp, 8
abi_ret(12)
#else
#error Unsupported Xtensa ABI
#endif
1: 1:
mov a6, a2 mov abi_arg0, abi_saved0
call4 do_syscall_trace_leave abi_call do_syscall_trace_leave
abi_ret_default j .Lsyscall_exit
ENDPROC(system_call) ENDPROC(system_call)
@ -1988,8 +2033,18 @@ ENDPROC(system_call)
ENTRY(_switch_to) ENTRY(_switch_to)
#if defined(__XTENSA_WINDOWED_ABI__)
abi_entry(XTENSA_SPILL_STACK_RESERVE) abi_entry(XTENSA_SPILL_STACK_RESERVE)
#elif defined(__XTENSA_CALL0_ABI__)
abi_entry(16)
s32i a12, sp, 0
s32i a13, sp, 4
s32i a14, sp, 8
s32i a15, sp, 12
#else
#error Unsupported Xtensa ABI
#endif
mov a11, a3 # and 'next' (a3) mov a11, a3 # and 'next' (a3)
l32i a4, a2, TASK_THREAD_INFO l32i a4, a2, TASK_THREAD_INFO
@ -2033,7 +2088,9 @@ ENTRY(_switch_to)
/* Flush register file. */ /* Flush register file. */
#if defined(__XTENSA_WINDOWED_ABI__)
spill_registers_kernel spill_registers_kernel
#endif
/* Set kernel stack (and leave critical section) /* Set kernel stack (and leave critical section)
* Note: It's save to set it here. The stack will not be overwritten * Note: It's save to set it here. The stack will not be overwritten
@ -2055,34 +2112,43 @@ ENTRY(_switch_to)
wsr a14, ps wsr a14, ps
rsync rsync
#if defined(__XTENSA_WINDOWED_ABI__)
abi_ret(XTENSA_SPILL_STACK_RESERVE) abi_ret(XTENSA_SPILL_STACK_RESERVE)
#elif defined(__XTENSA_CALL0_ABI__)
l32i a12, sp, 0
l32i a13, sp, 4
l32i a14, sp, 8
l32i a15, sp, 12
abi_ret(16)
#else
#error Unsupported Xtensa ABI
#endif
ENDPROC(_switch_to) ENDPROC(_switch_to)
ENTRY(ret_from_fork) ENTRY(ret_from_fork)
/* void schedule_tail (struct task_struct *prev) /* void schedule_tail (struct task_struct *prev)
* Note: prev is still in a6 (return value from fake call4 frame) * Note: prev is still in abi_arg0 (return value from fake call frame)
*/ */
call4 schedule_tail abi_call schedule_tail
mov a6, a1 mov abi_arg0, a1
call4 do_syscall_trace_leave abi_call do_syscall_trace_leave
j common_exception_return
j common_exception_return
ENDPROC(ret_from_fork) ENDPROC(ret_from_fork)
/* /*
* Kernel thread creation helper * Kernel thread creation helper
* On entry, set up by copy_thread: a2 = thread_fn, a3 = thread_fn arg * On entry, set up by copy_thread: abi_saved0 = thread_fn,
* left from _switch_to: a6 = prev * abi_saved1 = thread_fn arg. Left from _switch_to: abi_arg0 = prev
*/ */
ENTRY(ret_from_kernel_thread) ENTRY(ret_from_kernel_thread)
call4 schedule_tail abi_call schedule_tail
mov a6, a3 mov abi_arg0, abi_saved1
callx4 a2 abi_callx abi_saved0
j common_exception_return j common_exception_return
ENDPROC(ret_from_kernel_thread) ENDPROC(ret_from_kernel_thread)

24
arch/xtensa/kernel/head.S
View File

@ -15,6 +15,7 @@
* Kevin Chea * Kevin Chea
*/ */
#include <asm/asmmacro.h>
#include <asm/processor.h> #include <asm/processor.h>
#include <asm/page.h> #include <asm/page.h>
#include <asm/cacheasm.h> #include <asm/cacheasm.h>
@ -66,11 +67,13 @@ _SetupOCD:
* xt-gdb to single step via DEBUG exceptions received directly * xt-gdb to single step via DEBUG exceptions received directly
* by ocd. * by ocd.
*/ */
#if XCHAL_HAVE_WINDOWED
movi a1, 1 movi a1, 1
movi a0, 0 movi a0, 0
wsr a1, windowstart wsr a1, windowstart
wsr a0, windowbase wsr a0, windowbase
rsync rsync
#endif
movi a1, LOCKLEVEL movi a1, LOCKLEVEL
wsr a1, ps wsr a1, ps
@ -193,9 +196,10 @@ ENTRY(_startup)
movi a1, start_info movi a1, start_info
l32i a1, a1, 0 l32i a1, a1, 0
movi a2, PS_WOE_MASK | LOCKLEVEL /* Disable interrupts. */
# WOE=1, INTLEVEL=LOCKLEVEL, UM=0 /* Enable window exceptions if kernel is built with windowed ABI. */
wsr a2, ps # (enable reg-windows; progmode stack) movi a2, KERNEL_PS_WOE_MASK | LOCKLEVEL
wsr a2, ps
rsync rsync
#ifdef CONFIG_SMP #ifdef CONFIG_SMP
@ -267,13 +271,13 @@ ENTRY(_startup)
l32i a1, a1, 0 l32i a1, a1, 0
#endif #endif
movi a6, 0 movi abi_arg0, 0
xsr a6, excsave1 xsr abi_arg0, excsave1
/* init_arch kick-starts the linux kernel */ /* init_arch kick-starts the linux kernel */
call4 init_arch abi_call init_arch
call4 start_kernel abi_call start_kernel
should_never_return: should_never_return:
j should_never_return j should_never_return
@ -297,10 +301,10 @@ should_never_return:
s32i a3, a2, 0 s32i a3, a2, 0
memw memw
movi a6, 0 movi abi_arg0, 0
wsr a6, excsave1 wsr abi_arg0, excsave1
call4 secondary_start_kernel abi_call secondary_start_kernel
j should_never_return j should_never_return
#endif /* CONFIG_SMP */ #endif /* CONFIG_SMP */

38
arch/xtensa/kernel/mcount.S
View File

@ -17,11 +17,16 @@
/* /*
* Entry condition: * Entry condition:
* *
* a2: a0 of the caller * a2: a0 of the caller in windowed ABI
* a10: a0 of the caller in call0 ABI
*
* In call0 ABI the function _mcount is called with the special ABI:
* its argument is in a10 and all the usual argument registers (a2 - a7)
* must be preserved in addition to callee-saved a12 - a15.
*/ */
ENTRY(_mcount) ENTRY(_mcount)
#if defined(__XTENSA_WINDOWED_ABI__)
abi_entry_default abi_entry_default
movi a4, ftrace_trace_function movi a4, ftrace_trace_function
@ -42,7 +47,36 @@ ENTRY(_mcount)
callx4 a4 callx4 a4
abi_ret_default abi_ret_default
#elif defined(__XTENSA_CALL0_ABI__)
abi_entry_default
movi a9, ftrace_trace_function
l32i a9, a9, 0
movi a11, ftrace_stub
bne a9, a11, 1f
abi_ret_default
1: abi_entry(28)
s32i a0, sp, 0
s32i a2, sp, 4
s32i a3, sp, 8
s32i a4, sp, 12
s32i a5, sp, 16
s32i a6, sp, 20
s32i a7, sp, 24
addi a2, a10, -MCOUNT_INSN_SIZE
callx0 a9
l32i a0, sp, 0
l32i a2, sp, 4
l32i a3, sp, 8
l32i a4, sp, 12
l32i a5, sp, 16
l32i a6, sp, 20
l32i a7, sp, 24
abi_ret(28)
#else
#error Unsupported Xtensa ABI
#endif
ENDPROC(_mcount) ENDPROC(_mcount)
ENTRY(ftrace_stub) ENTRY(ftrace_stub)

27
arch/xtensa/kernel/process.c
View File

@ -211,11 +211,18 @@ int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
struct thread_info *ti; struct thread_info *ti;
#endif #endif
#if defined(__XTENSA_WINDOWED_ABI__)
/* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */ /* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */
SPILL_SLOT(childregs, 1) = (unsigned long)childregs; SPILL_SLOT(childregs, 1) = (unsigned long)childregs;
SPILL_SLOT(childregs, 0) = 0; SPILL_SLOT(childregs, 0) = 0;
p->thread.sp = (unsigned long)childregs; p->thread.sp = (unsigned long)childregs;
#elif defined(__XTENSA_CALL0_ABI__)
/* Reserve 16 bytes for the _switch_to stack frame. */
p->thread.sp = (unsigned long)childregs - 16;
#else
#error Unsupported Xtensa ABI
#endif
if (!(p->flags & (PF_KTHREAD | PF_IO_WORKER))) { if (!(p->flags & (PF_KTHREAD | PF_IO_WORKER))) {
struct pt_regs *regs = current_pt_regs(); struct pt_regs *regs = current_pt_regs();
@ -272,11 +279,25 @@ int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
p->thread.ra = MAKE_RA_FOR_CALL( p->thread.ra = MAKE_RA_FOR_CALL(
(unsigned long)ret_from_kernel_thread, 1); (unsigned long)ret_from_kernel_thread, 1);
/* pass parameters to ret_from_kernel_thread: /* pass parameters to ret_from_kernel_thread: */
* a2 = thread_fn, a3 = thread_fn arg #if defined(__XTENSA_WINDOWED_ABI__)
/*
* a2 = thread_fn, a3 = thread_fn arg.
* Window underflow will load registers from the
* spill slots on the stack on return from _switch_to.
*/ */
SPILL_SLOT(childregs, 3) = thread_fn_arg;
SPILL_SLOT(childregs, 2) = usp_thread_fn; SPILL_SLOT(childregs, 2) = usp_thread_fn;
SPILL_SLOT(childregs, 3) = thread_fn_arg;
#elif defined(__XTENSA_CALL0_ABI__)
/*
* a12 = thread_fn, a13 = thread_fn arg.
* _switch_to epilogue will load registers from the stack.
*/
((unsigned long *)p->thread.sp)[0] = usp_thread_fn;
((unsigned long *)p->thread.sp)[1] = thread_fn_arg;
#else
#error Unsupported Xtensa ABI
#endif
/* Childregs are only used when we're going to userspace /* Childregs are only used when we're going to userspace
* in which case start_thread will set them up. * in which case start_thread will set them up.

102
arch/xtensa/kernel/setup.c
View File

@ -37,14 +37,15 @@
#include <asm/bootparam.h> #include <asm/bootparam.h>
#include <asm/kasan.h> #include <asm/kasan.h>
#include <asm/mmu_context.h> #include <asm/mmu_context.h>
#include <asm/processor.h>
#include <asm/timex.h>
#include <asm/platform.h>
#include <asm/page.h> #include <asm/page.h>
#include <asm/setup.h>
#include <asm/param.h> #include <asm/param.h>
#include <asm/platform.h>
#include <asm/processor.h>
#include <asm/sections.h>
#include <asm/setup.h>
#include <asm/smp.h> #include <asm/smp.h>
#include <asm/sysmem.h> #include <asm/sysmem.h>
#include <asm/timex.h>
#if defined(CONFIG_VGA_CONSOLE) || defined(CONFIG_DUMMY_CONSOLE) #if defined(CONFIG_VGA_CONSOLE) || defined(CONFIG_DUMMY_CONSOLE)
struct screen_info screen_info = { struct screen_info screen_info = {
@ -271,49 +272,6 @@ void __init init_arch(bp_tag_t *bp_start)
* Initialize system. Setup memory and reserve regions. * Initialize system. Setup memory and reserve regions.
*/ */
extern char _end[];
extern char _stext[];
extern char _WindowVectors_text_start;
extern char _WindowVectors_text_end;
extern char _DebugInterruptVector_text_start;
extern char _DebugInterruptVector_text_end;
extern char _KernelExceptionVector_text_start;
extern char _KernelExceptionVector_text_end;
extern char _UserExceptionVector_text_start;
extern char _UserExceptionVector_text_end;
extern char _DoubleExceptionVector_text_start;
extern char _DoubleExceptionVector_text_end;
extern char _exception_text_start;
extern char _exception_text_end;
#if XCHAL_EXCM_LEVEL >= 2
extern char _Level2InterruptVector_text_start;
extern char _Level2InterruptVector_text_end;
#endif
#if XCHAL_EXCM_LEVEL >= 3
extern char _Level3InterruptVector_text_start;
extern char _Level3InterruptVector_text_end;
#endif
#if XCHAL_EXCM_LEVEL >= 4
extern char _Level4InterruptVector_text_start;
extern char _Level4InterruptVector_text_end;
#endif
#if XCHAL_EXCM_LEVEL >= 5
extern char _Level5InterruptVector_text_start;
extern char _Level5InterruptVector_text_end;
#endif
#if XCHAL_EXCM_LEVEL >= 6
extern char _Level6InterruptVector_text_start;
extern char _Level6InterruptVector_text_end;
#endif
#ifdef CONFIG_SMP
extern char _SecondaryResetVector_text_start;
extern char _SecondaryResetVector_text_end;
#endif
#ifdef CONFIG_XIP_KERNEL
extern char _xip_start[];
extern char _xip_end[];
#endif
static inline int __init_memblock mem_reserve(unsigned long start, static inline int __init_memblock mem_reserve(unsigned long start,
unsigned long end) unsigned long end)
{ {
@ -349,49 +307,51 @@ void __init setup_arch(char **cmdline_p)
#endif #endif
#ifdef CONFIG_VECTORS_ADDR #ifdef CONFIG_VECTORS_ADDR
mem_reserve(__pa(&_WindowVectors_text_start), #ifdef SUPPORT_WINDOWED
__pa(&_WindowVectors_text_end)); mem_reserve(__pa(_WindowVectors_text_start),
__pa(_WindowVectors_text_end));
#endif
mem_reserve(__pa(&_DebugInterruptVector_text_start), mem_reserve(__pa(_DebugInterruptVector_text_start),
__pa(&_DebugInterruptVector_text_end)); __pa(_DebugInterruptVector_text_end));
mem_reserve(__pa(&_KernelExceptionVector_text_start), mem_reserve(__pa(_KernelExceptionVector_text_start),
__pa(&_KernelExceptionVector_text_end)); __pa(_KernelExceptionVector_text_end));
mem_reserve(__pa(&_UserExceptionVector_text_start), mem_reserve(__pa(_UserExceptionVector_text_start),
__pa(&_UserExceptionVector_text_end)); __pa(_UserExceptionVector_text_end));
mem_reserve(__pa(&_DoubleExceptionVector_text_start), mem_reserve(__pa(_DoubleExceptionVector_text_start),
__pa(&_DoubleExceptionVector_text_end)); __pa(_DoubleExceptionVector_text_end));
mem_reserve(__pa(&_exception_text_start), mem_reserve(__pa(_exception_text_start),
__pa(&_exception_text_end)); __pa(_exception_text_end));
#if XCHAL_EXCM_LEVEL >= 2 #if XCHAL_EXCM_LEVEL >= 2
mem_reserve(__pa(&_Level2InterruptVector_text_start), mem_reserve(__pa(_Level2InterruptVector_text_start),
__pa(&_Level2InterruptVector_text_end)); __pa(_Level2InterruptVector_text_end));
#endif #endif
#if XCHAL_EXCM_LEVEL >= 3 #if XCHAL_EXCM_LEVEL >= 3
mem_reserve(__pa(&_Level3InterruptVector_text_start), mem_reserve(__pa(_Level3InterruptVector_text_start),
__pa(&_Level3InterruptVector_text_end)); __pa(_Level3InterruptVector_text_end));
#endif #endif
#if XCHAL_EXCM_LEVEL >= 4 #if XCHAL_EXCM_LEVEL >= 4
mem_reserve(__pa(&_Level4InterruptVector_text_start), mem_reserve(__pa(_Level4InterruptVector_text_start),
__pa(&_Level4InterruptVector_text_end)); __pa(_Level4InterruptVector_text_end));
#endif #endif
#if XCHAL_EXCM_LEVEL >= 5 #if XCHAL_EXCM_LEVEL >= 5
mem_reserve(__pa(&_Level5InterruptVector_text_start), mem_reserve(__pa(_Level5InterruptVector_text_start),
__pa(&_Level5InterruptVector_text_end)); __pa(_Level5InterruptVector_text_end));
#endif #endif
#if XCHAL_EXCM_LEVEL >= 6 #if XCHAL_EXCM_LEVEL >= 6
mem_reserve(__pa(&_Level6InterruptVector_text_start), mem_reserve(__pa(_Level6InterruptVector_text_start),
__pa(&_Level6InterruptVector_text_end)); __pa(_Level6InterruptVector_text_end));
#endif #endif
#endif /* CONFIG_VECTORS_ADDR */ #endif /* CONFIG_VECTORS_ADDR */
#ifdef CONFIG_SMP #ifdef CONFIG_SMP
mem_reserve(__pa(&_SecondaryResetVector_text_start), mem_reserve(__pa(_SecondaryResetVector_text_start),
__pa(&_SecondaryResetVector_text_end)); __pa(_SecondaryResetVector_text_end));
#endif #endif
parse_early_param(); parse_early_param();
bootmem_init(); bootmem_init();

10
arch/xtensa/kernel/signal.c
View File

@ -45,12 +45,13 @@ struct rt_sigframe
unsigned int window[4]; unsigned int window[4];
}; };
#if defined(USER_SUPPORT_WINDOWED)
/* /*
* Flush register windows stored in pt_regs to stack. * Flush register windows stored in pt_regs to stack.
* Returns 1 for errors. * Returns 1 for errors.
*/ */
int static int
flush_window_regs_user(struct pt_regs *regs) flush_window_regs_user(struct pt_regs *regs)
{ {
const unsigned long ws = regs->windowstart; const unsigned long ws = regs->windowstart;
@ -121,6 +122,13 @@ flush_window_regs_user(struct pt_regs *regs)
errout: errout:
return err; return err;
} }
#else
static int
flush_window_regs_user(struct pt_regs *regs)
{
return 0;
}
#endif
/* /*
* Note: We don't copy double exception 'regs', we have to finish double exc. * Note: We don't copy double exception 'regs', we have to finish double exc.

6
arch/xtensa/kernel/traps.c
View File

@ -97,7 +97,9 @@ static dispatch_init_table_t __initdata dispatch_init_table[] = {
/* EXCCAUSE_INSTRUCTION_FETCH unhandled */ /* EXCCAUSE_INSTRUCTION_FETCH unhandled */
/* EXCCAUSE_LOAD_STORE_ERROR unhandled*/ /* EXCCAUSE_LOAD_STORE_ERROR unhandled*/
{ EXCCAUSE_LEVEL1_INTERRUPT, 0, do_interrupt }, { EXCCAUSE_LEVEL1_INTERRUPT, 0, do_interrupt },
#ifdef SUPPORT_WINDOWED
{ EXCCAUSE_ALLOCA, USER|KRNL, fast_alloca }, { EXCCAUSE_ALLOCA, USER|KRNL, fast_alloca },
#endif
/* EXCCAUSE_INTEGER_DIVIDE_BY_ZERO unhandled */ /* EXCCAUSE_INTEGER_DIVIDE_BY_ZERO unhandled */
/* EXCCAUSE_PRIVILEGED unhandled */ /* EXCCAUSE_PRIVILEGED unhandled */
#if XCHAL_UNALIGNED_LOAD_EXCEPTION || XCHAL_UNALIGNED_STORE_EXCEPTION #if XCHAL_UNALIGNED_LOAD_EXCEPTION || XCHAL_UNALIGNED_STORE_EXCEPTION
@ -462,12 +464,10 @@ void secondary_trap_init(void)
void show_regs(struct pt_regs * regs) void show_regs(struct pt_regs * regs)
{ {
int i, wmask; int i;
show_regs_print_info(KERN_DEFAULT); show_regs_print_info(KERN_DEFAULT);
wmask = regs->wmask & ~1;
for (i = 0; i < 16; i++) { for (i = 0; i < 16; i++) {
if ((i % 8) == 0) if ((i % 8) == 0)
pr_info("a%02d:", i); pr_info("a%02d:", i);

55
arch/xtensa/kernel/vectors.S
View File

@ -226,6 +226,7 @@ ENTRY(_DoubleExceptionVector)
xsr a0, depc # get DEPC, save a0 xsr a0, depc # get DEPC, save a0
#ifdef SUPPORT_WINDOWED
movi a2, WINDOW_VECTORS_VADDR movi a2, WINDOW_VECTORS_VADDR
_bltu a0, a2, .Lfixup _bltu a0, a2, .Lfixup
addi a2, a2, WINDOW_VECTORS_SIZE addi a2, a2, WINDOW_VECTORS_SIZE
@ -275,6 +276,10 @@ _DoubleExceptionVector_WindowUnderflow:
l32i a0, a0, EXC_TABLE_FAST_USER l32i a0, a0, EXC_TABLE_FAST_USER
jx a0 jx a0
#else
j .Lfixup
#endif
/* /*
* We only allow the ITLB miss exception if we are in kernel space. * We only allow the ITLB miss exception if we are in kernel space.
* All other exceptions are unexpected and thus unrecoverable! * All other exceptions are unexpected and thus unrecoverable!
@ -343,6 +348,7 @@ _DoubleExceptionVector_WindowUnderflow:
l32i a0, a0, EXC_TABLE_FAST_USER l32i a0, a0, EXC_TABLE_FAST_USER
jx a0 jx a0
#ifdef SUPPORT_WINDOWED
/* /*
* Restart window OVERFLOW exception. * Restart window OVERFLOW exception.
* Currently: * Currently:
@ -475,9 +481,12 @@ _DoubleExceptionVector_handle_exception:
rsr a0, depc rsr a0, depc
rotw -3 rotw -3
j 1b j 1b
#endif
ENDPROC(_DoubleExceptionVector) ENDPROC(_DoubleExceptionVector)
#ifdef SUPPORT_WINDOWED
/* /*
* Fixup handler for TLB miss in double exception handler for window owerflow. * Fixup handler for TLB miss in double exception handler for window owerflow.
* We get here with windowbase set to the window that was being spilled and * We get here with windowbase set to the window that was being spilled and
@ -590,6 +599,8 @@ ENTRY(window_overflow_restore_a0_fixup)
ENDPROC(window_overflow_restore_a0_fixup) ENDPROC(window_overflow_restore_a0_fixup)
#endif
/* /*
* Debug interrupt vector * Debug interrupt vector
* *
@ -650,6 +661,25 @@ ENTRY(_Level\level\()InterruptVector)
irq_entry_level 5 irq_entry_level 5
irq_entry_level 6 irq_entry_level 6
#if XCHAL_EXCM_LEVEL >= 2
/*
* Continuation of medium priority interrupt dispatch code.
* On entry here, a0 contains PS, and EPC2 contains saved a0:
*/
__XTENSA_HANDLER
.align 4
_SimulateUserKernelVectorException:
addi a0, a0, (1 << PS_EXCM_BIT)
#if !XTENSA_FAKE_NMI
wsr a0, ps
#endif
bbsi.l a0, PS_UM_BIT, 1f # branch if user mode
xsr a0, excsave2 # restore a0
j _KernelExceptionVector # simulate kernel vector exception
1: xsr a0, excsave2 # restore a0
j _UserExceptionVector # simulate user vector exception
#endif
/* Window overflow and underflow handlers. /* Window overflow and underflow handlers.
* The handlers must be 64 bytes apart, first starting with the underflow * The handlers must be 64 bytes apart, first starting with the underflow
@ -668,6 +698,8 @@ ENTRY(_Level\level\()InterruptVector)
.section .WindowVectors.text, "ax" .section .WindowVectors.text, "ax"
#ifdef SUPPORT_WINDOWED
/* 4-Register Window Overflow Vector (Handler) */ /* 4-Register Window Overflow Vector (Handler) */
ENTRY_ALIGN64(_WindowOverflow4) ENTRY_ALIGN64(_WindowOverflow4)
@ -680,27 +712,6 @@ ENTRY_ALIGN64(_WindowOverflow4)
ENDPROC(_WindowOverflow4) ENDPROC(_WindowOverflow4)
#if XCHAL_EXCM_LEVEL >= 2
/* Not a window vector - but a convenient location
* (where we know there's space) for continuation of
* medium priority interrupt dispatch code.
* On entry here, a0 contains PS, and EPC2 contains saved a0:
*/
.align 4
_SimulateUserKernelVectorException:
addi a0, a0, (1 << PS_EXCM_BIT)
#if !XTENSA_FAKE_NMI
wsr a0, ps
#endif
bbsi.l a0, PS_UM_BIT, 1f # branch if user mode
xsr a0, excsave2 # restore a0
j _KernelExceptionVector # simulate kernel vector exception
1: xsr a0, excsave2 # restore a0
j _UserExceptionVector # simulate user vector exception
#endif
/* 4-Register Window Underflow Vector (Handler) */ /* 4-Register Window Underflow Vector (Handler) */
ENTRY_ALIGN64(_WindowUnderflow4) ENTRY_ALIGN64(_WindowUnderflow4)
@ -789,4 +800,6 @@ ENTRY_ALIGN64(_WindowUnderflow12)
ENDPROC(_WindowUnderflow12) ENDPROC(_WindowUnderflow12)
#endif
.text .text

12
arch/xtensa/kernel/vmlinux.lds.S
View File

@ -94,7 +94,9 @@ SECTIONS
. = ALIGN(PAGE_SIZE); . = ALIGN(PAGE_SIZE);
_vecbase = .; _vecbase = .;
#ifdef SUPPORT_WINDOWED
SECTION_VECTOR2 (.WindowVectors.text, WINDOW_VECTORS_VADDR) SECTION_VECTOR2 (.WindowVectors.text, WINDOW_VECTORS_VADDR)
#endif
#if XCHAL_EXCM_LEVEL >= 2 #if XCHAL_EXCM_LEVEL >= 2
SECTION_VECTOR2 (.Level2InterruptVector.text, INTLEVEL2_VECTOR_VADDR) SECTION_VECTOR2 (.Level2InterruptVector.text, INTLEVEL2_VECTOR_VADDR)
#endif #endif
@ -166,8 +168,10 @@ SECTIONS
__boot_reloc_table_start = ABSOLUTE(.); __boot_reloc_table_start = ABSOLUTE(.);
#if !MERGED_VECTORS #if !MERGED_VECTORS
#ifdef SUPPORT_WINDOWED
RELOCATE_ENTRY(_WindowVectors_text, RELOCATE_ENTRY(_WindowVectors_text,
.WindowVectors.text); .WindowVectors.text);
#endif
#if XCHAL_EXCM_LEVEL >= 2 #if XCHAL_EXCM_LEVEL >= 2
RELOCATE_ENTRY(_Level2InterruptVector_text, RELOCATE_ENTRY(_Level2InterruptVector_text,
.Level2InterruptVector.text); .Level2InterruptVector.text);
@ -229,14 +233,18 @@ SECTIONS
#if !MERGED_VECTORS #if !MERGED_VECTORS
/* The vectors are relocated to the real position at startup time */ /* The vectors are relocated to the real position at startup time */
#ifdef SUPPORT_WINDOWED
SECTION_VECTOR4 (_WindowVectors_text, SECTION_VECTOR4 (_WindowVectors_text,
.WindowVectors.text, .WindowVectors.text,
WINDOW_VECTORS_VADDR, WINDOW_VECTORS_VADDR,
.dummy) LAST)
#undef LAST
#define LAST .WindowVectors.text
#endif
SECTION_VECTOR4 (_DebugInterruptVector_text, SECTION_VECTOR4 (_DebugInterruptVector_text,
.DebugInterruptVector.text, .DebugInterruptVector.text,
DEBUG_VECTOR_VADDR, DEBUG_VECTOR_VADDR,
.WindowVectors.text) LAST)
#undef LAST #undef LAST
#define LAST .DebugInterruptVector.text #define LAST .DebugInterruptVector.text
#if XCHAL_EXCM_LEVEL >= 2 #if XCHAL_EXCM_LEVEL >= 2

17
arch/xtensa/lib/strncpy_user.S
View File

@ -45,7 +45,6 @@
# a9/ tmp # a9/ tmp
# a10/ tmp # a10/ tmp
# a11/ dst # a11/ dst
# a12/ tmp
.text .text
ENTRY(__strncpy_user) ENTRY(__strncpy_user)
@ -61,7 +60,7 @@ ENTRY(__strncpy_user)
bbsi.l a3, 0, .Lsrc1mod2 # if only 8-bit aligned bbsi.l a3, 0, .Lsrc1mod2 # if only 8-bit aligned
bbsi.l a3, 1, .Lsrc2mod4 # if only 16-bit aligned bbsi.l a3, 1, .Lsrc2mod4 # if only 16-bit aligned
.Lsrcaligned: # return here when src is word-aligned .Lsrcaligned: # return here when src is word-aligned
srli a12, a4, 2 # number of loop iterations with 4B per loop srli a10, a4, 2 # number of loop iterations with 4B per loop
movi a9, 3 movi a9, 3
bnone a11, a9, .Laligned bnone a11, a9, .Laligned
j .Ldstunaligned j .Ldstunaligned
@ -102,11 +101,11 @@ EX(10f) s8i a9, a11, 0 # store byte 0
.byte 0 # (0 mod 4 alignment for LBEG) .byte 0 # (0 mod 4 alignment for LBEG)
.Laligned: .Laligned:
#if XCHAL_HAVE_LOOPS #if XCHAL_HAVE_LOOPS
loopnez a12, .Loop1done loopnez a10, .Loop1done
#else #else
beqz a12, .Loop1done beqz a10, .Loop1done
slli a12, a12, 2 slli a10, a10, 2
add a12, a12, a11 # a12 = end of last 4B chunck add a10, a10, a11 # a10 = end of last 4B chunck
#endif #endif
.Loop1: .Loop1:
EX(11f) l32i a9, a3, 0 # get word from src EX(11f) l32i a9, a3, 0 # get word from src
@ -118,7 +117,7 @@ EX(10f) s32i a9, a11, 0 # store word to dst
bnone a9, a8, .Lz3 # if byte 3 is zero bnone a9, a8, .Lz3 # if byte 3 is zero
addi a11, a11, 4 # advance dst pointer addi a11, a11, 4 # advance dst pointer
#if !XCHAL_HAVE_LOOPS #if !XCHAL_HAVE_LOOPS
blt a11, a12, .Loop1 blt a11, a10, .Loop1
#endif #endif
.Loop1done: .Loop1done:
@ -185,7 +184,7 @@ EX(10f) s8i a9, a11, 2
loopnez a4, .Lunalignedend loopnez a4, .Lunalignedend
#else #else
beqz a4, .Lunalignedend beqz a4, .Lunalignedend
add a12, a11, a4 # a12 = ending address add a10, a11, a4 # a10 = ending address
#endif /* XCHAL_HAVE_LOOPS */ #endif /* XCHAL_HAVE_LOOPS */
.Lnextbyte: .Lnextbyte:
EX(11f) l8ui a9, a3, 0 EX(11f) l8ui a9, a3, 0
@ -194,7 +193,7 @@ EX(10f) s8i a9, a11, 0
beqz a9, .Lunalignedend beqz a9, .Lunalignedend
addi a11, a11, 1 addi a11, a11, 1
#if !XCHAL_HAVE_LOOPS #if !XCHAL_HAVE_LOOPS
blt a11, a12, .Lnextbyte blt a11, a10, .Lnextbyte
#endif #endif
.Lunalignedend: .Lunalignedend:

28
arch/xtensa/lib/usercopy.S
View File

@ -60,7 +60,12 @@
.text .text
ENTRY(__xtensa_copy_user) ENTRY(__xtensa_copy_user)
abi_entry_default #if !XCHAL_HAVE_LOOPS && defined(__XTENSA_CALL0_ABI__)
#define STACK_SIZE 4
#else
#define STACK_SIZE 0
#endif
abi_entry(STACK_SIZE)
# a2/ dst, a3/ src, a4/ len # a2/ dst, a3/ src, a4/ len
mov a5, a2 # copy dst so that a2 is return value mov a5, a2 # copy dst so that a2 is return value
mov a11, a4 # preserve original len for error case mov a11, a4 # preserve original len for error case
@ -75,7 +80,7 @@ ENTRY(__xtensa_copy_user)
__ssa8 a3 # set shift amount from byte offset __ssa8 a3 # set shift amount from byte offset
bnez a4, .Lsrcunaligned bnez a4, .Lsrcunaligned
movi a2, 0 # return success for len==0 movi a2, 0 # return success for len==0
abi_ret_default abi_ret(STACK_SIZE)
/* /*
* Destination is unaligned * Destination is unaligned
@ -127,7 +132,7 @@ EX(10f) s8i a6, a5, 0
#endif /* !XCHAL_HAVE_LOOPS */ #endif /* !XCHAL_HAVE_LOOPS */
.Lbytecopydone: .Lbytecopydone:
movi a2, 0 # return success for len bytes copied movi a2, 0 # return success for len bytes copied
abi_ret_default abi_ret(STACK_SIZE)
/* /*
* Destination and source are word-aligned. * Destination and source are word-aligned.
@ -187,7 +192,7 @@ EX(10f) l8ui a6, a3, 0
EX(10f) s8i a6, a5, 0 EX(10f) s8i a6, a5, 0
.L5: .L5:
movi a2, 0 # return success for len bytes copied movi a2, 0 # return success for len bytes copied
abi_ret_default abi_ret(STACK_SIZE)
/* /*
* Destination is aligned, Source is unaligned * Destination is aligned, Source is unaligned
@ -205,8 +210,14 @@ EX(10f) l32i a6, a3, 0 # load first word
loopnez a7, .Loop2done loopnez a7, .Loop2done
#else /* !XCHAL_HAVE_LOOPS */ #else /* !XCHAL_HAVE_LOOPS */
beqz a7, .Loop2done beqz a7, .Loop2done
#if defined(__XTENSA_CALL0_ABI__)
s32i a10, a1, 0
slli a10, a7, 4
add a10, a10, a3 # a10 = end of last 16B source chunk
#else
slli a12, a7, 4 slli a12, a7, 4
add a12, a12, a3 # a12 = end of last 16B source chunk add a12, a12, a3 # a12 = end of last 16B source chunk
#endif
#endif /* !XCHAL_HAVE_LOOPS */ #endif /* !XCHAL_HAVE_LOOPS */
.Loop2: .Loop2:
EX(10f) l32i a7, a3, 4 EX(10f) l32i a7, a3, 4
@ -224,7 +235,12 @@ EX(10f) s32i a8, a5, 8
EX(10f) s32i a9, a5, 12 EX(10f) s32i a9, a5, 12
addi a5, a5, 16 addi a5, a5, 16
#if !XCHAL_HAVE_LOOPS #if !XCHAL_HAVE_LOOPS
#if defined(__XTENSA_CALL0_ABI__)
blt a3, a10, .Loop2
l32i a10, a1, 0
#else
blt a3, a12, .Loop2 blt a3, a12, .Loop2
#endif
#endif /* !XCHAL_HAVE_LOOPS */ #endif /* !XCHAL_HAVE_LOOPS */
.Loop2done: .Loop2done:
bbci.l a4, 3, .L12 bbci.l a4, 3, .L12
@ -264,7 +280,7 @@ EX(10f) l8ui a6, a3, 0
EX(10f) s8i a6, a5, 0 EX(10f) s8i a6, a5, 0
.L15: .L15:
movi a2, 0 # return success for len bytes copied movi a2, 0 # return success for len bytes copied
abi_ret_default abi_ret(STACK_SIZE)
ENDPROC(__xtensa_copy_user) ENDPROC(__xtensa_copy_user)
@ -281,4 +297,4 @@ ENDPROC(__xtensa_copy_user)
10: 10:
sub a2, a5, a2 /* a2 <-- bytes copied */ sub a2, a5, a2 /* a2 <-- bytes copied */
sub a2, a11, a2 /* a2 <-- bytes not copied */ sub a2, a11, a2 /* a2 <-- bytes not copied */
abi_ret_default abi_ret(STACK_SIZE)