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

7 files changed, 166 insertions, 27 deletions

diff --git a/arch/xtensa/include/asm/asmmacro.h b/arch/xtensa/include/asm/asmmacro.h
index bfc89e11f469..809c507d1825 100644
--- a/arch/xtensa/include/asm/asmmacro.h
+++ b/arch/xtensa/include/asm/asmmacro.h
@@ -194,6 +194,12 @@
 #define XTENSA_STACK_ALIGNMENT		16
 
 #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_SPILL_STACK_RESERVE	32
 
@@ -206,8 +212,34 @@
 #define abi_ret(frame_size) 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__)
 
+/* 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 abi_entry(frame_size) __abi_entry (frame_size)
@@ -233,10 +265,43 @@
 
 #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
 #error Unsupported Xtensa ABI
 #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"
 
 #endif /* _XTENSA_ASMMACRO_H */
diff --git a/arch/xtensa/include/asm/atomic.h b/arch/xtensa/include/asm/atomic.h
index 4361fe4247e3..52da614f953c 100644
--- a/arch/xtensa/include/asm/atomic.h
+++ b/arch/xtensa/include/asm/atomic.h
@@ -25,15 +25,15 @@
  *
  * Locking interrupts looks like this:
  *
- *    rsil a15, TOPLEVEL
+ *    rsil a14, TOPLEVEL
  *    <code>
- *    wsr  a15, PS
+ *    wsr  a14, PS
  *    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
  * 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.
  */
 
@@ -185,15 +185,15 @@ static inline void arch_atomic_##op(int i, atomic_t * v)		\
 	unsigned int vval;						\
 									\
 	__asm__ __volatile__(						\
-			"       rsil    a15, "__stringify(TOPLEVEL)"\n"	\
+			"       rsil    a14, "__stringify(TOPLEVEL)"\n"	\
 			"       l32i    %[result], %[mem]\n"		\
 			"       " #op " %[result], %[result], %[i]\n"	\
 			"       s32i    %[result], %[mem]\n"		\
-			"       wsr     a15, ps\n"			\
+			"       wsr     a14, ps\n"			\
 			"       rsync\n"				\
 			: [result] "=&a" (vval), [mem] "+m" (*v)	\
 			: [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;						\
 									\
 	__asm__ __volatile__(						\
-			"       rsil    a15,"__stringify(TOPLEVEL)"\n"	\
+			"       rsil    a14,"__stringify(TOPLEVEL)"\n"	\
 			"       l32i    %[result], %[mem]\n"		\
 			"       " #op " %[result], %[result], %[i]\n"	\
 			"       s32i    %[result], %[mem]\n"		\
-			"       wsr     a15, ps\n"			\
+			"       wsr     a14, ps\n"			\
 			"       rsync\n"				\
 			: [result] "=&a" (vval), [mem] "+m" (*v)	\
 			: [i] "a" (i)					\
-			: "a15", "memory"				\
+			: "a14", "memory"				\
 			);						\
 									\
 	return vval;							\
@@ -223,16 +223,16 @@ static inline int arch_atomic_fetch_##op(int i, atomic_t * v)		\
 	unsigned int tmp, vval;						\
 									\
 	__asm__ __volatile__(						\
-			"       rsil    a15,"__stringify(TOPLEVEL)"\n"	\
+			"       rsil    a14,"__stringify(TOPLEVEL)"\n"	\
 			"       l32i    %[result], %[mem]\n"		\
 			"       " #op " %[tmp], %[result], %[i]\n"	\
 			"       s32i    %[tmp], %[mem]\n"		\
-			"       wsr     a15, ps\n"			\
+			"       wsr     a14, ps\n"			\
 			"       rsync\n"				\
 			: [result] "=&a" (vval), [tmp] "=&a" (tmp),	\
 			  [mem] "+m" (*v)				\
 			: [i] "a" (i)					\
-			: "a15", "memory"				\
+			: "a14", "memory"				\
 			);						\
 									\
 	return vval;							\
diff --git a/arch/xtensa/include/asm/cmpxchg.h b/arch/xtensa/include/asm/cmpxchg.h
index 3699e2818efb..eb87810357ad 100644
--- a/arch/xtensa/include/asm/cmpxchg.h
+++ b/arch/xtensa/include/asm/cmpxchg.h
@@ -52,16 +52,16 @@ __cmpxchg_u32(volatile int *p, int old, int new)
 	return new;
 #else
 	__asm__ __volatile__(
-			"       rsil    a15, "__stringify(TOPLEVEL)"\n"
+			"       rsil    a14, "__stringify(TOPLEVEL)"\n"
 			"       l32i    %[old], %[mem]\n"
 			"       bne     %[old], %[cmp], 1f\n"
 			"       s32i    %[new], %[mem]\n"
 			"1:\n"
-			"       wsr     a15, ps\n"
+			"       wsr     a14, ps\n"
 			"       rsync\n"
 			: [old] "=&a" (old), [mem] "+m" (*p)
 			: [cmp] "a" (old), [new] "r" (new)
-			: "a15", "memory");
+			: "a14", "memory");
 	return old;
 #endif
 }
@@ -116,10 +116,10 @@ static inline unsigned long __cmpxchg_local(volatile void *ptr,
 /*
  * 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
  * 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.
  */
 
@@ -157,14 +157,14 @@ static inline unsigned long xchg_u32(volatile int * m, unsigned long val)
 #else
 	unsigned long tmp;
 	__asm__ __volatile__(
-			"       rsil    a15, "__stringify(TOPLEVEL)"\n"
+			"       rsil    a14, "__stringify(TOPLEVEL)"\n"
 			"       l32i    %[tmp], %[mem]\n"
 			"       s32i    %[val], %[mem]\n"
-			"       wsr     a15, ps\n"
+			"       wsr     a14, ps\n"
 			"       rsync\n"
 			: [tmp] "=&a" (tmp), [mem] "+m" (*m)
 			: [val] "a" (val)
-			: "a15", "memory");
+			: "a14", "memory");
 	return tmp;
 #endif
 }
diff --git a/arch/xtensa/include/asm/core.h b/arch/xtensa/include/asm/core.h
index 5590b0f68837..9138077e567d 100644
--- a/arch/xtensa/include/asm/core.h
+++ b/arch/xtensa/include/asm/core.h
@@ -26,4 +26,15 @@
 #define XCHAL_SPANNING_WAY 0
 #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
diff --git a/arch/xtensa/include/asm/processor.h b/arch/xtensa/include/asm/processor.h
index ad15fbc57283..37d3e9887fe7 100644
--- a/arch/xtensa/include/asm/processor.h
+++ b/arch/xtensa/include/asm/processor.h
@@ -18,12 +18,6 @@
 #include <asm/types.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 */
 
 #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 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__
 
+#if defined(__XTENSA_WINDOWED_ABI__)
+
 /* Build a valid return address for the specified call winsize.
  * 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))
 
+#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
  * pointer sp. reg must be in the range [0..4).
  */
diff --git a/arch/xtensa/include/asm/sections.h b/arch/xtensa/include/asm/sections.h
new file mode 100644
index 000000000000..a8c42d08e281
--- /dev/null
+++ b/arch/xtensa/include/asm/sections.h
@@ -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
diff --git a/arch/xtensa/include/asm/traps.h b/arch/xtensa/include/asm/traps.h
index f720a57d0a5b..6fa47cd8e02d 100644
--- a/arch/xtensa/include/asm/traps.h
+++ b/arch/xtensa/include/asm/traps.h
@@ -56,6 +56,7 @@ void secondary_trap_init(void);
 
 static inline void spill_registers(void)
 {
+#if defined(__XTENSA_WINDOWED_ABI__)
 #if XCHAL_NUM_AREGS > 16
 	__asm__ __volatile__ (
 		"	call8	1f\n"
@@ -96,6 +97,7 @@ static inline void spill_registers(void)
 		"	mov	a12, a12\n"
 		: : : "memory");
 #endif
+#endif
 }
 
 struct debug_table {