xtensa: fixes for configs without loop option

Build-time fixes:
- make lbeg/lend/lcount save/restore conditional on kernel entry;
- don't clear lcount in platform_restart functions unconditionally.

Run-time fixes:
- use correct end of range register in __endla paired with __loopt, not
  the unused temporary register. This fixes .bss zero-initialization.
  Update comments in asmmacro.h;
- don't clobber a10 in the usercopy that leads to access to unmapped
  memory.

Cc: <stable@vger.kernel.org>
Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>

arch/xtensa/include/asm/asmmacro.h

@@ -35,9 +35,10 @@
  * __loop  as
  *	   restart loop. 'as' register must not have been modified!
  *
- * __endla ar, at, incr
+ * __endla ar, as, incr
  *	   ar	start address (modified)
- *	   as	scratch register used by macro
+ *	   as	scratch register used by __loops/__loopi macros or
+ *		end address used by __loopt macro
  *	   inc	increment
  */
 
@@ -97,7 +98,7 @@
 	.endm
 
 /*
- * loop from ar to ax
+ * loop from ar to as
  */
 
 	.macro	__loopt	ar, as, at, incr_log2

arch/xtensa/kernel/entry.S

@@ -367,8 +367,10 @@ common_exception:
 	s32i	a2, a1, PT_SYSCALL
 	movi	a2, 0
 	s32i	a3, a1, PT_EXCVADDR
+#if XCHAL_HAVE_LOOPS
 	xsr	a2, lcount
 	s32i	a2, a1, PT_LCOUNT
+#endif
 
 	/* It is now save to restore the EXC_TABLE_FIXUP variable. */
 
@@ -429,11 +431,12 @@ common_exception:
 	rsync				# PS.WOE => rsync => overflow
 
 	/* Save lbeg, lend */
-
+#if XCHAL_HAVE_LOOPS
 	rsr	a4, lbeg
 	rsr	a3, lend
 	s32i	a4, a1, PT_LBEG
 	s32i	a3, a1, PT_LEND
+#endif
 
 	/* Save SCOMPARE1 */
 
@@ -724,13 +727,14 @@ common_exception_exit:
 	wsr	a3, sar
 
 	/* Restore LBEG, LEND, LCOUNT */
-
+#if XCHAL_HAVE_LOOPS
 	l32i	a2, a1, PT_LBEG
 	l32i	a3, a1, PT_LEND
 	wsr	a2, lbeg
 	l32i	a2, a1, PT_LCOUNT
 	wsr	a3, lend
 	wsr	a2, lcount
+#endif
 
 	/* We control single stepping through the ICOUNTLEVEL register. */
 

arch/xtensa/kernel/head.S

@@ -249,7 +249,7 @@ ENTRY(_startup)
 
 	__loopt	a2, a3, a4, 2
 	s32i	a0, a2, 0
-	__endla	a2, a4, 4
+	__endla	a2, a3, 4
 
 #if XCHAL_DCACHE_IS_WRITEBACK
 

arch/xtensa/lib/usercopy.S

@@ -222,8 +222,8 @@ __xtensa_copy_user:
 	loopnez	a7, .Loop2done
 #else /* !XCHAL_HAVE_LOOPS */
 	beqz	a7, .Loop2done
-	slli	a10, a7, 4
+	slli	a12, a7, 4
-	add	a10, a10, a3	# a10 = end of last 16B source chunk
+	add	a12, a12, a3	# a12 = end of last 16B source chunk
 #endif /* !XCHAL_HAVE_LOOPS */
 .Loop2:
 	EX(l32i, a7, a3,  4, l_fixup)
@@ -241,7 +241,7 @@ __xtensa_copy_user:
 	EX(s32i, a9, a5, 12, s_fixup)
 	addi	a5, a5, 16
 #if !XCHAL_HAVE_LOOPS
-	blt	a3, a10, .Loop2
+	blt	a3, a12, .Loop2
 #endif /* !XCHAL_HAVE_LOOPS */
 .Loop2done:
 	bbci.l	a4, 3, .L12

arch/xtensa/platforms/iss/setup.c

@@ -61,7 +61,9 @@ void platform_restart(void)
 #if XCHAL_NUM_IBREAK > 0
 			     "wsr	a2, ibreakenable\n\t"
 #endif
+#if XCHAL_HAVE_LOOPS
 			     "wsr	a2, lcount\n\t"
+#endif
 			     "movi	a2, 0x1f\n\t"
 			     "wsr	a2, ps\n\t"
 			     "isync\n\t"

arch/xtensa/platforms/xt2000/setup.c

@@ -72,7 +72,9 @@ void platform_restart(void)
 #if XCHAL_NUM_IBREAK > 0
 			      "wsr	a2, ibreakenable\n\t"
 #endif
+#if XCHAL_HAVE_LOOPS
 			      "wsr	a2, lcount\n\t"
+#endif
 			      "movi	a2, 0x1f\n\t"
 			      "wsr	a2, ps\n\t"
 			      "isync\n\t"

arch/xtensa/platforms/xtfpga/setup.c

@@ -63,7 +63,9 @@ void platform_restart(void)
 #if XCHAL_NUM_IBREAK > 0
 			      "wsr	a2, ibreakenable\n\t"
 #endif
+#if XCHAL_HAVE_LOOPS
 			      "wsr	a2, lcount\n\t"
+#endif
 			      "movi	a2, 0x1f\n\t"
 			      "wsr	a2, ps\n\t"
 			      "isync\n\t"