selftests/powerpc: Add tests of PMU EBBs

The Power8 Performance Monitor Unit (PMU) has a new feature called Event
Based Branches (EBB). This commit adds tests of the kernel API for using
EBBs.

Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

1 files changed, 365 insertions, 0 deletions

diff --git a/tools/testing/selftests/powerpc/pmu/ebb/ebb_handler.S b/tools/testing/selftests/powerpc/pmu/ebb/ebb_handler.S
new file mode 100644
index 000000000000..14274ea206e5
--- /dev/null
+++ b/tools/testing/selftests/powerpc/pmu/ebb/ebb_handler.S
@@ -0,0 +1,365 @@
+/*
+ * Copyright 2014, Michael Ellerman, IBM Corp.
+ * Licensed under GPLv2.
+ */
+
+#include <ppc-asm.h>
+#include "reg.h"
+
+
+/* ppc-asm.h defines most of the reg aliases, but not r1/r2. */
+#define r1 1
+#define r2 2
+
+#define RFEBB   .long 0x4c000924
+
+/* Stack layout:
+ *
+ *                   ^
+ *  User stack       |
+ *  Back chain ------+	<- r1		<-------+
+ *  ...						|
+ *  Red zone / ABI Gap				|
+ *  ...						|
+ *  vr63	<+				|
+ *  vr0		 |				|
+ *  VSCR	 |				|
+ *  FSCR	 |				|
+ *  r31		 | Save area			|
+ *  r0		 |				|
+ *  XER		 |				|
+ *  CTR		 |				|
+ *  LR		 |				|
+ *  CCR		<+				|
+ *  ...		<+				|
+ *  LR		 | Caller frame			|
+ *  CCR		 |				|
+ *  Back chain	<+	<- updated r1	--------+
+ *
+ */
+
+#if defined(_CALL_ELF) && _CALL_ELF == 2
+#define ABIGAP		512
+#else
+#define ABIGAP		288
+#endif
+
+#define NR_GPR		32
+#define NR_SPR		6
+#define NR_VSR		64
+
+#define SAVE_AREA	((NR_GPR + NR_SPR) * 8 + (NR_VSR * 16))
+#define CALLER_FRAME	112
+
+#define STACK_FRAME	(ABIGAP + SAVE_AREA + CALLER_FRAME)
+
+#define CCR_SAVE	(CALLER_FRAME)
+#define LR_SAVE		(CCR_SAVE + 8)
+#define CTR_SAVE	(LR_SAVE  + 8)
+#define XER_SAVE	(CTR_SAVE + 8)
+#define GPR_SAVE(n)	(XER_SAVE + 8 + (8 * n))
+#define FSCR_SAVE	(GPR_SAVE(31) + 8)
+#define VSCR_SAVE	(FSCR_SAVE + 8)
+#define VSR_SAVE(n)	(VSCR_SAVE + 8 + (16 * n))
+
+#define SAVE_GPR(n)	std n,GPR_SAVE(n)(r1)
+#define REST_GPR(n)	ld  n,GPR_SAVE(n)(r1)
+#define TRASH_GPR(n)	lis n,0xaaaa
+
+#define SAVE_VSR(n, b)	li b, VSR_SAVE(n); stxvd2x n,b,r1
+#define LOAD_VSR(n, b)	li b, VSR_SAVE(n); lxvd2x  n,b,r1
+
+#define LOAD_REG_IMMEDIATE(reg,expr)	\
+	lis     reg,(expr)@highest;	\
+	ori     reg,reg,(expr)@higher;	\
+	rldicr  reg,reg,32,31;		\
+	oris    reg,reg,(expr)@h;	\
+	ori     reg,reg,(expr)@l;
+
+
+#if defined(_CALL_ELF) && _CALL_ELF == 2
+#define ENTRY_POINT(name) \
+	.type FUNC_NAME(name),@function; \
+	.globl FUNC_NAME(name); \
+	FUNC_NAME(name):
+
+#define RESTORE_TOC(name)	\
+	/* Restore our TOC pointer using our entry point */	\
+	LOAD_REG_IMMEDIATE(r12, name)				\
+0:	addis	r2,r12,(.TOC.-0b)@ha;				\
+	addi	r2,r2,(.TOC.-0b)@l;
+
+#else
+#define ENTRY_POINT(name) FUNC_START(name)
+#define RESTORE_TOC(name)	\
+	/* Restore our TOC pointer via our opd entry */	\
+	LOAD_REG_IMMEDIATE(r2, name)			\
+	ld      r2,8(r2);
+#endif
+
+    .text
+
+ENTRY_POINT(ebb_handler)
+    stdu    r1,-STACK_FRAME(r1)
+    SAVE_GPR(0)
+    mflr    r0
+    std     r0,LR_SAVE(r1)
+    mfcr    r0
+    std     r0,CCR_SAVE(r1)
+    mfctr   r0
+    std     r0,CTR_SAVE(r1)
+    mfxer   r0
+    std     r0,XER_SAVE(r1)
+    SAVE_GPR(2)
+    SAVE_GPR(3)
+    SAVE_GPR(4)
+    SAVE_GPR(5)
+    SAVE_GPR(6)
+    SAVE_GPR(7)
+    SAVE_GPR(8)
+    SAVE_GPR(9)
+    SAVE_GPR(10)
+    SAVE_GPR(11)
+    SAVE_GPR(12)
+    SAVE_GPR(13)
+    SAVE_GPR(14)
+    SAVE_GPR(15)
+    SAVE_GPR(16)
+    SAVE_GPR(17)
+    SAVE_GPR(18)
+    SAVE_GPR(19)
+    SAVE_GPR(20)
+    SAVE_GPR(21)
+    SAVE_GPR(22)
+    SAVE_GPR(23)
+    SAVE_GPR(24)
+    SAVE_GPR(25)
+    SAVE_GPR(26)
+    SAVE_GPR(27)
+    SAVE_GPR(28)
+    SAVE_GPR(29)
+    SAVE_GPR(30)
+    SAVE_GPR(31)
+    SAVE_VSR(0, r3)
+    mffs     f0
+    stfd     f0, FSCR_SAVE(r1)
+    mfvscr   f0
+    stfd     f0, VSCR_SAVE(r1)
+    SAVE_VSR(1,  r3)
+    SAVE_VSR(2,  r3)
+    SAVE_VSR(3,  r3)
+    SAVE_VSR(4,  r3)
+    SAVE_VSR(5,  r3)
+    SAVE_VSR(6,  r3)
+    SAVE_VSR(7,  r3)
+    SAVE_VSR(8,  r3)
+    SAVE_VSR(9,  r3)
+    SAVE_VSR(10, r3)
+    SAVE_VSR(11, r3)
+    SAVE_VSR(12, r3)
+    SAVE_VSR(13, r3)
+    SAVE_VSR(14, r3)
+    SAVE_VSR(15, r3)
+    SAVE_VSR(16, r3)
+    SAVE_VSR(17, r3)
+    SAVE_VSR(18, r3)
+    SAVE_VSR(19, r3)
+    SAVE_VSR(20, r3)
+    SAVE_VSR(21, r3)
+    SAVE_VSR(22, r3)
+    SAVE_VSR(23, r3)
+    SAVE_VSR(24, r3)
+    SAVE_VSR(25, r3)
+    SAVE_VSR(26, r3)
+    SAVE_VSR(27, r3)
+    SAVE_VSR(28, r3)
+    SAVE_VSR(29, r3)
+    SAVE_VSR(30, r3)
+    SAVE_VSR(31, r3)
+    SAVE_VSR(32, r3)
+    SAVE_VSR(33, r3)
+    SAVE_VSR(34, r3)
+    SAVE_VSR(35, r3)
+    SAVE_VSR(36, r3)
+    SAVE_VSR(37, r3)
+    SAVE_VSR(38, r3)
+    SAVE_VSR(39, r3)
+    SAVE_VSR(40, r3)
+    SAVE_VSR(41, r3)
+    SAVE_VSR(42, r3)
+    SAVE_VSR(43, r3)
+    SAVE_VSR(44, r3)
+    SAVE_VSR(45, r3)
+    SAVE_VSR(46, r3)
+    SAVE_VSR(47, r3)
+    SAVE_VSR(48, r3)
+    SAVE_VSR(49, r3)
+    SAVE_VSR(50, r3)
+    SAVE_VSR(51, r3)
+    SAVE_VSR(52, r3)
+    SAVE_VSR(53, r3)
+    SAVE_VSR(54, r3)
+    SAVE_VSR(55, r3)
+    SAVE_VSR(56, r3)
+    SAVE_VSR(57, r3)
+    SAVE_VSR(58, r3)
+    SAVE_VSR(59, r3)
+    SAVE_VSR(60, r3)
+    SAVE_VSR(61, r3)
+    SAVE_VSR(62, r3)
+    SAVE_VSR(63, r3)
+
+    TRASH_GPR(2)
+    TRASH_GPR(3)
+    TRASH_GPR(4)
+    TRASH_GPR(5)
+    TRASH_GPR(6)
+    TRASH_GPR(7)
+    TRASH_GPR(8)
+    TRASH_GPR(9)
+    TRASH_GPR(10)
+    TRASH_GPR(11)
+    TRASH_GPR(12)
+    TRASH_GPR(14)
+    TRASH_GPR(15)
+    TRASH_GPR(16)
+    TRASH_GPR(17)
+    TRASH_GPR(18)
+    TRASH_GPR(19)
+    TRASH_GPR(20)
+    TRASH_GPR(21)
+    TRASH_GPR(22)
+    TRASH_GPR(23)
+    TRASH_GPR(24)
+    TRASH_GPR(25)
+    TRASH_GPR(26)
+    TRASH_GPR(27)
+    TRASH_GPR(28)
+    TRASH_GPR(29)
+    TRASH_GPR(30)
+    TRASH_GPR(31)
+
+    RESTORE_TOC(ebb_handler)
+
+    /*
+     * r13 is our TLS pointer. We leave whatever value was in there when the
+     * EBB fired. That seems to be OK because once set the TLS pointer is not
+     * changed - but presumably that could change in future.
+     */
+
+    bl      ebb_hook
+    nop
+
+    /* r2 may be changed here but we don't care */
+
+    lfd      f0, FSCR_SAVE(r1)
+    mtfsf    0xff,f0
+    lfd      f0, VSCR_SAVE(r1)
+    mtvscr   f0
+    LOAD_VSR(0, r3)
+    LOAD_VSR(1,  r3)
+    LOAD_VSR(2,  r3)
+    LOAD_VSR(3,  r3)
+    LOAD_VSR(4,  r3)
+    LOAD_VSR(5,  r3)
+    LOAD_VSR(6,  r3)
+    LOAD_VSR(7,  r3)
+    LOAD_VSR(8,  r3)
+    LOAD_VSR(9,  r3)
+    LOAD_VSR(10, r3)
+    LOAD_VSR(11, r3)
+    LOAD_VSR(12, r3)
+    LOAD_VSR(13, r3)
+    LOAD_VSR(14, r3)
+    LOAD_VSR(15, r3)
+    LOAD_VSR(16, r3)
+    LOAD_VSR(17, r3)
+    LOAD_VSR(18, r3)
+    LOAD_VSR(19, r3)
+    LOAD_VSR(20, r3)
+    LOAD_VSR(21, r3)
+    LOAD_VSR(22, r3)
+    LOAD_VSR(23, r3)
+    LOAD_VSR(24, r3)
+    LOAD_VSR(25, r3)
+    LOAD_VSR(26, r3)
+    LOAD_VSR(27, r3)
+    LOAD_VSR(28, r3)
+    LOAD_VSR(29, r3)
+    LOAD_VSR(30, r3)
+    LOAD_VSR(31, r3)
+    LOAD_VSR(32, r3)
+    LOAD_VSR(33, r3)
+    LOAD_VSR(34, r3)
+    LOAD_VSR(35, r3)
+    LOAD_VSR(36, r3)
+    LOAD_VSR(37, r3)
+    LOAD_VSR(38, r3)
+    LOAD_VSR(39, r3)
+    LOAD_VSR(40, r3)
+    LOAD_VSR(41, r3)
+    LOAD_VSR(42, r3)
+    LOAD_VSR(43, r3)
+    LOAD_VSR(44, r3)
+    LOAD_VSR(45, r3)
+    LOAD_VSR(46, r3)
+    LOAD_VSR(47, r3)
+    LOAD_VSR(48, r3)
+    LOAD_VSR(49, r3)
+    LOAD_VSR(50, r3)
+    LOAD_VSR(51, r3)
+    LOAD_VSR(52, r3)
+    LOAD_VSR(53, r3)
+    LOAD_VSR(54, r3)
+    LOAD_VSR(55, r3)
+    LOAD_VSR(56, r3)
+    LOAD_VSR(57, r3)
+    LOAD_VSR(58, r3)
+    LOAD_VSR(59, r3)
+    LOAD_VSR(60, r3)
+    LOAD_VSR(61, r3)
+    LOAD_VSR(62, r3)
+    LOAD_VSR(63, r3)
+
+    ld      r0,XER_SAVE(r1)
+    mtxer   r0
+    ld      r0,CTR_SAVE(r1)
+    mtctr   r0
+    ld      r0,LR_SAVE(r1)
+    mtlr    r0
+    ld      r0,CCR_SAVE(r1)
+    mtcr    r0
+    REST_GPR(0)
+    REST_GPR(2)
+    REST_GPR(3)
+    REST_GPR(4)
+    REST_GPR(5)
+    REST_GPR(6)
+    REST_GPR(7)
+    REST_GPR(8)
+    REST_GPR(9)
+    REST_GPR(10)
+    REST_GPR(11)
+    REST_GPR(12)
+    REST_GPR(13)
+    REST_GPR(14)
+    REST_GPR(15)
+    REST_GPR(16)
+    REST_GPR(17)
+    REST_GPR(18)
+    REST_GPR(19)
+    REST_GPR(20)
+    REST_GPR(21)
+    REST_GPR(22)
+    REST_GPR(23)
+    REST_GPR(24)
+    REST_GPR(25)
+    REST_GPR(26)
+    REST_GPR(27)
+    REST_GPR(28)
+    REST_GPR(29)
+    REST_GPR(30)
+    REST_GPR(31)
+    addi    r1,r1,STACK_FRAME
+    RFEBB
+FUNC_END(ebb_handler)