powerpc/kprobes: Implement Optprobes

Current infrastructure of kprobe uses the unconditional trap instruction
to probe a running kernel. Optprobe allows kprobe to replace the trap
with a branch instruction to a detour buffer. Detour buffer contains
instructions to create an in memory pt_regs. Detour buffer also has a
call to optimized_callback() which in turn call the pre_handler(). After
the execution of the pre-handler, a call is made for instruction
emulation. The NIP is determined in advanced through dummy instruction
emulation and a branch instruction is created to the NIP at the end of
the trampoline.

To address the limitation of branch instruction in POWER architecture,
detour buffer slot is allocated from a reserved area. For the time
being, 64KB is reserved in memory for this purpose.

Instructions which can be emulated using analyse_instr() are the
candidates for optimization. Before optimization ensure that the address
range between the detour buffer allocated and the instruction being
probed is within +/- 32MB.

Signed-off-by: Anju T Sudhakar <anju@linux.vnet.ibm.com>
Signed-off-by: Naveen N. Rao <naveen.n.rao@linux.vnet.ibm.com>
Acked-by: Masami Hiramatsu <mhiramat@kernel.org>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>

arch/powerpc/Kconfig

@@ -100,6 +100,7 @@ config PPC
 	select HAVE_IOREMAP_PROT
 	select HAVE_EFFICIENT_UNALIGNED_ACCESS if !(CPU_LITTLE_ENDIAN && POWER7_CPU)
 	select HAVE_KPROBES
+	select HAVE_OPTPROBES if PPC64
 	select HAVE_ARCH_KGDB
 	select HAVE_KRETPROBES
 	select HAVE_ARCH_TRACEHOOK

arch/powerpc/include/asm/code-patching.h

@@ -35,6 +35,7 @@ int instr_is_branch_to_addr(const unsigned int *instr, unsigned long addr);
 unsigned long branch_target(const unsigned int *instr);
 unsigned int translate_branch(const unsigned int *dest,
 			      const unsigned int *src);
+extern bool is_conditional_branch(unsigned int instr);
 #ifdef CONFIG_PPC_BOOK3E_64
 void __patch_exception(int exc, unsigned long addr);
 #define patch_exception(exc, name) do { \

arch/powerpc/include/asm/kprobes.h

@@ -40,7 +40,23 @@ struct pt_regs;
 struct kprobe;
 
 typedef ppc_opcode_t kprobe_opcode_t;
-#define MAX_INSN_SIZE 1
+
+extern kprobe_opcode_t optinsn_slot;
+
+/* Optinsn template address */
+extern kprobe_opcode_t optprobe_template_entry[];
+extern kprobe_opcode_t optprobe_template_op_address[];
+extern kprobe_opcode_t optprobe_template_call_handler[];
+extern kprobe_opcode_t optprobe_template_insn[];
+extern kprobe_opcode_t optprobe_template_call_emulate[];
+extern kprobe_opcode_t optprobe_template_ret[];
+extern kprobe_opcode_t optprobe_template_end[];
+
+/* Fixed instruction size for powerpc */
+#define MAX_INSN_SIZE		1
+#define MAX_OPTIMIZED_LENGTH	sizeof(kprobe_opcode_t)	/* 4 bytes */
+#define MAX_OPTINSN_SIZE	(optprobe_template_end - optprobe_template_entry)
+#define RELATIVEJUMP_SIZE	sizeof(kprobe_opcode_t)	/* 4 bytes */
 
 #ifdef PPC64_ELF_ABI_v2
 /* PPC64 ABIv2 needs local entry point */
@@ -126,6 +142,12 @@ struct kprobe_ctlblk {
 	struct prev_kprobe prev_kprobe;
 };
 
+struct arch_optimized_insn {
+	kprobe_opcode_t copied_insn[1];
+	/* detour buffer */
+	kprobe_opcode_t *insn;
+};
+
 extern int kprobe_exceptions_notify(struct notifier_block *self,
 					unsigned long val, void *data);
 extern int kprobe_fault_handler(struct pt_regs *regs, int trapnr);

arch/powerpc/kernel/Makefile

@@ -100,6 +100,7 @@ obj-$(CONFIG_KGDB)		+= kgdb.o
 obj-$(CONFIG_BOOTX_TEXT)	+= btext.o
 obj-$(CONFIG_SMP)		+= smp.o
 obj-$(CONFIG_KPROBES)		+= kprobes.o
+obj-$(CONFIG_OPTPROBES)		+= optprobes.o optprobes_head.o
 obj-$(CONFIG_UPROBES)		+= uprobes.o
 obj-$(CONFIG_PPC_UDBG_16550)	+= legacy_serial.o udbg_16550.o
 obj-$(CONFIG_STACKTRACE)	+= stacktrace.o

arch/powerpc/kernel/optprobes.c

@@ -0,0 +1,348 @@
+/*
+ * Code for Kernel probes Jump optimization.
+ *
+ * Copyright 2017, Anju T, IBM Corp.
+ *
+ * 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; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/kprobes.h>
+#include <linux/jump_label.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/list.h>
+#include <asm/kprobes.h>
+#include <asm/ptrace.h>
+#include <asm/cacheflush.h>
+#include <asm/code-patching.h>
+#include <asm/sstep.h>
+#include <asm/ppc-opcode.h>
+
+#define TMPL_CALL_HDLR_IDX	\
+	(optprobe_template_call_handler - optprobe_template_entry)
+#define TMPL_EMULATE_IDX	\
+	(optprobe_template_call_emulate - optprobe_template_entry)
+#define TMPL_RET_IDX		\
+	(optprobe_template_ret - optprobe_template_entry)
+#define TMPL_OP_IDX		\
+	(optprobe_template_op_address - optprobe_template_entry)
+#define TMPL_INSN_IDX		\
+	(optprobe_template_insn - optprobe_template_entry)
+#define TMPL_END_IDX		\
+	(optprobe_template_end - optprobe_template_entry)
+
+DEFINE_INSN_CACHE_OPS(ppc_optinsn);
+
+static bool insn_page_in_use;
+
+static void *__ppc_alloc_insn_page(void)
+{
+	if (insn_page_in_use)
+		return NULL;
+	insn_page_in_use = true;
+	return &optinsn_slot;
+}
+
+static void __ppc_free_insn_page(void *page __maybe_unused)
+{
+	insn_page_in_use = false;
+}
+
+struct kprobe_insn_cache kprobe_ppc_optinsn_slots = {
+	.mutex = __MUTEX_INITIALIZER(kprobe_ppc_optinsn_slots.mutex),
+	.pages = LIST_HEAD_INIT(kprobe_ppc_optinsn_slots.pages),
+	/* insn_size initialized later */
+	.alloc = __ppc_alloc_insn_page,
+	.free = __ppc_free_insn_page,
+	.nr_garbage = 0,
+};
+
+/*
+ * Check if we can optimize this probe. Returns NIP post-emulation if this can
+ * be optimized and 0 otherwise.
+ */
+static unsigned long can_optimize(struct kprobe *p)
+{
+	struct pt_regs regs;
+	struct instruction_op op;
+	unsigned long nip = 0;
+
+	/*
+	 * kprobe placed for kretprobe during boot time
+	 * is not optimizing now.
+	 *
+	 * TODO: Optimize kprobe in kretprobe_trampoline
+	 */
+	if (p->addr == (kprobe_opcode_t *)&kretprobe_trampoline)
+		return 0;
+
+	/*
+	 * We only support optimizing kernel addresses, but not
+	 * module addresses.
+	 *
+	 * FIXME: Optimize kprobes placed in module addresses.
+	 */
+	if (!is_kernel_addr((unsigned long)p->addr))
+		return 0;
+
+	memset(&regs, 0, sizeof(struct pt_regs));
+	regs.nip = (unsigned long)p->addr;
+	regs.trap = 0x0;
+	regs.msr = MSR_KERNEL;
+
+	/*
+	 * Kprobe placed in conditional branch instructions are
+	 * not optimized, as we can't predict the nip prior with
+	 * dummy pt_regs and can not ensure that the return branch
+	 * from detour buffer falls in the range of address (i.e 32MB).
+	 * A branch back from trampoline is set up in the detour buffer
+	 * to the nip returned by the analyse_instr() here.
+	 *
+	 * Ensure that the instruction is not a conditional branch,
+	 * and that can be emulated.
+	 */
+	if (!is_conditional_branch(*p->ainsn.insn) &&
+			analyse_instr(&op, &regs, *p->ainsn.insn))
+		nip = regs.nip;
+
+	return nip;
+}
+
+static void optimized_callback(struct optimized_kprobe *op,
+			       struct pt_regs *regs)
+{
+	struct kprobe_ctlblk *kcb = get_kprobe_ctlblk();
+	unsigned long flags;
+
+	/* This is possible if op is under delayed unoptimizing */
+	if (kprobe_disabled(&op->kp))
+		return;
+
+	local_irq_save(flags);
+	hard_irq_disable();
+
+	if (kprobe_running()) {
+		kprobes_inc_nmissed_count(&op->kp);
+	} else {
+		__this_cpu_write(current_kprobe, &op->kp);
+		regs->nip = (unsigned long)op->kp.addr;
+		kcb->kprobe_status = KPROBE_HIT_ACTIVE;
+		opt_pre_handler(&op->kp, regs);
+		__this_cpu_write(current_kprobe, NULL);
+	}
+
+	/*
+	 * No need for an explicit __hard_irq_enable() here.
+	 * local_irq_restore() will re-enable interrupts,
+	 * if they were hard disabled.
+	 */
+	local_irq_restore(flags);
+}
+NOKPROBE_SYMBOL(optimized_callback);
+
+void arch_remove_optimized_kprobe(struct optimized_kprobe *op)
+{
+	if (op->optinsn.insn) {
+		free_ppc_optinsn_slot(op->optinsn.insn, 1);
+		op->optinsn.insn = NULL;
+	}
+}
+
+/*
+ * emulate_step() requires insn to be emulated as
+ * second parameter. Load register 'r4' with the
+ * instruction.
+ */
+void patch_imm32_load_insns(unsigned int val, kprobe_opcode_t *addr)
+{
+	/* addis r4,0,(insn)@h */
+	*addr++ = PPC_INST_ADDIS | ___PPC_RT(4) |
+		  ((val >> 16) & 0xffff);
+
+	/* ori r4,r4,(insn)@l */
+	*addr = PPC_INST_ORI | ___PPC_RA(4) | ___PPC_RS(4) |
+		(val & 0xffff);
+}
+
+/*
+ * Generate instructions to load provided immediate 64-bit value
+ * to register 'r3' and patch these instructions at 'addr'.
+ */
+void patch_imm64_load_insns(unsigned long val, kprobe_opcode_t *addr)
+{
+	/* lis r3,(op)@highest */
+	*addr++ = PPC_INST_ADDIS | ___PPC_RT(3) |
+		  ((val >> 48) & 0xffff);
+
+	/* ori r3,r3,(op)@higher */
+	*addr++ = PPC_INST_ORI | ___PPC_RA(3) | ___PPC_RS(3) |
+		  ((val >> 32) & 0xffff);
+
+	/* rldicr r3,r3,32,31 */
+	*addr++ = PPC_INST_RLDICR | ___PPC_RA(3) | ___PPC_RS(3) |
+		  __PPC_SH64(32) | __PPC_ME64(31);
+
+	/* oris r3,r3,(op)@h */
+	*addr++ = PPC_INST_ORIS | ___PPC_RA(3) | ___PPC_RS(3) |
+		  ((val >> 16) & 0xffff);
+
+	/* ori r3,r3,(op)@l */
+	*addr = PPC_INST_ORI | ___PPC_RA(3) | ___PPC_RS(3) |
+		(val & 0xffff);
+}
+
+int arch_prepare_optimized_kprobe(struct optimized_kprobe *op, struct kprobe *p)
+{
+	kprobe_opcode_t *buff, branch_op_callback, branch_emulate_step;
+	kprobe_opcode_t *op_callback_addr, *emulate_step_addr;
+	long b_offset;
+	unsigned long nip;
+
+	kprobe_ppc_optinsn_slots.insn_size = MAX_OPTINSN_SIZE;
+
+	nip = can_optimize(p);
+	if (!nip)
+		return -EILSEQ;
+
+	/* Allocate instruction slot for detour buffer */
+	buff = get_ppc_optinsn_slot();
+	if (!buff)
+		return -ENOMEM;
+
+	/*
+	 * OPTPROBE uses 'b' instruction to branch to optinsn.insn.
+	 *
+	 * The target address has to be relatively nearby, to permit use
+	 * of branch instruction in powerpc, because the address is specified
+	 * in an immediate field in the instruction opcode itself, ie 24 bits
+	 * in the opcode specify the address. Therefore the address should
+	 * be within 32MB on either side of the current instruction.
+	 */
+	b_offset = (unsigned long)buff - (unsigned long)p->addr;
+	if (!is_offset_in_branch_range(b_offset))
+		goto error;
+
+	/* Check if the return address is also within 32MB range */
+	b_offset = (unsigned long)(buff + TMPL_RET_IDX) -
+			(unsigned long)nip;
+	if (!is_offset_in_branch_range(b_offset))
+		goto error;
+
+	/* Setup template */
+	memcpy(buff, optprobe_template_entry,
+			TMPL_END_IDX * sizeof(kprobe_opcode_t));
+
+	/*
+	 * Fixup the template with instructions to:
+	 * 1. load the address of the actual probepoint
+	 */
+	patch_imm64_load_insns((unsigned long)op, buff + TMPL_OP_IDX);
+
+	/*
+	 * 2. branch to optimized_callback() and emulate_step()
+	 */
+	kprobe_lookup_name("optimized_callback", op_callback_addr);
+	kprobe_lookup_name("emulate_step", emulate_step_addr);
+	if (!op_callback_addr || !emulate_step_addr) {
+		WARN(1, "kprobe_lookup_name() failed\n");
+		goto error;
+	}
+
+	branch_op_callback = create_branch((unsigned int *)buff + TMPL_CALL_HDLR_IDX,
+				(unsigned long)op_callback_addr,
+				BRANCH_SET_LINK);
+
+	branch_emulate_step = create_branch((unsigned int *)buff + TMPL_EMULATE_IDX,
+				(unsigned long)emulate_step_addr,
+				BRANCH_SET_LINK);
+
+	if (!branch_op_callback || !branch_emulate_step)
+		goto error;
+
+	buff[TMPL_CALL_HDLR_IDX] = branch_op_callback;
+	buff[TMPL_EMULATE_IDX] = branch_emulate_step;
+
+	/*
+	 * 3. load instruction to be emulated into relevant register, and
+	 */
+	patch_imm32_load_insns(*p->ainsn.insn, buff + TMPL_INSN_IDX);
+
+	/*
+	 * 4. branch back from trampoline
+	 */
+	buff[TMPL_RET_IDX] = create_branch((unsigned int *)buff + TMPL_RET_IDX,
+				(unsigned long)nip, 0);
+
+	flush_icache_range((unsigned long)buff,
+			   (unsigned long)(&buff[TMPL_END_IDX]));
+
+	op->optinsn.insn = buff;
+
+	return 0;
+
+error:
+	free_ppc_optinsn_slot(buff, 0);
+	return -ERANGE;
+
+}
+
+int arch_prepared_optinsn(struct arch_optimized_insn *optinsn)
+{
+	return optinsn->insn != NULL;
+}
+
+/*
+ * On powerpc, Optprobes always replaces one instruction (4 bytes
+ * aligned and 4 bytes long). It is impossible to encounter another
+ * kprobe in this address range. So always return 0.
+ */
+int arch_check_optimized_kprobe(struct optimized_kprobe *op)
+{
+	return 0;
+}
+
+void arch_optimize_kprobes(struct list_head *oplist)
+{
+	struct optimized_kprobe *op;
+	struct optimized_kprobe *tmp;
+
+	list_for_each_entry_safe(op, tmp, oplist, list) {
+		/*
+		 * Backup instructions which will be replaced
+		 * by jump address
+		 */
+		memcpy(op->optinsn.copied_insn, op->kp.addr,
+					       RELATIVEJUMP_SIZE);
+		patch_instruction(op->kp.addr,
+			create_branch((unsigned int *)op->kp.addr,
+				      (unsigned long)op->optinsn.insn, 0));
+		list_del_init(&op->list);
+	}
+}
+
+void arch_unoptimize_kprobe(struct optimized_kprobe *op)
+{
+	arch_arm_kprobe(&op->kp);
+}
+
+void arch_unoptimize_kprobes(struct list_head *oplist,
+			     struct list_head *done_list)
+{
+	struct optimized_kprobe *op;
+	struct optimized_kprobe *tmp;
+
+	list_for_each_entry_safe(op, tmp, oplist, list) {
+		arch_unoptimize_kprobe(op);
+		list_move(&op->list, done_list);
+	}
+}
+
+int arch_within_optimized_kprobe(struct optimized_kprobe *op,
+				 unsigned long addr)
+{
+	return ((unsigned long)op->kp.addr <= addr &&
+		(unsigned long)op->kp.addr + RELATIVEJUMP_SIZE > addr);
+}

arch/powerpc/kernel/optprobes_head.S

@@ -0,0 +1,135 @@
+/*
+ * Code to prepare detour buffer for optprobes in Kernel.
+ *
+ * Copyright 2017, Anju T, IBM Corp.
+ *
+ * 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; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <asm/ppc_asm.h>
+#include <asm/ptrace.h>
+#include <asm/asm-offsets.h>
+
+#define	OPT_SLOT_SIZE	65536
+
+	.balign	4
+
+	/*
+	 * Reserve an area to allocate slots for detour buffer.
+	 * This is part of .text section (rather than vmalloc area)
+	 * as this needs to be within 32MB of the probed address.
+	 */
+	.global optinsn_slot
+optinsn_slot:
+	.space	OPT_SLOT_SIZE
+
+	/*
+	 * Optprobe template:
+	 * This template gets copied into one of the slots in optinsn_slot
+	 * and gets fixed up with real optprobe structures et al.
+	 */
+	.global optprobe_template_entry
+optprobe_template_entry:
+	/* Create an in-memory pt_regs */
+	stdu	r1,-INT_FRAME_SIZE(r1)
+	SAVE_GPR(0,r1)
+	/* Save the previous SP into stack */
+	addi	r0,r1,INT_FRAME_SIZE
+	std	r0,GPR1(r1)
+	SAVE_10GPRS(2,r1)
+	SAVE_10GPRS(12,r1)
+	SAVE_10GPRS(22,r1)
+	/* Save SPRS */
+	mfmsr	r5
+	std	r5,_MSR(r1)
+	li	r5,0x700
+	std	r5,_TRAP(r1)
+	li	r5,0
+	std	r5,ORIG_GPR3(r1)
+	std	r5,RESULT(r1)
+	mfctr	r5
+	std	r5,_CTR(r1)
+	mflr	r5
+	std	r5,_LINK(r1)
+	mfspr	r5,SPRN_XER
+	std	r5,_XER(r1)
+	mfcr	r5
+	std	r5,_CCR(r1)
+	lbz     r5,PACASOFTIRQEN(r13)
+	std     r5,SOFTE(r1)
+	mfdar	r5
+	std	r5,_DAR(r1)
+	mfdsisr	r5
+	std	r5,_DSISR(r1)
+
+	.global optprobe_template_op_address
+optprobe_template_op_address:
+	/*
+	 * Parameters to optimized_callback():
+	 * 1. optimized_kprobe structure in r3
+	 */
+	nop
+	nop
+	nop
+	nop
+	nop
+	/* 2. pt_regs pointer in r4 */
+	addi	r4,r1,STACK_FRAME_OVERHEAD
+
+	.global optprobe_template_call_handler
+optprobe_template_call_handler:
+	/* Branch to optimized_callback() */
+	nop
+
+	/*
+	 * Parameters for instruction emulation:
+	 * 1. Pass SP in register r3.
+	 */
+	addi	r3,r1,STACK_FRAME_OVERHEAD
+
+	.global optprobe_template_insn
+optprobe_template_insn:
+	/* 2, Pass instruction to be emulated in r4 */
+	nop
+	nop
+
+	.global optprobe_template_call_emulate
+optprobe_template_call_emulate:
+	/* Branch to emulate_step()  */
+	nop
+
+	/*
+	 * All done.
+	 * Now, restore the registers...
+	 */
+	ld	r5,_MSR(r1)
+	mtmsr	r5
+	ld	r5,_CTR(r1)
+	mtctr	r5
+	ld	r5,_LINK(r1)
+	mtlr	r5
+	ld	r5,_XER(r1)
+	mtxer	r5
+	ld	r5,_CCR(r1)
+	mtcr	r5
+	ld	r5,_DAR(r1)
+	mtdar	r5
+	ld	r5,_DSISR(r1)
+	mtdsisr	r5
+	REST_GPR(0,r1)
+	REST_10GPRS(2,r1)
+	REST_10GPRS(12,r1)
+	REST_10GPRS(22,r1)
+	/* Restore the previous SP */
+	addi	r1,r1,INT_FRAME_SIZE
+
+	.global optprobe_template_ret
+optprobe_template_ret:
+	/* ... and jump back from trampoline */
+	nop
+
+	.global optprobe_template_end
+optprobe_template_end:

arch/powerpc/lib/code-patching.c

@@ -54,6 +54,27 @@ bool is_offset_in_branch_range(long offset)
 	return (offset >= -0x2000000 && offset <= 0x1fffffc && !(offset & 0x3));
 }
 
+/*
+ * Helper to check if a given instruction is a conditional branch
+ * Derived from the conditional checks in analyse_instr()
+ */
+bool __kprobes is_conditional_branch(unsigned int instr)
+{
+	unsigned int opcode = instr >> 26;
+
+	if (opcode == 16)       /* bc, bca, bcl, bcla */
+		return true;
+	if (opcode == 19) {
+		switch ((instr >> 1) & 0x3ff) {
+		case 16:        /* bclr, bclrl */
+		case 528:       /* bcctr, bcctrl */
+		case 560:       /* bctar, bctarl */
+			return true;
+		}
+	}
+	return false;
+}
+
 unsigned int create_branch(const unsigned int *addr,
 			   unsigned long target, int flags)
 {