perf.c - platform/external/strace.git - Git at Google

 /*
  * Copyright (c) 2013 Ben Noordhuis <info@bnoordhuis.nl>
  * Copyright (c) 2013-2015 Dmitry V. Levin <ldv@altlinux.org>
  * Copyright (c) 2016 Eugene Syromyatnikov <evgsyr@gmail.com>
  * Copyright (c) 2015-2018 The strace developers.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  * 1. Redistributions of source code must retain the above copyright
  *    notice, this list of conditions and the following disclaimer.
  * 2. Redistributions in binary form must reproduce the above copyright
  *    notice, this list of conditions and the following disclaimer in the
  *    documentation and/or other materials provided with the distribution.
  * 3. The name of the author may not be used to endorse or promote products
  *    derived from this software without specific prior written permission.
  *
  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */

 #include "defs.h"

 #include "perf_event_struct.h"
 #include "print_fields.h"

 #include "xlat/hw_breakpoint_len.h"
 #include "xlat/hw_breakpoint_type.h"
 #include "xlat/perf_attr_size.h"
 #include "xlat/perf_branch_sample_type.h"
 #include "xlat/perf_event_open_flags.h"
 #include "xlat/perf_event_read_format.h"
 #include "xlat/perf_event_sample_format.h"
 #include "xlat/perf_hw_cache_id.h"
 #include "xlat/perf_hw_cache_op_id.h"
 #include "xlat/perf_hw_cache_op_result_id.h"
 #include "xlat/perf_hw_id.h"
 #include "xlat/perf_sw_ids.h"
 #include "xlat/perf_type_id.h"

 struct pea_desc {
 	struct perf_event_attr *attr;
 	uint32_t size;
 };

 static void
 free_pea_desc(void *pea_desc_ptr)
 {
 	struct pea_desc *desc = pea_desc_ptr;

 	free(desc->attr);
 	free(desc);
 }

 int
 fetch_perf_event_attr(struct tcb *const tcp, const kernel_ulong_t addr)
 {
 	struct pea_desc *desc;
 	struct perf_event_attr *attr;
 	uint32_t size;

 	if (umove(tcp, addr + offsetof(struct perf_event_attr, size), &size)) {
 		printaddr(addr);
 		return 1;
 	}

 	if (size > sizeof(*attr))
 		size = sizeof(*attr);

 	if (!size)
 		size = PERF_ATTR_SIZE_VER0;

 	/*
 	 * Kernel (rightfully) deems invalid attribute structures with size less
 	 * than first published format size, and we do the same.
 	 */
 	if (size < PERF_ATTR_SIZE_VER0) {
 		printaddr(addr);
 		return 1;
 	}

 	if (abbrev(tcp))
 		size = offsetofend(struct perf_event_attr, config);

 	/* Size should be multiple of 8, but kernel doesn't check for it */
 	/* size &= ~7; */

 	attr = xcalloc(1, sizeof(*attr));

 	if (umoven_or_printaddr(tcp, addr, size, attr)) {
 		free(attr);

 		return 1;
 	}

 	desc = xmalloc(sizeof(*desc));

 	desc->attr = attr;
 	desc->size = size;

 	set_tcb_priv_data(tcp, desc, free_pea_desc);

 	return 0;
 }

 void
 print_perf_event_attr(struct tcb *const tcp, const kernel_ulong_t addr)
 {
 	static const char *precise_ip_desc[] = {
 		"arbitrary skid",
 		"constant skid",
 		"requested to have 0 skid",
 		"must have 0 skid",
 	};

 	struct pea_desc *desc;
 	struct perf_event_attr *attr;
 	uint32_t size;
 	uint32_t new_size;
 	int use_new_size = 0;

 	/*
 	 * Amusingly, kernel accepts structures with only part of the field
 	 * present, so we making check like this (instead of checking
 	 * offsetofend against size) in order to print fields as kernel sees
 	 * them. This also should work great on big endian architectures.
 	 */
 	#define _PERF_CHECK_FIELD(_field) \
 		do { \
 			if (offsetof(struct perf_event_attr, _field) >= size) \
 				goto print_perf_event_attr_out; \
 		} while (0)

 	desc = get_tcb_priv_data(tcp);

 	attr = desc->attr;
 	size = desc->size;

 	/* The only error which expected to change size field currently */
 	if (tcp->u_error == E2BIG) {
 		if (umove(tcp, addr + offsetof(struct perf_event_attr, size),
 		    &new_size))
 			use_new_size = -1;
 		else
 			use_new_size = 1;
 	}

 	PRINT_FIELD_XVAL("{", *attr, type, perf_type_id, "PERF_TYPE_???");
 	PRINT_FIELD_XVAL(", ", *attr, size, perf_attr_size,
 			 "PERF_ATTR_SIZE_???");

 	if (use_new_size) {
 		tprints(" => ");

 		if (use_new_size > 0)
 			printxval(perf_attr_size, new_size,
 				  "PERF_ATTR_SIZE_???");
 		else
 			tprints("???");
 	}

 	switch (attr->type) {
 	case PERF_TYPE_HARDWARE:
 		PRINT_FIELD_XVAL(", ", *attr, config, perf_hw_id,
 				 "PERF_COUNT_HW_???");
 		break;
 	case PERF_TYPE_SOFTWARE:
 		PRINT_FIELD_XVAL(", ", *attr, config, perf_sw_ids,
 				 "PERF_COUNT_SW_???");
 		break;
 	case PERF_TYPE_TRACEPOINT:
 		/*
 		 * "The value to use in config can be obtained from under
 		 * debugfs tracing/events/../../id if ftrace is enabled
 		 * in the kernel."
 		 */
 		PRINT_FIELD_U(", ", *attr, config);
 		break;
 	case PERF_TYPE_HW_CACHE:
 		/*
 		 * (perf_hw_cache_id) | (perf_hw_cache_op_id << 8) |
 		 * (perf_hw_cache_op_result_id << 16)
 		 */
 		tprints(", config=");
 		printxval(perf_hw_cache_id, attr->config & 0xFF,
 			  "PERF_COUNT_HW_CACHE_???");
 		tprints("|");
 		printxval(perf_hw_cache_op_id, (attr->config >> 8) & 0xFF,
 			   "PERF_COUNT_HW_CACHE_OP_???");
 		tprints("<<8|");
 		/*
 		 * Current code (see set_ext_hw_attr in arch/x86/events/core.c,
 		 * tile_map_cache_event in arch/tile/kernel/perf_event.c,
 		 * arc_pmu_cache_event in arch/arc/kernel/perf_event.c,
 		 * hw_perf_cache_event in arch/blackfin/kernel/perf_event.c,
 		 * _hw_perf_cache_event in arch/metag/kernel/perf/perf_event.c,
 		 * mipspmu_map_cache_event in arch/mips/kernel/perf_event_mipsxx.c,
 		 * hw_perf_cache_event in arch/powerpc/perf/core-book3s.c,
 		 * hw_perf_cache_event in arch/powerpc/perf/core-fsl-emb.c,
 		 * hw_perf_cache_event in arch/sh/kernel/perf_event.c,
 		 * sparc_map_cache_event in arch/sparc/kernel/perf_event.c,
 		 * xtensa_pmu_cache_event in arch/xtensa/kernel/perf_event.c,
 		 * armpmu_map_cache_event in drivers/perf/arm_pmu.c) assumes
 		 * that cache result is 8 bits in size.
 		 */
 		printxval(perf_hw_cache_op_result_id,
 			  (attr->config >> 16) & 0xFF,
 			  "PERF_COUNT_HW_CACHE_RESULT_???");
 		tprints("<<16");
 		if (attr->config >> 24) {
 			tprintf("|%#" PRIx64 "<<24", attr->config >> 24);
 			tprints_comment("PERF_COUNT_HW_CACHE_???");
 		}
 		break;
 	case PERF_TYPE_RAW:
 		/*
 		 * "If type is PERF_TYPE_RAW, then a custom "raw" config
 		 * value is needed. Most CPUs support events that are not
 		 * covered by the "generalized" events. These are
 		 * implementation defined; see your CPU manual (for example the
 		 * Intel Volume 3B documentation or the AMD BIOS and Kernel
 		 * Developer Guide). The libpfm4 library can be used to
 		 * translate from the name in the architectural manuals
 		 * to the raw hex value perf_event_open() expects in this
 		 * field."
 		 */
 	case PERF_TYPE_BREAKPOINT:
 		/*
 		 * "If type is PERF_TYPE_BREAKPOINT, then leave config set
 		 * to zero. Its parameters are set in other places."
 		 */
 	default:
 		PRINT_FIELD_X(", ", *attr, config);
 		break;
 	}

 	if (abbrev(tcp))
 		goto print_perf_event_attr_out;

 	if (attr->freq)
 		PRINT_FIELD_U(", ", *attr, sample_freq);
 	else
 		PRINT_FIELD_U(", ", *attr, sample_period);

 	PRINT_FIELD_FLAGS(", ", *attr, sample_type, perf_event_sample_format,
 			  "PERF_SAMPLE_???");
 	PRINT_FIELD_FLAGS(", ", *attr, read_format, perf_event_read_format,
 			  "PERF_FORMAT_???");

 	tprintf(", disabled=%u"
 		", inherit=%u"
 		", pinned=%u"
 		", exclusive=%u"
 		", exclusive_user=%u"
 		", exclude_kernel=%u"
 		", exclude_hv=%u"
 		", exclude_idle=%u"
 		", mmap=%u"
 		", comm=%u"
 		", freq=%u"
 		", inherit_stat=%u"
 		", enable_on_exec=%u"
 		", task=%u"
 		", watermark=%u"
 		", precise_ip=%u",
 		attr->disabled,
 		attr->inherit,
 		attr->pinned,
 		attr->exclusive,
 		attr->exclude_user,
 		attr->exclude_kernel,
 		attr->exclude_hv,
 		attr->exclude_idle,
 		attr->mmap,
 		attr->comm,
 		attr->freq,
 		attr->inherit_stat,
 		attr->enable_on_exec,
 		attr->task,
 		attr->watermark,
 		attr->precise_ip);
 	tprints_comment(precise_ip_desc[attr->precise_ip]);
 	tprintf(", mmap_data=%u"
 		", sample_id_all=%u"
 		", exclude_host=%u"
 		", exclude_guest=%u"
 		", exclude_callchain_kernel=%u"
 		", exclude_callchain_user=%u"
 		", mmap2=%u"
 		", comm_exec=%u"
 		", use_clockid=%u"
 		", context_switch=%u"
 		", write_backward=%u"
 		", namespaces=%u",
 		attr->mmap_data,
 		attr->sample_id_all,
 		attr->exclude_host,
 		attr->exclude_guest,
 		attr->exclude_callchain_kernel,
 		attr->exclude_callchain_user,
 		attr->mmap2,
 		attr->comm_exec,
 		attr->use_clockid,
 		attr->context_switch,
 		attr->write_backward,
 		attr->namespaces);

 	/*
 	 * Print it only in case it is non-zero, since it may contain flags we
 	 * are not aware about.
 	 */
 	if (attr->__reserved_1) {
 		tprintf(", __reserved_1=%#" PRIx64,
 			(uint64_t) attr->__reserved_1);
 		tprints_comment("Bits 63..29");
 	}

 	if (attr->watermark)
 		PRINT_FIELD_U(", ", *attr, wakeup_watermark);
 	else
 		PRINT_FIELD_U(", ", *attr, wakeup_events);

 	if (attr->type == PERF_TYPE_BREAKPOINT)
 		/* Any combination of R/W with X is deemed invalid */
 		PRINT_FIELD_XVAL(", ", *attr, bp_type, hw_breakpoint_type,
 				 (attr->bp_type <=
 					(HW_BREAKPOINT_X | HW_BREAKPOINT_RW))
 						? "HW_BREAKPOINT_INVALID"
 						: "HW_BREAKPOINT_???");

 	if (attr->type == PERF_TYPE_BREAKPOINT)
 		PRINT_FIELD_X(", ", *attr, bp_addr);
 	else
 		PRINT_FIELD_X(", ", *attr, config1);

 	/*
 	 * Fields after bp_addr/config1 are optional and may not present; check
 	 * against size is needed.
 	 */

 	_PERF_CHECK_FIELD(bp_len);
 	if (attr->type == PERF_TYPE_BREAKPOINT)
 		PRINT_FIELD_U(", ", *attr, bp_len);
 	else
 		PRINT_FIELD_X(", ", *attr, config2);

 	_PERF_CHECK_FIELD(branch_sample_type);
 	if (attr->sample_type & PERF_SAMPLE_BRANCH_STACK) {
 		PRINT_FIELD_FLAGS(", ", *attr, branch_sample_type,
 				  perf_branch_sample_type,
 				  "PERF_SAMPLE_BRANCH_???");
 	}

 	_PERF_CHECK_FIELD(sample_regs_user);
 	/*
 	 * "This bit mask defines the set of user CPU registers to dump on
 	 * samples. The layout of the register mask is architecture-specific and
 	 * described in the kernel header
 	 * arch/ARCH/include/uapi/asm/perf_regs.h."
 	 */
 	PRINT_FIELD_X(", ", *attr, sample_regs_user);

 	_PERF_CHECK_FIELD(sample_stack_user);
 	/*
 	 * "size of the user stack to dump if PERF_SAMPLE_STACK_USER is
 	 * specified."
 	 */
 	if (attr->sample_type & PERF_SAMPLE_STACK_USER)
 		PRINT_FIELD_X(", ", *attr, sample_stack_user);

 	if (attr->use_clockid) {
 		_PERF_CHECK_FIELD(clockid);
 		PRINT_FIELD_XVAL(", ", *attr, clockid, clocknames, "CLOCK_???");
 	}

 	_PERF_CHECK_FIELD(sample_regs_intr);
 	PRINT_FIELD_X(", ", *attr, sample_regs_intr);

 	_PERF_CHECK_FIELD(aux_watermark);
 	PRINT_FIELD_U(", ", *attr, aux_watermark);

 	_PERF_CHECK_FIELD(sample_max_stack);
 	PRINT_FIELD_U(", ", *attr, sample_max_stack);

 	/* _PERF_CHECK_FIELD(__reserved_2);
 	PRINT_FIELD_U(", ", *attr, __reserved2); */

 print_perf_event_attr_out:
 	if ((attr->size && (attr->size > size)) ||
 	    (!attr->size && (size < PERF_ATTR_SIZE_VER0)))
 		tprints(", ...");

 	tprints("}");
 }

 SYS_FUNC(perf_event_open)
 {
 	/*
 	 * We try to copy out the whole structure on entering in order to check
 	 * size value on exiting. We do not check the rest of the fields because
 	 * they shouldn't be changed, but copy the whole structure instead
 	 * of just size field because they could.
 	 */
 	if (entering(tcp)) {
 		if (!fetch_perf_event_attr(tcp, tcp->u_arg[0]))
 			return 0;
 	} else {
 		print_perf_event_attr(tcp, tcp->u_arg[0]);
 	}

 	tprintf(", %d, %d, ",
 		(int) tcp->u_arg[1],
 		(int) tcp->u_arg[2]);
 	printfd(tcp, tcp->u_arg[3]);
 	tprints(", ");
 	printflags64(perf_event_open_flags, tcp->u_arg[4], "PERF_FLAG_???");

 	return RVAL_DECODED | RVAL_FD;
 }
	/*
	* Copyright (c) 2013 Ben Noordhuis <info@bnoordhuis.nl>
	* Copyright (c) 2013-2015 Dmitry V. Levin <ldv@altlinux.org>
	* Copyright (c) 2016 Eugene Syromyatnikov <evgsyr@gmail.com>
	* Copyright (c) 2015-2018 The strace developers.
	* All rights reserved.
	*
	* Redistribution and use in source and binary forms, with or without
	* modification, are permitted provided that the following conditions
	* are met:
	* 1. Redistributions of source code must retain the above copyright
	* notice, this list of conditions and the following disclaimer.
	* 2. Redistributions in binary form must reproduce the above copyright
	* notice, this list of conditions and the following disclaimer in the
	* documentation and/or other materials provided with the distribution.
	* 3. The name of the author may not be used to endorse or promote products
	* derived from this software without specific prior written permission.
	*
	* THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
	* IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
	* OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
	* IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
	* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
	* NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
	* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
	* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
	* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
	* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
	*/

	#include "defs.h"

	#include "perf_event_struct.h"
	#include "print_fields.h"

	#include "xlat/hw_breakpoint_len.h"
	#include "xlat/hw_breakpoint_type.h"
	#include "xlat/perf_attr_size.h"
	#include "xlat/perf_branch_sample_type.h"
	#include "xlat/perf_event_open_flags.h"
	#include "xlat/perf_event_read_format.h"
	#include "xlat/perf_event_sample_format.h"
	#include "xlat/perf_hw_cache_id.h"
	#include "xlat/perf_hw_cache_op_id.h"
	#include "xlat/perf_hw_cache_op_result_id.h"
	#include "xlat/perf_hw_id.h"
	#include "xlat/perf_sw_ids.h"
	#include "xlat/perf_type_id.h"

	struct pea_desc {
	struct perf_event_attr *attr;
	uint32_t size;
	};

	static void
	free_pea_desc(void *pea_desc_ptr)
	{
	struct pea_desc *desc = pea_desc_ptr;

	free(desc->attr);
	free(desc);
	}

	int
	fetch_perf_event_attr(struct tcb *const tcp, const kernel_ulong_t addr)
	{
	struct pea_desc *desc;
	struct perf_event_attr *attr;
	uint32_t size;

	if (umove(tcp, addr + offsetof(struct perf_event_attr, size), &size)) {
	printaddr(addr);
	return 1;
	}

	if (size > sizeof(*attr))
	size = sizeof(*attr);

	if (!size)
	size = PERF_ATTR_SIZE_VER0;

	/*
	* Kernel (rightfully) deems invalid attribute structures with size less
	* than first published format size, and we do the same.
	*/
	if (size < PERF_ATTR_SIZE_VER0) {
	printaddr(addr);
	return 1;
	}

	if (abbrev(tcp))
	size = offsetofend(struct perf_event_attr, config);

	/* Size should be multiple of 8, but kernel doesn't check for it */
	/* size &= ~7; */

	attr = xcalloc(1, sizeof(*attr));

	if (umoven_or_printaddr(tcp, addr, size, attr)) {
	free(attr);

	return 1;
	}

	desc = xmalloc(sizeof(*desc));

	desc->attr = attr;
	desc->size = size;

	set_tcb_priv_data(tcp, desc, free_pea_desc);

	return 0;
	}

	void
	print_perf_event_attr(struct tcb *const tcp, const kernel_ulong_t addr)
	{
	static const char *precise_ip_desc[] = {
	"arbitrary skid",
	"constant skid",
	"requested to have 0 skid",
	"must have 0 skid",
	};

	struct pea_desc *desc;
	struct perf_event_attr *attr;
	uint32_t size;
	uint32_t new_size;
	int use_new_size = 0;

	/*
	* Amusingly, kernel accepts structures with only part of the field
	* present, so we making check like this (instead of checking
	* offsetofend against size) in order to print fields as kernel sees
	* them. This also should work great on big endian architectures.
	*/
	#define _PERF_CHECK_FIELD(_field) \
	do { \
	if (offsetof(struct perf_event_attr, _field) >= size) \
	goto print_perf_event_attr_out; \
	} while (0)

	desc = get_tcb_priv_data(tcp);

	attr = desc->attr;
	size = desc->size;

	/* The only error which expected to change size field currently */
	if (tcp->u_error == E2BIG) {
	if (umove(tcp, addr + offsetof(struct perf_event_attr, size),
	&new_size))
	use_new_size = -1;
	else
	use_new_size = 1;
	}

	PRINT_FIELD_XVAL("{", *attr, type, perf_type_id, "PERF_TYPE_???");
	PRINT_FIELD_XVAL(", ", *attr, size, perf_attr_size,
	"PERF_ATTR_SIZE_???");

	if (use_new_size) {
	tprints(" => ");

	if (use_new_size > 0)
	printxval(perf_attr_size, new_size,
	"PERF_ATTR_SIZE_???");
	else
	tprints("???");
	}

	switch (attr->type) {
	case PERF_TYPE_HARDWARE:
	PRINT_FIELD_XVAL(", ", *attr, config, perf_hw_id,
	"PERF_COUNT_HW_???");
	break;
	case PERF_TYPE_SOFTWARE:
	PRINT_FIELD_XVAL(", ", *attr, config, perf_sw_ids,
	"PERF_COUNT_SW_???");
	break;
	case PERF_TYPE_TRACEPOINT:
	/*
	* "The value to use in config can be obtained from under
	* debugfs tracing/events/../../id if ftrace is enabled
	* in the kernel."
	*/
	PRINT_FIELD_U(", ", *attr, config);
	break;
	case PERF_TYPE_HW_CACHE:
	/*
	* (perf_hw_cache_id) \| (perf_hw_cache_op_id << 8) \|
	* (perf_hw_cache_op_result_id << 16)
	*/
	tprints(", config=");
	printxval(perf_hw_cache_id, attr->config & 0xFF,
	"PERF_COUNT_HW_CACHE_???");
	tprints("\|");
	printxval(perf_hw_cache_op_id, (attr->config >> 8) & 0xFF,
	"PERF_COUNT_HW_CACHE_OP_???");
	tprints("<<8\|");
	/*
	* Current code (see set_ext_hw_attr in arch/x86/events/core.c,
	* tile_map_cache_event in arch/tile/kernel/perf_event.c,
	* arc_pmu_cache_event in arch/arc/kernel/perf_event.c,
	* hw_perf_cache_event in arch/blackfin/kernel/perf_event.c,
	* _hw_perf_cache_event in arch/metag/kernel/perf/perf_event.c,
	* mipspmu_map_cache_event in arch/mips/kernel/perf_event_mipsxx.c,
	* hw_perf_cache_event in arch/powerpc/perf/core-book3s.c,
	* hw_perf_cache_event in arch/powerpc/perf/core-fsl-emb.c,
	* hw_perf_cache_event in arch/sh/kernel/perf_event.c,
	* sparc_map_cache_event in arch/sparc/kernel/perf_event.c,
	* xtensa_pmu_cache_event in arch/xtensa/kernel/perf_event.c,
	* armpmu_map_cache_event in drivers/perf/arm_pmu.c) assumes
	* that cache result is 8 bits in size.
	*/
	printxval(perf_hw_cache_op_result_id,
	(attr->config >> 16) & 0xFF,
	"PERF_COUNT_HW_CACHE_RESULT_???");
	tprints("<<16");
	if (attr->config >> 24) {
	tprintf("\|%#" PRIx64 "<<24", attr->config >> 24);
	tprints_comment("PERF_COUNT_HW_CACHE_???");
	}
	break;
	case PERF_TYPE_RAW:
	/*
	* "If type is PERF_TYPE_RAW, then a custom "raw" config
	* value is needed. Most CPUs support events that are not
	* covered by the "generalized" events. These are
	* implementation defined; see your CPU manual (for example the
	* Intel Volume 3B documentation or the AMD BIOS and Kernel
	* Developer Guide). The libpfm4 library can be used to
	* translate from the name in the architectural manuals
	* to the raw hex value perf_event_open() expects in this
	* field."
	*/
	case PERF_TYPE_BREAKPOINT:
	/*
	* "If type is PERF_TYPE_BREAKPOINT, then leave config set
	* to zero. Its parameters are set in other places."
	*/
	default:
	PRINT_FIELD_X(", ", *attr, config);
	break;
	}

	if (abbrev(tcp))
	goto print_perf_event_attr_out;

	if (attr->freq)
	PRINT_FIELD_U(", ", *attr, sample_freq);
	else
	PRINT_FIELD_U(", ", *attr, sample_period);

	PRINT_FIELD_FLAGS(", ", *attr, sample_type, perf_event_sample_format,
	"PERF_SAMPLE_???");
	PRINT_FIELD_FLAGS(", ", *attr, read_format, perf_event_read_format,
	"PERF_FORMAT_???");

	tprintf(", disabled=%u"
	", inherit=%u"
	", pinned=%u"
	", exclusive=%u"
	", exclusive_user=%u"
	", exclude_kernel=%u"
	", exclude_hv=%u"
	", exclude_idle=%u"
	", mmap=%u"
	", comm=%u"
	", freq=%u"
	", inherit_stat=%u"
	", enable_on_exec=%u"
	", task=%u"
	", watermark=%u"
	", precise_ip=%u",
	attr->disabled,
	attr->inherit,
	attr->pinned,
	attr->exclusive,
	attr->exclude_user,
	attr->exclude_kernel,
	attr->exclude_hv,
	attr->exclude_idle,
	attr->mmap,
	attr->comm,
	attr->freq,
	attr->inherit_stat,
	attr->enable_on_exec,
	attr->task,
	attr->watermark,
	attr->precise_ip);
	tprints_comment(precise_ip_desc[attr->precise_ip]);
	tprintf(", mmap_data=%u"
	", sample_id_all=%u"
	", exclude_host=%u"
	", exclude_guest=%u"
	", exclude_callchain_kernel=%u"
	", exclude_callchain_user=%u"
	", mmap2=%u"
	", comm_exec=%u"
	", use_clockid=%u"
	", context_switch=%u"
	", write_backward=%u"
	", namespaces=%u",
	attr->mmap_data,
	attr->sample_id_all,
	attr->exclude_host,
	attr->exclude_guest,
	attr->exclude_callchain_kernel,
	attr->exclude_callchain_user,
	attr->mmap2,
	attr->comm_exec,
	attr->use_clockid,
	attr->context_switch,
	attr->write_backward,
	attr->namespaces);

	/*
	* Print it only in case it is non-zero, since it may contain flags we
	* are not aware about.
	*/
	if (attr->__reserved_1) {
	tprintf(", __reserved_1=%#" PRIx64,
	(uint64_t) attr->__reserved_1);
	tprints_comment("Bits 63..29");
	}

	if (attr->watermark)
	PRINT_FIELD_U(", ", *attr, wakeup_watermark);
	else
	PRINT_FIELD_U(", ", *attr, wakeup_events);

	if (attr->type == PERF_TYPE_BREAKPOINT)
	/* Any combination of R/W with X is deemed invalid */
	PRINT_FIELD_XVAL(", ", *attr, bp_type, hw_breakpoint_type,
	(attr->bp_type <=
	(HW_BREAKPOINT_X \| HW_BREAKPOINT_RW))
	? "HW_BREAKPOINT_INVALID"
	: "HW_BREAKPOINT_???");

	if (attr->type == PERF_TYPE_BREAKPOINT)
	PRINT_FIELD_X(", ", *attr, bp_addr);
	else
	PRINT_FIELD_X(", ", *attr, config1);

	/*
	* Fields after bp_addr/config1 are optional and may not present; check
	* against size is needed.
	*/

	_PERF_CHECK_FIELD(bp_len);
	if (attr->type == PERF_TYPE_BREAKPOINT)
	PRINT_FIELD_U(", ", *attr, bp_len);
	else
	PRINT_FIELD_X(", ", *attr, config2);

	_PERF_CHECK_FIELD(branch_sample_type);
	if (attr->sample_type & PERF_SAMPLE_BRANCH_STACK) {
	PRINT_FIELD_FLAGS(", ", *attr, branch_sample_type,
	perf_branch_sample_type,
	"PERF_SAMPLE_BRANCH_???");
	}

	_PERF_CHECK_FIELD(sample_regs_user);
	/*
	* "This bit mask defines the set of user CPU registers to dump on
	* samples. The layout of the register mask is architecture-specific and
	* described in the kernel header
	* arch/ARCH/include/uapi/asm/perf_regs.h."
	*/
	PRINT_FIELD_X(", ", *attr, sample_regs_user);

	_PERF_CHECK_FIELD(sample_stack_user);
	/*
	* "size of the user stack to dump if PERF_SAMPLE_STACK_USER is
	* specified."
	*/
	if (attr->sample_type & PERF_SAMPLE_STACK_USER)
	PRINT_FIELD_X(", ", *attr, sample_stack_user);

	if (attr->use_clockid) {
	_PERF_CHECK_FIELD(clockid);
	PRINT_FIELD_XVAL(", ", *attr, clockid, clocknames, "CLOCK_???");
	}

	_PERF_CHECK_FIELD(sample_regs_intr);
	PRINT_FIELD_X(", ", *attr, sample_regs_intr);

	_PERF_CHECK_FIELD(aux_watermark);
	PRINT_FIELD_U(", ", *attr, aux_watermark);

	_PERF_CHECK_FIELD(sample_max_stack);
	PRINT_FIELD_U(", ", *attr, sample_max_stack);

	/* _PERF_CHECK_FIELD(__reserved_2);
	PRINT_FIELD_U(", ", attr, __reserved2); /

	print_perf_event_attr_out:
	if ((attr->size && (attr->size > size)) \|\|
	(!attr->size && (size < PERF_ATTR_SIZE_VER0)))
	tprints(", ...");

	tprints("}");
	}

	SYS_FUNC(perf_event_open)
	{
	/*
	* We try to copy out the whole structure on entering in order to check
	* size value on exiting. We do not check the rest of the fields because
	* they shouldn't be changed, but copy the whole structure instead
	* of just size field because they could.
	*/
	if (entering(tcp)) {
	if (!fetch_perf_event_attr(tcp, tcp->u_arg[0]))
	return 0;
	} else {
	print_perf_event_attr(tcp, tcp->u_arg[0]);
	}

	tprintf(", %d, %d, ",
	(int) tcp->u_arg[1],
	(int) tcp->u_arg[2]);
	printfd(tcp, tcp->u_arg[3]);
	tprints(", ");
	printflags64(perf_event_open_flags, tcp->u_arg[4], "PERF_FLAG_???");

	return RVAL_DECODED \| RVAL_FD;
	}