btrace, linux: add perf event buffer abstraction

[deliverable/binutils-gdb.git] / gdb / nat / linux-btrace.c

/* Linux-dependent part of branch trace support for GDB, and GDBserver.

   Copyright (C) 2013-2015 Free Software Foundation, Inc.

   Contributed by Intel Corp. <markus.t.metzger@intel.com>

   This file is part of GDB.

   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 3 of the License, or
   (at your option) any later version.

   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with this program.  If not, see <http://www.gnu.org/licenses/>.  */

#include "common-defs.h"
#include "linux-btrace.h"
#include "common-regcache.h"
#include "gdb_wait.h"
#include "x86-cpuid.h"

#ifdef HAVE_SYS_SYSCALL_H
#include <sys/syscall.h>
#endif

#if HAVE_LINUX_PERF_EVENT_H && defined(SYS_perf_event_open)

#include <stdint.h>
#include <unistd.h>
#include <sys/mman.h>
#include <sys/user.h>
#include <sys/ptrace.h>
#include <sys/types.h>
#include <signal.h>

/* A branch trace record in perf_event.  */
struct perf_event_bts
{
  /* The linear address of the branch source.  */
  uint64_t from;

  /* The linear address of the branch destination.  */
  uint64_t to;
};

/* A perf_event branch trace sample.  */
struct perf_event_sample
{
  /* The perf_event sample header.  */
  struct perf_event_header header;

  /* The perf_event branch tracing payload.  */
  struct perf_event_bts bts;
};

/* Return non-zero if there is new data in PEVENT; zero otherwise.  */

static int
perf_event_new_data (const struct perf_event_buffer *pev)
{
  return *pev->data_head != pev->last_head;
}

/* Check whether an address is in the kernel.  */

static inline int
perf_event_is_kernel_addr (const struct btrace_target_info *tinfo,
			   uint64_t addr)
{
  uint64_t mask;

  /* If we don't know the size of a pointer, we can't check.  Let's assume it's
     not a kernel address in this case.  */
  if (tinfo->ptr_bits == 0)
    return 0;

  /* A bit mask for the most significant bit in an address.  */
  mask = (uint64_t) 1 << (tinfo->ptr_bits - 1);

  /* Check whether the most significant bit in the address is set.  */
  return (addr & mask) != 0;
}

/* Check whether a perf event record should be skipped.  */

static inline int
perf_event_skip_record (const struct btrace_target_info *tinfo,
			const struct perf_event_bts *bts)
{
  /* The hardware may report branches from kernel into user space.  Branches
     from user into kernel space will be suppressed.  We filter the former to
     provide a consistent branch trace excluding kernel.  */
  return perf_event_is_kernel_addr (tinfo, bts->from);
}

/* Perform a few consistency checks on a perf event sample record.  This is
   meant to catch cases when we get out of sync with the perf event stream.  */

static inline int
perf_event_sample_ok (const struct perf_event_sample *sample)
{
  if (sample->header.type != PERF_RECORD_SAMPLE)
    return 0;

  if (sample->header.size != sizeof (*sample))
    return 0;

  return 1;
}

/* Branch trace is collected in a circular buffer [begin; end) as pairs of from
   and to addresses (plus a header).

   Start points into that buffer at the next sample position.
   We read the collected samples backwards from start.

   While reading the samples, we convert the information into a list of blocks.
   For two adjacent samples s1 and s2, we form a block b such that b.begin =
   s1.to and b.end = s2.from.

   In case the buffer overflows during sampling, one sample may have its lower
   part at the end and its upper part at the beginning of the buffer.  */

static VEC (btrace_block_s) *
perf_event_read_bts (struct btrace_target_info* tinfo, const uint8_t *begin,
		     const uint8_t *end, const uint8_t *start,
		     unsigned long long size)
{
  VEC (btrace_block_s) *btrace = NULL;
  struct perf_event_sample sample;
  unsigned long long read = 0;
  struct btrace_block block = { 0, 0 };
  struct regcache *regcache;

  gdb_assert (begin <= start);
  gdb_assert (start <= end);

  /* The first block ends at the current pc.  */
  regcache = get_thread_regcache_for_ptid (tinfo->ptid);
  block.end = regcache_read_pc (regcache);

  /* The buffer may contain a partial record as its last entry (i.e. when the
     buffer size is not a multiple of the sample size).  */
  read = sizeof (sample) - 1;

  for (; read < size; read += sizeof (sample))
    {
      const struct perf_event_sample *psample;

      /* Find the next perf_event sample in a backwards traversal.  */
      start -= sizeof (sample);

      /* If we're still inside the buffer, we're done.  */
      if (begin <= start)
	psample = (const struct perf_event_sample *) start;
      else
	{
	  int missing;

	  /* We're to the left of the ring buffer, we will wrap around and
	     reappear at the very right of the ring buffer.  */

	  missing = (begin - start);
	  start = (end - missing);

	  /* If the entire sample is missing, we're done.  */
	  if (missing == sizeof (sample))
	    psample = (const struct perf_event_sample *) start;
	  else
	    {
	      uint8_t *stack;

	      /* The sample wrapped around.  The lower part is at the end and
		 the upper part is at the beginning of the buffer.  */
	      stack = (uint8_t *) &sample;

	      /* Copy the two parts so we have a contiguous sample.  */
	      memcpy (stack, start, missing);
	      memcpy (stack + missing, begin, sizeof (sample) - missing);

	      psample = &sample;
	    }
	}

      if (!perf_event_sample_ok (psample))
	{
	  warning (_("Branch trace may be incomplete."));
	  break;
	}

      if (perf_event_skip_record (tinfo, &psample->bts))
	continue;

      /* We found a valid sample, so we can complete the current block.  */
      block.begin = psample->bts.to;

      VEC_safe_push (btrace_block_s, btrace, &block);

      /* Start the next block.  */
      block.end = psample->bts.from;
    }

  /* Push the last block (i.e. the first one of inferior execution), as well.
     We don't know where it ends, but we know where it starts.  If we're
     reading delta trace, we can fill in the start address later on.
     Otherwise we will prune it.  */
  block.begin = 0;
  VEC_safe_push (btrace_block_s, btrace, &block);

  return btrace;
}

/* Check whether the kernel supports BTS.  */

static int
kernel_supports_bts (void)
{
  struct perf_event_attr attr;
  pid_t child, pid;
  int status, file;

  errno = 0;
  child = fork ();
  switch (child)
    {
    case -1:
      warning (_("test bts: cannot fork: %s."), strerror (errno));
      return 0;

    case 0:
      status = ptrace (PTRACE_TRACEME, 0, NULL, NULL);
      if (status != 0)
	{
	  warning (_("test bts: cannot PTRACE_TRACEME: %s."),
		   strerror (errno));
	  _exit (1);
	}

      status = raise (SIGTRAP);
      if (status != 0)
	{
	  warning (_("test bts: cannot raise SIGTRAP: %s."),
		   strerror (errno));
	  _exit (1);
	}

      _exit (1);

    default:
      pid = waitpid (child, &status, 0);
      if (pid != child)
	{
	  warning (_("test bts: bad pid %ld, error: %s."),
		   (long) pid, strerror (errno));
	  return 0;
	}

      if (!WIFSTOPPED (status))
	{
	  warning (_("test bts: expected stop. status: %d."),
		   status);
	  return 0;
	}

      memset (&attr, 0, sizeof (attr));

      attr.type = PERF_TYPE_HARDWARE;
      attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS;
      attr.sample_period = 1;
      attr.sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_ADDR;
      attr.exclude_kernel = 1;
      attr.exclude_hv = 1;
      attr.exclude_idle = 1;

      file = syscall (SYS_perf_event_open, &attr, child, -1, -1, 0);
      if (file >= 0)
	close (file);

      kill (child, SIGKILL);
      ptrace (PTRACE_KILL, child, NULL, NULL);

      pid = waitpid (child, &status, 0);
      if (pid != child)
	{
	  warning (_("test bts: bad pid %ld, error: %s."),
		   (long) pid, strerror (errno));
	  if (!WIFSIGNALED (status))
	    warning (_("test bts: expected killed. status: %d."),
		     status);
	}

      return (file >= 0);
    }
}

/* Check whether an Intel cpu supports BTS.  */

static int
intel_supports_bts (void)
{
  unsigned int cpuid, model, family;

  if (!x86_cpuid (1, &cpuid, NULL, NULL, NULL))
    return 0;

  family = (cpuid >> 8) & 0xf;
  model = (cpuid >> 4) & 0xf;

  switch (family)
    {
    case 0x6:
      model += (cpuid >> 12) & 0xf0;

      switch (model)
	{
	case 0x1a: /* Nehalem */
	case 0x1f:
	case 0x1e:
	case 0x2e:
	case 0x25: /* Westmere */
	case 0x2c:
	case 0x2f:
	case 0x2a: /* Sandy Bridge */
	case 0x2d:
	case 0x3a: /* Ivy Bridge */

	  /* AAJ122: LBR, BTM, or BTS records may have incorrect branch
	     "from" information afer an EIST transition, T-states, C1E, or
	     Adaptive Thermal Throttling.  */
	  return 0;
	}
    }

  return 1;
}

/* Check whether the cpu supports BTS.  */

static int
cpu_supports_bts (void)
{
  unsigned int ebx, ecx, edx;

  if (!x86_cpuid (0, NULL, &ebx, &ecx, &edx))
    return 0;

  if (ebx == signature_INTEL_ebx && ecx == signature_INTEL_ecx
      && edx == signature_INTEL_edx)
    return intel_supports_bts ();

  /* Don't know about others.  Let's assume they do.  */
  return 1;
}

/* Check whether the linux target supports BTS.  */

static int
linux_supports_bts (void)
{
  static int cached;

  if (cached == 0)
    {
      if (!kernel_supports_bts ())
	cached = -1;
      else if (!cpu_supports_bts ())
	cached = -1;
      else
	cached = 1;
    }

  return cached > 0;
}

/* See linux-btrace.h.  */

int
linux_supports_btrace (struct target_ops *ops, enum btrace_format format)
{
  switch (format)
    {
    case BTRACE_FORMAT_NONE:
      return 0;

    case BTRACE_FORMAT_BTS:
      return linux_supports_bts ();
    }

  internal_error (__FILE__, __LINE__, _("Unknown branch trace format"));
}

/* See linux-btrace.h.  */

struct btrace_target_info *
linux_enable_btrace (ptid_t ptid)
{
  struct perf_event_mmap_page *header;
  struct btrace_target_info *tinfo;
  int pid, pg;

  tinfo = xzalloc (sizeof (*tinfo));
  tinfo->ptid = ptid;

  tinfo->attr.size = sizeof (tinfo->attr);
  tinfo->attr.type = PERF_TYPE_HARDWARE;
  tinfo->attr.config = PERF_COUNT_HW_BRANCH_INSTRUCTIONS;
  tinfo->attr.sample_period = 1;

  /* We sample from and to address.  */
  tinfo->attr.sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_ADDR;

  tinfo->attr.exclude_kernel = 1;
  tinfo->attr.exclude_hv = 1;
  tinfo->attr.exclude_idle = 1;

  tinfo->ptr_bits = 0;

  pid = ptid_get_lwp (ptid);
  if (pid == 0)
    pid = ptid_get_pid (ptid);

  errno = 0;
  tinfo->file = syscall (SYS_perf_event_open, &tinfo->attr, pid, -1, -1, 0);
  if (tinfo->file < 0)
    goto err;

  /* We try to allocate as much buffer as we can get.
     We could allow the user to specify the size of the buffer, but then
     we'd leave this search for the maximum buffer size to him.  */
  for (pg = 4; pg >= 0; --pg)
    {
      /* The number of pages we request needs to be a power of two.  */
      header = mmap (NULL, ((1 << pg) + 1) * PAGE_SIZE, PROT_READ, MAP_SHARED,
		     tinfo->file, 0);
      if (header != MAP_FAILED)
	break;
    }

  if (header == MAP_FAILED)
    goto err_file;

  tinfo->header = header;
  tinfo->bts.mem = ((const uint8_t *) header) + PAGE_SIZE;
  tinfo->bts.size = (1 << pg) * PAGE_SIZE;
  tinfo->bts.data_head = &header->data_head;
  tinfo->bts.last_head = 0;

  return tinfo;

 err_file:
  /* We were not able to allocate any buffer.  */
  close (tinfo->file);

 err:
  xfree (tinfo);
  return NULL;
}

/* See linux-btrace.h.  */

enum btrace_error
linux_disable_btrace (struct btrace_target_info *tinfo)
{
  munmap((void *) tinfo->header, tinfo->bts.size + PAGE_SIZE);
  close (tinfo->file);
  xfree (tinfo);

  return BTRACE_ERR_NONE;
}

/* Read branch trace data in BTS format for the thread given by TINFO into
   BTRACE using the TYPE reading method.  */

static enum btrace_error
linux_read_bts (struct btrace_data_bts *btrace,
		struct btrace_target_info *tinfo,
		enum btrace_read_type type)
{
  struct perf_event_buffer *pevent;
  const uint8_t *begin, *end, *start;
  unsigned long long data_head, data_tail, buffer_size, size;
  unsigned int retries = 5;

  pevent = &tinfo->bts;

  /* For delta reads, we return at least the partial last block containing
     the current PC.  */
  if (type == BTRACE_READ_NEW && !perf_event_new_data (pevent))
    return BTRACE_ERR_NONE;

  buffer_size = pevent->size;
  data_tail = pevent->last_head;

  /* We may need to retry reading the trace.  See below.  */
  while (retries--)
    {
      data_head = *pevent->data_head;

      /* Delete any leftover trace from the previous iteration.  */
      VEC_free (btrace_block_s, btrace->blocks);

      if (type == BTRACE_READ_DELTA)
	{
	  /* Determine the number of bytes to read and check for buffer
	     overflows.  */

	  /* Check for data head overflows.  We might be able to recover from
	     those but they are very unlikely and it's not really worth the
	     effort, I think.  */
	  if (data_head < data_tail)
	    return BTRACE_ERR_OVERFLOW;

	  /* If the buffer is smaller than the trace delta, we overflowed.  */
	  size = data_head - data_tail;
	  if (buffer_size < size)
	    return BTRACE_ERR_OVERFLOW;
	}
      else
	{
	  /* Read the entire buffer.  */
	  size = buffer_size;

	  /* Adjust the size if the buffer has not overflowed, yet.  */
	  if (data_head < size)
	    size = data_head;
	}

      /* Data_head keeps growing; the buffer itself is circular.  */
      begin = pevent->mem;
      start = begin + data_head % buffer_size;

      if (data_head <= buffer_size)
	end = start;
      else
	end = begin + pevent->size;

      btrace->blocks = perf_event_read_bts (tinfo, begin, end, start, size);

      /* The stopping thread notifies its ptracer before it is scheduled out.
	 On multi-core systems, the debugger might therefore run while the
	 kernel might be writing the last branch trace records.

	 Let's check whether the data head moved while we read the trace.  */
      if (data_head == *pevent->data_head)
	break;
    }

  pevent->last_head = data_head;

  /* Prune the incomplete last block (i.e. the first one of inferior execution)
     if we're not doing a delta read.  There is no way of filling in its zeroed
     BEGIN element.  */
  if (!VEC_empty (btrace_block_s, btrace->blocks)
      && type != BTRACE_READ_DELTA)
    VEC_pop (btrace_block_s, btrace->blocks);

  return BTRACE_ERR_NONE;
}

/* See linux-btrace.h.  */

enum btrace_error
linux_read_btrace (struct btrace_data *btrace,
		   struct btrace_target_info *tinfo,
		   enum btrace_read_type type)
{
  /* We read btrace in BTS format.  */
  btrace->format = BTRACE_FORMAT_BTS;
  btrace->variant.bts.blocks = NULL;

  return linux_read_bts (&btrace->variant.bts, tinfo, type);
}

#else /* !HAVE_LINUX_PERF_EVENT_H */

/* See linux-btrace.h.  */

int
linux_supports_btrace (struct target_ops *ops, enum btrace_format format)
{
  return 0;
}

/* See linux-btrace.h.  */

struct btrace_target_info *
linux_enable_btrace (ptid_t ptid)
{
  return NULL;
}

/* See linux-btrace.h.  */

enum btrace_error
linux_disable_btrace (struct btrace_target_info *tinfo)
{
  return BTRACE_ERR_NOT_SUPPORTED;
}

/* See linux-btrace.h.  */

enum btrace_error
linux_read_btrace (struct btrace_data *btrace,
		   struct btrace_target_info *tinfo,
		   enum btrace_read_type type)
{
  return BTRACE_ERR_NOT_SUPPORTED;
}

#endif /* !HAVE_LINUX_PERF_EVENT_H */