gcc/libgomp/config/linux/allocator.c

/* Copyright (C) 2022-2026 Free Software Foundation, Inc.
   Contributed by Jakub Jelinek <jakub@redhat.com>.

   This file is part of the GNU Offloading and Multi Processing Library
   (libgomp).

   Libgomp 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, or (at your option)
   any later version.

   Libgomp 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.

   Under Section 7 of GPL version 3, you are granted additional
   permissions described in the GCC Runtime Library Exception, version
   3.1, as published by the Free Software Foundation.

   You should have received a copy of the GNU General Public License and
   a copy of the GCC Runtime Library Exception along with this program;
   see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
   <http://www.gnu.org/licenses/>.  */

/* This file contains wrappers for the system allocation routines.  Most
   places in the OpenMP API do not make any provision for failure, so in
   general we cannot allow memory allocation to fail.  */

#define _GNU_SOURCE
#include "libgomp.h"
#if defined(PLUGIN_SUPPORT) && defined(LIBGOMP_USE_PTHREADS)
#define LIBGOMP_USE_MEMKIND
#define LIBGOMP_USE_LIBNUMA
#endif

/* Implement malloc routines that can handle pinned memory on Linux.
   
   Given that pinned memory is typically used to help host <-> device memory
   transfers, we attempt to allocate such memory using a device (really:
   libgomp plugin), but fall back to mmap plus mlock if no suitable device is
   available.

   It's possible to use mlock on any heap memory, but using munlock is
   problematic if there are multiple pinned allocations on the same page.
   Tracking all that manually would be possible, but adds overhead. This may
   be worth it if there are a lot of small allocations getting pinned, but
   this seems less likely in a HPC application.

   Instead we optimize for large pinned allocations, and use mmap to ensure
   that two pinned allocations don't share the same page.  This also means
   that large allocations don't pin extra pages by being poorly aligned.  */

#define _GNU_SOURCE
#include <sys/mman.h>
#include <unistd.h>
#include <string.h>
#include <assert.h>
#include "libgomp.h"
#ifdef HAVE_INTTYPES_H
# include <inttypes.h>  /* For PRIu64.  */
#endif

static int using_device_for_page_locked
  = /* uninitialized */ -1;


static gomp_simple_alloc_ctx_p pin_ctx = NULL;
static pthread_once_t ctxlock = PTHREAD_ONCE_INIT;

static void
linux_init_pin_ctx ()
{
  pin_ctx = gomp_simple_alloc_init_context ();
}

static void *
linux_memspace_alloc (omp_memspace_handle_t memspace, size_t size, int pin,
		      bool init0)
{
  void *addr = NULL;

  if (memspace == ompx_gnu_managed_mem_space)
    addr = gomp_managed_alloc (size);
  else if (pin)
    {
      int using_device = __atomic_load_n (&using_device_for_page_locked,
					  MEMMODEL_RELAXED);
      if (using_device != 0)
	{
	  using_device = gomp_page_locked_host_alloc (&addr, size);
	  int using_device_old
	    = __atomic_exchange_n (&using_device_for_page_locked,
				   using_device, MEMMODEL_RELAXED);
	  assert (using_device_old == -1
		  /* We shouldn't have concurrently changed our mind.  */
		  || using_device_old == using_device);
	}
      if (using_device == 0)
	{
	  static int pagesize = 0;
	  static void *addrhint = NULL;

	  if (!pagesize)
	    pagesize = sysconf(_SC_PAGE_SIZE);

	  while (1)
	    {
	      addr = gomp_simple_alloc (pin_ctx, size);
	      if (addr)
		break;

	      /* Round up to a whole page.  */
	      size_t misalignment = size % pagesize;
	      size_t mmap_size = (misalignment > 0
				  ? size + pagesize - misalignment
				  : size);
	      void *newpage = mmap (addrhint, mmap_size, PROT_READ | PROT_WRITE,
				 MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
	      if (newpage == MAP_FAILED)
		break;
	      else
		{
		  if (mlock (newpage, size))
		    {
#ifdef HAVE_INTTYPES_H
		      gomp_debug (0, "libgomp: failed to pin %"PRIu64" bytes"
				  " of memory (ulimit too low?)\n",
				  (uint64_t) size);
#else
		      gomp_debug (0, "libgomp: failed to pin %lu bytes of"
				  " memory (ulimit too low?)\n",
				  (unsigned long) size);
#endif
		      munmap (newpage, size);
		      break;
		    }

		  addrhint = newpage + mmap_size;

		  pthread_once (&ctxlock, linux_init_pin_ctx);
		  gomp_simple_alloc_register_memory (pin_ctx, newpage,
						     mmap_size);
		}
	    }
	}
    }
  else
    addr = malloc (size);

  if (addr && init0)
    memset (addr, 0, size);

  return addr;
}

static void *
linux_memspace_calloc (omp_memspace_handle_t memspace, size_t size, int pin)
{
  if (memspace == ompx_gnu_managed_mem_space)
    {
      void *ret = gomp_managed_alloc (size);
      if (!ret)
	return NULL;
      memset (ret, 0, size);
      return ret;
    }
  else if (pin)
    return linux_memspace_alloc (memspace, size, pin, true);
  else
    return calloc (1, size);
}

static void
linux_memspace_free (omp_memspace_handle_t memspace, void *addr, size_t size,
		     int pin)
{
  if (memspace == ompx_gnu_managed_mem_space)
    gomp_managed_free (addr);
  else if (pin)
    {
      int using_device
	= __atomic_load_n (&using_device_for_page_locked,
			   MEMMODEL_RELAXED);
      if (using_device == 1)
	gomp_page_locked_host_free (addr);
      else
	/* The "simple" allocator does not (currently) munmap locked pages
	   (meaning that the number of locked pages never decreases), but it
	   can reuse the freed memory in subsequent gomp_simple_alloc calls.  */
	gomp_simple_free (pin_ctx, addr);
    }
  else
    free (addr);
}

static void *
linux_memspace_realloc (omp_memspace_handle_t memspace, void *addr,
			size_t oldsize, size_t size, int oldpin, int pin)
{
  if (memspace == ompx_gnu_managed_mem_space)
    /* Realloc is not implemented for device Managed Memory.  */
    ;
  else if (oldpin && pin)
    {
      int using_device
	= __atomic_load_n (&using_device_for_page_locked,
		       MEMMODEL_RELAXED);
      /* The device plugin API does not support realloc,
	 but the gomp_simple_alloc allocator does.  */
      if (using_device == 0)
	{
	  /* This can fail if there is insufficient pinned memory free.  */
	  void *newaddr = gomp_simple_realloc (pin_ctx, addr, size);
	  if (newaddr)
	    return newaddr;
	}
    }
  else if (oldpin || pin)
    /* Moving from pinned to unpinned memory cannot be done in-place.  */
    ;
  else
    return realloc (addr, size);

  /* In-place reallocation failed.  Fall back to copy.  */
  void *newaddr = linux_memspace_alloc (memspace, size, pin, false);
  if (newaddr)
    {
      memcpy (newaddr, addr, oldsize < size ? oldsize : size);
      linux_memspace_free (memspace, addr, oldsize, oldpin);
    }

  return newaddr;
}

static int
linux_memspace_validate (omp_memspace_handle_t, unsigned, int)
{
  /* Everything should be accepted on Linux, including pinning and
     non-standard memspaces.  */
  return 1;
}

#define MEMSPACE_ALLOC(MEMSPACE, SIZE, PIN) \
  linux_memspace_alloc (MEMSPACE, SIZE, PIN, false)
#define MEMSPACE_CALLOC(MEMSPACE, SIZE, PIN) \
  linux_memspace_calloc (MEMSPACE, SIZE, PIN)
#define MEMSPACE_REALLOC(MEMSPACE, ADDR, OLDSIZE, SIZE, OLDPIN, PIN) \
  linux_memspace_realloc (MEMSPACE, ADDR, OLDSIZE, SIZE, OLDPIN, PIN)
#define MEMSPACE_FREE(MEMSPACE, ADDR, SIZE, PIN) \
  linux_memspace_free (MEMSPACE, ADDR, SIZE, PIN)
#define MEMSPACE_VALIDATE(MEMSPACE, ACCESS, PIN) \
  linux_memspace_validate (MEMSPACE, ACCESS, PIN)

#include "../../allocator.c"