index : vpp	master stable/1606 stable/1609 stable/1701 stable/1704 stable/1707 stable/1710 stable/1801 stable/1804 stable/1807 stable/1810 stable/1901 stable/1904 stable/1908 stable/2001 stable/2005 stable/2009 stable/2101 stable/2106 stable/2110 stable/2202 stable/2206 stable/2210 stable/2302 stable/2306 stable/2310 stable/2402 stable/2406 stable/2410 stable/test
	Vector Packet Processing	Grokmirror user

summaryrefslogtreecommitdiffstats

log msg author committer range

path: root/test

Age	Commit message (Expand)	Author	Files	Lines
2017-08-16	SNAT: Make proto optional in nat64_bib_dump (VPP-942)	Matus Fabian	2	-20/+8
2017-08-16	NAT64: Fallback to 3-tuple key for non TCP/UDP sessions (VPP-884)	Matus Fabian	1	-18/+37
2017-08-15	Support proxy ARP on mirrored TAP interfaces	Neale Ranns	2	-0/+70
2017-08-14	SNAT: fix overlapping address space test	Matus Fabian	1	-2/+2
2017-08-14	Fix VCL LD_PRELOAD lib location in socket_test.sh	Dave Wallace	1	-9/+16
2017-08-14	make test: properly handle ctrl-c	Klement Sekera	3	-3/+36
2017-08-12	make test: fix broken DEBUG=gdb* options	Klement Sekera	2	-44/+56
2017-08-11	Dedicated SW Interface Event	Neale Ranns	1	-7/+2
2017-08-11	Add VPP Communications Library (VCL)	Dave Wallace	1	-0/+637
2017-08-10	make test: bump default test timeout to 10 minutes	Klement Sekera	1	-1/+1
2017-08-10	make test: detect hung tests	Klement Sekera	6	-29/+148
2017-08-10	acl-plugin: all TCP sessions treated as transient (VPP-932)	Andrew Yourtchenko	1	-10/+140
2017-08-09	Add PPPoE Plugin	Hongjun Ni	3	-0/+710
2017-08-08	L2 over MPLS	Neale Ranns	9	-112/+331
2017-08-07	make test: kill all remaining subprocesses on exit	Klement Sekera	2	-1/+26
2017-08-03	DHCP Client: receive unicast ACKs	Neale Ranns	1	-21/+71
2017-08-01	Tests for recursive load-balancing with no choices.	Neale Ranns	2	-5/+68
2017-08-01	P2P Ethernet	Pavel Kotucek	3	-0/+586
2017-07-24	SPAN:add l2 mirror	Eyal Bari	2	-33/+322
2017-07-21	SNAT: in2out translation as an output feature hairpinning (VPP-913)	Matus Fabian	1	-1/+1
2017-07-20	SNAT: in2out translation as an output feature (VPP-903)	Matus Fabian	2	-0/+169
2017-07-18	make test: improve console output messages	Klement Sekera	1	-3/+3
2017-07-18	DHCP client option 61 "client_id"	Neale Ranns	2	-2/+34
2017-07-18	DHCP client - remove interface address when DHCP de-configured	Neale Ranns	2	-21/+152
2017-07-18	TEST:add l2bd nd term tests	Eyal Bari	1	-6/+119
2017-07-07	SNAT: Fallback to 3-tuple key for non TCP/UDP sessions (VPP-884)	Matus Fabian	1	-1/+246
2017-07-04	TEST:add L2BD arp term tests	Eyal Bari	3	-0/+273
2017-07-03	SNAT: fix failing test_session_limit_per_user (VPP-896)	Matus Fabian	1	-0/+2
2017-06-27	L2-LEARN:fix l2fib entry seq num not updated on hit (VPP-888)	Eyal Bari	1	-29/+29
2017-06-27	FLOWPROBE: Add flowstartns, flowendns and tcpcontrolbits	Ole Troan	1	-10/+96
2017-06-21	acl-plugin: CLI to clear all sessions	Andrew Yourtchenko	1	-0/+37
2017-06-21	NAT64: custom prefix	Matus Fabian	2	-7/+164
2017-06-21	ARP: ignore non-connected routes and non-interface sources when determing if ...	Neale Ranns	1	-2/+70
2017-06-21	L2-VTR: add vtr tests	Eyal Bari	4	-65/+455
2017-06-20	SNAT: unknow protocol hairpinning fix	Matus Fabian	1	-2/+57
2017-06-19	L2FWD:fix seq_num overwritten + validate l2fib entries when forwarding	Eyal Bari	1	-3/+0
2017-06-19	SNAT: NAT packet with unknown L4 protocol if match 1:1 NAT	Matus Fabian	1	-1/+49
2017-06-19	acl-plugin: bihash-based ACL lookup	Andrew Yourtchenko	3	-0/+18
2017-06-19	NAT64: Hairpinning (VPP-699)	Matus Fabian	1	-14/+271
2017-06-13	Flowprobe - tests speed-up	Pavel Kotucek	1	-183/+151
2017-06-13	SNAT: fix 1:1 NAT without port hairpinning TCP checksum update	Matus Fabian	1	-0/+17
2017-06-12	L2FIB:fix crash in show with deleted subif entries	Eyal Bari	1	-0/+3
2017-06-09	NAT64: ICMP error support	Matus Fabian	1	-0/+90
2017-06-08	NAT64: Add NAT64 support for snat plugin (VPP-699)	Matus Fabian	2	-5/+530
2017-06-07	make test: improve debugability	Klement Sekera	2	-2/+3
2017-06-06	Packets recieved on VLAN-0 map to the main interface	Neale Ranns	2	-2/+106
2017-06-06	Rework vxlan-gpe to support FIB 2.0 and bypass mode	Hongjun Ni	2	-0/+348
2017-06-05	SNAT: fix ICMP hairpinning	Matus Fabian	1	-2/+164
2017-06-01	IP Mcast - recalculate on interface up/dowm	Neale Ranns	1	-18/+24
2017-06-01	make test: add suddenly missing dependency (six module)	Klement Sekera	1	-1/+1

• [next]

© 2016 FD.io a Linux Foundation Collaborative Project. All Rights Reserved.

Linux Foundation is a registered trademark of The Linux Foundation. Linux is a registered trademark of Linus Torvalds.

Please see our privacy policy and terms of use

/*
 * Copyright (c) 2016-2019 Cisco and/or its affiliates.
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at:
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include <svm/fifo_segment.h>
#include <vppinfra/mem.h>

static inline void *
fsh_alloc_aligned (fifo_segment_header_t *fsh, uword size, uword align)
{
  uword cur_pos, cur_pos_align, new_pos;

  cur_pos = clib_atomic_load_relax_n (&fsh->byte_index);
  cur_pos_align = round_pow2_u64 (cur_pos, align);
  size = round_pow2_u64 (size, align);
  new_pos = cur_pos_align + size;

  if (new_pos >= fsh->max_byte_index)
    return 0;

  while (!clib_atomic_cmp_and_swap_acq_relax (&fsh->byte_index, &cur_pos,
					      &new_pos, 1 /* weak */))
    {
      cur_pos_align = round_pow2_u64 (cur_pos, align);
      new_pos = cur_pos_align + size;
      if (new_pos >= fsh->max_byte_index)
	return 0;
    }
  return uword_to_pointer ((u8 *) fsh + cur_pos_align, void *);
}

static inline void *
fsh_alloc (fifo_segment_header_t *fsh, uword size)
{
  return fsh_alloc_aligned (fsh, size, 8);
}

static inline fifo_segment_slice_t *
fsh_slice_get (fifo_segment_header_t * fsh, u32 slice_index)
{
  return &fsh->slices[slice_index];
}

static inline fifo_slice_private_t *
fs_slice_private_get (fifo_segment_t *fs, u32 slice_index)
{
  ASSERT (slice_index < fs->n_slices);
  return &fs->slices[slice_index];
}

static char *fifo_segment_mem_status_strings[] = {
#define _(sym,str) str,
  foreach_segment_mem_status
#undef _
};

static inline uword
fsh_n_free_bytes (fifo_segment_header_t * fsh)
{
  uword cur_pos = clib_atomic_load_relax_n (&fsh->byte_index);
  ASSERT (fsh->max_byte_index > cur_pos);
  return fsh->max_byte_index - cur_pos;
}

static inline void
fsh_cached_bytes_add (fifo_segment_header_t * fsh, uword size)
{
  clib_atomic_fetch_add_rel (&fsh->n_cached_bytes, size);
}

static inline void
fsh_cached_bytes_sub (fifo_segment_header_t * fsh, uword size)
{
  clib_atomic_fetch_sub_rel (&fsh->n_cached_bytes, size);
}

static inline uword
fsh_n_cached_bytes (fifo_segment_header_t * fsh)
{
  uword n_cached = clib_atomic_load_relax_n (&fsh->n_cached_bytes);
  return n_cached;
}

static inline void
fsh_active_fifos_update (fifo_segment_header_t * fsh, int inc)
{
  clib_atomic_fetch_add_rel (&fsh->n_active_fifos, inc);
}

static inline u32
fsh_n_active_fifos (fifo_segment_header_t * fsh)
{
  return clib_atomic_load_relax_n (&fsh->n_active_fifos);
}

static inline uword
fsh_virtual_mem (fifo_segment_header_t * fsh)
{
  fifo_segment_slice_t *fss;
  uword total_vm = 0;
  int i;

  for (i = 0; i < fsh->n_slices; i++)
    {
      fss = fsh_slice_get (fsh, i);
      total_vm += clib_atomic_load_relax_n (&fss->virtual_mem);
    }
  return total_vm;
}

void
fsh_virtual_mem_update (fifo_segment_header_t * fsh, u32 slice_index,
			int n_bytes)
{
  fifo_segment_slice_t *fss = fsh_slice_get (fsh, slice_index);
  fss->virtual_mem += n_bytes;
}

static inline int
fss_chunk_fl_index_is_valid (fifo_segment_slice_t *fss, u32 fl_index)
{
  return (fl_index < FS_CHUNK_VEC_LEN);
}

#define FS_CL_HEAD_MASK	 0xFFFFFFFFFFFF
#define FS_CL_HEAD_TMASK 0xFFFF000000000000
#define FS_CL_HEAD_TINC	 (1ULL << 48)

static svm_fifo_chunk_t *
fss_chunk_free_list_head (fifo_segment_header_t *fsh,
			  fifo_segment_slice_t *fss, u32 fl_index)
{
  fs_sptr_t headsp = clib_atomic_load_relax_n (&fss->free_chunks[fl_index]);
  return fs_chunk_ptr (fsh, headsp & FS_CL_HEAD_MASK);
}

static void
fss_chunk_free_list_push (fifo_segment_header_t *fsh,
			  fifo_segment_slice_t *fss, u32 fl_index,
			  svm_fifo_chunk_t *c)
{
  fs_sptr_t old_head, new_head, csp;

  csp = fs_chunk_sptr (fsh, c);
  ASSERT (csp <= FS_CL_HEAD_MASK);
  old_head = clib_atomic_load_relax_n (&fss->free_chunks[fl_index]);

  do
    {
      c->next = old_head & FS_CL_HEAD_MASK;
      new_head = csp + ((old_head + FS_CL_HEAD_TINC) & FS_CL_HEAD_TMASK);
    }
  while (!clib_atomic_cmp_and_swap_acq_relax (
    &fss->free_chunks[fl_index], &old_head, &new_head, 1 /* weak */));
}

static void
fss_chunk_free_list_push_list (fifo_segment_header_t *fsh,
			       fifo_segment_slice_t *fss, u32 fl_index,
			       svm_fifo_chunk_t *head, svm_fifo_chunk_t *tail)
{
  fs_sptr_t old_head, new_head, headsp;

  headsp = fs_chunk_sptr (fsh, head);
  ASSERT (headsp <= FS_CL_HEAD_MASK);
  old_head = clib_atomic_load_relax_n (&fss->free_chunks[fl_index]);

  do
    {
      tail->next = old_head & FS_CL_HEAD_MASK;
      new_head = headsp + ((old_head + FS_CL_HEAD_TINC) & FS_CL_HEAD_TMASK);
    }
  while (!clib_atomic_cmp_and_swap_acq_relax (
    &fss->free_chunks[fl_index], &old_head, &new_head, 1 /* weak */));
}

static svm_fifo_chunk_t *
fss_chunk_free_list_pop (fifo_segment_header_t *fsh, fifo_segment_slice_t *fss,
			 u32 fl_index)
{
  fs_sptr_t old_head, new_head;
  svm_fifo_chunk_t *c;

  ASSERT (fss_chunk_fl_index_is_valid (fss, fl_index));

  old_head = clib_atomic_load_relax_n (&fss->free_chunks[fl_index]);

  /* Lock-free stacks are affected by ABA if a side allocates a chunk and
   * shortly thereafter frees it. To circumvent that, reuse the upper bits
   * of the head of the list shared pointer, i.e., offset to where the chunk
   * is, as a tag. The tag is incremented with each push/pop operation and
   * therefore collisions can only happen if an element is popped and pushed
   * exactly after a complete wrap of the tag (16 bits). It's unlikely either
   * of the sides will be descheduled for that long */
  do
    {
      if (!(old_head & FS_CL_HEAD_MASK))
	return 0;
      c = fs_chunk_ptr (fsh, old_head & FS_CL_HEAD_MASK);
      new_head = c->next + ((old_head + FS_CL_HEAD_TINC) & FS_CL_HEAD_TMASK);
    }
  while (!clib_atomic_cmp_and_swap_acq_relax (
    &fss->free_chunks[fl_index], &old_head, &new_head, 1 /* weak */));

  return c;
}

static void
fss_fifo_free_list_push (fifo_segment_header_t *fsh, fifo_segment_slice_t *fss,
			 svm_fifo_shared_t *sf)
{
  sf->next = fss->free_fifos;
  fss->free_fifos = fs_sptr (fsh, sf);
}

static void
fss_fifo_free_list_push_list (fifo_segment_header_t *fsh,
			      fifo_segment_slice_t *fss,
			      svm_fifo_shared_t *head, svm_fifo_shared_t *tail)
{
  tail->next = fss->free_fifos;
  fss->free_fifos = fs_sptr (fsh, head);
}

svm_fifo_shared_t *
fss_fifo_free_list_pop (fifo_segment_header_t *fsh, fifo_segment_slice_t *fss)
{
  svm_fifo_shared_t *sf;
  sf = fs_ptr (fsh, fss->free_fifos);
  fss->free_fifos = sf->next;
  return sf;
}

static inline void
pfss_fifo_add_active_list (fifo_slice_private_t *pfss, svm_fifo_t *f)
{
  if (pfss->active_fifos)
    {
      pfss->active_fifos->prev = f;
      f->next = pfss->active_fifos;
    }
  pfss->active_fifos = f;
}

static inline void
pfss_fifo_del_active_list (fifo_slice_private_t *pfss, svm_fifo_t *f)
{
  if (f->flags & SVM_FIFO_F_LL_TRACKED)
    {
      if (f->prev)
	f->prev->next = f->next;
      else
	pfss->active_fifos = f->next;
      if (f->next)
	f->next->prev = f->prev;
    }
}

static inline uword
fss_fl_chunk_bytes (fifo_segment_slice_t * fss)
{
  return clib_atomic_load_relax_n (&fss->n_fl_chunk_bytes);
}

static inline void
fss_fl_chunk_bytes_add (fifo_segment_slice_t * fss, uword size)
{
  clib_atomic_fetch_add_relax (&fss->n_fl_chunk_bytes, size);
}

static inline void
fss_fl_chunk_bytes_sub (fifo_segment_slice_t * fss, uword size)
{
  clib_atomic_fetch_sub_relax (&fss->n_fl_chunk_bytes, size);
}

/**
 * Initialize fifo segment shared header
 */
int
fifo_segment_init (fifo_segment_t * fs)
{
  u32 align = 8, offset = 2 * 4096, slices_sz, i;
  uword max_fifo, seg_start, seg_sz;
  fifo_segment_header_t *fsh;
  ssvm_shared_header_t *sh;
  void *seg_data;

  /* TODO remove ssvm heap entirely */
  sh = fs->ssvm.sh;

  seg_data = (u8 *) sh + offset;
  seg_sz = sh->ssvm_size - offset;

  fs->n_slices = clib_max (fs->n_slices, 1);
  slices_sz = sizeof (fifo_segment_slice_t) * fs->n_slices;

  seg_start = round_pow2_u64 (pointer_to_uword (seg_data), align);
  fsh = uword_to_pointer (seg_start, void *);
  CLIB_MEM_UNPOISON (fsh, seg_sz);
  memset (fsh, 0, sizeof (*fsh) + slices_sz);

  fsh->byte_index = sizeof (*fsh) + slices_sz;
  fsh->max_byte_index = seg_sz;
  fsh->n_slices = fs->n_slices;
  max_fifo = clib_min ((seg_sz - slices_sz) / 2, FIFO_SEGMENT_MAX_FIFO_SIZE);
  fsh->max_log2_fifo_size = min_log2 (max_fifo);
  fsh->n_cached_bytes = 0;
  fsh->n_reserved_bytes = fsh->byte_index;
  fsh->start_byte_index = fsh->byte_index;
  ASSERT (fsh->max_byte_index <= sh->ssvm_size - offset);

  fs->max_byte_index = fsh->max_byte_index;
  fs->h = fsh;
  sh->opaque[0] = (void *) ((u8 *) fsh - (u8 *) fs->ssvm.sh);

  /* Allow random offsets */
  fs->ssvm.sh->ssvm_va = 0;

  vec_validate (fs->slices, fs->n_slices - 1);
  for (i = 0; i < fs->n_slices; i++)
    fs->slices[i].fifos =
      clib_mem_bulk_init (sizeof (svm_fifo_t), CLIB_CACHE_LINE_BYTES, 32);

  sh->ready = 1;
  return (0);
}

/**
 * Create a fifo segment and initialize as master
 */
int
fifo_segment_create (fifo_segment_main_t * sm, fifo_segment_create_args_t * a)
{
  fifo_segment_t *fs;
  uword baseva;
  int rv;

  /* Allocate a fresh segment */
  pool_get_zero (sm->segments, fs);

  baseva = a->segment_type == SSVM_SEGMENT_PRIVATE ? ~0ULL : sm->next_baseva;
  fs->ssvm.ssvm_size = a->segment_size;
  fs->ssvm.is_server = 1;
  fs->ssvm.my_pid = getpid ();
  fs->ssvm.name = format (0, "%s%c", a->segment_name, 0);
  fs->ssvm.requested_va = baseva;

  if ((rv = ssvm_server_init (&fs->ssvm, a->segment_type)))
    {
      pool_put (sm->segments, fs);
      return (rv);
    }

  /* Note: requested_va updated due to seg base addr randomization */
  sm->next_baseva = fs->ssvm.sh->ssvm_va + fs->ssvm.ssvm_size;

  fifo_segment_init (fs);
  vec_add1 (a->new_segment_indices, fs - sm->segments);
  return (0);
}

/**
 * Attach as slave to a fifo segment
 */
int
fifo_segment_attach (fifo_segment_main_t * sm, fifo_segment_create_args_t * a)
{
  fifo_segment_header_t *fsh;
  fifo_segment_t *fs;
  int rv;

  pool_get_zero (sm->segments, fs);

  fs->ssvm.ssvm_size = a->segment_size;
  fs->ssvm.my_pid = getpid ();
  fs->ssvm.name = format (0, "%s%c", a->segment_name, 0);
  fs->ssvm.requested_va = 0;
  if (a->segment_type == SSVM_SEGMENT_MEMFD)
    fs->ssvm.fd = a->memfd_fd;
  else
    fs->ssvm.attach_timeout = sm->timeout_in_seconds;

  if ((rv = ssvm_client_init (&fs->ssvm, a->segment_type)))
    {
      pool_put (sm->segments, fs);
      return (rv);
    }

  /* Probably a segment without fifos */
  if (!fs->ssvm.sh->opaque[0])
    goto done;

  fsh = fs->h = (void *) fs->ssvm.sh + (uword) fs->ssvm.sh->opaque[0];
  fs->max_byte_index = fsh->max_byte_index;
  vec_validate (fs->slices, 0);
  fs->slices[0].fifos =
    clib_mem_bulk_init (sizeof (svm_fifo_t), CLIB_CACHE_LINE_BYTES, 32);

done:
  vec_add1 (a->new_segment_indices, fs - sm->segments);
  return (0);
}

void
fifo_segment_delete (fifo_segment_main_t * sm, fifo_segment_t * s)
{
  fifo_segment_cleanup (s);
  ssvm_delete (&s->ssvm);
  clib_memset (s, 0xfe, sizeof (*s));
  pool_put (sm->segments, s);
}

u32
fifo_segment_index (fifo_segment_main_t * sm, fifo_segment_t * s)
{
  return s - sm->segments;
}

fifo_segment_t *
fifo_segment_get_segment (fifo_segment_main_t * sm, u32 segment_index)
{
  return pool_elt_at_index (sm->segments, segment_index);
}

fifo_segment_t *
fifo_segment_get_segment_if_valid (fifo_segment_main_t *sm, u32 segment_index)
{
  if (pool_is_free_index (sm->segments, segment_index))
    return 0;
  return pool_elt_at_index (sm->segments, segment_index);
}

void
fifo_segment_info (fifo_segment_t * seg, char **address, size_t * size)
{
  *address = (char *) seg->ssvm.sh->ssvm_va;
  *size = seg->ssvm.ssvm_size;
}

void
fifo_segment_main_init (fifo_segment_main_t * sm, u64 baseva,
			u32 timeout_in_seconds)
{
  sm->next_baseva = baseva;
  sm->timeout_in_seconds = timeout_in_seconds;
}

static inline u32
fs_freelist_for_size (u32 size)
{
  if (PREDICT_FALSE (size < FIFO_SEGMENT_MIN_FIFO_SIZE))
    return 0;
  return clib_min (max_log2 (size) - FIFO_SEGMENT_MIN_LOG2_FIFO_SIZE,
		   FS_CHUNK_VEC_LEN - 1);
}

static inline u32
fs_freelist_index_to_size (u32 fl_index)
{
  return 1 << (fl_index + FIFO_SEGMENT_MIN_LOG2_FIFO_SIZE);
}

static inline int
fs_chunk_size_is_valid (fifo_segment_header_t * fsh, u32 size)
{
  /*
   * 4K minimum. It's not likely that anything good will happen
   * with a smaller FIFO.
   */
  return size >= FIFO_SEGMENT_MIN_FIFO_SIZE &&
	 size <= (1ULL << fsh->max_log2_fifo_size);
}

svm_fifo_chunk_t *
fs_try_alloc_multi_chunk (fifo_segment_header_t * fsh,
			  fifo_segment_slice_t * fss, u32 data_bytes)
{
  u32 fl_index, fl_size, n_alloc = 0, req_bytes = data_bytes;
  svm_fifo_chunk_t *c, *first = 0, *next;

  fl_index = fs_freelist_for_size (req_bytes);
  if (fl_index > 0)
    fl_index -= 1;

  fl_size = fs_freelist_index_to_size (fl_index);

  while (req_bytes)
    {
      c = fss_chunk_free_list_pop (fsh, fss, fl_index);
      if (c)
	{
	  c->next = fs_chunk_sptr (fsh, first);
	  first = c;
	  n_alloc += fl_size;
	  req_bytes -= clib_min (fl_size, req_bytes);
	}
      else
	{
	  /* Failed to allocate with smaller chunks */
	  if (fl_index == 0)
	    {
	      /* Free all chunks if any allocated */
	      c = first;
	      while (c)
		{
		  fl_index = fs_freelist_for_size (c->length);
		  next = fs_chunk_ptr (fsh, c->next);
		  fss_chunk_free_list_push (fsh, fss, fl_index, c);
		  c = next;
		}
	      n_alloc = 0;
	      first = 0;
	      /* As last attempt, try allocating a chunk larger than
	       * the requested size, if possible */
	      fl_index = fs_freelist_for_size (data_bytes) + 1;
	      if (!fss_chunk_fl_index_is_valid (fss, fl_index))
		return 0;
	      first = fss_chunk_free_list_pop (fsh, fss, fl_index);
	      if (first)
		{
		  first->next = 0;
		  n_alloc = fs_freelist_index_to_size (fl_index);
		  goto done;
		}
	      return 0;
	    }
	  fl_index -= 1;
	  fl_size = fl_size >> 1;
	}
    }

done:
  fss_fl_chunk_bytes_sub (fss, n_alloc);
  fsh_cached_bytes_sub (fsh, n_alloc);
  return first;
}

static int
fsh_try_alloc_fifo_hdr_batch (fifo_segment_header_t * fsh,
			      fifo_segment_slice_t * fss, u32 batch_size)
{
  svm_fifo_shared_t *f, *head = 0, *tail;
  uword size;
  u8 *fmem;
  int i;

  ASSERT (batch_size != 0);

  size = (uword) sizeof (*f) * batch_size;

  fmem = fsh_alloc_aligned (fsh, size, CLIB_CACHE_LINE_BYTES);
  if (fmem == 0)
    return -1;

  /* Carve fifo hdr space */
  tail = f = (svm_fifo_shared_t *) fmem;
  for (i = 0; i < batch_size; i++)
    {
      clib_memset (f, 0, sizeof (*f));
      f->next = fs_sptr (fsh, head);
      head = f;
      fmem += sizeof (*f);
      f = (svm_fifo_shared_t *) fmem;
    }

  fss_fifo_free_list_push_list (fsh, fss, head, tail);

  return 0;
}

static int
fsh_try_alloc_chunk_batch (fifo_segment_header_t * fsh,
			   fifo_segment_slice_t * fss,
			   u32 fl_index, u32 batch_size)
{
  svm_fifo_chunk_t *c, *head = 0, *tail;
  uword size, total_chunk_bytes;
  u32 rounded_data_size;
  u8 *cmem;
  int i;

  ASSERT (batch_size != 0);

  rounded_data_size = fs_freelist_index_to_size (fl_index);
  total_chunk_bytes = (uword) batch_size *rounded_data_size;
  size = (uword) (sizeof (*c) + rounded_data_size) * batch_size;

  cmem = fsh_alloc_aligned (fsh, size, 8 /* chunk hdr is 24B */);
  if (cmem == 0)
    return -1;

  /* Carve fifo + chunk space */
  tail = c = (svm_fifo_chunk_t *) cmem;
  for (i = 0; i < batch_size; i++)
    {
      c->start_byte = 0;
      c->length = rounded_data_size;
      c->next = fs_chunk_sptr (fsh, head);
      head = c;
      cmem += sizeof (*c) + rounded_data_size;
      c = (svm_fifo_chunk_t *) cmem;
    }

  fss_chunk_free_list_push_list (fsh, fss, fl_index, head, tail);
  fss->num_chunks[fl_index] += batch_size;
  fss_fl_chunk_bytes_add (fss, total_chunk_bytes);
  fsh_cached_bytes_add (fsh, total_chunk_bytes);

  return 0;
}

static int
fs_try_alloc_fifo_batch (fifo_segment_header_t * fsh,
			 fifo_segment_slice_t * fss,
			 u32 fl_index, u32 batch_size)
{
  if (fsh_try_alloc_fifo_hdr_batch (fsh, fss, batch_size))
    return 0;
  return fsh_try_alloc_chunk_batch (fsh, fss, fl_index, batch_size);
}

static svm_fifo_shared_t *
fsh_try_alloc_fifo_hdr (fifo_segment_header_t *fsh, fifo_segment_slice_t *fss)
{
  svm_fifo_shared_t *sf;

  if (!fss->free_fifos)
    {
      if (fsh_try_alloc_fifo_hdr_batch (fsh, fss,
					FIFO_SEGMENT_ALLOC_BATCH_SIZE))
	return 0;
    }

  sf = fss_fifo_free_list_pop (fsh, fss);
  clib_memset (sf, 0, sizeof (*sf));

  return sf;
}

static svm_fifo_chunk_t *
fsh_try_alloc_chunk (fifo_segment_header_t * fsh,
		     fifo_segment_slice_t * fss, u32 data_bytes)
{
  svm_fifo_chunk_t *c;
  u32 fl_index;

  fl_index = fs_freelist_for_size (data_bytes);

free_list:
  c = fss_chunk_free_list_pop (fsh, fss, fl_index);
  if (c)
    {
      c->next = 0;
      fss_fl_chunk_bytes_sub (fss, fs_freelist_index_to_size (fl_index));
      fsh_cached_bytes_sub (fsh, fs_freelist_index_to_size (fl_index));
    }
  else
    {
      u32 chunk_size, batch = FIFO_SEGMENT_ALLOC_BATCH_SIZE;
      uword n_free;

      chunk_size = fs_freelist_index_to_size (fl_index);
      n_free = fsh_n_free_bytes (fsh);

      if (chunk_size <= n_free)
	{
	  batch = chunk_size * batch <= n_free ? batch : 1;
	  if (!fsh_try_alloc_chunk_batch (fsh, fss, fl_index, batch))
	    goto free_list;
	}
      /* Failed to allocate larger chunk, try to allocate multi-chunk
       * that is close to what was actually requested */
      if (data_bytes <= fss_fl_chunk_bytes (fss))
	{
	  c = fs_try_alloc_multi_chunk (fsh, fss, data_bytes);
	  if (c)
	    goto done;
	  batch = n_free / FIFO_SEGMENT_MIN_FIFO_SIZE;
	  if (!batch || fsh_try_alloc_chunk_batch (fsh, fss, 0, batch))
	    goto done;
	}
      if (data_bytes <= fss_fl_chunk_bytes (fss) + n_free)
	{
	  u32 min_size = FIFO_SEGMENT_MIN_FIFO_SIZE;
	  if (n_free < min_size)
	    goto done;
	  batch = (data_bytes - fss_fl_chunk_bytes (fss)) / min_size;
	  batch = clib_min (batch + 1, n_free / min_size);
	  if (fsh_try_alloc_chunk_batch (fsh, fss, 0, batch))
	    goto done;
	  c = fs_try_alloc_multi_chunk (fsh, fss, data_bytes);
	}
    }

done:

  return c;
}

/**
 * Try to allocate new fifo
 *
 * Tries the following steps in order:
 * - grab fifo and chunk from freelists
 * - batch fifo and chunk allocation
 * - single fifo allocation
 * - grab multiple fifo chunks from freelists
 */
static svm_fifo_shared_t *
fs_try_alloc_fifo (fifo_segment_header_t *fsh, u32 slice_index, u32 data_bytes)
{
  fifo_segment_slice_t *fss;
  u32 fl_index, min_size;
  svm_fifo_chunk_t *c;
  svm_fifo_shared_t *sf = 0;

  fss = fsh_slice_get (fsh, slice_index);
  min_size = clib_max ((fsh->pct_first_alloc * data_bytes) / 100, 4096);
  fl_index = fs_freelist_for_size (min_size);

  if (!fss_chunk_fl_index_is_valid (fss, fl_index))
    return 0;

  sf = fsh_try_alloc_fifo_hdr (fsh, fss);
  if (!sf)
    return 0;

  c = fsh_try_alloc_chunk (fsh, fss, min_size);
  if (!c)
    {
      fss_fifo_free_list_push (fsh, fss, sf);
      return 0;
    }

  sf->start_chunk = fs_chunk_sptr (fsh, c);
  while (c->next)
    c = fs_chunk_ptr (fsh, c->next);
  sf->end_chunk = fs_chunk_sptr (fsh, c);
  sf->size = data_bytes;
  sf->slice_index = slice_index;

  return sf;
}

svm_fifo_chunk_t *
fsh_alloc_chunk (fifo_segment_header_t * fsh, u32 slice_index, u32 chunk_size)
{
  fifo_segment_slice_t *fss;
  svm_fifo_chunk_t *c;

  fss = fsh_slice_get (fsh, slice_index);
  c = fsh_try_alloc_chunk (fsh, fss, chunk_size);

  return c;
}

static void
fsh_slice_collect_chunks (fifo_segment_header_t * fsh,
			  fifo_segment_slice_t * fss, svm_fifo_chunk_t * c)
{
  u32 n_collect = 0, fl_index;
  svm_fifo_chunk_t *next;

  while (c)
    {
      CLIB_MEM_UNPOISON (c, sizeof (*c));
      next = fs_chunk_ptr (fsh, c->next);
      fl_index = fs_freelist_for_size (c->length);
      fss_chunk_free_list_push (fsh, fss, fl_index, c);
      n_collect += fs_freelist_index_to_size (fl_index);
      c = next;
    }

  fss_fl_chunk_bytes_add (fss, n_collect);
  fsh_cached_bytes_add (fsh, n_collect);
}

void
fsh_collect_chunks (fifo_segment_header_t * fsh, u32 slice_index,
		    svm_fifo_chunk_t * c)
{
  fifo_segment_slice_t *fss;
  fss = fsh_slice_get (fsh, slice_index);
  fsh_slice_collect_chunks (fsh, fss, c);
}

svm_fifo_t *
fs_fifo_alloc (fifo_segment_t *fs, u32 slice_index)
{
  fifo_slice_private_t *pfss = &fs->slices[slice_index];
  svm_fifo_t *f;

  f = clib_mem_bulk_alloc (pfss->fifos);
  clib_memset (f, 0, sizeof (*f));
  return f;
}

void
fs_fifo_free (fifo_segment_t *fs, svm_fifo_t *f, u32 slice_index)
{
  fifo_slice_private_t *pfss;

  if (CLIB_DEBUG)
    clib_memset (f, 0xfc, sizeof (*f));

  pfss = &fs->slices[slice_index];
  clib_mem_bulk_free (pfss->fifos, f);
}

void
fifo_segment_cleanup (fifo_segment_t *fs)
{
  int slice_index;
  svm_msg_q_t *mq = 0;

  for (slice_index = 0; slice_index < fs->n_slices; slice_index++)
    clib_mem_bulk_destroy (fs->slices[slice_index].fifos);

  vec_free (fs->slices);

  vec_foreach (fs->mqs, mq)
    svm_msg_q_cleanup (mq);

  vec_free (fs->mqs);
}

/**
 * Allocate fifo in fifo segment
 */
svm_fifo_t *
fifo_segment_alloc_fifo_w_slice (fifo_segment_t * fs, u32 slice_index,
				 u32 data_bytes, fifo_segment_ftype_t ftype)
{
  fifo_segment_header_t *fsh = fs->h;
  fifo_slice_private_t *pfss;
  fifo_segment_slice_t *fss;
  svm_fifo_shared_t *sf;
  svm_fifo_t *f = 0;

  ASSERT (slice_index < fs->n_slices);

  if (PREDICT_FALSE (data_bytes > 1 << fsh->max_log2_fifo_size))
    return 0;

  sf = fs_try_alloc_fifo (fsh, slice_index, data_bytes);
  if (!sf)
    goto done;

  f = fs_fifo_alloc (fs, slice_index);
  f->fs_hdr = fsh;
  f->shr = sf;

  svm_fifo_init (f, data_bytes);

  fss = fsh_slice_get (fsh, slice_index);
  pfss = fs_slice_private_get (fs, slice_index);

  /* If rx fifo type add to active fifos list. When cleaning up segment,
   * we need a list of active sessions that should be disconnected. Since
   * both rx and tx fifos keep pointers to the session, it's enough to track
   * only one. */
  if (ftype == FIFO_SEGMENT_RX_FIFO)
    {
      pfss_fifo_add_active_list (pfss, f);
      f->flags |= SVM_FIFO_F_LL_TRACKED;
    }

  fsh_active_fifos_update (fsh, 1);
  fss->virtual_mem += svm_fifo_size (f);

done:
  return (f);
}

svm_fifo_t *
fifo_segment_alloc_fifo_w_offset (fifo_segment_t *fs, uword offset)
{
  svm_fifo_t *f = fs_fifo_alloc (fs, 0);
  svm_fifo_shared_t *sf;

  sf = (svm_fifo_shared_t *) ((u8 *) fs->h + offset);
  f->fs_hdr = fs->h;
  f->shr = sf;

  f->ooos_list_head = OOO_SEGMENT_INVALID_INDEX;
  f->segment_index = SVM_FIFO_INVALID_INDEX;
  f->refcnt = 1;
  return f;
}

/**
 * Free fifo allocated in fifo segment
 */
void
fifo_segment_free_fifo (fifo_segment_t * fs, svm_fifo_t * f)
{
  fifo_segment_header_t *fsh = fs->h;
  fifo_slice_private_t *pfss;
  fifo_segment_slice_t *fss;
  svm_fifo_shared_t *sf;

  ASSERT (f->refcnt > 0);

  if (--f->refcnt > 0)
    return;

  /*
   * Cleanup shared state
   */

  sf = f->shr;
  fss = fsh_slice_get (fsh, sf->slice_index);
  pfss = fs_slice_private_get (fs, sf->slice_index);

  /* Free fifo chunks */
  fsh_slice_collect_chunks (fsh, fss, fs_chunk_ptr (fsh, f->shr->start_chunk));

  sf->start_chunk = sf->end_chunk = 0;
  sf->head_chunk = sf->tail_chunk = 0;

  /* Add to free list */
  fss_fifo_free_list_push (fsh, fss, sf);

  fss->virtual_mem -= svm_fifo_size (f);

  /*
   *  Cleanup private state
   */

  /* Remove from active list. Only rx fifos are tracked */
  if (f->flags & SVM_FIFO_F_LL_TRACKED)
    {
      pfss_fifo_del_active_list (pfss, f);
      f->flags &= ~SVM_FIFO_F_LL_TRACKED;
    }

  svm_fifo_free_chunk_lookup (f);
  svm_fifo_free_ooo_data (f);

  if (CLIB_DEBUG)
    {
      sf->master_session_index = ~0;
      f->master_thread_index = ~0;
    }

  f->ooo_enq = f->ooo_deq = 0;
  f->prev = 0;

  fs_fifo_free (fs, f, f->shr->slice_index);

  fsh_active_fifos_update (fsh, -1);
}

void
fifo_segment_free_client_fifo (fifo_segment_t *fs, svm_fifo_t *f)
{
  fs_fifo_free (fs, f, 0 /* clients attach fifos in slice 0 */);
}

void
fifo_segment_detach_fifo (fifo_segment_t *fs, svm_fifo_t **f)
{
  fifo_slice_private_t *pfss;
  fifo_segment_slice_t *fss;
  svm_fifo_t *of = *f;
  u32 slice_index;

  slice_index = of->master_thread_index;
  fss = fsh_slice_get (fs->h, slice_index);
  pfss = fs_slice_private_get (fs, slice_index);
  fss->virtual_mem -= svm_fifo_size (of);
  if (of->flags & SVM_FIFO_F_LL_TRACKED)
    pfss_fifo_del_active_list (pfss, of);

  /* Collect chunks that were provided in return for those detached */
  fsh_slice_collect_chunks (fs->h, fss, of->chunks_at_attach);
  of->chunks_at_attach = 0;

  /* Collect hdr that was provided in return for the detached */
  fss_fifo_free_list_push (fs->h, fss, of->hdr_at_attach);
  of->hdr_at_attach = 0;

  clib_mem_bulk_free (pfss->fifos, *f);
  *f = 0;
}

void
fifo_segment_attach_fifo (fifo_segment_t *fs, svm_fifo_t **f, u32 slice_index)
{
  svm_fifo_chunk_t *c, *nc, *pc = 0;
  fifo_slice_private_t *pfss;
  fifo_segment_slice_t *fss;
  svm_fifo_t *nf, *of;

  nf = fs_fifo_alloc (fs, slice_index);
  clib_memcpy_fast (nf, *f, sizeof (*nf));

  fss = fsh_slice_get (fs->h, slice_index);
  pfss = fs_slice_private_get (fs, slice_index);
  fss->virtual_mem += svm_fifo_size (nf);
  nf->next = nf->prev = 0;
  if (nf->flags & SVM_FIFO_F_LL_TRACKED)
    pfss_fifo_add_active_list (pfss, nf);

  /* Allocate shared hdr and chunks to be collected at detach in return
   * for those that are being attached now */
  of = *f;
  of->hdr_at_attach = fsh_try_alloc_fifo_hdr (fs->h, fss);

  c = fs_chunk_ptr (fs->h, nf->shr->start_chunk);
  of->chunks_at_attach = pc = fsh_try_alloc_chunk (fs->h, fss, c->length);

  while ((c = fs_chunk_ptr (fs->h, c->next)))
    {
      nc = fsh_try_alloc_chunk (fs->h, fss, c->length);
      pc->next = fs_chunk_sptr (fs->h, nc);
      pc = nc;
    }

  nf->shr->slice_index = slice_index;
  *f = nf;
}

uword
fifo_segment_fifo_offset (svm_fifo_t *f)
{
  return (u8 *) f->shr - (u8 *) f->fs_hdr;
}

svm_fifo_chunk_t *
fifo_segment_alloc_chunk_w_slice (fifo_segment_t *fs, u32 slice_index,
				  u32 chunk_size)
{
  fifo_segment_header_t *fsh = fs->h;
  fifo_segment_slice_t *fss;

  fss = fsh_slice_get (fsh, slice_index);
  return fsh_try_alloc_chunk (fsh, fss, chunk_size);
}

void
fifo_segment_collect_chunk (fifo_segment_t *fs, u32 slice_index,
			    svm_fifo_chunk_t *c)
{
  fsh_collect_chunks (fs->h, slice_index, c);
}

uword
fifo_segment_chunk_offset (fifo_segment_t *fs, svm_fifo_chunk_t *c)
{
  return (u8 *) c - (u8 *) fs->h;
}

svm_msg_q_t *
fifo_segment_msg_q_alloc (fifo_segment_t *fs, u32 mq_index,
			  svm_msg_q_cfg_t *cfg)
{
  fifo_segment_header_t *fsh = fs->h;
  svm_msg_q_shared_t *smq;
  svm_msg_q_t *mq;
  void *base;
  u32 size;

  if (!fs->mqs)
    {
      u32 n_mqs = clib_max (fs->h->n_mqs, 1);
      vec_validate (fs->mqs, n_mqs - 1);
    }

  size = svm_msg_q_size_to_alloc (cfg);
  base = fsh_alloc_aligned (fsh, size, 8);
  fsh->n_reserved_bytes += size;

  smq = svm_msg_q_init (base, cfg);
  mq = vec_elt_at_index (fs->mqs, mq_index);
  svm_msg_q_attach (mq, smq);

  return mq;
}

svm_msg_q_t *
fifo_segment_msg_q_attach (fifo_segment_t *fs, uword offset, u32 mq_index)
{
  svm_msg_q_t *mq;

  if (!fs->mqs)
    {
      u32 n_mqs = clib_max (fs->h->n_mqs, 1);
      vec_validate (fs->mqs, n_mqs - 1);
    }

  mq = vec_elt_at_index (fs->mqs, mq_index);

  if (!mq->q.shr)
    {
      svm_msg_q_shared_t *smq;
      smq = (svm_msg_q_shared_t *) ((u8 *) fs->h + offset);
      svm_msg_q_attach (mq, smq);
    }

  ASSERT (fifo_segment_msg_q_offset (fs, mq_index) == offset);

  return mq;
}

void
fifo_segment_msg_qs_discover (fifo_segment_t *fs, int *fds, u32 n_fds)
{
  svm_msg_q_shared_t *smq;
  u32 n_mqs, size, i;
  uword offset = 0, n_alloced;
  svm_msg_q_t *mq;

  n_mqs = fs->h->n_mqs;
  if (n_fds && n_mqs != n_fds)
    {
      clib_warning ("expected %u fds got %u", n_mqs, n_fds);
      return;
    }

  vec_validate (fs->mqs, n_mqs - 1);
  n_alloced = fs->h->n_reserved_bytes - fs->h->start_byte_index;
  ASSERT (n_alloced % n_mqs == 0);
  size = n_alloced / n_mqs;

  offset = fs->h->start_byte_index;
  for (i = 0; i < n_mqs; i++)
    {
      mq = vec_elt_at_index (fs->mqs, i);
      smq = (svm_msg_q_shared_t *) ((u8 *) fs->h + offset);
      svm_msg_q_attach (mq, smq);
      if (n_fds)
	svm_msg_q_set_eventfd (mq, fds[i]);
      offset += size;
    }
}

uword
fifo_segment_msg_q_offset (fifo_segment_t *fs, u32 mq_index)
{
  svm_msg_q_t *mq = vec_elt_at_index (fs->mqs, mq_index);

  if (mq->q.shr == 0)
    return ~0ULL;

  return (uword) ((u8 *) mq->q.shr - (u8 *) fs->h) -
	 sizeof (svm_msg_q_shared_t);
}

int
fifo_segment_prealloc_fifo_hdrs (fifo_segment_t * fs, u32 slice_index,
				 u32 batch_size)
{
  fifo_segment_header_t *fsh = fs->h;
  fifo_segment_slice_t *fss;

  fss = fsh_slice_get (fsh, slice_index);
  return fsh_try_alloc_fifo_hdr_batch (fsh, fss, batch_size);
}

int
fifo_segment_prealloc_fifo_chunks (fifo_segment_t * fs, u32 slice_index,
				   u32 chunk_size, u32 batch_size)
{
  fifo_segment_header_t *fsh = fs->h;
  fifo_segment_slice_t *fss;
  u32 fl_index;

  if (!fs_chunk_size_is_valid (fsh, chunk_size))
    {
      clib_warning ("chunk size out of range %d", chunk_size);
      return -1;
    }

  fl_index = fs_freelist_for_size (chunk_size);
  fss = fsh_slice_get (fsh, slice_index);

  return fsh_try_alloc_chunk_batch (fsh, fss, fl_index, batch_size);
}

/**
 * Pre-allocates fifo pairs in fifo segment
 */
void
fifo_segment_preallocate_fifo_pairs (fifo_segment_t * fs,
				     u32 rx_fifo_size, u32 tx_fifo_size,
				     u32 * n_fifo_pairs)
{
  u32 rx_rounded_data_size, tx_rounded_data_size, pair_size, pairs_to_alloc;
  u32 hdrs, pairs_per_slice, alloc_now;
  fifo_segment_header_t *fsh = fs->h;
  int rx_fl_index, tx_fl_index, i;
  fifo_segment_slice_t *fss;
  uword space_available;

  /* Parameter check */
  if (rx_fifo_size == 0 || tx_fifo_size == 0 || *n_fifo_pairs == 0)
    return;

  if (!fs_chunk_size_is_valid (fsh, rx_fifo_size))
    {
      clib_warning ("rx fifo_size out of range %d", rx_fifo_size);
      return;
    }

  if (!fs_chunk_size_is_valid (fsh, tx_fifo_size))
    {
      clib_warning ("tx fifo_size out of range %d", tx_fifo_size);
      return;
    }

  rx_rounded_data_size = (1 << (max_log2 (rx_fifo_size)));
  rx_fl_index = fs_freelist_for_size (rx_fifo_size);
  tx_rounded_data_size = (1 << (max_log2 (tx_fifo_size)));
  tx_fl_index = fs_freelist_for_size (tx_fifo_size);

  hdrs = sizeof (svm_fifo_t) + sizeof (svm_fifo_chunk_t);

  /* Calculate space requirements */
  pair_size = 2 * hdrs + rx_rounded_data_size + tx_rounded_data_size;
  space_available = fsh_n_free_bytes (fsh);
  pairs_to_alloc = space_available / pair_size;
  pairs_to_alloc = clib_min (pairs_to_alloc, *n_fifo_pairs);
  pairs_per_slice = pairs_to_alloc / fs->n_slices;
  pairs_per_slice += pairs_to_alloc % fs->n_slices ? 1 : 0;

  if (!pairs_per_slice)
    return;

  for (i = 0; i < fs->n_slices; i++)
    {
      alloc_now = clib_min (pairs_per_slice, *n_fifo_pairs);
      if (0 == alloc_now)
	break;

      fss = fsh_slice_get (fsh, i);
      if (fs_try_alloc_fifo_batch (fsh, fss, rx_fl_index, alloc_now))
	clib_warning ("rx prealloc failed: pairs %u", alloc_now);
      if (fs_try_alloc_fifo_batch (fsh, fss, tx_fl_index, alloc_now))
	clib_warning ("tx prealloc failed: pairs %u", alloc_now);

      /* Account for the pairs allocated */
      *n_fifo_pairs -= alloc_now;
    }
}

/**
 * Get number of active fifos
 */
u32
fifo_segment_num_fifos (fifo_segment_t * fs)
{
  return fsh_n_active_fifos (fs->h);
}

static u32
fs_slice_num_free_fifos (fifo_segment_header_t *fsh, fifo_segment_slice_t *fss)
{
  svm_fifo_shared_t *f;
  u32 count = 0;

  f = fs_ptr (fsh, fss->free_fifos);
  if (f == 0)
    return 0;

  while (f)
    {
      f = fs_ptr (fsh, f->next);
      count++;
    }
  return count;
}

u32
fifo_segment_num_free_fifos (fifo_segment_t * fs)
{
  fifo_segment_header_t *fsh = fs->h;
  fifo_segment_slice_t *fss;
  int slice_index;
  u32 count = 0;

  for (slice_index = 0; slice_index < fs->n_slices; slice_index++)
    {
      fss = fsh_slice_get (fsh, slice_index);
      count += fs_slice_num_free_fifos (fsh, fss);
    }
  return count;
}

static u32
fs_slice_num_free_chunks (fifo_segment_header_t *fsh,
			  fifo_segment_slice_t *fss, u32 size)
{
  u32 count = 0, rounded_size, fl_index;
  svm_fifo_chunk_t *c;
  int i;

  /* Count all free chunks? */
  if (size == ~0)
    {
      for (i = 0; i < FS_CHUNK_VEC_LEN; i++)
	{
	  c = fss_chunk_free_list_head (fsh, fss, i);
	  if (c == 0)
	    continue;

	  while (c)
	    {
	      c = fs_chunk_ptr (fsh, c->next);
	      count++;
	    }
	}
      return count;
    }

  rounded_size = (1 << (max_log2 (size)));
  fl_index = fs_freelist_for_size (rounded_size);

  if (fl_index >= FS_CHUNK_VEC_LEN)
    return 0;

  c = fss_chunk_free_list_head (fsh, fss, fl_index);
  if (c == 0)
    return 0;

  while (c)
    {
      c = fs_chunk_ptr (fsh, c->next);
      count++;
    }
  return count;
}

u32
fifo_segment_num_free_chunks (fifo_segment_t * fs, u32 size)
{
  fifo_segment_header_t *fsh = fs->h;
  fifo_segment_slice_t *fss;
  int slice_index;
  u32 count = 0;

  for (slice_index = 0; slice_index < fs->n_slices; slice_index++)
    {
      fss = fsh_slice_get (fsh, slice_index);
      count += fs_slice_num_free_chunks (fsh, fss, size);
    }
  return count;
}

uword
fifo_segment_size (fifo_segment_t * fs)
{
  return fs->h->max_byte_index - fs->h->n_reserved_bytes;
}

u8
fsh_has_reached_mem_limit (fifo_segment_header_t * fsh)
{
  return (fsh->flags & FIFO_SEGMENT_F_MEM_LIMIT) ? 1 : 0;
}

void
fsh_reset_mem_limit (fifo_segment_header_t * fsh)
{
  fsh->flags &= ~FIFO_SEGMENT_F_MEM_LIMIT;
}

void *
fifo_segment_alloc (fifo_segment_t *fs, uword size)
{
  void *rv = fsh_alloc (fs->h, size);
  /* Mark externally allocated bytes as reserved. This helps
   * @ref fifo_segment_size report bytes used only for fifos */
  fs->h->n_reserved_bytes += size;
  return rv;
}

uword
fifo_segment_free_bytes (fifo_segment_t * fs)
{
  return fsh_n_free_bytes (fs->h);
}

uword
fifo_segment_cached_bytes (fifo_segment_t * fs)
{
  return fsh_n_cached_bytes (fs->h);
}

uword
fifo_segment_available_bytes (fifo_segment_t * fs)
{
  return fsh_n_free_bytes (fs->h) + fsh_n_cached_bytes (fs->h);
}

uword
fifo_segment_fl_chunk_bytes (fifo_segment_t * fs)
{
  fifo_segment_header_t *fsh = fs->h;
  fifo_segment_slice_t *fss;
  uword n_bytes = 0;
  int slice_index;

  for (slice_index = 0; slice_index < fs->n_slices; slice_index++)
    {
      fss = fsh_slice_get (fsh, slice_index);
      n_bytes += fss_fl_chunk_bytes (fss);
    }

  return n_bytes;
}

u8
fifo_segment_has_fifos (fifo_segment_t * fs)
{
  return (fsh_n_active_fifos (fs->h) != 0);
}

svm_fifo_t *
fifo_segment_get_slice_fifo_list (fifo_segment_t * fs, u32 slice_index)
{
  fifo_slice_private_t *pfss;

  pfss = fs_slice_private_get (fs, slice_index);
  return pfss->active_fifos;
}

u8
fifo_segment_get_mem_usage (fifo_segment_t * fs)
{
  uword size, in_use;

  size = fifo_segment_size (fs);
  in_use =
    size - fifo_segment_free_bytes (fs) - fifo_segment_cached_bytes (fs);
  return (in_use * 100) / size;
}

fifo_segment_mem_status_t
fifo_segment_determine_status (fifo_segment_header_t * fsh, u8 usage)
{
  if (!fsh->high_watermark || !fsh->low_watermark)
    return MEMORY_PRESSURE_NO_PRESSURE;

  /* once the no-memory is detected, the status continues
   * until memory usage gets below the high watermark
   */
  if (fsh_has_reached_mem_limit (fsh))
    {
      if (usage >= fsh->high_watermark)
	return MEMORY_PRESSURE_NO_MEMORY;
      else
	fsh_reset_mem_limit (fsh);
    }

  if (usage >= fsh->high_watermark)
    return MEMORY_PRESSURE_HIGH_PRESSURE;

  else if (usage >= fsh->low_watermark)
    return MEMORY_PRESSURE_LOW_PRESSURE;

  return MEMORY_PRESSURE_NO_PRESSURE;
}

fifo_segment_mem_status_t
fifo_segment_get_mem_status (fifo_segment_t * fs)
{
  fifo_segment_header_t *fsh = fs->h;
  u8 usage = fifo_segment_get_mem_usage (fs);

  return fifo_segment_determine_status (fsh, usage);
}

u8 *
format_fifo_segment_type (u8 * s, va_list * args)
{
  fifo_segment_t *sp;
  sp = va_arg (*args, fifo_segment_t *);
  ssvm_segment_type_t st = ssvm_type (&sp->ssvm);

  if (st == SSVM_SEGMENT_PRIVATE)
    s = format (s, "%s", "private");
  else if (st == SSVM_SEGMENT_MEMFD)
    s = format (s, "%s", "memfd");
  else if (st == SSVM_SEGMENT_SHM)
    s = format (s, "%s", "shm");
  else
    s = format (s, "%s", "unknown");
  return s;
}

/**
 * Segment format function
 */
u8 *
format_fifo_segment (u8 * s, va_list * args)
{
  u32 count, indent, active_fifos, free_fifos;
  fifo_segment_t *fs = va_arg (*args, fifo_segment_t *);
  int verbose __attribute__ ((unused)) = va_arg (*args, int);
  uword est_chunk_bytes, est_free_seg_bytes, free_chunks;
  uword chunk_bytes = 0, free_seg_bytes, chunk_size;
  uword tracked_cached_bytes;
  uword fifo_hdr = 0, reserved;
  fifo_segment_header_t *fsh;
  fifo_segment_slice_t *fss;
  svm_fifo_chunk_t *c;
  u32 slice_index;
  char *address;
  size_t size;
  int i;
  uword allocated, in_use, virt;
  f64 usage;
  fifo_segment_mem_status_t mem_st;

  indent = format_get_indent (s) + 2;

  if (fs == 0)
    {
      s = format (s, "%-20s%10s%15s%15s%15s%15s", "Name", "Type",
		  "HeapSize (M)", "ActiveFifos", "FreeFifos", "Address");
      return s;
    }

  fifo_segment_info (fs, &address, &size);
  active_fifos = fifo_segment_num_fifos (fs);
  free_fifos = fifo_segment_num_free_fifos (fs);

  s = format (s, "%-20v%10U%15llu%15u%15u%15llx", ssvm_name (&fs->ssvm),
	      format_fifo_segment_type, fs, size >> 20ULL, active_fifos,
	      free_fifos, address);

  if (!verbose)
    return s;

  fsh = fs->h;

  free_chunks = fifo_segment_num_free_chunks (fs, ~0);
  if (free_chunks)
    s =
      format (s, "\n\n%UFree/Allocated chunks by size:\n", format_white_space,
	      indent + 2);
  else
    s = format (s, "\n");

  for (slice_index = 0; slice_index < fs->n_slices; slice_index++)
    {
      fss = fsh_slice_get (fsh, slice_index);
      for (i = 0; i < FS_CHUNK_VEC_LEN; i++)
	{
	  c = fss_chunk_free_list_head (fsh, fss, i);
	  if (c == 0 && fss->num_chunks[i] == 0)
	    continue;
	  count = 0;
	  while (c)
	    {
	      c = fs_chunk_ptr (fsh, c->next);
	      count++;
	    }

	  chunk_size = fs_freelist_index_to_size (i);
	  s = format (s, "%U%-5u kB: %u/%u\n", format_white_space, indent + 2,
		      chunk_size >> 10, count, fss->num_chunks[i]);

	  chunk_bytes += count * chunk_size;
	}
    }

  fifo_hdr = free_fifos * sizeof (svm_fifo_t);
  est_chunk_bytes = fifo_segment_fl_chunk_bytes (fs);
  est_free_seg_bytes = fifo_segment_free_bytes (fs);
  free_seg_bytes = fifo_segment_free_bytes (fs);
  tracked_cached_bytes = fifo_segment_cached_bytes (fs);
  allocated = fifo_segment_size (fs);
  in_use = fifo_segment_size (fs) - est_free_seg_bytes - tracked_cached_bytes;
  usage = (100.0 * in_use) / allocated;
  mem_st = fifo_segment_get_mem_status (fs);
  virt = fsh_virtual_mem (fsh);
  reserved = fsh->n_reserved_bytes;

  s = format (s, "\n%Useg free bytes: %U (%lu) estimated: %U (%lu) reserved:"
	      " %U (%lu)\n", format_white_space, indent + 2,
	      format_memory_size, free_seg_bytes, free_seg_bytes,
	      format_memory_size, est_free_seg_bytes, est_free_seg_bytes,
	      format_memory_size, reserved, reserved);
  s = format (s, "%Uchunk free bytes: %U (%lu) estimated: %U (%lu) tracked:"
	      " %U (%lu)\n", format_white_space, indent + 2,
	      format_memory_size, chunk_bytes, chunk_bytes,
	      format_memory_size, est_chunk_bytes, est_chunk_bytes,
	      format_memory_size, tracked_cached_bytes, tracked_cached_bytes);
  s = format (s, "%Ufifo active: %u hdr free: %u bytes: %U (%u) \n",
	      format_white_space, indent + 2, fsh->n_active_fifos, free_fifos,
	      format_memory_size, fifo_hdr, fifo_hdr);
  s = format (s, "%Usegment usage: %.2f%% (%U / %U) virt: %U status: %s\n",
	      format_white_space, indent + 2, usage, format_memory_size,
	      in_use, format_memory_size, allocated, format_memory_size, virt,
	      fifo_segment_mem_status_strings[mem_st]);
  s = format (s, "\n");

  return s;
}

/*
 * fd.io coding-style-patch-verification: ON
 *
 * Local Variables:
 * eval: (c-set-style "gnu")
 * End:
 */