/* * Copyright (c) 2015 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. */ /* * main.c: main vector processing loop * * Copyright (c) 2008 Eliot Dresselhaus * * Permission is hereby granted, free of charge, to any person obtaining * a copy of this software and associated documentation files (the * "Software"), to deal in the Software without restriction, including * without limitation the rights to use, copy, modify, merge, publish, * distribute, sublicense, and/or sell copies of the Software, and to * permit persons to whom the Software is furnished to do so, subject to * the following conditions: * * The above copyright notice and this permission notice shall be * included in all copies or substantial portions of the Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ #include #include #include #include #include #include #include CJ_GLOBAL_LOG_PROTOTYPE; /* Actually allocate a few extra slots of vector data to support speculative vector enqueues which overflow vector data in next frame. */ #define VLIB_FRAME_SIZE_ALLOC (VLIB_FRAME_SIZE + 4) u32 wraps; always_inline u32 vlib_frame_bytes (u32 n_scalar_bytes, u32 n_vector_bytes) { u32 n_bytes; /* Make room for vlib_frame_t plus scalar arguments. */ n_bytes = vlib_frame_vector_byte_offset (n_scalar_bytes); /* Make room for vector arguments. Allocate a few extra slots of vector data to support speculative vector enqueues which overflow vector data in next frame. */ #define VLIB_FRAME_SIZE_EXTRA 4 n_bytes += (VLIB_FRAME_SIZE + VLIB_FRAME_SIZE_EXTRA) * n_vector_bytes; /* Magic number is first 32bit number after vector data. Used to make sure that vector data is never overrun. */ #define VLIB_FRAME_MAGIC (0xabadc0ed) n_bytes += sizeof (u32); /* Pad to cache line. */ n_bytes = round_pow2 (n_bytes, CLIB_CACHE_LINE_BYTES); return n_bytes; } always_inline u32 * vlib_frame_find_magic (vlib_frame_t * f, vlib_node_t * node) { void *p = f; p += vlib_frame_vector_byte_offset (node->scalar_size); p += (VLIB_FRAME_SIZE + VLIB_FRAME_SIZE_EXTRA) * node->vector_size; return p; } static vlib_frame_size_t * get_frame_size_info (vlib_node_main_t * nm, u32 n_scalar_bytes, u32 n_vector_bytes) { uword key = (n_scalar_bytes << 16) | n_vector_bytes; uword *p, i; p = hash_get (nm->frame_size_hash, key); if (p) i = p[0]; else { i = vec_len (nm->frame_sizes); vec_validate (nm->frame_sizes, i); hash_set (nm->frame_size_hash, key, i); } return vec_elt_at_index (nm->frame_sizes, i); } static u32 vlib_frame_alloc_to_node (vlib_main_t * vm, u32 to_node_index, u32 frame_flags) { vlib_node_main_t *nm = &vm->node_main; vlib_frame_size_t *fs; vlib_node_t *to_node; vlib_frame_t *f; u32 fi, l, n, scalar_size, vector_size; to_node = vlib_get_node (vm, to_node_index); scalar_size = to_node->scalar_size; vector_size = to_node->vector_size; fs = get_frame_size_info (nm, scalar_size, vector_size); n = vlib_frame_bytes (scalar_size, vector_size); if ((l = vec_len (fs->free_frame_indices)) > 0) { /* Allocate from end of free list. */ fi = fs->free_frame_indices[l - 1]; f = vlib_get_frame_no_check (vm, fi); _vec_len (fs->free_frame_indices) = l - 1; } else { f = clib_mem_alloc_aligned_no_fail (n, VLIB_FRAME_ALIGN); fi = vlib_frame_index_no_check (vm, f); } /* Poison frame when debugging. */ if (CLIB_DEBUG > 0) memset (f, 0xfe, n); /* Insert magic number. */ { u32 *magic; magic = vlib_frame_find_magic (f, to_node); *magic = VLIB_FRAME_MAGIC; } f->frame_flags = VLIB_FRAME_IS_ALLOCATED | frame_flags; f->n_vectors = 0; f->scalar_size = scalar_size; f->vector_size = vector_size; fs->n_alloc_frames += 1; return fi; } /* Allocate a frame for from FROM_NODE to TO_NODE via TO_NEXT_INDEX. Returns frame index. */ static u32 vlib_frame_alloc (vlib_main_t * vm, vlib_node_runtime_t * from_node_runtime, u32 to_next_index) { vlib_node_t *from_node; from_node = vlib_get_node (vm, from_node_runtime->node_index); ASSERT (to_next_index < vec_len (from_node->next_nodes)); return vlib_frame_alloc_to_node (vm, from_node->next_nodes[to_next_index], /* frame_flags */ 0); } vlib_frame_t * vlib_get_frame_to_node (vlib_main_t * vm, u32 to_node_index) { u32 fi = vlib_frame_alloc_to_node (vm, to_node_index, /* frame_flags */ VLIB_FRAME_FREE_AFTER_DISPATCH); return vlib_get_frame (vm, fi); } void vlib_put_frame_to_node (vlib_main_t * vm, u32 to_node_index, vlib_frame_t * f) { vlib_pending_frame_t *p; vlib_node_t *to_node; if (f->n_vectors == 0) return; to_node = vlib_get_node (vm, to_node_index); vec_add2 (vm->node_main.pending_frames, p, 1); f->frame_flags |= VLIB_FRAME_PENDING; p->frame_index = vlib_frame_index (vm, f); p->node_runtime_index = to_node->runtime_index; p->next_frame_index = VLIB_PENDING_FRAME_NO_NEXT_FRAME; } /* Free given frame. */ void vlib_frame_free (vlib_main_t * vm, vlib_node_runtime_t * r, vlib_frame_t * f) { vlib_node_main_t *nm = &vm->node_main; vlib_node_t *node; vlib_frame_size_t *fs; u32 frame_index; ASSERT (f->frame_flags & VLIB_FRAME_IS_ALLOCATED); node = vlib_get_node (vm, r->node_index); fs = get_frame_size_info (nm, node->scalar_size, node->vector_size); frame_index = vlib_frame_index (vm, f); ASSERT (f->frame_flags & VLIB_FRAME_IS_ALLOCATED); /* No next frames may point to freed frame. */ if (CLIB_DEBUG > 0) { vlib_next_frame_t *nf; vec_foreach (nf, vm->node_main.next_frames) ASSERT (nf->frame_index != frame_index); } f->frame_flags &= ~VLIB_FRAME_IS_ALLOCATED; vec_add1 (fs->free_frame_indices, frame_index); ASSERT (fs->n_alloc_frames > 0); fs->n_alloc_frames -= 1; } static clib_error_t * show_frame_stats (vlib_main_t * vm, unformat_input_t * input, vlib_cli_command_t * cmd) { vlib_node_main_t *nm = &vm->node_main; vlib_frame_size_t *fs; vlib_cli_output (vm, "%=6s%=12s%=12s", "Size", "# Alloc", "# Free"); vec_foreach (fs, nm->frame_sizes) { u32 n_alloc = fs->n_alloc_frames; u32 n_free = vec_len (fs->free_frame_indices); if (n_alloc + n_free > 0) vlib_cli_output (vm, "%=6d%=12d%=12d", fs - nm->frame_sizes, n_alloc, n_free); } return 0; } /* *INDENT-OFF* */ VLIB_CLI_COMMAND (show_frame_stats_cli, static) = { .path = "show vlib frame-allocation", .short_help = "Show node dispatch frame statistics", .function = show_frame_stats, }; /* *INDENT-ON* */ /* Change ownership of enqueue rights to given next node. */ static void vlib_next_frame_change_ownership (vlib_main_t * vm, vlib_node_runtime_t * node_runtime, u32 next_index) { vlib_node_main_t *nm = &vm->node_main; vlib_next_frame_t *next_frame; vlib_node_t *node, *next_node; node = vec_elt (nm->nodes, node_runtime->node_index); /* Only internal & input nodes are allowed to call other nodes. */ ASSERT (node->type == VLIB_NODE_TYPE_INTERNAL || node->type == VLIB_NODE_TYPE_INPUT || node->type == VLIB_NODE_TYPE_PROCESS); ASSERT (vec_len (node->next_nodes) == node_runtime->n_next_nodes); next_frame = vlib_node_runtime_get_next_frame (vm, node_runtime, next_index); next_node = vec_elt (nm->nodes, node->next_nodes[next_index]); if (next_node->owner_node_index != VLIB_INVALID_NODE_INDEX) { /* Get frame from previous owner. */ vlib_next_frame_t *owner_next_frame; vlib_next_frame_t tmp; owner_next_frame = vlib_node_get_next_frame (vm, next_node->owner_node_index, next_node->owner_next_index); /* Swap target next frame with owner's. */ tmp = owner_next_frame[0]; owner_next_frame[0] = next_frame[0]; next_frame[0] = tmp; /* * If next_frame is already pending, we have to track down * all pending frames and fix their next_frame_index fields. */ if (next_frame->flags & VLIB_FRAME_PENDING) { vlib_pending_frame_t *p; if (next_frame->frame_index != ~0) { vec_foreach (p, nm->pending_frames) { if (p->frame_index == next_frame->frame_index) { p->next_frame_index = next_frame - vm->node_main.next_frames; } } } } } else { /* No previous owner. Take ownership. */ next_frame->flags |= VLIB_FRAME_OWNER; } /* Record new owner. */ next_node->owner_node_index = node->index; next_node->owner_next_index = next_index; /* Now we should be owner. */ ASSERT (next_frame->flags & VLIB_FRAME_OWNER); } /* Make sure that magic number is still there. Otherwise, it is likely that caller has overrun frame arguments. */ always_inline void validate_frame_magic (vlib_main_t * vm, vlib_frame_t * f, vlib_node_t * n, uword next_index) { vlib_node_t *next_node = vlib_get_node (vm, n->next_nodes[next_index]); u32 *magic = vlib_frame_find_magic (f, next_node); ASSERT (VLIB_FRAME_MAGIC == magic[0]); } vlib_frame_t * vlib_get_next_frame_internal (vlib_main_t * vm, vlib_node_runtime_t * node, u32 next_index, u32 allocate_new_next_frame) { vlib_frame_t *f; vlib_next_frame_t *nf; u32 n_used; nf = vlib_node_runtime_get_next_frame (vm, node, next_index); /* Make sure this next frame owns right to enqueue to destination frame. */ if (PREDICT_FALSE (!(nf->flags & VLIB_FRAME_OWNER))) vlib_next_frame_change_ownership (vm, node, next_index); /* ??? Don't need valid flag: can use frame_index == ~0 */ if (PREDICT_FALSE (!(nf->flags & VLIB_FRAME_IS_ALLOCATED))) { nf->frame_index = vlib_frame_alloc (vm, node, next_index); nf->flags |= VLIB_FRAME_IS_ALLOCATED; } f = vlib_get_frame (vm, nf->frame_index); /* Has frame been removed from pending vector (e.g. finished dispatching)? If so we can reuse frame. */ if ((nf->flags & VLIB_FRAME_PENDING) && !(f->frame_flags & VLIB_FRAME_PENDING)) { nf->flags &= ~VLIB_FRAME_PENDING; f->n_vectors = 0; } /* Allocate new frame if current one is already full. */ n_used = f->n_vectors; if (n_used >= VLIB_FRAME_SIZE || (allocate_new_next_frame && n_used > 0)) { /* Old frame may need to be freed after dispatch, since we'll have two redundant frames from node -> next node. */ if (!(nf->flags & VLIB_FRAME_NO_FREE_AFTER_DISPATCH)) { vlib_frame_t *f_old = vlib_get_frame (vm, nf->frame_index); f_old->frame_flags |= VLIB_FRAME_FREE_AFTER_DISPATCH; } /* Allocate new frame to replace full one. */ nf->frame_index = vlib_frame_alloc (vm, node, next_index); f = vlib_get_frame (vm, nf->frame_index); n_used = f->n_vectors; } /* Should have free vectors in frame now. */ ASSERT (n_used < VLIB_FRAME_SIZE); if (CLIB_DEBUG > 0) { validate_frame_magic (vm, f, vlib_get_node (vm, node->node_index), next_index); } return f; } static void vlib_put_next_frame_validate (vlib_main_t * vm, vlib_node_runtime_t * rt, u32 next_index, u32 n_vectors_left) { vlib_node_main_t *nm = &vm->node_main; vlib_next_frame_t *nf; vlib_frame_t *f; vlib_node_runtime_t *next_rt; vlib_node_t *next_node; u32 n_before, n_after; nf = vlib_node_runtime_get_next_frame (vm, rt, next_index); f = vlib_get_frame (vm, nf->frame_index); ASSERT (n_vectors_left <= VLIB_FRAME_SIZE); n_after = VLIB_FRAME_SIZE - n_vectors_left; n_before = f->n_vectors; ASSERT (n_after >= n_before); next_rt = vec_elt_at_index (nm->nodes_by_type[VLIB_NODE_TYPE_INTERNAL], nf->node_runtime_index); next_node = vlib_get_node (vm, next_rt->node_index); if (n_after > 0 && next_node->validate_frame) { u8 *msg = next_node->validate_frame (vm, rt, f); if (msg) { clib_warning ("%v", msg); ASSERT (0); } vec_free (msg); } } void vlib_put_next_frame (vlib_main_t * vm, vlib_node_runtime_t * r, u32 next_index, u32 n_vectors_left) { vlib_node_main_t *nm = &vm->node_main; vlib_next_frame_t *nf; vlib_frame_t *f; u32 n_vectors_in_frame; if (buffer_main.callbacks_registered == 0 && CLIB_DEBUG > 0) vlib_put_next_frame_validate (vm, r, next_index, n_vectors_left); nf = vlib_node_runtime_get_next_frame (vm, r, next_index); f = vlib_get_frame (vm, nf->frame_index); /* Make sure that magic number is still there. Otherwise, caller has overrun frame meta data. */ if (CLIB_DEBUG > 0) { vlib_node_t *node = vlib_get_node (vm, r->node_index); validate_frame_magic (vm, f, node, next_index); } /* Convert # of vectors left -> number of vectors there. */ ASSERT (n_vectors_left <= VLIB_FRAME_SIZE); n_vectors_in_frame = VLIB_FRAME_SIZE - n_vectors_left; f->n_vectors = n_vectors_in_frame; /* If vectors were added to frame, add to pending vector. */ if (PREDICT_TRUE (n_vectors_in_frame > 0)) { vlib_pending_frame_t *p; u32 v0, v1; r->cached_next_index = next_index; if (!(f->frame_flags & VLIB_FRAME_PENDING)) { __attribute__ ((unused)) vlib_node_t *node; vlib_node_t *next_node; vlib_node_runtime_t *next_runtime; node = vlib_get_node (vm, r->node_index); next_node = vlib_get_next_node (vm, r->node_index, next_index); next_runtime = vlib_node_get_runtime (vm, next_node->index); vec_add2 (nm->pending_frames, p, 1); p->frame_index = nf->frame_index; p->node_runtime_index = nf->node_runtime_index; p->next_frame_index = nf - nm->next_frames; nf->flags |= VLIB_FRAME_PENDING; f->frame_flags |= VLIB_FRAME_PENDING; /* * If we're going to dispatch this frame on another thread, * force allocation of a new frame. Otherwise, we create * a dangling frame reference. Each thread has its own copy of * the next_frames vector. */ if (0 && r->thread_index != next_runtime->thread_index) { nf->frame_index = ~0; nf->flags &= ~(VLIB_FRAME_PENDING | VLIB_FRAME_IS_ALLOCATED); } } /* Copy trace flag from next_frame and from runtime. */ nf->flags |= (nf->flags & VLIB_NODE_FLAG_TRACE) | (r-> flags & VLIB_NODE_FLAG_TRACE); v0 = nf->vectors_since_last_overflow; v1 = v0 + n_vectors_in_frame; nf->vectors_since_last_overflow = v1; if (PREDICT_FALSE (v1 < v0)) { vlib_node_t *node = vlib_get_node (vm, r->node_index); vec_elt (node->n_vectors_by_next_node, next_index) += v0; } } } /* Sync up runtime (32 bit counters) and main node stats (64 bit counters). */ never_inline void vlib_node_runtime_sync_stats (vlib_main_t * vm, vlib_node_runtime_t * r, uword n_calls, uword n_vectors, uword n_clocks) { vlib_node_t *n = vlib_get_node (vm, r->node_index); n->stats_total.calls += n_calls + r->calls_since_last_overflow; n->stats_total.vectors += n_vectors + r->vectors_since_last_overflow; n->stats_total.clocks += n_clocks + r->clocks_since_last_overflow; n->stats_total.max_clock = r->max_clock; n->stats_total.max_clock_n = r->max_clock_n; r->calls_since_last_overflow = 0; r->vectors_since_last_overflow = 0; r->clocks_since_last_overflow = 0; } always_inline void __attribute__ ((unused)) vlib_process_sync_stats (vlib_main_t * vm, vlib_process_t * p, uword n_calls, uword n_vectors, uword n_clocks) { vlib_node_runtime_t *rt = &p->node_runtime; vlib_node_t *n = vlib_get_node (vm, rt->node_index); vlib_node_runtime_sync_stats (vm, rt, n_calls, n_vectors, n_clocks); n->stats_total.suspends += p->n_suspends; p->n_suspends = 0; } void vlib_node_sync_stats (vlib_main_t * vm, vlib_node_t * n) { vlib_node_runtime_t *rt; if (n->type == VLIB_NODE_TYPE_PROCESS) { /* Nothing to do for PROCESS nodes except in main thread */ if (vm != &vlib_global_main) return; vlib_process_t *p = vlib_get_process_from_node (vm, n); n->stats_total.suspends += p->n_suspends; p->n_suspends = 0; rt = &p->node_runtime; } else rt = vec_elt_at_index (vm->node_main.nodes_by_type[n->type], n->runtime_index); vlib_node_runtime_sync_stats (vm, rt, 0, 0, 0); /* Sync up runtime next frame vector counters with main node structure. */ { vlib_next_frame_t *nf; uword i; for (i = 0; i < rt->n_next_nodes; i++) { nf = vlib_node_runtime_get_next_frame (vm, rt, i); vec_elt (n->n_vectors_by_next_node, i) += nf->vectors_since_last_overflow; nf->vectors_since_last_overflow = 0; } } } always_inline u32 vlib_node_runtime_update_stats (vlib_main_t * vm, vlib_node_runtime_t * node, uword n_calls, uword n_vectors, uword n_clocks) { u32 ca0, ca1, v0, v1, cl0, cl1, r; cl0 = cl1 = node->clocks_since_last_overflow; ca0 = ca1 = node->calls_since_last_overflow; v0 = v1 = node->vectors_since_last_overflow; ca1 = ca0 + n_calls; v1 = v0 + n_vectors; cl1 = cl0 + n_clocks; node->calls_since_last_overflow = ca1; node->clocks_since_last_overflow = cl1; node->vectors_since_last_overflow = v1; node->max_clock_n = node->max_clock > n_clocks ? node->max_clock_n : n_vectors; node->max_clock = node->max_clock > n_clocks ? node->max_clock : n_clocks; r = vlib_node_runtime_update_main_loop_vector_stats (vm, node, n_vectors); if (PREDICT_FALSE (ca1 < ca0 || v1 < v0 || cl1 < cl0)) { node->calls_since_last_overflow = ca0; node->clocks_since_last_overflow = cl0; node->vectors_since_last_overflow = v0; vlib_node_runtime_sync_stats (vm, node, n_calls, n_vectors, n_clocks); } return r; } always_inline void vlib_process_update_stats (vlib_main_t * vm, vlib_process_t * p, uword n_calls, uword n_vectors, uword n_clocks) { vlib_node_runtime_update_stats (vm, &p->node_runtime, n_calls, n_vectors, n_clocks); } static clib_error_t * vlib_cli_elog_clear (vlib_main_t * vm, unformat_input_t * input, vlib_cli_command_t * cmd) { elog_reset_buffer (&vm->elog_main); return 0; } /* *INDENT-OFF* */ VLIB_CLI_COMMAND (elog_clear_cli, static) = { .path = "event-logger clear", .short_help = "Clear the event log", .function = vlib_cli_elog_clear, }; 
/*
 * Copyright (c) 2015 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.
 */
/*
  Copyright (c) 2001, 2002, 2003, 2005 Eliot Dresselhaus

  Permission is hereby granted, free of charge, to any person obtaining
  a copy of this software and associated documentation files (the
  "Software"), to deal in the Software without restriction, including
  without limitation the rights to use, copy, modify, merge, publish,
  distribute, sublicense, and/or sell copies of the Software, and to
  permit persons to whom the Software is furnished to do so, subject to
  the following conditions:

  The above copyright notice and this permission notice shall be
  included in all copies or substantial portions of the Software.

  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/

#include <vppinfra/bitmap.h>
#include <vppinfra/bitops.h>	/* for next_with_same_number_of_set_bits */
#include <vppinfra/error.h>	/* for ASSERT */
#include <vppinfra/mem.h>
#include <vppinfra/os.h>	/* for os_panic */
#include <vppinfra/vec.h>
#include <vppinfra/zvec.h>

/* Consider coding as bitmap, coding = 2^c_0 + 2^c_1 + ... + 2^c_n
   With c_0 < c_1 < ... < c_n.  coding == 0 represents c_n = BITS (uword).

   Unsigned integers i = 0 ... are represented as follows:

       0 <= i < 2^c_0       	(i << 1) | (1 << 0) binary:   i 1
   2^c_0 <= i < 2^c_0 + 2^c_1   (i << 2) | (1 << 1) binary: i 1 0
   ...                                              binary: i 0 ... 0

   Smaller numbers use less bits.  Coding is chosen so that encoding
   of given histogram of typical values gives smallest number of bits.
   The number and position of coding bits c_i are used to best fit the
   histogram of typical values.
*/

/* Decode given compressed data.  Return number of compressed data
   bits used. */
uword
zvec_decode (uword coding, uword zdata, uword * n_zdata_bits)
{
  uword c, d, result, n_bits;
  uword explicit_end, implicit_end;

  result = 0;
  n_bits = 0;
  while (1)
    {
      c = first_set (coding);
      implicit_end = c == coding;
      explicit_end = (zdata & 1) & ~implicit_end;
      d = (zdata >> explicit_end) & (c - 1);
      if (explicit_end | implicit_end)
	{
	  result += d;
	  n_bits += min_log2 (c) + explicit_end;
	  break;
	}
      n_bits += 1;
      result += c;
      coding ^= c;
      zdata >>= 1;
    }

  if (coding == 0)
    n_bits = BITS (uword);

  *n_zdata_bits = n_bits;
  return result;
}

uword
zvec_encode (uword coding, uword data, uword * n_result_bits)
{
  uword c, shift, result;
  uword explicit_end, implicit_end;

  /* Data must be in range.  Note special coding == 0
     would break for data - 1 <= coding. */
  ASSERT (data <= coding - 1);

  shift = 0;
  while (1)
    {
      c = first_set (coding);
      implicit_end = c == coding;
      explicit_end = ((data & (c - 1)) == data);
      if (explicit_end | implicit_end)
	{
	  uword t = explicit_end & ~implicit_end;
	  result = ((data << t) | t) << shift;
	  *n_result_bits =
	    /* data bits */ (c == 0 ? BITS (uword) : min_log2 (c))
	    /* shift bits */  + shift + t;
	  return result;
	}
      data -= c;
      coding ^= c;
      shift++;
    }

  /* Never reached. */
  ASSERT (0);
  return ~0;
}

always_inline uword
get_data (void *data, uword data_bytes, uword is_signed)
{
  if (data_bytes == 1)
    return is_signed ? zvec_signed_to_unsigned (*(i8 *) data) : *(u8 *) data;
  else if (data_bytes == 2)
    return is_signed ? zvec_signed_to_unsigned (*(i16 *) data) : *(u16 *)
      data;
  else if (data_bytes == 4)
    return is_signed ? zvec_signed_to_unsigned (*(i32 *) data) : *(u32 *)
      data;
  else if (data_bytes == 8)
    return is_signed ? zvec_signed_to_unsigned (*(i64 *) data) : *(u64 *)
      data;
  else
    {
      os_panic ();
      return ~0;
    }
}

always_inline void
put_data (void *data, uword data_bytes, uword is_signed, uword x)
{
  if (data_bytes == 1)
    {
      if (is_signed)
	*(i8 *) data = zvec_unsigned_to_signed (x);
      else
	*(u8 *) data = x;
    }
  else if (data_bytes == 2)
    {
      if (is_signed)
	*(i16 *) data = zvec_unsigned_to_signed (x);
      else
	*(u16 *) data = x;
    }
  else if (data_bytes == 4)
    {
      if (is_signed)
	*(i32 *) data = zvec_unsigned_to_signed (x);
      else
	*(u32 *) data = x;
    }
  else if (data_bytes == 8)
    {
      if (is_signed)
	*(i64 *) data = zvec_unsigned_to_signed (x);
      else
	*(u64 *) data = x;
    }
  else
    {
      os_panic ();
    }
}

always_inline uword *
zvec_encode_inline (uword * zvec,
		    uword * zvec_n_bits,
		    uword coding,
		    void *data,
		    uword data_stride,
		    uword n_data, uword data_bytes, uword is_signed)
{
  uword i;

  i = *zvec_n_bits;
  while (n_data >= 1)
    {
      uword d0, z0, l0;

      d0 = get_data (data + 0 * data_stride, data_bytes, is_signed);
      data += 1 * data_stride;
      n_data -= 1;

      z0 = zvec_encode (coding, d0, &l0);
      zvec = clib_bitmap_set_multiple (zvec, i, z0,