1 /* SPDX-License-Identifier: BSD-3-Clause
5 #include <rte_memory.h>
7 #include "dpaax_iova_table.h"
8 #include "dpaax_logs.h"
10 /* Global dpaax logger identifier */
13 /* Global table reference */
14 struct dpaax_iova_table *dpaax_iova_table_p;
16 static int dpaax_handle_memevents(void);
18 /* A structure representing the device-tree node available in /proc/device-tree.
25 /* A ntohll equivalent routine
26 * XXX: This is only applicable for 64 bit environment.
29 rotate_8(unsigned char *arr)
33 uint32_t *second_half;
35 first_half = (uint32_t *)(arr);
36 second_half = (uint32_t *)(arr + 4);
39 *first_half = *second_half;
42 *first_half = ntohl(*first_half);
43 *second_half = ntohl(*second_half);
47 * Memory layout for DPAAx platforms (LS1043, LS1046, LS1088, LS2088, LX2160)
48 * are populated by Uboot and available in device tree:
49 * /proc/device-tree/memory@<address>/reg <= register.
50 * Entries are of the form:
51 * (<8 byte start addr><8 byte length>)(..more similar blocks of start,len>)..
54 * OUT populate number of entries found in memory node
56 * Pointer to array of reg_node elements, count size
58 static struct reg_node *
59 read_memory_node(unsigned int *count)
64 struct stat statbuf = {0};
65 char file_data[MEM_NODE_FILE_LEN];
66 struct reg_node *nodes = NULL;
70 ret = glob(MEM_NODE_PATH_GLOB, 0, NULL, &result);
72 DPAAX_ERR("Unable to glob device-tree memory node: (%s)(%d)",
73 MEM_NODE_PATH_GLOB, ret);
77 if (result.gl_pathc != 1) {
78 /* Either more than one memory@<addr> node found, or none.
79 * In either case, cannot work ahead.
81 DPAAX_ERR("Found (%zu) entries in device-tree. Not supported!",
86 DPAAX_DEBUG("Opening and parsing device-tree node: (%s)",
88 fd = open(result.gl_pathv[0], O_RDONLY);
90 DPAAX_ERR("Unable to open the device-tree node: (%s)(fd=%d)",
91 MEM_NODE_PATH_GLOB, fd);
95 /* Stat to get the file size */
96 ret = fstat(fd, &statbuf);
98 DPAAX_ERR("Unable to get device-tree memory node size.");
102 DPAAX_DEBUG("Size of device-tree mem node: %lu", statbuf.st_size);
103 if (statbuf.st_size > MEM_NODE_FILE_LEN) {
104 DPAAX_WARN("More memory nodes available than assumed.");
105 DPAAX_WARN("System may not work properly!");
108 ret = read(fd, file_data, statbuf.st_size > MEM_NODE_FILE_LEN ?
109 MEM_NODE_FILE_LEN : statbuf.st_size);
111 DPAAX_ERR("Unable to read device-tree memory node: (%d)", ret);
115 /* The reg node should be multiple of 16 bytes, 8 bytes each for addr
118 *count = (statbuf.st_size / 16);
119 if ((*count) <= 0 || (statbuf.st_size % 16 != 0)) {
120 DPAAX_ERR("Invalid memory node values or count. (size=%lu)",
125 /* each entry is of 16 bytes, and size/16 is total count of entries */
126 nodes = malloc(sizeof(struct reg_node) * (*count));
128 DPAAX_ERR("Failure in allocating working memory.");
131 memset(nodes, 0, sizeof(struct reg_node) * (*count));
133 for (i = 0, j = 0; i < (statbuf.st_size) && j < (*count); i += 16, j++) {
134 memcpy(&nodes[j], file_data + i, 16);
135 /* Rotate (ntohl) each 8 byte entry */
136 rotate_8((unsigned char *)(&(nodes[j].addr)));
137 rotate_8((unsigned char *)(&(nodes[j].len)));
140 DPAAX_DEBUG("Device-tree memory node data:");
142 DPAAX_DEBUG("\n %08" PRIx64 " %08zu", nodes[j].addr, nodes[j].len);
153 dpaax_iova_table_populate(void)
156 unsigned int i, node_count;
157 size_t tot_memory_size, total_table_size;
158 struct reg_node *nodes;
159 struct dpaax_iovat_element *entry;
161 /* dpaax_iova_table_p is a singleton - only one instance should be
164 if (dpaax_iova_table_p) {
165 DPAAX_DEBUG("Multiple allocation attempt for IOVA Table (%p)",
167 /* This can be an error case as well - some path not cleaning
168 * up table - but, for now, it is assumed that if IOVA Table
169 * pointer is valid, table is allocated.
174 nodes = read_memory_node(&node_count);
175 if (nodes == NULL || node_count <= 0) {
176 DPAAX_WARN("PA->VA translation not available;");
177 DPAAX_WARN("Expect performance impact.");
182 for (i = 0; i < node_count; i++)
183 tot_memory_size += nodes[i].len;
185 DPAAX_DEBUG("Total available PA memory size: %zu", tot_memory_size);
187 /* Total table size = meta data + tot_memory_size/8 */
188 total_table_size = sizeof(struct dpaax_iova_table) +
189 (sizeof(struct dpaax_iovat_element) * node_count) +
190 ((tot_memory_size / DPAAX_MEM_SPLIT) * sizeof(uint64_t));
192 /* TODO: This memory doesn't need to shared but needs to be always
193 * pinned to RAM (no swap out) - using hugepage rather than malloc
195 dpaax_iova_table_p = rte_zmalloc(NULL, total_table_size, 0);
196 if (dpaax_iova_table_p == NULL) {
197 DPAAX_WARN("Unable to allocate memory for PA->VA Table;");
198 DPAAX_WARN("PA->VA translation not available;");
199 DPAAX_WARN("Expect performance impact.");
204 /* Initialize table */
205 dpaax_iova_table_p->count = node_count;
206 entry = dpaax_iova_table_p->entries;
208 DPAAX_DEBUG("IOVA Table entries: (entry start = %p)", (void *)entry);
209 DPAAX_DEBUG("\t(entry),(start),(len),(next)");
211 for (i = 0; i < node_count; i++) {
212 /* dpaax_iova_table_p
213 * | dpaax_iova_table_p->entries
217 * +------+------+-------+---+----------+---------+---
218 * |iova_ |entry | entry | | pages | pages |
219 * |table | 1 | 2 |...| entry 1 | entry2 |
220 * +-----'+.-----+-------+---+;---------+;--------+---
222 * `~~~~~~|~~~~~>pages /
226 entry[i].start = nodes[i].addr;
227 entry[i].len = nodes[i].len;
229 entry[i].pages = entry[i-1].pages +
230 ((entry[i-1].len/DPAAX_MEM_SPLIT));
232 entry[i].pages = (uint64_t *)((unsigned char *)entry +
233 (sizeof(struct dpaax_iovat_element) *
236 DPAAX_DEBUG("\t(%u),(%8"PRIx64"),(%8zu),(%8p)",
237 i, entry[i].start, entry[i].len, entry[i].pages);
240 /* Release memory associated with nodes array - not required now */
243 DPAAX_DEBUG("Adding mem-event handler\n");
244 ret = dpaax_handle_memevents();
246 DPAAX_ERR("Unable to add mem-event handler");
247 DPAAX_WARN("Cases with non-buffer pool mem won't work!");
254 dpaax_iova_table_depopulate(void)
256 if (dpaax_iova_table_p == NULL)
259 rte_free(dpaax_iova_table_p->entries);
260 dpaax_iova_table_p = NULL;
262 DPAAX_DEBUG("IOVA Table cleanedup");
266 dpaax_iova_table_update(phys_addr_t paddr, void *vaddr, size_t length)
270 size_t req_length = length, e_offset;
271 struct dpaax_iovat_element *entry;
272 uintptr_t align_vaddr;
273 phys_addr_t align_paddr;
275 align_paddr = paddr & DPAAX_MEM_SPLIT_MASK;
276 align_vaddr = ((uintptr_t)vaddr & DPAAX_MEM_SPLIT_MASK);
278 /* Check if paddr is available in table */
279 entry = dpaax_iova_table_p->entries;
280 for (i = 0; i < dpaax_iova_table_p->count; i++) {
281 if (align_paddr < entry[i].start) {
282 /* Address lower than start, but not found in previous
283 * iteration shouldn't exist.
285 DPAAX_ERR("Add: Incorrect entry for PA->VA Table"
286 "(%"PRIu64")", paddr);
287 DPAAX_ERR("Add: Lowest address: %"PRIu64"",
292 if (align_paddr > (entry[i].start + entry[i].len))
295 /* align_paddr >= start && align_paddr < (start + len) */
299 e_offset = ((align_paddr - entry[i].start) / DPAAX_MEM_SPLIT);
300 /* TODO: Whatif something already exists at this
301 * location - is that an error? For now, ignoring the
304 entry[i].pages[e_offset] = align_vaddr;
305 DPAAX_DEBUG("Added: vaddr=%zu for Phy:%"PRIu64" at %zu"
306 " remaining len %zu", align_vaddr,
307 align_paddr, e_offset, req_length);
309 /* Incoming request can be larger than the
310 * DPAAX_MEM_SPLIT size - in which case, multiple
311 * entries in entry->pages[] are filled up.
313 if (req_length <= DPAAX_MEM_SPLIT)
315 align_paddr += DPAAX_MEM_SPLIT;
316 align_vaddr += DPAAX_MEM_SPLIT;
317 req_length -= DPAAX_MEM_SPLIT;
324 /* There might be case where the incoming physical address is
325 * beyond the address discovered in the memory node of
326 * device-tree. Specially if some malloc'd area is used by EAL
327 * and the memevent handlers passes that across. But, this is
328 * not necessarily an error.
330 DPAAX_DEBUG("Add: Unable to find slot for vaddr:(%p),"
336 DPAAX_DEBUG("Add: Found slot at (%"PRIu64")[(%zu)] for vaddr:(%p),"
337 " phy(%"PRIu64"), len(%zu)", entry[i].start, e_offset,
338 vaddr, paddr, length);
342 /* dpaax_iova_table_dump
343 * Dump the table, with its entries, on screen. Only works in Debug Mode
344 * Not for weak hearted - the tables can get quite large
347 dpaax_iova_table_dump(void)
350 struct dpaax_iovat_element *entry;
352 /* In case DEBUG is not enabled, some 'if' conditions might misbehave
353 * as they have nothing else in them except a DPAAX_DEBUG() which if
354 * tuned out would leave 'if' naked.
356 if (rte_log_get_global_level() < RTE_LOG_DEBUG) {
357 DPAAX_ERR("Set log level to Debug for PA->Table dump!");
361 DPAAX_DEBUG(" === Start of PA->VA Translation Table ===");
362 if (dpaax_iova_table_p == NULL)
363 DPAAX_DEBUG("\tNULL");
365 entry = dpaax_iova_table_p->entries;
366 for (i = 0; i < dpaax_iova_table_p->count; i++) {
367 DPAAX_DEBUG("\t(%16i),(%16"PRIu64"),(%16zu),(%16p)",
368 i, entry[i].start, entry[i].len, entry[i].pages);
369 DPAAX_DEBUG("\t\t (PA), (VA)");
370 for (j = 0; j < (entry->len/DPAAX_MEM_SPLIT); j++) {
371 if (entry[i].pages[j] == 0)
373 DPAAX_DEBUG("\t\t(%16"PRIx64"),(%16"PRIx64")",
374 (entry[i].start + (j * sizeof(uint64_t))),
378 DPAAX_DEBUG(" === End of PA->VA Translation Table ===");
382 dpaax_memevent_cb(enum rte_mem_event type, const void *addr, size_t len,
383 void *arg __rte_unused)
385 struct rte_memseg_list *msl;
386 struct rte_memseg *ms;
387 size_t cur_len = 0, map_len = 0;
388 phys_addr_t phys_addr;
392 DPAAX_DEBUG("Called with addr=%p, len=%zu", addr, len);
394 msl = rte_mem_virt2memseg_list(addr);
396 while (cur_len < len) {
397 const void *va = RTE_PTR_ADD(addr, cur_len);
399 ms = rte_mem_virt2memseg(va, msl);
400 phys_addr = rte_mem_virt2phy(ms->addr);
401 virt_addr = ms->addr;
404 DPAAX_DEBUG("Request for %s, va=%p, virt_addr=%p,"
405 "iova=%"PRIu64", map_len=%zu",
406 type == RTE_MEM_EVENT_ALLOC ?
408 va, virt_addr, phys_addr, map_len);
410 if (type == RTE_MEM_EVENT_ALLOC)
411 ret = dpaax_iova_table_update(phys_addr, virt_addr,
414 /* In case of mem_events for MEM_EVENT_FREE, complete
415 * hugepage is released and its PA entry is set to 0.
417 ret = dpaax_iova_table_update(phys_addr, 0, map_len);
420 DPAAX_ERR("PA-Table entry update failed. "
421 "Map=%d, addr=%p, len=%zu, err:(%d)",
422 type, va, map_len, ret);
431 dpaax_memevent_walk_memsegs(const struct rte_memseg_list *msl __rte_unused,
432 const struct rte_memseg *ms, size_t len,
433 void *arg __rte_unused)
435 DPAAX_DEBUG("Walking for %p (pa=%"PRIu64") and len %zu",
436 ms->addr, ms->phys_addr, len);
437 dpaax_iova_table_update(rte_mem_virt2phy(ms->addr), ms->addr, len);
442 dpaax_handle_memevents(void)
444 /* First, walk through all memsegs and pin them, before installing
445 * handler. This assures that all memseg which have already been
446 * identified/allocated by EAL, are already part of PA->VA Table. This
447 * is especially for cases where application allocates memory before
448 * the EAL or this is an externally allocated memory passed to EAL.
450 rte_memseg_contig_walk_thread_unsafe(dpaax_memevent_walk_memsegs, NULL);
452 return rte_mem_event_callback_register("dpaax_memevents_cb",
453 dpaax_memevent_cb, NULL);
458 dpaax_logger = rte_log_register("pmd.common.dpaax");
459 if (dpaax_logger >= 0)
460 rte_log_set_level(dpaax_logger, RTE_LOG_NOTICE);