2 * Copyright (c) 2013 The FreeBSD Foundation
5 * This software was developed by Konstantin Belousov <kib@FreeBSD.org>
6 * under sponsorship from the FreeBSD Foundation.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
35 #include <sys/param.h>
37 #include <sys/kernel.h>
38 #include <sys/malloc.h>
39 #include <sys/memdesc.h>
40 #include <sys/module.h>
42 #include <sys/taskqueue.h>
46 #include <machine/bus.h>
47 #include <contrib/dev/acpica/include/acpi.h>
48 #include <contrib/dev/acpica/include/accommon.h>
49 #include <dev/acpica/acpivar.h>
51 #include <vm/vm_extern.h>
52 #include <vm/vm_kern.h>
53 #include <vm/vm_page.h>
54 #include <vm/vm_map.h>
55 #include <machine/cpu.h>
56 #include <x86/include/busdma_impl.h>
57 #include <x86/iommu/intel_reg.h>
58 #include <x86/iommu/busdma_dmar.h>
59 #include <x86/iommu/intel_dmar.h>
62 dmar_qi_seq_processed(const struct dmar_unit *unit,
63 const struct dmar_qi_genseq *pseq)
66 return (pseq->gen < unit->inv_waitd_gen ||
67 (pseq->gen == unit->inv_waitd_gen &&
68 pseq->seq <= unit->inv_waitd_seq_hw));
72 dmar_enable_qi(struct dmar_unit *unit)
76 DMAR_ASSERT_LOCKED(unit);
77 unit->hw_gcmd |= DMAR_GCMD_QIE;
78 dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd);
79 DMAR_WAIT_UNTIL(((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_QIES)
85 dmar_disable_qi(struct dmar_unit *unit)
89 DMAR_ASSERT_LOCKED(unit);
90 unit->hw_gcmd &= ~DMAR_GCMD_QIE;
91 dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd);
92 DMAR_WAIT_UNTIL(((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_QIES)
98 dmar_qi_advance_tail(struct dmar_unit *unit)
101 DMAR_ASSERT_LOCKED(unit);
102 dmar_write4(unit, DMAR_IQT_REG, unit->inv_queue_tail);
106 dmar_qi_ensure(struct dmar_unit *unit, int descr_count)
111 DMAR_ASSERT_LOCKED(unit);
112 bytes = descr_count << DMAR_IQ_DESCR_SZ_SHIFT;
114 if (bytes <= unit->inv_queue_avail)
117 head = dmar_read4(unit, DMAR_IQH_REG);
118 head &= DMAR_IQH_MASK;
119 unit->inv_queue_avail = head - unit->inv_queue_tail -
121 if (head <= unit->inv_queue_tail)
122 unit->inv_queue_avail += unit->inv_queue_size;
123 if (bytes <= unit->inv_queue_avail)
127 * No space in the queue, do busy wait. Hardware must
128 * make a progress. But first advance the tail to
129 * inform the descriptor streamer about entries we
130 * might have already filled, otherwise they could
131 * clog the whole queue..
133 dmar_qi_advance_tail(unit);
134 unit->inv_queue_full++;
137 unit->inv_queue_avail -= bytes;
141 dmar_qi_emit(struct dmar_unit *unit, uint64_t data1, uint64_t data2)
144 DMAR_ASSERT_LOCKED(unit);
145 *(volatile uint64_t *)(unit->inv_queue + unit->inv_queue_tail) = data1;
146 unit->inv_queue_tail += DMAR_IQ_DESCR_SZ / 2;
147 KASSERT(unit->inv_queue_tail <= unit->inv_queue_size,
148 ("tail overflow 0x%x 0x%jx", unit->inv_queue_tail,
149 (uintmax_t)unit->inv_queue_size));
150 unit->inv_queue_tail &= unit->inv_queue_size - 1;
151 *(volatile uint64_t *)(unit->inv_queue + unit->inv_queue_tail) = data2;
152 unit->inv_queue_tail += DMAR_IQ_DESCR_SZ / 2;
153 KASSERT(unit->inv_queue_tail <= unit->inv_queue_size,
154 ("tail overflow 0x%x 0x%jx", unit->inv_queue_tail,
155 (uintmax_t)unit->inv_queue_size));
156 unit->inv_queue_tail &= unit->inv_queue_size - 1;
160 dmar_qi_emit_wait_descr(struct dmar_unit *unit, uint32_t seq, bool intr,
161 bool memw, bool fence)
164 DMAR_ASSERT_LOCKED(unit);
165 dmar_qi_emit(unit, DMAR_IQ_DESCR_WAIT_ID |
166 (intr ? DMAR_IQ_DESCR_WAIT_IF : 0) |
167 (memw ? DMAR_IQ_DESCR_WAIT_SW : 0) |
168 (fence ? DMAR_IQ_DESCR_WAIT_FN : 0) |
169 (memw ? DMAR_IQ_DESCR_WAIT_SD(seq) : 0),
170 memw ? unit->inv_waitd_seq_hw_phys : 0);
174 dmar_qi_emit_wait_seq(struct dmar_unit *unit, struct dmar_qi_genseq *pseq,
177 struct dmar_qi_genseq gsec;
180 KASSERT(pseq != NULL, ("wait descriptor with no place for seq"));
181 DMAR_ASSERT_LOCKED(unit);
182 if (unit->inv_waitd_seq == 0xffffffff) {
183 gsec.gen = unit->inv_waitd_gen;
184 gsec.seq = unit->inv_waitd_seq;
185 dmar_qi_ensure(unit, 1);
186 dmar_qi_emit_wait_descr(unit, gsec.seq, false, true, false);
187 dmar_qi_advance_tail(unit);
188 while (!dmar_qi_seq_processed(unit, &gsec))
190 unit->inv_waitd_gen++;
191 unit->inv_waitd_seq = 1;
193 seq = unit->inv_waitd_seq++;
194 pseq->gen = unit->inv_waitd_gen;
197 dmar_qi_ensure(unit, 1);
198 dmar_qi_emit_wait_descr(unit, seq, true, true, false);
203 dmar_qi_wait_for_seq(struct dmar_unit *unit, const struct dmar_qi_genseq *gseq,
207 DMAR_ASSERT_LOCKED(unit);
208 unit->inv_seq_waiters++;
209 while (!dmar_qi_seq_processed(unit, gseq)) {
210 if (cold || nowait) {
213 msleep(&unit->inv_seq_waiters, &unit->lock, 0,
217 unit->inv_seq_waiters--;
221 dmar_qi_invalidate_locked(struct dmar_domain *domain, dmar_gaddr_t base,
222 dmar_gaddr_t size, struct dmar_qi_genseq *pseq, bool emit_wait)
224 struct dmar_unit *unit;
229 DMAR_ASSERT_LOCKED(unit);
230 for (; size > 0; base += isize, size -= isize) {
231 am = calc_am(unit, base, size, &isize);
232 dmar_qi_ensure(unit, 1);
233 dmar_qi_emit(unit, DMAR_IQ_DESCR_IOTLB_INV |
234 DMAR_IQ_DESCR_IOTLB_PAGE | DMAR_IQ_DESCR_IOTLB_DW |
235 DMAR_IQ_DESCR_IOTLB_DR |
236 DMAR_IQ_DESCR_IOTLB_DID(domain->domain),
239 dmar_qi_emit_wait_seq(unit, pseq, emit_wait);
240 dmar_qi_advance_tail(unit);
244 dmar_qi_invalidate_ctx_glob_locked(struct dmar_unit *unit)
246 struct dmar_qi_genseq gseq;
248 DMAR_ASSERT_LOCKED(unit);
249 dmar_qi_ensure(unit, 2);
250 dmar_qi_emit(unit, DMAR_IQ_DESCR_CTX_INV | DMAR_IQ_DESCR_CTX_GLOB, 0);
251 dmar_qi_emit_wait_seq(unit, &gseq, true);
252 dmar_qi_advance_tail(unit);
253 dmar_qi_wait_for_seq(unit, &gseq, false);
257 dmar_qi_invalidate_iotlb_glob_locked(struct dmar_unit *unit)
259 struct dmar_qi_genseq gseq;
261 DMAR_ASSERT_LOCKED(unit);
262 dmar_qi_ensure(unit, 2);
263 dmar_qi_emit(unit, DMAR_IQ_DESCR_IOTLB_INV | DMAR_IQ_DESCR_IOTLB_GLOB |
264 DMAR_IQ_DESCR_IOTLB_DW | DMAR_IQ_DESCR_IOTLB_DR, 0);
265 dmar_qi_emit_wait_seq(unit, &gseq, true);
266 dmar_qi_advance_tail(unit);
267 dmar_qi_wait_for_seq(unit, &gseq, false);
271 dmar_qi_invalidate_iec_glob(struct dmar_unit *unit)
273 struct dmar_qi_genseq gseq;
275 DMAR_ASSERT_LOCKED(unit);
276 dmar_qi_ensure(unit, 2);
277 dmar_qi_emit(unit, DMAR_IQ_DESCR_IEC_INV, 0);
278 dmar_qi_emit_wait_seq(unit, &gseq, true);
279 dmar_qi_advance_tail(unit);
280 dmar_qi_wait_for_seq(unit, &gseq, false);
284 dmar_qi_invalidate_iec(struct dmar_unit *unit, u_int start, u_int cnt)
286 struct dmar_qi_genseq gseq;
289 DMAR_ASSERT_LOCKED(unit);
290 KASSERT(start < unit->irte_cnt && start < start + cnt &&
291 start + cnt <= unit->irte_cnt,
292 ("inv iec overflow %d %d %d", unit->irte_cnt, start, cnt));
293 for (; cnt > 0; cnt -= c, start += c) {
294 l = ffs(start | cnt) - 1;
296 dmar_qi_ensure(unit, 1);
297 dmar_qi_emit(unit, DMAR_IQ_DESCR_IEC_INV |
298 DMAR_IQ_DESCR_IEC_IDX | DMAR_IQ_DESCR_IEC_IIDX(start) |
299 DMAR_IQ_DESCR_IEC_IM(l), 0);
301 dmar_qi_ensure(unit, 1);
302 dmar_qi_emit_wait_seq(unit, &gseq, true);
303 dmar_qi_advance_tail(unit);
306 * The caller of the function, in particular,
307 * dmar_ir_program_irte(), may be called from the context
308 * where the sleeping is forbidden (in fact, the
309 * intr_table_lock mutex may be held, locked from
310 * intr_shuffle_irqs()). Wait for the invalidation completion
311 * using the busy wait.
313 * The impact on the interrupt input setup code is small, the
314 * expected overhead is comparable with the chipset register
315 * read. It is more harmful for the parallel DMA operations,
316 * since we own the dmar unit lock until whole invalidation
317 * queue is processed, which includes requests possibly issued
318 * before our request.
320 dmar_qi_wait_for_seq(unit, &gseq, true);
324 dmar_qi_intr(void *arg)
326 struct dmar_unit *unit;
329 KASSERT(unit->qi_enabled, ("dmar%d: QI is not enabled", unit->unit));
330 taskqueue_enqueue(unit->qi_taskqueue, &unit->qi_task);
331 return (FILTER_HANDLED);
335 dmar_qi_task(void *arg, int pending __unused)
337 struct dmar_unit *unit;
338 struct dmar_map_entry *entry;
345 entry = TAILQ_FIRST(&unit->tlb_flush_entries);
348 if (!dmar_qi_seq_processed(unit, &entry->gseq))
350 TAILQ_REMOVE(&unit->tlb_flush_entries, entry, dmamap_link);
352 dmar_domain_free_entry(entry, (entry->flags &
353 DMAR_MAP_ENTRY_QI_NF) == 0);
356 ics = dmar_read4(unit, DMAR_ICS_REG);
357 if ((ics & DMAR_ICS_IWC) != 0) {
359 dmar_write4(unit, DMAR_ICS_REG, ics);
361 if (unit->inv_seq_waiters > 0)
362 wakeup(&unit->inv_seq_waiters);
367 dmar_init_qi(struct dmar_unit *unit)
373 if (!DMAR_HAS_QI(unit) || (unit->hw_cap & DMAR_CAP_CM) != 0)
375 unit->qi_enabled = 1;
376 TUNABLE_INT_FETCH("hw.dmar.qi", &unit->qi_enabled);
377 if (!unit->qi_enabled)
380 TAILQ_INIT(&unit->tlb_flush_entries);
381 TASK_INIT(&unit->qi_task, 0, dmar_qi_task, unit);
382 unit->qi_taskqueue = taskqueue_create_fast("dmarqf", M_WAITOK,
383 taskqueue_thread_enqueue, &unit->qi_taskqueue);
384 taskqueue_start_threads(&unit->qi_taskqueue, 1, PI_AV,
385 "dmar%d qi taskq", unit->unit);
387 unit->inv_waitd_gen = 0;
388 unit->inv_waitd_seq = 1;
390 qi_sz = DMAR_IQA_QS_DEF;
391 TUNABLE_INT_FETCH("hw.dmar.qi_size", &qi_sz);
392 if (qi_sz > DMAR_IQA_QS_MAX)
393 qi_sz = DMAR_IQA_QS_MAX;
394 unit->inv_queue_size = (1ULL << qi_sz) * PAGE_SIZE;
395 /* Reserve one descriptor to prevent wraparound. */
396 unit->inv_queue_avail = unit->inv_queue_size - DMAR_IQ_DESCR_SZ;
398 /* The invalidation queue reads by DMARs are always coherent. */
399 unit->inv_queue = kmem_alloc_contig(kernel_arena, unit->inv_queue_size,
400 M_WAITOK | M_ZERO, 0, dmar_high, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
401 unit->inv_waitd_seq_hw_phys = pmap_kextract(
402 (vm_offset_t)&unit->inv_waitd_seq_hw);
405 dmar_write8(unit, DMAR_IQT_REG, 0);
406 iqa = pmap_kextract(unit->inv_queue);
408 dmar_write8(unit, DMAR_IQA_REG, iqa);
409 dmar_enable_qi(unit);
410 ics = dmar_read4(unit, DMAR_ICS_REG);
411 if ((ics & DMAR_ICS_IWC) != 0) {
413 dmar_write4(unit, DMAR_ICS_REG, ics);
415 dmar_enable_qi_intr(unit);
422 dmar_fini_qi(struct dmar_unit *unit)
424 struct dmar_qi_genseq gseq;
426 if (unit->qi_enabled)
428 taskqueue_drain(unit->qi_taskqueue, &unit->qi_task);
429 taskqueue_free(unit->qi_taskqueue);
430 unit->qi_taskqueue = NULL;
434 dmar_qi_ensure(unit, 1);
435 dmar_qi_emit_wait_seq(unit, &gseq, true);
436 dmar_qi_advance_tail(unit);
437 dmar_qi_wait_for_seq(unit, &gseq, false);
438 /* only after the quisce, disable queue */
439 dmar_disable_qi_intr(unit);
440 dmar_disable_qi(unit);
441 KASSERT(unit->inv_seq_waiters == 0,
442 ("dmar%d: waiters on disabled queue", unit->unit));
445 kmem_free(kernel_arena, unit->inv_queue, unit->inv_queue_size);
447 unit->inv_queue_size = 0;
448 unit->qi_enabled = 0;
452 dmar_enable_qi_intr(struct dmar_unit *unit)
456 DMAR_ASSERT_LOCKED(unit);
457 KASSERT(DMAR_HAS_QI(unit), ("dmar%d: QI is not supported", unit->unit));
458 iectl = dmar_read4(unit, DMAR_IECTL_REG);
459 iectl &= ~DMAR_IECTL_IM;
460 dmar_write4(unit, DMAR_IECTL_REG, iectl);
464 dmar_disable_qi_intr(struct dmar_unit *unit)
468 DMAR_ASSERT_LOCKED(unit);
469 KASSERT(DMAR_HAS_QI(unit), ("dmar%d: QI is not supported", unit->unit));
470 iectl = dmar_read4(unit, DMAR_IECTL_REG);
471 dmar_write4(unit, DMAR_IECTL_REG, iectl | DMAR_IECTL_IM);