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("$FreeBSD$");
33 #include <sys/param.h>
35 #include <sys/kernel.h>
37 #include <sys/malloc.h>
38 #include <sys/memdesc.h>
39 #include <sys/mutex.h>
41 #include <sys/queue.h>
43 #include <sys/rwlock.h>
44 #include <sys/sched.h>
45 #include <sys/sf_buf.h>
46 #include <sys/sysctl.h>
47 #include <sys/systm.h>
48 #include <sys/taskqueue.h>
52 #include <dev/pci/pcivar.h>
54 #include <vm/vm_extern.h>
55 #include <vm/vm_kern.h>
56 #include <vm/vm_object.h>
57 #include <vm/vm_page.h>
58 #include <vm/vm_map.h>
59 #include <vm/vm_pageout.h>
60 #include <machine/bus.h>
61 #include <machine/cpu.h>
62 #include <machine/intr_machdep.h>
63 #include <x86/include/apicvar.h>
64 #include <x86/include/busdma_impl.h>
65 #include <x86/iommu/intel_reg.h>
66 #include <x86/iommu/busdma_dmar.h>
67 #include <x86/iommu/intel_dmar.h>
70 dmar_nd2mask(u_int nd)
72 static const u_int masks[] = {
80 0x0000, /* nd == 7 reserved */
83 KASSERT(nd <= 6, ("number of domains %d", nd));
87 static const struct sagaw_bits_tag {
93 {.agaw = 30, .cap = DMAR_CAP_SAGAW_2LVL, .awlvl = DMAR_CTX2_AW_2LVL,
95 {.agaw = 39, .cap = DMAR_CAP_SAGAW_3LVL, .awlvl = DMAR_CTX2_AW_3LVL,
97 {.agaw = 48, .cap = DMAR_CAP_SAGAW_4LVL, .awlvl = DMAR_CTX2_AW_4LVL,
99 {.agaw = 57, .cap = DMAR_CAP_SAGAW_5LVL, .awlvl = DMAR_CTX2_AW_5LVL,
101 {.agaw = 64, .cap = DMAR_CAP_SAGAW_6LVL, .awlvl = DMAR_CTX2_AW_6LVL,
106 dmar_pglvl_supported(struct dmar_unit *unit, int pglvl)
110 for (i = 0; i < nitems(sagaw_bits); i++) {
111 if (sagaw_bits[i].pglvl != pglvl)
113 if ((DMAR_CAP_SAGAW(unit->hw_cap) & sagaw_bits[i].cap) != 0)
120 domain_set_agaw(struct dmar_domain *domain, int mgaw)
125 sagaw = DMAR_CAP_SAGAW(domain->dmar->hw_cap);
126 for (i = 0; i < nitems(sagaw_bits); i++) {
127 if (sagaw_bits[i].agaw >= mgaw) {
128 domain->agaw = sagaw_bits[i].agaw;
129 domain->pglvl = sagaw_bits[i].pglvl;
130 domain->awlvl = sagaw_bits[i].awlvl;
134 device_printf(domain->dmar->dev,
135 "context request mgaw %d: no agaw found, sagaw %x\n",
141 * Find a best fit mgaw for the given maxaddr:
142 * - if allow_less is false, must find sagaw which maps all requested
143 * addresses (used by identity mappings);
144 * - if allow_less is true, and no supported sagaw can map all requested
145 * address space, accept the biggest sagaw, whatever is it.
148 dmar_maxaddr2mgaw(struct dmar_unit *unit, dmar_gaddr_t maxaddr, bool allow_less)
152 for (i = 0; i < nitems(sagaw_bits); i++) {
153 if ((1ULL << sagaw_bits[i].agaw) >= maxaddr &&
154 (DMAR_CAP_SAGAW(unit->hw_cap) & sagaw_bits[i].cap) != 0)
157 if (allow_less && i == nitems(sagaw_bits)) {
160 } while ((DMAR_CAP_SAGAW(unit->hw_cap) & sagaw_bits[i].cap)
163 if (i < nitems(sagaw_bits))
164 return (sagaw_bits[i].agaw);
165 KASSERT(0, ("no mgaw for maxaddr %jx allow_less %d",
166 (uintmax_t) maxaddr, allow_less));
171 * Calculate the total amount of page table pages needed to map the
172 * whole bus address space on the context with the selected agaw.
175 pglvl_max_pages(int pglvl)
180 for (res = 0, i = pglvl; i > 0; i--) {
188 * Return true if the page table level lvl supports the superpage for
192 domain_is_sp_lvl(struct dmar_domain *domain, int lvl)
195 static const int sagaw_sp[] = {
202 alvl = domain->pglvl - lvl - 1;
203 cap_sps = DMAR_CAP_SPS(domain->dmar->hw_cap);
204 return (alvl < nitems(sagaw_sp) && (sagaw_sp[alvl] & cap_sps) != 0);
208 pglvl_page_size(int total_pglvl, int lvl)
211 static const dmar_gaddr_t pg_sz[] = {
212 (dmar_gaddr_t)DMAR_PAGE_SIZE,
213 (dmar_gaddr_t)DMAR_PAGE_SIZE << DMAR_NPTEPGSHIFT,
214 (dmar_gaddr_t)DMAR_PAGE_SIZE << (2 * DMAR_NPTEPGSHIFT),
215 (dmar_gaddr_t)DMAR_PAGE_SIZE << (3 * DMAR_NPTEPGSHIFT),
216 (dmar_gaddr_t)DMAR_PAGE_SIZE << (4 * DMAR_NPTEPGSHIFT),
217 (dmar_gaddr_t)DMAR_PAGE_SIZE << (5 * DMAR_NPTEPGSHIFT)
220 KASSERT(lvl >= 0 && lvl < total_pglvl,
221 ("total %d lvl %d", total_pglvl, lvl));
222 rlvl = total_pglvl - lvl - 1;
223 KASSERT(rlvl < nitems(pg_sz), ("sizeof pg_sz lvl %d", lvl));
224 return (pg_sz[rlvl]);
228 domain_page_size(struct dmar_domain *domain, int lvl)
231 return (pglvl_page_size(domain->pglvl, lvl));
235 calc_am(struct dmar_unit *unit, dmar_gaddr_t base, dmar_gaddr_t size,
236 dmar_gaddr_t *isizep)
241 for (am = DMAR_CAP_MAMV(unit->hw_cap);; am--) {
242 isize = 1ULL << (am + DMAR_PAGE_SHIFT);
243 if ((base & (isize - 1)) == 0 && size >= isize)
252 dmar_haddr_t dmar_high;
254 int dmar_tbl_pagecnt;
257 dmar_pgalloc(vm_object_t obj, vm_pindex_t idx, int flags)
262 zeroed = (flags & DMAR_PGF_ZERO) != 0 ? VM_ALLOC_ZERO : 0;
263 aflags = zeroed | VM_ALLOC_NOBUSY | VM_ALLOC_SYSTEM | VM_ALLOC_NODUMP |
264 ((flags & DMAR_PGF_WAITOK) != 0 ? VM_ALLOC_WAITFAIL :
267 if ((flags & DMAR_PGF_OBJL) == 0)
268 VM_OBJECT_WLOCK(obj);
269 m = vm_page_lookup(obj, idx);
270 if ((flags & DMAR_PGF_NOALLOC) != 0 || m != NULL) {
271 if ((flags & DMAR_PGF_OBJL) == 0)
272 VM_OBJECT_WUNLOCK(obj);
275 m = vm_page_alloc_contig(obj, idx, aflags, 1, 0,
276 dmar_high, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
277 if ((flags & DMAR_PGF_OBJL) == 0)
278 VM_OBJECT_WUNLOCK(obj);
280 if (zeroed && (m->flags & PG_ZERO) == 0)
282 atomic_add_int(&dmar_tbl_pagecnt, 1);
285 if ((flags & DMAR_PGF_WAITOK) == 0)
292 dmar_pgfree(vm_object_t obj, vm_pindex_t idx, int flags)
296 if ((flags & DMAR_PGF_OBJL) == 0)
297 VM_OBJECT_WLOCK(obj);
298 m = vm_page_lookup(obj, idx);
301 atomic_subtract_int(&dmar_tbl_pagecnt, 1);
303 if ((flags & DMAR_PGF_OBJL) == 0)
304 VM_OBJECT_WUNLOCK(obj);
308 dmar_map_pgtbl(vm_object_t obj, vm_pindex_t idx, int flags,
314 if ((flags & DMAR_PGF_OBJL) == 0)
315 VM_OBJECT_WLOCK(obj);
316 m = vm_page_lookup(obj, idx);
317 if (m == NULL && (flags & DMAR_PGF_ALLOC) != 0) {
318 m = dmar_pgalloc(obj, idx, flags | DMAR_PGF_OBJL);
323 if ((flags & DMAR_PGF_OBJL) == 0)
324 VM_OBJECT_WUNLOCK(obj);
327 /* Sleepable allocations cannot fail. */
328 if ((flags & DMAR_PGF_WAITOK) != 0)
329 VM_OBJECT_WUNLOCK(obj);
331 *sf = sf_buf_alloc(m, SFB_CPUPRIVATE | ((flags & DMAR_PGF_WAITOK)
332 == 0 ? SFB_NOWAIT : 0));
336 VM_OBJECT_ASSERT_WLOCKED(obj);
337 dmar_pgfree(obj, m->pindex, flags | DMAR_PGF_OBJL);
339 if ((flags & DMAR_PGF_OBJL) == 0)
340 VM_OBJECT_WUNLOCK(obj);
343 if ((flags & (DMAR_PGF_WAITOK | DMAR_PGF_OBJL)) ==
344 (DMAR_PGF_WAITOK | DMAR_PGF_OBJL))
345 VM_OBJECT_WLOCK(obj);
346 else if ((flags & (DMAR_PGF_WAITOK | DMAR_PGF_OBJL)) == 0)
347 VM_OBJECT_WUNLOCK(obj);
348 return ((void *)sf_buf_kva(*sf));
352 dmar_unmap_pgtbl(struct sf_buf *sf)
360 dmar_flush_transl_to_ram(struct dmar_unit *unit, void *dst, size_t sz)
363 if (DMAR_IS_COHERENT(unit))
366 * If DMAR does not snoop paging structures accesses, flush
367 * CPU cache to memory.
369 pmap_invalidate_cache_range((uintptr_t)dst, (uintptr_t)dst + sz,
374 dmar_flush_pte_to_ram(struct dmar_unit *unit, dmar_pte_t *dst)
377 dmar_flush_transl_to_ram(unit, dst, sizeof(*dst));
381 dmar_flush_ctx_to_ram(struct dmar_unit *unit, dmar_ctx_entry_t *dst)
384 dmar_flush_transl_to_ram(unit, dst, sizeof(*dst));
388 dmar_flush_root_to_ram(struct dmar_unit *unit, dmar_root_entry_t *dst)
391 dmar_flush_transl_to_ram(unit, dst, sizeof(*dst));
395 * Load the root entry pointer into the hardware, busily waiting for
399 dmar_load_root_entry_ptr(struct dmar_unit *unit)
401 vm_page_t root_entry;
405 * Access to the GCMD register must be serialized while the
406 * command is submitted.
408 DMAR_ASSERT_LOCKED(unit);
410 VM_OBJECT_RLOCK(unit->ctx_obj);
411 root_entry = vm_page_lookup(unit->ctx_obj, 0);
412 VM_OBJECT_RUNLOCK(unit->ctx_obj);
413 dmar_write8(unit, DMAR_RTADDR_REG, VM_PAGE_TO_PHYS(root_entry));
414 dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd | DMAR_GCMD_SRTP);
415 DMAR_WAIT_UNTIL(((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_RTPS)
421 * Globally invalidate the context entries cache, busily waiting for
425 dmar_inv_ctx_glob(struct dmar_unit *unit)
430 * Access to the CCMD register must be serialized while the
431 * command is submitted.
433 DMAR_ASSERT_LOCKED(unit);
434 KASSERT(!unit->qi_enabled, ("QI enabled"));
437 * The DMAR_CCMD_ICC bit in the upper dword should be written
438 * after the low dword write is completed. Amd64
439 * dmar_write8() does not have this issue, i386 dmar_write8()
440 * writes the upper dword last.
442 dmar_write8(unit, DMAR_CCMD_REG, DMAR_CCMD_ICC | DMAR_CCMD_CIRG_GLOB);
443 DMAR_WAIT_UNTIL(((dmar_read4(unit, DMAR_CCMD_REG + 4) & DMAR_CCMD_ICC32)
449 * Globally invalidate the IOTLB, busily waiting for the completion.
452 dmar_inv_iotlb_glob(struct dmar_unit *unit)
456 DMAR_ASSERT_LOCKED(unit);
457 KASSERT(!unit->qi_enabled, ("QI enabled"));
459 reg = 16 * DMAR_ECAP_IRO(unit->hw_ecap);
460 /* See a comment about DMAR_CCMD_ICC in dmar_inv_ctx_glob. */
461 dmar_write8(unit, reg + DMAR_IOTLB_REG_OFF, DMAR_IOTLB_IVT |
462 DMAR_IOTLB_IIRG_GLB | DMAR_IOTLB_DR | DMAR_IOTLB_DW);
463 DMAR_WAIT_UNTIL(((dmar_read4(unit, reg + DMAR_IOTLB_REG_OFF + 4) &
464 DMAR_IOTLB_IVT32) == 0));
469 * Flush the chipset write buffers. See 11.1 "Write Buffer Flushing"
470 * in the architecture specification.
473 dmar_flush_write_bufs(struct dmar_unit *unit)
477 DMAR_ASSERT_LOCKED(unit);
480 * DMAR_GCMD_WBF is only valid when CAP_RWBF is reported.
482 KASSERT((unit->hw_cap & DMAR_CAP_RWBF) != 0,
483 ("dmar%d: no RWBF", unit->unit));
485 dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd | DMAR_GCMD_WBF);
486 DMAR_WAIT_UNTIL(((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_WBFS)
492 dmar_enable_translation(struct dmar_unit *unit)
496 DMAR_ASSERT_LOCKED(unit);
497 unit->hw_gcmd |= DMAR_GCMD_TE;
498 dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd);
499 DMAR_WAIT_UNTIL(((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_TES)
505 dmar_disable_translation(struct dmar_unit *unit)
509 DMAR_ASSERT_LOCKED(unit);
510 unit->hw_gcmd &= ~DMAR_GCMD_TE;
511 dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd);
512 DMAR_WAIT_UNTIL(((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_TES)
518 dmar_load_irt_ptr(struct dmar_unit *unit)
523 DMAR_ASSERT_LOCKED(unit);
524 irta = unit->irt_phys;
525 if (DMAR_X2APIC(unit))
526 irta |= DMAR_IRTA_EIME;
527 s = fls(unit->irte_cnt) - 2;
528 KASSERT(unit->irte_cnt >= 2 && s <= DMAR_IRTA_S_MASK &&
529 powerof2(unit->irte_cnt),
530 ("IRTA_REG_S overflow %x", unit->irte_cnt));
532 dmar_write8(unit, DMAR_IRTA_REG, irta);
533 dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd | DMAR_GCMD_SIRTP);
534 DMAR_WAIT_UNTIL(((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_IRTPS)
540 dmar_enable_ir(struct dmar_unit *unit)
544 DMAR_ASSERT_LOCKED(unit);
545 unit->hw_gcmd |= DMAR_GCMD_IRE;
546 unit->hw_gcmd &= ~DMAR_GCMD_CFI;
547 dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd);
548 DMAR_WAIT_UNTIL(((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_IRES)
554 dmar_disable_ir(struct dmar_unit *unit)
558 DMAR_ASSERT_LOCKED(unit);
559 unit->hw_gcmd &= ~DMAR_GCMD_IRE;
560 dmar_write4(unit, DMAR_GCMD_REG, unit->hw_gcmd);
561 DMAR_WAIT_UNTIL(((dmar_read4(unit, DMAR_GSTS_REG) & DMAR_GSTS_IRES)
567 u_int f_done, f_inproc, f_wakeup; \
569 f_done = 1 << (barrier_id * 3); \
570 f_inproc = 1 << (barrier_id * 3 + 1); \
571 f_wakeup = 1 << (barrier_id * 3 + 2)
574 dmar_barrier_enter(struct dmar_unit *dmar, u_int barrier_id)
579 if ((dmar->barrier_flags & f_done) != 0) {
584 if ((dmar->barrier_flags & f_inproc) != 0) {
585 while ((dmar->barrier_flags & f_inproc) != 0) {
586 dmar->barrier_flags |= f_wakeup;
587 msleep(&dmar->barrier_flags, &dmar->lock, 0,
590 KASSERT((dmar->barrier_flags & f_done) != 0,
591 ("dmar%d barrier %d missing done", dmar->unit, barrier_id));
596 dmar->barrier_flags |= f_inproc;
602 dmar_barrier_exit(struct dmar_unit *dmar, u_int barrier_id)
606 DMAR_ASSERT_LOCKED(dmar);
607 KASSERT((dmar->barrier_flags & (f_done | f_inproc)) == f_inproc,
608 ("dmar%d barrier %d missed entry", dmar->unit, barrier_id));
609 dmar->barrier_flags |= f_done;
610 if ((dmar->barrier_flags & f_wakeup) != 0)
611 wakeup(&dmar->barrier_flags);
612 dmar->barrier_flags &= ~(f_inproc | f_wakeup);
616 int dmar_match_verbose;
617 int dmar_batch_coalesce = 100;
618 struct timespec dmar_hw_timeout = {
623 static const uint64_t d = 1000000000;
626 dmar_update_timeout(uint64_t newval)
629 /* XXXKIB not atomic */
630 dmar_hw_timeout.tv_sec = newval / d;
631 dmar_hw_timeout.tv_nsec = newval % d;
635 dmar_get_timeout(void)
638 return ((uint64_t)dmar_hw_timeout.tv_sec * d +
639 dmar_hw_timeout.tv_nsec);
643 dmar_timeout_sysctl(SYSCTL_HANDLER_ARGS)
648 val = dmar_get_timeout();
649 error = sysctl_handle_long(oidp, &val, 0, req);
650 if (error != 0 || req->newptr == NULL)
652 dmar_update_timeout(val);
656 static SYSCTL_NODE(_hw, OID_AUTO, dmar, CTLFLAG_RD, NULL, "");
657 SYSCTL_INT(_hw_dmar, OID_AUTO, tbl_pagecnt, CTLFLAG_RD,
658 &dmar_tbl_pagecnt, 0,
659 "Count of pages used for DMAR pagetables");
660 SYSCTL_INT(_hw_dmar, OID_AUTO, match_verbose, CTLFLAG_RWTUN,
661 &dmar_match_verbose, 0,
662 "Verbose matching of the PCI devices to DMAR paths");
663 SYSCTL_INT(_hw_dmar, OID_AUTO, batch_coalesce, CTLFLAG_RWTUN,
664 &dmar_batch_coalesce, 0,
665 "Number of qi batches between interrupt");
666 SYSCTL_PROC(_hw_dmar, OID_AUTO, timeout,
667 CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, 0,
668 dmar_timeout_sysctl, "QU",
669 "Timeout for command wait, in nanoseconds");
672 SYSCTL_INT(_hw_dmar, OID_AUTO, check_free, CTLFLAG_RWTUN,
674 "Check the GPA RBtree for free_down and free_after validity");