Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 9 Nov 2024 20:19:15 GMT
From:      Konstantin Belousov <kib@FreeBSD.org>
To:        src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-branches@FreeBSD.org
Subject:   git: fa6787221742 - stable/14 - AMD IOMMU driver
Message-ID:  <202411092019.4A9KJFgD033481@gitrepo.freebsd.org>

next in thread | raw e-mail | index | archive | help
The branch stable/14 has been updated by kib:

URL: https://cgit.FreeBSD.org/src/commit/?id=fa6787221742a00d410920a4a917bf2f9b1ed192

commit fa6787221742a00d410920a4a917bf2f9b1ed192
Author:     Konstantin Belousov <kib@FreeBSD.org>
AuthorDate: 2024-05-12 10:20:11 +0000
Commit:     Konstantin Belousov <kib@FreeBSD.org>
CommitDate: 2024-11-09 19:04:18 +0000

    AMD IOMMU driver
    
    (cherry picked from commit 0f5116d7efe33c81f0b24b56eec78af37898f500)
---
 sys/conf/files.x86          |    6 +
 sys/x86/iommu/amd_cmd.c     |  360 +++++++++++++
 sys/x86/iommu/amd_ctx.c     |  639 +++++++++++++++++++++++
 sys/x86/iommu/amd_drv.c     | 1205 +++++++++++++++++++++++++++++++++++++++++++
 sys/x86/iommu/amd_event.c   |  323 ++++++++++++
 sys/x86/iommu/amd_idpgtbl.c |  396 ++++++++++++++
 sys/x86/iommu/amd_intrmap.c |  391 ++++++++++++++
 sys/x86/iommu/amd_iommu.h   |  243 +++++++++
 8 files changed, 3563 insertions(+)

diff --git a/sys/conf/files.x86 b/sys/conf/files.x86
index c6d705e9715d..33da95a65ba4 100644
--- a/sys/conf/files.x86
+++ b/sys/conf/files.x86
@@ -344,6 +344,12 @@ x86/cpufreq/hwpstate_amd.c	optional	cpufreq
 x86/cpufreq/hwpstate_intel.c	optional	cpufreq
 x86/cpufreq/p4tcc.c		optional	cpufreq
 x86/cpufreq/powernow.c		optional	cpufreq
+x86/iommu/amd_cmd.c		optional	acpi iommu pci
+x86/iommu/amd_ctx.c		optional	acpi iommu pci
+x86/iommu/amd_drv.c		optional	acpi iommu pci
+x86/iommu/amd_event.c		optional	acpi iommu pci
+x86/iommu/amd_idpgtbl.c		optional	acpi iommu pci
+x86/iommu/amd_intrmap.c		optional	acpi iommu pci
 x86/iommu/intel_ctx.c		optional	acpi iommu pci
 x86/iommu/intel_drv.c		optional	acpi iommu pci
 x86/iommu/intel_fault.c		optional	acpi iommu pci
diff --git a/sys/x86/iommu/amd_cmd.c b/sys/x86/iommu/amd_cmd.c
new file mode 100644
index 000000000000..bbc2a8e0ad9f
--- /dev/null
+++ b/sys/x86/iommu/amd_cmd.c
@@ -0,0 +1,360 @@
+/*-
+ * SPDX-License-Identifier: BSD-2-Clause
+ *
+ * Copyright (c) 2024 The FreeBSD Foundation
+ *
+ * This software was developed by Konstantin Belousov <kib@FreeBSD.org>
+ * under sponsorship from the FreeBSD Foundation.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include "opt_acpi.h"
+
+#include <sys/param.h>
+#include <sys/bus.h>
+#include <sys/kernel.h>
+#include <sys/malloc.h>
+#include <sys/memdesc.h>
+#include <sys/module.h>
+#include <sys/rman.h>
+#include <sys/taskqueue.h>
+#include <sys/time.h>
+#include <sys/tree.h>
+#include <sys/vmem.h>
+#include <vm/vm.h>
+#include <vm/vm_extern.h>
+#include <vm/vm_kern.h>
+#include <vm/vm_page.h>
+#include <vm/vm_map.h>
+#include <contrib/dev/acpica/include/acpi.h>
+#include <contrib/dev/acpica/include/accommon.h>
+#include <dev/acpica/acpivar.h>
+#include <dev/pci/pcireg.h>
+#include <machine/bus.h>
+#include <machine/cpu.h>
+#include <x86/include/busdma_impl.h>
+#include <dev/iommu/busdma_iommu.h>
+#include <x86/iommu/amd_reg.h>
+#include <x86/iommu/x86_iommu.h>
+#include <x86/iommu/amd_iommu.h>
+
+static void
+amdiommu_enable_cmdbuf(struct amdiommu_unit *unit)
+{
+	AMDIOMMU_ASSERT_LOCKED(unit);
+
+	unit->hw_ctrl |= AMDIOMMU_CTRL_CMDBUF_EN;
+	amdiommu_write8(unit, AMDIOMMU_CTRL, unit->hw_ctrl);
+}
+
+static void
+amdiommu_disable_cmdbuf(struct amdiommu_unit *unit)
+{
+	AMDIOMMU_ASSERT_LOCKED(unit);
+
+	unit->hw_ctrl &= ~AMDIOMMU_CTRL_CMDBUF_EN;
+	amdiommu_write8(unit, AMDIOMMU_CTRL, unit->hw_ctrl);
+}
+
+
+static void
+amdiommu_enable_qi_intr(struct iommu_unit *iommu)
+{
+	struct amdiommu_unit *unit;
+
+	unit = IOMMU2AMD(iommu);
+	AMDIOMMU_ASSERT_LOCKED(unit);
+	unit->hw_ctrl |= AMDIOMMU_CTRL_COMWINT_EN;
+	amdiommu_write8(unit, AMDIOMMU_CTRL, unit->hw_ctrl);
+	amdiommu_write8(unit, AMDIOMMU_CMDEV_STATUS,
+	    AMDIOMMU_CMDEVS_COMWAITINT);
+}
+
+static void
+amdiommu_disable_qi_intr(struct iommu_unit *iommu)
+{
+	struct amdiommu_unit *unit;
+
+	unit = IOMMU2AMD(iommu);
+	AMDIOMMU_ASSERT_LOCKED(unit);
+	unit->hw_ctrl &= ~AMDIOMMU_CTRL_COMWINT_EN;
+	amdiommu_write8(unit, AMDIOMMU_CTRL, unit->hw_ctrl);
+}
+
+static void
+amdiommu_cmd_advance_tail(struct iommu_unit *iommu)
+{
+	struct amdiommu_unit *unit;
+
+	unit = IOMMU2AMD(iommu);
+	AMDIOMMU_ASSERT_LOCKED(unit);
+	amdiommu_write8(unit, AMDIOMMU_CMDBUF_TAIL, unit->x86c.inv_queue_tail);
+}
+
+static void
+amdiommu_cmd_ensure(struct iommu_unit *iommu, int descr_count)
+{
+	struct amdiommu_unit *unit;
+	uint64_t head;
+	int bytes;
+
+	unit = IOMMU2AMD(iommu);
+	AMDIOMMU_ASSERT_LOCKED(unit);
+	bytes = descr_count << AMDIOMMU_CMD_SZ_SHIFT;
+	for (;;) {
+		if (bytes <= unit->x86c.inv_queue_avail)
+			break;
+		/* refill */
+		head = amdiommu_read8(unit, AMDIOMMU_CMDBUF_HEAD);
+		head &= AMDIOMMU_CMDPTR_MASK;
+		unit->x86c.inv_queue_avail = head - unit->x86c.inv_queue_tail -
+		    AMDIOMMU_CMD_SZ;
+		if (head <= unit->x86c.inv_queue_tail)
+			unit->x86c.inv_queue_avail += unit->x86c.inv_queue_size;
+		if (bytes <= unit->x86c.inv_queue_avail)
+			break;
+
+		/*
+		 * No space in the queue, do busy wait.  Hardware must
+		 * make a progress.  But first advance the tail to
+		 * inform the descriptor streamer about entries we
+		 * might have already filled, otherwise they could
+		 * clog the whole queue..
+		 *
+		 * See dmar_qi_invalidate_locked() for a discussion
+		 * about data race prevention.
+		 */
+		amdiommu_cmd_advance_tail(iommu);
+		unit->x86c.inv_queue_full++;
+		cpu_spinwait();
+	}
+	unit->x86c.inv_queue_avail -= bytes;
+}
+
+static void
+amdiommu_cmd_emit(struct amdiommu_unit *unit, const struct
+    amdiommu_cmd_generic *cmd)
+{
+	AMDIOMMU_ASSERT_LOCKED(unit);
+
+	memcpy(unit->x86c.inv_queue + unit->x86c.inv_queue_tail, cmd,
+	    sizeof(*cmd));
+	unit->x86c.inv_queue_tail += AMDIOMMU_CMD_SZ;
+	KASSERT(unit->x86c.inv_queue_tail <= unit->x86c.inv_queue_size,
+	    ("tail overflow 0x%x 0x%jx", unit->x86c.inv_queue_tail,
+	    (uintmax_t)unit->x86c.inv_queue_size));
+	unit->x86c.inv_queue_tail &= unit->x86c.inv_queue_size - 1;
+}
+
+static void
+amdiommu_cmd_emit_wait_descr(struct iommu_unit *iommu, uint32_t seq,
+    bool intr, bool memw, bool fence)
+{
+	struct amdiommu_unit *unit;
+	struct amdiommu_cmd_completion_wait c;
+
+	unit = IOMMU2AMD(iommu);
+	AMDIOMMU_ASSERT_LOCKED(unit);
+
+	bzero(&c, sizeof(c));
+	c.op = AMDIOMMU_CMD_COMPLETION_WAIT;
+	if (memw) {
+		uint32_t x;
+
+		c.s = 1;
+		x = unit->x86c.inv_waitd_seq_hw_phys;
+		x >>= 3;
+		c.address0 = x;
+		x = unit->x86c.inv_waitd_seq_hw_phys >> 32;
+		c.address1 = x;
+		c.data0 = seq;
+	}
+	if (fence)
+		c.f = 1;
+	if (intr)
+		c.i = 1;
+	amdiommu_cmd_emit(unit, (struct amdiommu_cmd_generic *)&c);
+}
+
+static void
+amdiommu_qi_invalidate_emit(struct iommu_domain *adomain, iommu_gaddr_t base,
+    iommu_gaddr_t size, struct iommu_qi_genseq *pseq, bool emit_wait)
+{
+	struct amdiommu_domain *domain;
+	struct amdiommu_unit *unit;
+	struct amdiommu_cmd_invalidate_iommu_pages c;
+	u_int isize;
+
+	domain = IODOM2DOM(adomain);
+	unit = domain->unit;
+	AMDIOMMU_ASSERT_LOCKED(unit);
+	bzero(&c, sizeof(c));
+	c.op = AMDIOMMU_CMD_INVALIDATE_IOMMU_PAGES;
+	c.domainid = domain->domain;
+	isize = IOMMU_PAGE_SIZE; /* XXXKIB handle superpages */
+
+	for (; size > 0; base += isize, size -= isize) {
+		amdiommu_cmd_ensure(AMD2IOMMU(unit), 1);
+		c.s = 0;
+		c.pde = 1;
+		c.address = base >> IOMMU_PAGE_SHIFT;
+		amdiommu_cmd_emit(unit, (struct amdiommu_cmd_generic *)&c);
+	}
+	iommu_qi_emit_wait_seq(AMD2IOMMU(unit), pseq, emit_wait);
+}
+
+void
+amdiommu_qi_invalidate_all_pages_locked_nowait(struct amdiommu_domain *domain)
+{
+	struct amdiommu_unit *unit;
+	struct amdiommu_cmd_invalidate_iommu_pages c;
+
+	unit = domain->unit;
+	AMDIOMMU_ASSERT_LOCKED(unit);
+	bzero(&c, sizeof(c));
+	c.op = AMDIOMMU_CMD_INVALIDATE_IOMMU_PAGES;
+	c.domainid = domain->domain;
+
+	/*
+	 * The magic specified in the note for INVALIDATE_IOMMU_PAGES
+	 * description.
+	 */
+	c.s = 1;
+	c.pde = 1;
+	c.address = 0x7ffffffffffff;
+
+	amdiommu_cmd_ensure(AMD2IOMMU(unit), 1);
+	amdiommu_cmd_emit(unit, (struct amdiommu_cmd_generic *)&c);
+}
+
+void
+amdiommu_qi_invalidate_wait_sync(struct iommu_unit *iommu)
+{
+	struct iommu_qi_genseq gseq;
+
+	amdiommu_cmd_ensure(iommu, 1);
+	iommu_qi_emit_wait_seq(iommu, &gseq, true);
+	IOMMU2AMD(iommu)->x86c.inv_seq_waiters++;
+	amdiommu_cmd_advance_tail(iommu);
+	iommu_qi_wait_for_seq(iommu, &gseq, true);
+}
+
+void
+amdiommu_qi_invalidate_ctx_locked_nowait(struct amdiommu_ctx *ctx)
+{
+	struct amdiommu_cmd_invalidate_devtab_entry c;
+
+	amdiommu_cmd_ensure(AMD2IOMMU(CTX2AMD(ctx)), 1);
+	bzero(&c, sizeof(c));
+	c.op = AMDIOMMU_CMD_INVALIDATE_DEVTAB_ENTRY;
+	c.devid = ctx->context.rid;
+	amdiommu_cmd_emit(CTX2AMD(ctx), (struct amdiommu_cmd_generic *)&c);
+}
+
+
+void
+amdiommu_qi_invalidate_ctx_locked(struct amdiommu_ctx *ctx)
+{
+	amdiommu_qi_invalidate_ctx_locked_nowait(ctx);
+	amdiommu_qi_invalidate_wait_sync(AMD2IOMMU(CTX2AMD(ctx)));
+}
+
+void
+amdiommu_qi_invalidate_ir_locked_nowait(struct amdiommu_unit *unit,
+    uint16_t devid)
+{
+	struct amdiommu_cmd_invalidate_interrupt_table c;
+
+	AMDIOMMU_ASSERT_LOCKED(unit);
+
+	amdiommu_cmd_ensure(AMD2IOMMU(unit), 1);
+	bzero(&c, sizeof(c));
+	c.op = AMDIOMMU_CMD_INVALIDATE_INTERRUPT_TABLE;
+	c.devid = devid;
+	amdiommu_cmd_emit(unit, (struct amdiommu_cmd_generic *)&c);
+}
+
+void
+amdiommu_qi_invalidate_ir_locked(struct amdiommu_unit *unit, uint16_t devid)
+{
+	amdiommu_qi_invalidate_ir_locked_nowait(unit, devid);
+	amdiommu_qi_invalidate_wait_sync(AMD2IOMMU(unit));
+}
+
+static void
+amdiommu_qi_task(void *arg, int pending __unused)
+{
+	struct amdiommu_unit *unit;
+
+	unit = IOMMU2AMD(arg);
+	iommu_qi_drain_tlb_flush(AMD2IOMMU(unit));
+
+	AMDIOMMU_LOCK(unit);
+	if (unit->x86c.inv_seq_waiters > 0)
+		wakeup(&unit->x86c.inv_seq_waiters);
+	AMDIOMMU_UNLOCK(unit);
+}
+
+int
+amdiommu_init_cmd(struct amdiommu_unit *unit)
+{
+	uint64_t qi_sz, rv;
+
+	unit->x86c.qi_buf_maxsz = ilog2(AMDIOMMU_CMDBUF_MAX / PAGE_SIZE);
+	unit->x86c.qi_cmd_sz = AMDIOMMU_CMD_SZ;
+	iommu_qi_common_init(AMD2IOMMU(unit), amdiommu_qi_task);
+	get_x86_iommu()->qi_ensure = amdiommu_cmd_ensure;
+	get_x86_iommu()->qi_emit_wait_descr = amdiommu_cmd_emit_wait_descr;
+	get_x86_iommu()->qi_advance_tail = amdiommu_cmd_advance_tail;
+	get_x86_iommu()->qi_invalidate_emit = amdiommu_qi_invalidate_emit;
+
+	rv = pmap_kextract((uintptr_t)unit->x86c.inv_queue);
+
+	/*
+	 * See the description of the ComLen encoding for Command
+	 * buffer Base Address Register.
+	 */
+	qi_sz = ilog2(unit->x86c.inv_queue_size / PAGE_SIZE) + 8;
+	rv |= qi_sz << AMDIOMMU_CMDBUF_BASE_SZSHIFT;
+
+	AMDIOMMU_LOCK(unit);
+	amdiommu_write8(unit, AMDIOMMU_CMDBUF_BASE, rv);
+	amdiommu_enable_cmdbuf(unit);
+	amdiommu_enable_qi_intr(AMD2IOMMU(unit));
+	AMDIOMMU_UNLOCK(unit);
+
+	return (0);
+}
+
+static void
+amdiommu_fini_cmd_helper(struct iommu_unit *iommu)
+{
+	amdiommu_disable_cmdbuf(IOMMU2AMD(iommu));
+	amdiommu_disable_qi_intr(iommu);
+}
+
+void
+amdiommu_fini_cmd(struct amdiommu_unit *unit)
+{
+	iommu_qi_common_fini(AMD2IOMMU(unit), amdiommu_fini_cmd_helper);
+}
diff --git a/sys/x86/iommu/amd_ctx.c b/sys/x86/iommu/amd_ctx.c
new file mode 100644
index 000000000000..b3e85350a995
--- /dev/null
+++ b/sys/x86/iommu/amd_ctx.c
@@ -0,0 +1,639 @@
+/*-
+ * SPDX-License-Identifier: BSD-2-Clause
+ *
+ * Copyright (c) 2024 The FreeBSD Foundation
+ *
+ * This software was developed by Konstantin Belousov <kib@FreeBSD.org>
+ * under sponsorship from the FreeBSD Foundation.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/malloc.h>
+#include <sys/bus.h>
+#include <sys/interrupt.h>
+#include <sys/kernel.h>
+#include <sys/ktr.h>
+#include <sys/limits.h>
+#include <sys/lock.h>
+#include <sys/memdesc.h>
+#include <sys/mutex.h>
+#include <sys/proc.h>
+#include <sys/rwlock.h>
+#include <sys/rman.h>
+#include <sys/sysctl.h>
+#include <sys/taskqueue.h>
+#include <sys/tree.h>
+#include <sys/uio.h>
+#include <sys/vmem.h>
+#include <vm/vm.h>
+#include <vm/vm_extern.h>
+#include <vm/vm_kern.h>
+#include <vm/vm_object.h>
+#include <vm/vm_page.h>
+#include <vm/vm_pager.h>
+#include <vm/vm_map.h>
+#include <contrib/dev/acpica/include/acpi.h>
+#include <contrib/dev/acpica/include/accommon.h>
+#include <dev/pci/pcireg.h>
+#include <dev/pci/pcivar.h>
+#include <machine/atomic.h>
+#include <machine/bus.h>
+#include <machine/md_var.h>
+#include <machine/specialreg.h>
+#include <x86/include/busdma_impl.h>
+#include <dev/iommu/busdma_iommu.h>
+#include <x86/iommu/amd_reg.h>
+#include <x86/iommu/x86_iommu.h>
+#include <x86/iommu/amd_iommu.h>
+
+static MALLOC_DEFINE(M_AMDIOMMU_CTX, "amdiommu_ctx", "AMD IOMMU Context");
+static MALLOC_DEFINE(M_AMDIOMMU_DOMAIN, "amdiommu_dom", "AMD IOMMU Domain");
+
+static void amdiommu_unref_domain_locked(struct amdiommu_unit *unit,
+    struct amdiommu_domain *domain);
+
+static struct amdiommu_dte *
+amdiommu_get_dtep(struct amdiommu_ctx *ctx)
+{
+	return (&CTX2AMD(ctx)->dev_tbl[ctx->context.rid]);
+}
+
+void
+amdiommu_domain_unload_entry(struct iommu_map_entry *entry, bool free,
+    bool cansleep)
+{
+	struct amdiommu_domain *domain;
+	struct amdiommu_unit *unit;
+
+	domain = IODOM2DOM(entry->domain);
+	unit = DOM2AMD(domain);
+
+	/*
+	 * If "free" is false, then the IOTLB invalidation must be performed
+	 * synchronously.  Otherwise, the caller might free the entry before
+	 * dmar_qi_task() is finished processing it.
+	 */
+	if (free) {
+		AMDIOMMU_LOCK(unit);
+		iommu_qi_invalidate_locked(&domain->iodom, entry, true);
+		AMDIOMMU_UNLOCK(unit);
+	} else {
+		iommu_qi_invalidate_sync(&domain->iodom, entry->start,
+		    entry->end - entry->start, cansleep);
+		iommu_domain_free_entry(entry, false);
+	}
+}
+
+static bool
+amdiommu_domain_unload_emit_wait(struct amdiommu_domain *domain,
+    struct iommu_map_entry *entry)
+{
+	return (true); /* XXXKIB */
+}
+
+void
+amdiommu_domain_unload(struct iommu_domain *iodom,
+    struct iommu_map_entries_tailq *entries, bool cansleep)
+{
+	struct amdiommu_domain *domain;
+	struct amdiommu_unit *unit;
+	struct iommu_map_entry *entry, *entry1;
+	int error __diagused;
+
+	domain = IODOM2DOM(iodom);
+	unit = DOM2AMD(domain);
+
+	TAILQ_FOREACH_SAFE(entry, entries, dmamap_link, entry1) {
+		KASSERT((entry->flags & IOMMU_MAP_ENTRY_MAP) != 0,
+		    ("not mapped entry %p %p", domain, entry));
+		error = iodom->ops->unmap(iodom, entry,
+		    cansleep ? IOMMU_PGF_WAITOK : 0);
+		KASSERT(error == 0, ("unmap %p error %d", domain, error));
+	}
+	if (TAILQ_EMPTY(entries))
+		return;
+
+	AMDIOMMU_LOCK(unit);
+	while ((entry = TAILQ_FIRST(entries)) != NULL) {
+		TAILQ_REMOVE(entries, entry, dmamap_link);
+		iommu_qi_invalidate_locked(&domain->iodom, entry,
+		    amdiommu_domain_unload_emit_wait(domain, entry));
+	}
+	AMDIOMMU_UNLOCK(unit);
+}
+
+static void
+amdiommu_domain_destroy(struct amdiommu_domain *domain)
+{
+	struct iommu_domain *iodom;
+	struct amdiommu_unit *unit;
+
+	iodom = DOM2IODOM(domain);
+
+	KASSERT(TAILQ_EMPTY(&domain->iodom.unload_entries),
+	    ("unfinished unloads %p", domain));
+	KASSERT(LIST_EMPTY(&iodom->contexts),
+	    ("destroying dom %p with contexts", domain));
+	KASSERT(domain->ctx_cnt == 0,
+	    ("destroying dom %p with ctx_cnt %d", domain, domain->ctx_cnt));
+	KASSERT(domain->refs == 0,
+	    ("destroying dom %p with refs %d", domain, domain->refs));
+
+	if ((domain->iodom.flags & IOMMU_DOMAIN_GAS_INITED) != 0) {
+		AMDIOMMU_DOMAIN_LOCK(domain);
+		iommu_gas_fini_domain(iodom);
+		AMDIOMMU_DOMAIN_UNLOCK(domain);
+	}
+	if ((domain->iodom.flags & IOMMU_DOMAIN_PGTBL_INITED) != 0) {
+		if (domain->pgtbl_obj != NULL)
+			AMDIOMMU_DOMAIN_PGLOCK(domain);
+		amdiommu_domain_free_pgtbl(domain);
+	}
+	iommu_domain_fini(iodom);
+	unit = DOM2AMD(domain);
+	free_unr(unit->domids, domain->domain);
+	free(domain, M_AMDIOMMU_DOMAIN);
+}
+
+static iommu_gaddr_t
+lvl2addr(int lvl)
+{
+	int x;
+
+	x = IOMMU_PAGE_SHIFT + IOMMU_NPTEPGSHIFT * lvl;
+	/* Level 6 has only 8 bits for page table index */
+	if (x >= NBBY * sizeof(uint64_t))
+		return (-1ull);
+	return (1ull < (1ull << x));
+}
+
+static void
+amdiommu_domain_init_pglvl(struct amdiommu_unit *unit,
+    struct amdiommu_domain *domain)
+{
+	iommu_gaddr_t end;
+	int hats, i;
+	uint64_t efr_hats;
+
+	end = DOM2IODOM(domain)->end;
+	for (i = AMDIOMMU_PGTBL_MAXLVL; i > 1; i--) {
+		if (lvl2addr(i) >= end && lvl2addr(i - 1) < end)
+			break;
+	}
+	domain->pglvl = i;
+
+	efr_hats = unit->efr & AMDIOMMU_EFR_HATS_MASK;
+	switch (efr_hats) {
+	case AMDIOMMU_EFR_HATS_6LVL:
+		hats = 6;
+		break;
+	case AMDIOMMU_EFR_HATS_5LVL:
+		hats = 5;
+		break;
+	case AMDIOMMU_EFR_HATS_4LVL:
+		hats = 4;
+		break;
+	default:
+		printf("amdiommu%d: HATS %#jx (reserved) ignoring\n",
+		    unit->iommu.unit, (uintmax_t)efr_hats);
+		return;
+	}
+	if (hats >= domain->pglvl)
+		return;
+
+	printf("amdiommu%d: domain %d HATS %d pglvl %d reducing to HATS\n",
+	    unit->iommu.unit, domain->domain, hats, domain->pglvl);
+	domain->pglvl = hats;
+	domain->iodom.end = lvl2addr(hats);
+}
+
+static struct amdiommu_domain *
+amdiommu_domain_alloc(struct amdiommu_unit *unit, bool id_mapped)
+{
+	struct amdiommu_domain *domain;
+	struct iommu_domain *iodom;
+	int error, id;
+
+	id = alloc_unr(unit->domids);
+	if (id == -1)
+		return (NULL);
+	domain = malloc(sizeof(*domain), M_AMDIOMMU_DOMAIN, M_WAITOK | M_ZERO);
+	iodom = DOM2IODOM(domain);
+	domain->domain = id;
+	LIST_INIT(&iodom->contexts);
+	iommu_domain_init(AMD2IOMMU(unit), iodom, &amdiommu_domain_map_ops);
+
+	domain->unit = unit;
+
+	domain->iodom.end = id_mapped ? ptoa(Maxmem) : BUS_SPACE_MAXADDR;
+	amdiommu_domain_init_pglvl(unit, domain);
+	iommu_gas_init_domain(DOM2IODOM(domain));
+
+	if (id_mapped) {
+		domain->iodom.flags |= IOMMU_DOMAIN_IDMAP;
+	} else {
+		error = amdiommu_domain_alloc_pgtbl(domain);
+		if (error != 0)
+			goto fail;
+		/* Disable local apic region access */
+		error = iommu_gas_reserve_region(iodom, 0xfee00000,
+		    0xfeefffff + 1, &iodom->msi_entry);
+		if (error != 0)
+			goto fail;
+	}
+
+	return (domain);
+
+fail:
+	amdiommu_domain_destroy(domain);
+	return (NULL);
+}
+
+static struct amdiommu_ctx *
+amdiommu_ctx_alloc(struct amdiommu_domain *domain, uint16_t rid)
+{
+	struct amdiommu_ctx *ctx;
+
+	ctx = malloc(sizeof(*ctx), M_AMDIOMMU_CTX, M_WAITOK | M_ZERO);
+	ctx->context.domain = DOM2IODOM(domain);
+	ctx->context.tag = malloc(sizeof(struct bus_dma_tag_iommu),
+	    M_AMDIOMMU_CTX, M_WAITOK | M_ZERO);
+	ctx->context.rid = rid;
+	ctx->context.refs = 1;
+	return (ctx);
+}
+
+static void
+amdiommu_ctx_link(struct amdiommu_ctx *ctx)
+{
+	struct amdiommu_domain *domain;
+
+	domain = CTX2DOM(ctx);
+	IOMMU_ASSERT_LOCKED(domain->iodom.iommu);
+	KASSERT(domain->refs >= domain->ctx_cnt,
+	    ("dom %p ref underflow %d %d", domain, domain->refs,
+	    domain->ctx_cnt));
+	domain->refs++;
+	domain->ctx_cnt++;
+	LIST_INSERT_HEAD(&domain->iodom.contexts, &ctx->context, link);
+}
+
+static void
+amdiommu_ctx_unlink(struct amdiommu_ctx *ctx)
+{
+	struct amdiommu_domain *domain;
+
+	domain = CTX2DOM(ctx);
+	IOMMU_ASSERT_LOCKED(domain->iodom.iommu);
+	KASSERT(domain->refs > 0,
+	    ("domain %p ctx dtr refs %d", domain, domain->refs));
+	KASSERT(domain->ctx_cnt >= domain->refs,
+	    ("domain %p ctx dtr refs %d ctx_cnt %d", domain,
+	    domain->refs, domain->ctx_cnt));
+	domain->refs--;
+	domain->ctx_cnt--;
+	LIST_REMOVE(&ctx->context, link);
+}
+
+struct amdiommu_ctx *
+amdiommu_find_ctx_locked(struct amdiommu_unit *unit, uint16_t rid)
+{
+	struct amdiommu_domain *domain;
+	struct iommu_ctx *ctx;
+
+	AMDIOMMU_ASSERT_LOCKED(unit);
+
+	LIST_FOREACH(domain, &unit->domains, link) {
+		LIST_FOREACH(ctx, &domain->iodom.contexts, link) {
+			if (ctx->rid == rid)
+				return (IOCTX2CTX(ctx));
+		}
+	}
+	return (NULL);
+}
+
+struct amdiommu_domain *
+amdiommu_find_domain(struct amdiommu_unit *unit, uint16_t rid)
+{
+	struct amdiommu_domain *domain;
+	struct iommu_ctx *ctx;
+
+	AMDIOMMU_LOCK(unit);
+	LIST_FOREACH(domain, &unit->domains, link) {
+		LIST_FOREACH(ctx, &domain->iodom.contexts, link) {
+			if (ctx->rid == rid)
+				break;
+		}
+	}
+	AMDIOMMU_UNLOCK(unit);
+	return (domain);
+}
+
+static void
+amdiommu_free_ctx_locked(struct amdiommu_unit *unit, struct amdiommu_ctx *ctx)
+{
+	struct amdiommu_dte *dtep;
+	struct amdiommu_domain *domain;
+
+	AMDIOMMU_ASSERT_LOCKED(unit);
+	KASSERT(ctx->context.refs >= 1,
+	    ("amdiommu %p ctx %p refs %u", unit, ctx, ctx->context.refs));
+
+	/*
+	 * If our reference is not last, only the dereference should
+	 * be performed.
+	 */
+	if (ctx->context.refs > 1) {
+		ctx->context.refs--;
+		AMDIOMMU_UNLOCK(unit);
+		return;
+	}
+
+	KASSERT((ctx->context.flags & IOMMU_CTX_DISABLED) == 0,
+	    ("lost ref on disabled ctx %p", ctx));
+
+	/*
+	 * Otherwise, the device table entry must be cleared before
+	 * the page table is destroyed.
+	 */
+	dtep = amdiommu_get_dtep(ctx);
+	dtep->v = 0;
+	atomic_thread_fence_rel();
+	memset(dtep, 0, sizeof(*dtep));
+
+	domain = CTX2DOM(ctx);
+	amdiommu_qi_invalidate_ctx_locked_nowait(ctx);
+	amdiommu_qi_invalidate_ir_locked_nowait(unit, ctx->context.rid);
+	amdiommu_qi_invalidate_all_pages_locked_nowait(domain);
+	amdiommu_qi_invalidate_wait_sync(AMD2IOMMU(CTX2AMD(ctx)));
+
+	if (unit->irte_enabled)
+		amdiommu_ctx_fini_irte(ctx);
+
+	amdiommu_ctx_unlink(ctx);
+	free(ctx->context.tag, M_AMDIOMMU_CTX);
+	free(ctx, M_AMDIOMMU_CTX);
+	amdiommu_unref_domain_locked(unit, domain);
+}
+
+static void
+amdiommu_free_ctx(struct amdiommu_ctx *ctx)
+{
+	struct amdiommu_unit *unit;
+
+	unit = CTX2AMD(ctx);
+	AMDIOMMU_LOCK(unit);
+	amdiommu_free_ctx_locked(unit, ctx);
+}
+
+static void
+amdiommu_unref_domain_locked(struct amdiommu_unit *unit,
+    struct amdiommu_domain *domain)
+{
+	AMDIOMMU_ASSERT_LOCKED(unit);
+	KASSERT(domain->refs >= 1,
+	    ("amdiommu%d domain %p refs %u", unit->iommu.unit, domain,
+	    domain->refs));
+	KASSERT(domain->refs > domain->ctx_cnt,
+	    ("amdiommu%d domain %p refs %d ctx_cnt %d", unit->iommu.unit,
+	    domain, domain->refs, domain->ctx_cnt));
+
+	if (domain->refs > 1) {
+		domain->refs--;
+		AMDIOMMU_UNLOCK(unit);
+		return;
+	}
+
+	LIST_REMOVE(domain, link);
+	AMDIOMMU_UNLOCK(unit);
+
+	taskqueue_drain(unit->iommu.delayed_taskqueue,
+	    &domain->iodom.unload_task);
+	amdiommu_domain_destroy(domain);
+}
+
+static void
+dte_entry_init_one(struct amdiommu_dte *dtep, struct amdiommu_ctx *ctx,
+    vm_page_t pgtblr, uint8_t dte, uint32_t edte)
+{
+	struct amdiommu_domain *domain;
+	struct amdiommu_unit *unit;
+
+	domain = CTX2DOM(ctx);
+	unit = DOM2AMD(domain);
+
+	dtep->tv = 1;
+	/* dtep->had not used for now */
+	dtep->ir = 1;
+	dtep->iw = 1;
+	dtep->domainid = domain->domain;
+	dtep->pioctl = AMDIOMMU_DTE_PIOCTL_DIS;
+
+	/* fill device interrupt passing hints from IVHD. */
+	dtep->initpass = (dte & ACPI_IVHD_INIT_PASS) != 0;
+	dtep->eintpass = (dte & ACPI_IVHD_EINT_PASS) != 0;
+	dtep->nmipass = (dte & ACPI_IVHD_NMI_PASS) != 0;
+	dtep->sysmgt = (dte & ACPI_IVHD_SYSTEM_MGMT) >> 4;
+	dtep->lint0pass = (dte & ACPI_IVHD_LINT0_PASS) != 0;
+	dtep->lint1pass = (dte & ACPI_IVHD_LINT1_PASS) != 0;
+
+	if (unit->irte_enabled) {
+		dtep->iv = 1;
+		dtep->i = 0;
+		dtep->inttablen = ilog2(unit->irte_nentries);
+		dtep->intrroot = pmap_kextract(unit->irte_x2apic ?
+		    (vm_offset_t)ctx->irtx2 :
+		    (vm_offset_t)ctx->irtb) >> 6;
+
+		dtep->intctl = AMDIOMMU_DTE_INTCTL_MAP;
+	}
+
+	if ((DOM2IODOM(domain)->flags & IOMMU_DOMAIN_IDMAP) != 0) {
+		dtep->pgmode = AMDIOMMU_DTE_PGMODE_1T1;
+	} else {
+		MPASS(domain->pglvl > 0 && domain->pglvl <=
+		    AMDIOMMU_PGTBL_MAXLVL);
+		dtep->pgmode = domain->pglvl;
+		dtep->ptroot = VM_PAGE_TO_PHYS(pgtblr) >> 12;
+	}
+
+	atomic_thread_fence_rel();
+	dtep->v = 1;
+}
+
+static void
+dte_entry_init(struct amdiommu_ctx *ctx, bool move, uint8_t dte, uint32_t edte)
+{
+	struct amdiommu_dte *dtep;
+	struct amdiommu_unit *unit;
+	struct amdiommu_domain *domain;
+	int i;
+
+	domain = CTX2DOM(ctx);
+	unit = DOM2AMD(domain);
+
+	dtep = amdiommu_get_dtep(ctx);
+	KASSERT(dtep->v == 0,
+	    ("amdiommu%d initializing valid dte @%p %#jx",
+	    CTX2AMD(ctx)->iommu.unit, dtep, (uintmax_t)(*(uint64_t *)dtep)));
+
+	if (iommu_is_buswide_ctx(AMD2IOMMU(unit),
+	    PCI_RID2BUS(ctx->context.rid))) {
+		MPASS(!move);
+		for (i = 0; i <= PCI_BUSMAX; i++) {
+			dte_entry_init_one(&dtep[i], ctx, domain->pgtblr,
+			    dte, edte);
+		}
+	} else {
+		dte_entry_init_one(dtep, ctx, domain->pgtblr, dte, edte);
+	}
+}
+
+struct amdiommu_ctx *
+amdiommu_get_ctx_for_dev(struct amdiommu_unit *unit, device_t dev, uint16_t rid,
+    int dev_domain, bool id_mapped, bool rmrr_init, uint8_t dte, uint32_t edte)
+{
+	struct amdiommu_domain *domain, *domain1;
+	struct amdiommu_ctx *ctx, *ctx1;
+	int bus, slot, func;
+
+	if (dev != NULL) {
+		bus = pci_get_bus(dev);
+		slot = pci_get_slot(dev);
+		func = pci_get_function(dev);
+	} else {
+		bus = PCI_RID2BUS(rid);
+		slot = PCI_RID2SLOT(rid);
+		func = PCI_RID2FUNC(rid);
+	}
+	AMDIOMMU_LOCK(unit);
+	KASSERT(!iommu_is_buswide_ctx(AMD2IOMMU(unit), bus) ||
+	    (slot == 0 && func == 0),
+	    ("iommu%d pci%d:%d:%d get_ctx for buswide", AMD2IOMMU(unit)->unit,
+	    bus, slot, func));
+	ctx = amdiommu_find_ctx_locked(unit, rid);
+	if (ctx == NULL) {
+		/*
+		 * Perform the allocations which require sleep or have
+		 * higher chance to succeed if the sleep is allowed.
+		 */
+		AMDIOMMU_UNLOCK(unit);
+		domain1 = amdiommu_domain_alloc(unit, id_mapped);
+		if (domain1 == NULL)
+			return (NULL);
+		if (!id_mapped) {
+			/*
+			 * XXXKIB IVMD seems to be less significant
+			 * and less used on AMD than RMRR on Intel.
+			 * Not implemented for now.
+			 */
+		}
+		ctx1 = amdiommu_ctx_alloc(domain1, rid);
+		amdiommu_ctx_init_irte(ctx1);
+		AMDIOMMU_LOCK(unit);
+
+		/*
+		 * Recheck the contexts, other thread might have
+		 * already allocated needed one.
+		 */
+		ctx = amdiommu_find_ctx_locked(unit, rid);
+		if (ctx == NULL) {
+			domain = domain1;
+			ctx = ctx1;
+			amdiommu_ctx_link(ctx);
+			ctx->context.tag->owner = dev;
+			iommu_device_tag_init(CTX2IOCTX(ctx), dev);
+
+			LIST_INSERT_HEAD(&unit->domains, domain, link);
+			dte_entry_init(ctx, false, dte, edte);
+			amdiommu_qi_invalidate_ctx_locked(ctx);
+			if (dev != NULL) {
+				device_printf(dev,
+			    "amdiommu%d pci%d:%d:%d:%d rid %x domain %d "
*** 2653 LINES SKIPPED ***



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?202411092019.4A9KJFgD033481>