xref: /freebsd/sys/x86/iommu/x86_iommu.h (revision 29e227047065c6b1d08fdb72a21f33560637332b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2013-2015, 2024 The FreeBSD Foundation
5  *
6  * This software was developed by Konstantin Belousov <kib@FreeBSD.org>
7  * under sponsorship from the FreeBSD Foundation.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  */
30 
31 #ifndef __X86_IOMMU_X86_IOMMU_H
32 #define	__X86_IOMMU_X86_IOMMU_H
33 
34 /* Both Intel and AMD are not too crazy to have different sizes. */
35 typedef struct iommu_pte {
36 	uint64_t pte;
37 } iommu_pte_t;
38 
39 #define	IOMMU_PAGE_SIZE		PAGE_SIZE
40 #define	IOMMU_PAGE_MASK		(IOMMU_PAGE_SIZE - 1)
41 #define	IOMMU_PAGE_SHIFT	PAGE_SHIFT
42 #define	IOMMU_NPTEPG		(IOMMU_PAGE_SIZE / sizeof(iommu_pte_t))
43 #define	IOMMU_NPTEPGSHIFT 	9
44 #define	IOMMU_PTEMASK		(IOMMU_NPTEPG - 1)
45 
46 struct sf_buf;
47 struct vm_object;
48 
49 struct vm_page *iommu_pgalloc(struct vm_object *obj, vm_pindex_t idx,
50     int flags);
51 void iommu_pgfree(struct vm_object *obj, vm_pindex_t idx, int flags);
52 void *iommu_map_pgtbl(struct vm_object *obj, vm_pindex_t idx, int flags,
53     struct sf_buf **sf);
54 void iommu_unmap_pgtbl(struct sf_buf *sf);
55 
56 extern iommu_haddr_t iommu_high;
57 extern int iommu_tbl_pagecnt;
58 
59 SYSCTL_DECL(_hw_iommu);
60 SYSCTL_DECL(_hw_iommu_dmar);
61 
62 struct x86_unit_common;
63 
64 struct x86_iommu {
65 	struct x86_unit_common *(*get_x86_common)(struct
66 	    iommu_unit *iommu);
67 	void (*qi_ensure)(struct iommu_unit *unit, int descr_count);
68 	void (*qi_emit_wait_descr)(struct iommu_unit *unit, uint32_t seq,
69 	    bool, bool, bool);
70 	void (*qi_advance_tail)(struct iommu_unit *unit);
71 	void (*qi_invalidate_emit)(struct iommu_domain *idomain,
72 	    iommu_gaddr_t base, iommu_gaddr_t size, struct iommu_qi_genseq *
73 	    pseq, bool emit_wait);
74 	void (*domain_unload_entry)(struct iommu_map_entry *entry, bool free,
75 	    bool cansleep);
76 	void (*domain_unload)(struct iommu_domain *iodom,
77 		struct iommu_map_entries_tailq *entries, bool cansleep);
78 	struct iommu_ctx *(*get_ctx)(struct iommu_unit *iommu,
79 	    device_t dev, uint16_t rid, bool id_mapped, bool rmrr_init);
80 	void (*free_ctx_locked)(struct iommu_unit *iommu,
81 	    struct iommu_ctx *context);
82 	void (*free_ctx)(struct iommu_ctx *context);
83 	struct iommu_unit *(*find)(device_t dev, bool verbose);
84 	int (*alloc_msi_intr)(device_t src, u_int *cookies, u_int count);
85 	int (*map_msi_intr)(device_t src, u_int cpu, u_int vector,
86 	    u_int cookie, uint64_t *addr, uint32_t *data);
87 	int (*unmap_msi_intr)(device_t src, u_int cookie);
88 	int (*map_ioapic_intr)(u_int ioapic_id, u_int cpu, u_int vector,
89 	    bool edge, bool activehi, int irq, u_int *cookie, uint32_t *hi,
90 	    uint32_t *lo);
91 	int (*unmap_ioapic_intr)(u_int ioapic_id, u_int *cookie);
92 };
93 void set_x86_iommu(struct x86_iommu *);
94 struct x86_iommu *get_x86_iommu(void);
95 
96 struct iommu_msi_data {
97 	int irq;
98 	int irq_rid;
99 	struct resource *irq_res;
100 	void *intr_handle;
101 	int (*handler)(void *);
102 	int msi_data_reg;
103 	int msi_addr_reg;
104 	int msi_uaddr_reg;
105 	uint64_t msi_addr;
106 	uint32_t msi_data;
107 	void (*enable_intr)(struct iommu_unit *);
108 	void (*disable_intr)(struct iommu_unit *);
109 	const char *name;
110 };
111 
112 #define	IOMMU_MAX_MSI	3
113 
114 struct x86_unit_common {
115 	uint32_t qi_buf_maxsz;
116 	uint32_t qi_cmd_sz;
117 
118 	char *inv_queue;
119 	vm_size_t inv_queue_size;
120 	uint32_t inv_queue_avail;
121 	uint32_t inv_queue_tail;
122 
123 	/*
124 	 * Hw writes there on completion of wait descriptor
125 	 * processing.  Intel writes 4 bytes, while AMD does the
126 	 * 8-bytes write.  Due to little-endian, and use of 4-byte
127 	 * sequence numbers, the difference does not matter for us.
128 	 */
129 	volatile uint64_t inv_waitd_seq_hw;
130 
131 	uint64_t inv_waitd_seq_hw_phys;
132 	uint32_t inv_waitd_seq; /* next sequence number to use for wait descr */
133 	u_int inv_waitd_gen;	/* seq number generation AKA seq overflows */
134 	u_int inv_seq_waiters;	/* count of waiters for seq */
135 	u_int inv_queue_full;	/* informational counter */
136 
137 	/*
138 	 * Delayed freeing of map entries queue processing:
139 	 *
140 	 * tlb_flush_head and tlb_flush_tail are used to implement a FIFO
141 	 * queue that supports concurrent dequeues and enqueues.  However,
142 	 * there can only be a single dequeuer (accessing tlb_flush_head) and
143 	 * a single enqueuer (accessing tlb_flush_tail) at a time.  Since the
144 	 * unit's qi_task is the only dequeuer, it can access tlb_flush_head
145 	 * without any locking.  In contrast, there may be multiple enqueuers,
146 	 * so the enqueuers acquire the iommu unit lock to serialize their
147 	 * accesses to tlb_flush_tail.
148 	 *
149 	 * In this FIFO queue implementation, the key to enabling concurrent
150 	 * dequeues and enqueues is that the dequeuer never needs to access
151 	 * tlb_flush_tail and the enqueuer never needs to access
152 	 * tlb_flush_head.  In particular, tlb_flush_head and tlb_flush_tail
153 	 * are never NULL, so neither a dequeuer nor an enqueuer ever needs to
154 	 * update both.  Instead, tlb_flush_head always points to a "zombie"
155 	 * struct, which previously held the last dequeued item.  Thus, the
156 	 * zombie's next field actually points to the struct holding the first
157 	 * item in the queue.  When an item is dequeued, the current zombie is
158 	 * finally freed, and the struct that held the just dequeued item
159 	 * becomes the new zombie.  When the queue is empty, tlb_flush_tail
160 	 * also points to the zombie.
161 	 */
162 	struct iommu_map_entry *tlb_flush_head;
163 	struct iommu_map_entry *tlb_flush_tail;
164 	struct task qi_task;
165 	struct taskqueue *qi_taskqueue;
166 
167 	struct iommu_msi_data intrs[IOMMU_MAX_MSI];
168 };
169 
170 void iommu_domain_free_entry(struct iommu_map_entry *entry, bool free);
171 
172 void iommu_qi_emit_wait_seq(struct iommu_unit *unit, struct iommu_qi_genseq *
173     pseq, bool emit_wait);
174 void iommu_qi_wait_for_seq(struct iommu_unit *unit, const struct
175     iommu_qi_genseq *gseq, bool nowait);
176 void iommu_qi_drain_tlb_flush(struct iommu_unit *unit);
177 void iommu_qi_invalidate_locked(struct iommu_domain *domain,
178     struct iommu_map_entry *entry, bool emit_wait);
179 void iommu_qi_invalidate_sync(struct iommu_domain *domain, iommu_gaddr_t base,
180     iommu_gaddr_t size, bool cansleep);
181 void iommu_qi_common_init(struct iommu_unit *unit, task_fn_t taskfunc);
182 void iommu_qi_common_fini(struct iommu_unit *unit, void (*disable_qi)(
183     struct iommu_unit *));
184 
185 int iommu_alloc_irq(struct iommu_unit *unit, int idx);
186 void iommu_release_intr(struct iommu_unit *unit, int idx);
187 
188 void iommu_device_tag_init(struct iommu_ctx *ctx, device_t dev);
189 
190 int pglvl_pgtbl_pte_off(int pglvl, iommu_gaddr_t base, int lvl);
191 vm_pindex_t pglvl_pgtbl_get_pindex(int pglvl, iommu_gaddr_t base, int lvl);
192 vm_pindex_t pglvl_max_pages(int pglvl);
193 iommu_gaddr_t pglvl_page_size(int total_pglvl, int lvl);
194 
195 #endif
196