1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2007-2010 Advanced Micro Devices, Inc.
8 #define pr_fmt(fmt) "AMD-Vi: " fmt
9 #define dev_fmt(fmt) pr_fmt(fmt)
11 #include <linux/pci.h>
12 #include <linux/acpi.h>
13 #include <linux/list.h>
14 #include <linux/bitmap.h>
15 #include <linux/slab.h>
16 #include <linux/syscore_ops.h>
17 #include <linux/interrupt.h>
18 #include <linux/msi.h>
19 #include <linux/irq.h>
20 #include <linux/amd-iommu.h>
21 #include <linux/export.h>
22 #include <linux/kmemleak.h>
23 #include <linux/cc_platform.h>
24 #include <linux/iopoll.h>
25 #include <asm/pci-direct.h>
26 #include <asm/iommu.h>
29 #include <asm/x86_init.h>
30 #include <asm/io_apic.h>
31 #include <asm/irq_remapping.h>
32 #include <asm/set_memory.h>
35 #include <linux/crash_dump.h>
37 #include "amd_iommu.h"
38 #include "../irq_remapping.h"
39 #include "../iommu-pages.h"
42 * definitions for the ACPI scanning code
44 #define IVRS_HEADER_LENGTH 48
46 #define ACPI_IVHD_TYPE_MAX_SUPPORTED 0x40
47 #define ACPI_IVMD_TYPE_ALL 0x20
48 #define ACPI_IVMD_TYPE 0x21
49 #define ACPI_IVMD_TYPE_RANGE 0x22
51 #define IVHD_DEV_ALL 0x01
52 #define IVHD_DEV_SELECT 0x02
53 #define IVHD_DEV_SELECT_RANGE_START 0x03
54 #define IVHD_DEV_RANGE_END 0x04
55 #define IVHD_DEV_ALIAS 0x42
56 #define IVHD_DEV_ALIAS_RANGE 0x43
57 #define IVHD_DEV_EXT_SELECT 0x46
58 #define IVHD_DEV_EXT_SELECT_RANGE 0x47
59 #define IVHD_DEV_SPECIAL 0x48
60 #define IVHD_DEV_ACPI_HID 0xf0
62 #define UID_NOT_PRESENT 0
63 #define UID_IS_INTEGER 1
64 #define UID_IS_CHARACTER 2
66 #define IVHD_SPECIAL_IOAPIC 1
67 #define IVHD_SPECIAL_HPET 2
69 #define IVHD_FLAG_HT_TUN_EN_MASK 0x01
70 #define IVHD_FLAG_PASSPW_EN_MASK 0x02
71 #define IVHD_FLAG_RESPASSPW_EN_MASK 0x04
72 #define IVHD_FLAG_ISOC_EN_MASK 0x08
74 #define IVMD_FLAG_EXCL_RANGE 0x08
75 #define IVMD_FLAG_IW 0x04
76 #define IVMD_FLAG_IR 0x02
77 #define IVMD_FLAG_UNITY_MAP 0x01
79 #define ACPI_DEVFLAG_INITPASS 0x01
80 #define ACPI_DEVFLAG_EXTINT 0x02
81 #define ACPI_DEVFLAG_NMI 0x04
82 #define ACPI_DEVFLAG_SYSMGT1 0x10
83 #define ACPI_DEVFLAG_SYSMGT2 0x20
84 #define ACPI_DEVFLAG_LINT0 0x40
85 #define ACPI_DEVFLAG_LINT1 0x80
86 #define ACPI_DEVFLAG_ATSDIS 0x10000000
88 #define IVRS_GET_SBDF_ID(seg, bus, dev, fn) (((seg & 0xffff) << 16) | ((bus & 0xff) << 8) \
89 | ((dev & 0x1f) << 3) | (fn & 0x7))
92 * ACPI table definitions
94 * These data structures are laid over the table to parse the important values
99 * structure describing one IOMMU in the ACPI table. Typically followed by one
100 * or more ivhd_entrys.
113 /* Following only valid on IVHD type 11h and 40h */
114 u64 efr_reg; /* Exact copy of MMIO_EXT_FEATURES */
116 } __attribute__((packed));
119 * A device entry describing which devices a specific IOMMU translates and
120 * which requestor ids they use.
126 struct_group(ext_hid,
134 } __attribute__((packed));
137 * An AMD IOMMU memory definition structure. It defines things like exclusion
138 * ranges for devices and regions that should be unity mapped.
150 } __attribute__((packed));
153 bool amd_iommu_irq_remap __read_mostly;
155 enum io_pgtable_fmt amd_iommu_pgtable = AMD_IOMMU_V1;
156 /* Guest page table level */
157 int amd_iommu_gpt_level = PAGE_MODE_4_LEVEL;
159 int amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
160 static int amd_iommu_xt_mode = IRQ_REMAP_XAPIC_MODE;
162 static bool amd_iommu_detected;
163 static bool amd_iommu_disabled __initdata;
164 static bool amd_iommu_force_enable __initdata;
165 static bool amd_iommu_irtcachedis;
166 static int amd_iommu_target_ivhd_type;
168 /* Global EFR and EFR2 registers */
172 /* SNP is enabled on the system? */
173 bool amd_iommu_snp_en;
174 EXPORT_SYMBOL(amd_iommu_snp_en);
176 LIST_HEAD(amd_iommu_pci_seg_list); /* list of all PCI segments */
177 LIST_HEAD(amd_iommu_list); /* list of all AMD IOMMUs in the
180 /* Array to assign indices to IOMMUs*/
181 struct amd_iommu *amd_iommus[MAX_IOMMUS];
183 /* Number of IOMMUs present in the system */
184 static int amd_iommus_present;
186 /* IOMMUs have a non-present cache? */
187 bool amd_iommu_np_cache __read_mostly;
188 bool amd_iommu_iotlb_sup __read_mostly = true;
190 static bool amd_iommu_pc_present __read_mostly;
191 bool amdr_ivrs_remap_support __read_mostly;
193 bool amd_iommu_force_isolation __read_mostly;
196 * AMD IOMMU allows up to 2^16 different protection domains. This is a bitmap
197 * to know which ones are already in use.
199 unsigned long *amd_iommu_pd_alloc_bitmap;
201 enum iommu_init_state {
211 IOMMU_CMDLINE_DISABLED,
214 /* Early ioapic and hpet maps from kernel command line */
215 #define EARLY_MAP_SIZE 4
216 static struct devid_map __initdata early_ioapic_map[EARLY_MAP_SIZE];
217 static struct devid_map __initdata early_hpet_map[EARLY_MAP_SIZE];
218 static struct acpihid_map_entry __initdata early_acpihid_map[EARLY_MAP_SIZE];
220 static int __initdata early_ioapic_map_size;
221 static int __initdata early_hpet_map_size;
222 static int __initdata early_acpihid_map_size;
224 static bool __initdata cmdline_maps;
226 static enum iommu_init_state init_state = IOMMU_START_STATE;
228 static int amd_iommu_enable_interrupts(void);
229 static int __init iommu_go_to_state(enum iommu_init_state state);
230 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg);
232 static bool amd_iommu_pre_enabled = true;
234 static u32 amd_iommu_ivinfo __initdata;
236 bool translation_pre_enabled(struct amd_iommu *iommu)
238 return (iommu->flags & AMD_IOMMU_FLAG_TRANS_PRE_ENABLED);
241 static void clear_translation_pre_enabled(struct amd_iommu *iommu)
243 iommu->flags &= ~AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
246 static void init_translation_status(struct amd_iommu *iommu)
250 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
251 if (ctrl & (1<<CONTROL_IOMMU_EN))
252 iommu->flags |= AMD_IOMMU_FLAG_TRANS_PRE_ENABLED;
255 static inline unsigned long tbl_size(int entry_size, int last_bdf)
257 unsigned shift = PAGE_SHIFT +
258 get_order((last_bdf + 1) * entry_size);
263 int amd_iommu_get_num_iommus(void)
265 return amd_iommus_present;
269 * Iterate through all the IOMMUs to get common EFR
270 * masks among all IOMMUs and warn if found inconsistency.
272 static __init void get_global_efr(void)
274 struct amd_iommu *iommu;
276 for_each_iommu(iommu) {
277 u64 tmp = iommu->features;
278 u64 tmp2 = iommu->features2;
280 if (list_is_first(&iommu->list, &amd_iommu_list)) {
282 amd_iommu_efr2 = tmp2;
286 if (amd_iommu_efr == tmp &&
287 amd_iommu_efr2 == tmp2)
291 "Found inconsistent EFR/EFR2 %#llx,%#llx (global %#llx,%#llx) on iommu%d (%04x:%02x:%02x.%01x).\n",
292 tmp, tmp2, amd_iommu_efr, amd_iommu_efr2,
293 iommu->index, iommu->pci_seg->id,
294 PCI_BUS_NUM(iommu->devid), PCI_SLOT(iommu->devid),
295 PCI_FUNC(iommu->devid));
297 amd_iommu_efr &= tmp;
298 amd_iommu_efr2 &= tmp2;
301 pr_info("Using global IVHD EFR:%#llx, EFR2:%#llx\n", amd_iommu_efr, amd_iommu_efr2);
305 * For IVHD type 0x11/0x40, EFR is also available via IVHD.
306 * Default to IVHD EFR since it is available sooner
307 * (i.e. before PCI init).
309 static void __init early_iommu_features_init(struct amd_iommu *iommu,
310 struct ivhd_header *h)
312 if (amd_iommu_ivinfo & IOMMU_IVINFO_EFRSUP) {
313 iommu->features = h->efr_reg;
314 iommu->features2 = h->efr_reg2;
316 if (amd_iommu_ivinfo & IOMMU_IVINFO_DMA_REMAP)
317 amdr_ivrs_remap_support = true;
320 /* Access to l1 and l2 indexed register spaces */
322 static u32 iommu_read_l1(struct amd_iommu *iommu, u16 l1, u8 address)
326 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
327 pci_read_config_dword(iommu->dev, 0xfc, &val);
331 static void iommu_write_l1(struct amd_iommu *iommu, u16 l1, u8 address, u32 val)
333 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16 | 1 << 31));
334 pci_write_config_dword(iommu->dev, 0xfc, val);
335 pci_write_config_dword(iommu->dev, 0xf8, (address | l1 << 16));
338 static u32 iommu_read_l2(struct amd_iommu *iommu, u8 address)
342 pci_write_config_dword(iommu->dev, 0xf0, address);
343 pci_read_config_dword(iommu->dev, 0xf4, &val);
347 static void iommu_write_l2(struct amd_iommu *iommu, u8 address, u32 val)
349 pci_write_config_dword(iommu->dev, 0xf0, (address | 1 << 8));
350 pci_write_config_dword(iommu->dev, 0xf4, val);
353 /****************************************************************************
355 * AMD IOMMU MMIO register space handling functions
357 * These functions are used to program the IOMMU device registers in
358 * MMIO space required for that driver.
360 ****************************************************************************/
363 * This function set the exclusion range in the IOMMU. DMA accesses to the
364 * exclusion range are passed through untranslated
366 static void iommu_set_exclusion_range(struct amd_iommu *iommu)
368 u64 start = iommu->exclusion_start & PAGE_MASK;
369 u64 limit = (start + iommu->exclusion_length - 1) & PAGE_MASK;
372 if (!iommu->exclusion_start)
375 entry = start | MMIO_EXCL_ENABLE_MASK;
376 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
377 &entry, sizeof(entry));
380 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
381 &entry, sizeof(entry));
384 static void iommu_set_cwwb_range(struct amd_iommu *iommu)
386 u64 start = iommu_virt_to_phys((void *)iommu->cmd_sem);
387 u64 entry = start & PM_ADDR_MASK;
389 if (!check_feature(FEATURE_SNP))
393 * Re-purpose Exclusion base/limit registers for Completion wait
394 * write-back base/limit.
396 memcpy_toio(iommu->mmio_base + MMIO_EXCL_BASE_OFFSET,
397 &entry, sizeof(entry));
400 * Default to 4 Kbytes, which can be specified by setting base
401 * address equal to the limit address.
403 memcpy_toio(iommu->mmio_base + MMIO_EXCL_LIMIT_OFFSET,
404 &entry, sizeof(entry));
407 /* Programs the physical address of the device table into the IOMMU hardware */
408 static void iommu_set_device_table(struct amd_iommu *iommu)
411 u32 dev_table_size = iommu->pci_seg->dev_table_size;
412 void *dev_table = (void *)get_dev_table(iommu);
414 BUG_ON(iommu->mmio_base == NULL);
416 entry = iommu_virt_to_phys(dev_table);
417 entry |= (dev_table_size >> 12) - 1;
418 memcpy_toio(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET,
419 &entry, sizeof(entry));
422 /* Generic functions to enable/disable certain features of the IOMMU. */
423 void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
427 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
428 ctrl |= (1ULL << bit);
429 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
432 static void iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
436 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
437 ctrl &= ~(1ULL << bit);
438 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
441 static void iommu_set_inv_tlb_timeout(struct amd_iommu *iommu, int timeout)
445 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
446 ctrl &= ~CTRL_INV_TO_MASK;
447 ctrl |= (timeout << CONTROL_INV_TIMEOUT) & CTRL_INV_TO_MASK;
448 writeq(ctrl, iommu->mmio_base + MMIO_CONTROL_OFFSET);
451 /* Function to enable the hardware */
452 static void iommu_enable(struct amd_iommu *iommu)
454 iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
457 static void iommu_disable(struct amd_iommu *iommu)
459 if (!iommu->mmio_base)
462 /* Disable command buffer */
463 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
465 /* Disable event logging and event interrupts */
466 iommu_feature_disable(iommu, CONTROL_EVT_INT_EN);
467 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
469 /* Disable IOMMU GA_LOG */
470 iommu_feature_disable(iommu, CONTROL_GALOG_EN);
471 iommu_feature_disable(iommu, CONTROL_GAINT_EN);
473 /* Disable IOMMU PPR logging */
474 iommu_feature_disable(iommu, CONTROL_PPRLOG_EN);
475 iommu_feature_disable(iommu, CONTROL_PPRINT_EN);
477 /* Disable IOMMU hardware itself */
478 iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
480 /* Clear IRTE cache disabling bit */
481 iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS);
485 * mapping and unmapping functions for the IOMMU MMIO space. Each AMD IOMMU in
486 * the system has one.
488 static u8 __iomem * __init iommu_map_mmio_space(u64 address, u64 end)
490 if (!request_mem_region(address, end, "amd_iommu")) {
491 pr_err("Can not reserve memory region %llx-%llx for mmio\n",
493 pr_err("This is a BIOS bug. Please contact your hardware vendor\n");
497 return (u8 __iomem *)ioremap(address, end);
500 static void __init iommu_unmap_mmio_space(struct amd_iommu *iommu)
502 if (iommu->mmio_base)
503 iounmap(iommu->mmio_base);
504 release_mem_region(iommu->mmio_phys, iommu->mmio_phys_end);
507 static inline u32 get_ivhd_header_size(struct ivhd_header *h)
523 /****************************************************************************
525 * The functions below belong to the first pass of AMD IOMMU ACPI table
526 * parsing. In this pass we try to find out the highest device id this
527 * code has to handle. Upon this information the size of the shared data
528 * structures is determined later.
530 ****************************************************************************/
533 * This function calculates the length of a given IVHD entry
535 static inline int ivhd_entry_length(u8 *ivhd)
537 u32 type = ((struct ivhd_entry *)ivhd)->type;
540 return 0x04 << (*ivhd >> 6);
541 } else if (type == IVHD_DEV_ACPI_HID) {
542 /* For ACPI_HID, offset 21 is uid len */
543 return *((u8 *)ivhd + 21) + 22;
549 * After reading the highest device id from the IOMMU PCI capability header
550 * this function looks if there is a higher device id defined in the ACPI table
552 static int __init find_last_devid_from_ivhd(struct ivhd_header *h)
554 u8 *p = (void *)h, *end = (void *)h;
555 struct ivhd_entry *dev;
556 int last_devid = -EINVAL;
558 u32 ivhd_size = get_ivhd_header_size(h);
561 pr_err("Unsupported IVHD type %#x\n", h->type);
569 dev = (struct ivhd_entry *)p;
572 /* Use maximum BDF value for DEV_ALL */
574 case IVHD_DEV_SELECT:
575 case IVHD_DEV_RANGE_END:
577 case IVHD_DEV_EXT_SELECT:
578 /* all the above subfield types refer to device ids */
579 if (dev->devid > last_devid)
580 last_devid = dev->devid;
585 p += ivhd_entry_length(p);
593 static int __init check_ivrs_checksum(struct acpi_table_header *table)
596 u8 checksum = 0, *p = (u8 *)table;
598 for (i = 0; i < table->length; ++i)
601 /* ACPI table corrupt */
602 pr_err(FW_BUG "IVRS invalid checksum\n");
610 * Iterate over all IVHD entries in the ACPI table and find the highest device
611 * id which we need to handle. This is the first of three functions which parse
612 * the ACPI table. So we check the checksum here.
614 static int __init find_last_devid_acpi(struct acpi_table_header *table, u16 pci_seg)
616 u8 *p = (u8 *)table, *end = (u8 *)table;
617 struct ivhd_header *h;
618 int last_devid, last_bdf = 0;
620 p += IVRS_HEADER_LENGTH;
622 end += table->length;
624 h = (struct ivhd_header *)p;
625 if (h->pci_seg == pci_seg &&
626 h->type == amd_iommu_target_ivhd_type) {
627 last_devid = find_last_devid_from_ivhd(h);
631 if (last_devid > last_bdf)
632 last_bdf = last_devid;
641 /****************************************************************************
643 * The following functions belong to the code path which parses the ACPI table
644 * the second time. In this ACPI parsing iteration we allocate IOMMU specific
645 * data structures, initialize the per PCI segment device/alias/rlookup table
646 * and also basically initialize the hardware.
648 ****************************************************************************/
650 /* Allocate per PCI segment device table */
651 static inline int __init alloc_dev_table(struct amd_iommu_pci_seg *pci_seg)
653 pci_seg->dev_table = iommu_alloc_pages(GFP_KERNEL | GFP_DMA32,
654 get_order(pci_seg->dev_table_size));
655 if (!pci_seg->dev_table)
661 static inline void free_dev_table(struct amd_iommu_pci_seg *pci_seg)
663 iommu_free_pages(pci_seg->dev_table,
664 get_order(pci_seg->dev_table_size));
665 pci_seg->dev_table = NULL;
668 /* Allocate per PCI segment IOMMU rlookup table. */
669 static inline int __init alloc_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
671 pci_seg->rlookup_table = iommu_alloc_pages(GFP_KERNEL,
672 get_order(pci_seg->rlookup_table_size));
673 if (pci_seg->rlookup_table == NULL)
679 static inline void free_rlookup_table(struct amd_iommu_pci_seg *pci_seg)
681 iommu_free_pages(pci_seg->rlookup_table,
682 get_order(pci_seg->rlookup_table_size));
683 pci_seg->rlookup_table = NULL;
686 static inline int __init alloc_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
688 pci_seg->irq_lookup_table = iommu_alloc_pages(GFP_KERNEL,
689 get_order(pci_seg->rlookup_table_size));
690 kmemleak_alloc(pci_seg->irq_lookup_table,
691 pci_seg->rlookup_table_size, 1, GFP_KERNEL);
692 if (pci_seg->irq_lookup_table == NULL)
698 static inline void free_irq_lookup_table(struct amd_iommu_pci_seg *pci_seg)
700 kmemleak_free(pci_seg->irq_lookup_table);
701 iommu_free_pages(pci_seg->irq_lookup_table,
702 get_order(pci_seg->rlookup_table_size));
703 pci_seg->irq_lookup_table = NULL;
706 static int __init alloc_alias_table(struct amd_iommu_pci_seg *pci_seg)
710 pci_seg->alias_table = iommu_alloc_pages(GFP_KERNEL,
711 get_order(pci_seg->alias_table_size));
712 if (!pci_seg->alias_table)
716 * let all alias entries point to itself
718 for (i = 0; i <= pci_seg->last_bdf; ++i)
719 pci_seg->alias_table[i] = i;
724 static void __init free_alias_table(struct amd_iommu_pci_seg *pci_seg)
726 iommu_free_pages(pci_seg->alias_table,
727 get_order(pci_seg->alias_table_size));
728 pci_seg->alias_table = NULL;
732 * Allocates the command buffer. This buffer is per AMD IOMMU. We can
733 * write commands to that buffer later and the IOMMU will execute them
736 static int __init alloc_command_buffer(struct amd_iommu *iommu)
738 iommu->cmd_buf = iommu_alloc_pages(GFP_KERNEL,
739 get_order(CMD_BUFFER_SIZE));
741 return iommu->cmd_buf ? 0 : -ENOMEM;
745 * Interrupt handler has processed all pending events and adjusted head
746 * and tail pointer. Reset overflow mask and restart logging again.
748 void amd_iommu_restart_log(struct amd_iommu *iommu, const char *evt_type,
749 u8 cntrl_intr, u8 cntrl_log,
750 u32 status_run_mask, u32 status_overflow_mask)
754 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
755 if (status & status_run_mask)
758 pr_info_ratelimited("IOMMU %s log restarting\n", evt_type);
760 iommu_feature_disable(iommu, cntrl_log);
761 iommu_feature_disable(iommu, cntrl_intr);
763 writel(status_overflow_mask, iommu->mmio_base + MMIO_STATUS_OFFSET);
765 iommu_feature_enable(iommu, cntrl_intr);
766 iommu_feature_enable(iommu, cntrl_log);
770 * This function restarts event logging in case the IOMMU experienced
771 * an event log buffer overflow.
773 void amd_iommu_restart_event_logging(struct amd_iommu *iommu)
775 amd_iommu_restart_log(iommu, "Event", CONTROL_EVT_INT_EN,
776 CONTROL_EVT_LOG_EN, MMIO_STATUS_EVT_RUN_MASK,
777 MMIO_STATUS_EVT_OVERFLOW_MASK);
781 * This function restarts event logging in case the IOMMU experienced
784 void amd_iommu_restart_ga_log(struct amd_iommu *iommu)
786 amd_iommu_restart_log(iommu, "GA", CONTROL_GAINT_EN,
787 CONTROL_GALOG_EN, MMIO_STATUS_GALOG_RUN_MASK,
788 MMIO_STATUS_GALOG_OVERFLOW_MASK);
792 * This function resets the command buffer if the IOMMU stopped fetching
795 static void amd_iommu_reset_cmd_buffer(struct amd_iommu *iommu)
797 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
799 writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
800 writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
801 iommu->cmd_buf_head = 0;
802 iommu->cmd_buf_tail = 0;
804 iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
808 * This function writes the command buffer address to the hardware and
811 static void iommu_enable_command_buffer(struct amd_iommu *iommu)
815 BUG_ON(iommu->cmd_buf == NULL);
817 entry = iommu_virt_to_phys(iommu->cmd_buf);
818 entry |= MMIO_CMD_SIZE_512;
820 memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
821 &entry, sizeof(entry));
823 amd_iommu_reset_cmd_buffer(iommu);
827 * This function disables the command buffer
829 static void iommu_disable_command_buffer(struct amd_iommu *iommu)
831 iommu_feature_disable(iommu, CONTROL_CMDBUF_EN);
834 static void __init free_command_buffer(struct amd_iommu *iommu)
836 iommu_free_pages(iommu->cmd_buf, get_order(CMD_BUFFER_SIZE));
839 void *__init iommu_alloc_4k_pages(struct amd_iommu *iommu, gfp_t gfp,
842 int order = get_order(size);
843 void *buf = iommu_alloc_pages(gfp, order);
846 check_feature(FEATURE_SNP) &&
847 set_memory_4k((unsigned long)buf, (1 << order))) {
848 iommu_free_pages(buf, order);
855 /* allocates the memory where the IOMMU will log its events to */
856 static int __init alloc_event_buffer(struct amd_iommu *iommu)
858 iommu->evt_buf = iommu_alloc_4k_pages(iommu, GFP_KERNEL,
861 return iommu->evt_buf ? 0 : -ENOMEM;
864 static void iommu_enable_event_buffer(struct amd_iommu *iommu)
868 BUG_ON(iommu->evt_buf == NULL);
870 entry = iommu_virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
872 memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
873 &entry, sizeof(entry));
875 /* set head and tail to zero manually */
876 writel(0x00, iommu->mmio_base + MMIO_EVT_HEAD_OFFSET);
877 writel(0x00, iommu->mmio_base + MMIO_EVT_TAIL_OFFSET);
879 iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
883 * This function disables the event log buffer
885 static void iommu_disable_event_buffer(struct amd_iommu *iommu)
887 iommu_feature_disable(iommu, CONTROL_EVT_LOG_EN);
890 static void __init free_event_buffer(struct amd_iommu *iommu)
892 iommu_free_pages(iommu->evt_buf, get_order(EVT_BUFFER_SIZE));
895 static void free_ga_log(struct amd_iommu *iommu)
897 #ifdef CONFIG_IRQ_REMAP
898 iommu_free_pages(iommu->ga_log, get_order(GA_LOG_SIZE));
899 iommu_free_pages(iommu->ga_log_tail, get_order(8));
903 #ifdef CONFIG_IRQ_REMAP
904 static int iommu_ga_log_enable(struct amd_iommu *iommu)
912 entry = iommu_virt_to_phys(iommu->ga_log) | GA_LOG_SIZE_512;
913 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_BASE_OFFSET,
914 &entry, sizeof(entry));
915 entry = (iommu_virt_to_phys(iommu->ga_log_tail) &
916 (BIT_ULL(52)-1)) & ~7ULL;
917 memcpy_toio(iommu->mmio_base + MMIO_GA_LOG_TAIL_OFFSET,
918 &entry, sizeof(entry));
919 writel(0x00, iommu->mmio_base + MMIO_GA_HEAD_OFFSET);
920 writel(0x00, iommu->mmio_base + MMIO_GA_TAIL_OFFSET);
923 iommu_feature_enable(iommu, CONTROL_GAINT_EN);
924 iommu_feature_enable(iommu, CONTROL_GALOG_EN);
926 for (i = 0; i < MMIO_STATUS_TIMEOUT; ++i) {
927 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
928 if (status & (MMIO_STATUS_GALOG_RUN_MASK))
933 if (WARN_ON(i >= MMIO_STATUS_TIMEOUT))
939 static int iommu_init_ga_log(struct amd_iommu *iommu)
941 if (!AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
944 iommu->ga_log = iommu_alloc_pages(GFP_KERNEL, get_order(GA_LOG_SIZE));
948 iommu->ga_log_tail = iommu_alloc_pages(GFP_KERNEL, get_order(8));
949 if (!iommu->ga_log_tail)
957 #endif /* CONFIG_IRQ_REMAP */
959 static int __init alloc_cwwb_sem(struct amd_iommu *iommu)
961 iommu->cmd_sem = iommu_alloc_4k_pages(iommu, GFP_KERNEL, 1);
963 return iommu->cmd_sem ? 0 : -ENOMEM;
966 static void __init free_cwwb_sem(struct amd_iommu *iommu)
969 iommu_free_page((void *)iommu->cmd_sem);
972 static void iommu_enable_xt(struct amd_iommu *iommu)
974 #ifdef CONFIG_IRQ_REMAP
976 * XT mode (32-bit APIC destination ID) requires
977 * GA mode (128-bit IRTE support) as a prerequisite.
979 if (AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir) &&
980 amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
981 iommu_feature_enable(iommu, CONTROL_XT_EN);
982 #endif /* CONFIG_IRQ_REMAP */
985 static void iommu_enable_gt(struct amd_iommu *iommu)
987 if (!check_feature(FEATURE_GT))
990 iommu_feature_enable(iommu, CONTROL_GT_EN);
993 /* sets a specific bit in the device table entry. */
994 static void __set_dev_entry_bit(struct dev_table_entry *dev_table,
997 int i = (bit >> 6) & 0x03;
998 int _bit = bit & 0x3f;
1000 dev_table[devid].data[i] |= (1UL << _bit);
1003 static void set_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit)
1005 struct dev_table_entry *dev_table = get_dev_table(iommu);
1007 return __set_dev_entry_bit(dev_table, devid, bit);
1010 static int __get_dev_entry_bit(struct dev_table_entry *dev_table,
1013 int i = (bit >> 6) & 0x03;
1014 int _bit = bit & 0x3f;
1016 return (dev_table[devid].data[i] & (1UL << _bit)) >> _bit;
1019 static int get_dev_entry_bit(struct amd_iommu *iommu, u16 devid, u8 bit)
1021 struct dev_table_entry *dev_table = get_dev_table(iommu);
1023 return __get_dev_entry_bit(dev_table, devid, bit);
1026 static bool __copy_device_table(struct amd_iommu *iommu)
1028 u64 int_ctl, int_tab_len, entry = 0;
1029 struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1030 struct dev_table_entry *old_devtb = NULL;
1031 u32 lo, hi, devid, old_devtb_size;
1032 phys_addr_t old_devtb_phys;
1033 u16 dom_id, dte_v, irq_v;
1036 /* Each IOMMU use separate device table with the same size */
1037 lo = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET);
1038 hi = readl(iommu->mmio_base + MMIO_DEV_TABLE_OFFSET + 4);
1039 entry = (((u64) hi) << 32) + lo;
1041 old_devtb_size = ((entry & ~PAGE_MASK) + 1) << 12;
1042 if (old_devtb_size != pci_seg->dev_table_size) {
1043 pr_err("The device table size of IOMMU:%d is not expected!\n",
1049 * When SME is enabled in the first kernel, the entry includes the
1050 * memory encryption mask(sme_me_mask), we must remove the memory
1051 * encryption mask to obtain the true physical address in kdump kernel.
1053 old_devtb_phys = __sme_clr(entry) & PAGE_MASK;
1055 if (old_devtb_phys >= 0x100000000ULL) {
1056 pr_err("The address of old device table is above 4G, not trustworthy!\n");
1059 old_devtb = (cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) && is_kdump_kernel())
1060 ? (__force void *)ioremap_encrypted(old_devtb_phys,
1061 pci_seg->dev_table_size)
1062 : memremap(old_devtb_phys, pci_seg->dev_table_size, MEMREMAP_WB);
1067 pci_seg->old_dev_tbl_cpy = iommu_alloc_pages(GFP_KERNEL | GFP_DMA32,
1068 get_order(pci_seg->dev_table_size));
1069 if (pci_seg->old_dev_tbl_cpy == NULL) {
1070 pr_err("Failed to allocate memory for copying old device table!\n");
1071 memunmap(old_devtb);
1075 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
1076 pci_seg->old_dev_tbl_cpy[devid] = old_devtb[devid];
1077 dom_id = old_devtb[devid].data[1] & DEV_DOMID_MASK;
1078 dte_v = old_devtb[devid].data[0] & DTE_FLAG_V;
1080 if (dte_v && dom_id) {
1081 pci_seg->old_dev_tbl_cpy[devid].data[0] = old_devtb[devid].data[0];
1082 pci_seg->old_dev_tbl_cpy[devid].data[1] = old_devtb[devid].data[1];
1083 __set_bit(dom_id, amd_iommu_pd_alloc_bitmap);
1084 /* If gcr3 table existed, mask it out */
1085 if (old_devtb[devid].data[0] & DTE_FLAG_GV) {
1086 tmp = DTE_GCR3_VAL_B(~0ULL) << DTE_GCR3_SHIFT_B;
1087 tmp |= DTE_GCR3_VAL_C(~0ULL) << DTE_GCR3_SHIFT_C;
1088 pci_seg->old_dev_tbl_cpy[devid].data[1] &= ~tmp;
1089 tmp = DTE_GCR3_VAL_A(~0ULL) << DTE_GCR3_SHIFT_A;
1091 pci_seg->old_dev_tbl_cpy[devid].data[0] &= ~tmp;
1095 irq_v = old_devtb[devid].data[2] & DTE_IRQ_REMAP_ENABLE;
1096 int_ctl = old_devtb[devid].data[2] & DTE_IRQ_REMAP_INTCTL_MASK;
1097 int_tab_len = old_devtb[devid].data[2] & DTE_INTTABLEN_MASK;
1098 if (irq_v && (int_ctl || int_tab_len)) {
1099 if ((int_ctl != DTE_IRQ_REMAP_INTCTL) ||
1100 (int_tab_len != DTE_INTTABLEN)) {
1101 pr_err("Wrong old irq remapping flag: %#x\n", devid);
1102 memunmap(old_devtb);
1106 pci_seg->old_dev_tbl_cpy[devid].data[2] = old_devtb[devid].data[2];
1109 memunmap(old_devtb);
1114 static bool copy_device_table(void)
1116 struct amd_iommu *iommu;
1117 struct amd_iommu_pci_seg *pci_seg;
1119 if (!amd_iommu_pre_enabled)
1122 pr_warn("Translation is already enabled - trying to copy translation structures\n");
1125 * All IOMMUs within PCI segment shares common device table.
1126 * Hence copy device table only once per PCI segment.
1128 for_each_pci_segment(pci_seg) {
1129 for_each_iommu(iommu) {
1130 if (pci_seg->id != iommu->pci_seg->id)
1132 if (!__copy_device_table(iommu))
1141 void amd_iommu_apply_erratum_63(struct amd_iommu *iommu, u16 devid)
1145 sysmgt = get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1) |
1146 (get_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2) << 1);
1149 set_dev_entry_bit(iommu, devid, DEV_ENTRY_IW);
1153 * This function takes the device specific flags read from the ACPI
1154 * table and sets up the device table entry with that information
1156 static void __init set_dev_entry_from_acpi(struct amd_iommu *iommu,
1157 u16 devid, u32 flags, u32 ext_flags)
1159 if (flags & ACPI_DEVFLAG_INITPASS)
1160 set_dev_entry_bit(iommu, devid, DEV_ENTRY_INIT_PASS);
1161 if (flags & ACPI_DEVFLAG_EXTINT)
1162 set_dev_entry_bit(iommu, devid, DEV_ENTRY_EINT_PASS);
1163 if (flags & ACPI_DEVFLAG_NMI)
1164 set_dev_entry_bit(iommu, devid, DEV_ENTRY_NMI_PASS);
1165 if (flags & ACPI_DEVFLAG_SYSMGT1)
1166 set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT1);
1167 if (flags & ACPI_DEVFLAG_SYSMGT2)
1168 set_dev_entry_bit(iommu, devid, DEV_ENTRY_SYSMGT2);
1169 if (flags & ACPI_DEVFLAG_LINT0)
1170 set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT0_PASS);
1171 if (flags & ACPI_DEVFLAG_LINT1)
1172 set_dev_entry_bit(iommu, devid, DEV_ENTRY_LINT1_PASS);
1174 amd_iommu_apply_erratum_63(iommu, devid);
1176 amd_iommu_set_rlookup_table(iommu, devid);
1179 int __init add_special_device(u8 type, u8 id, u32 *devid, bool cmd_line)
1181 struct devid_map *entry;
1182 struct list_head *list;
1184 if (type == IVHD_SPECIAL_IOAPIC)
1186 else if (type == IVHD_SPECIAL_HPET)
1191 list_for_each_entry(entry, list, list) {
1192 if (!(entry->id == id && entry->cmd_line))
1195 pr_info("Command-line override present for %s id %d - ignoring\n",
1196 type == IVHD_SPECIAL_IOAPIC ? "IOAPIC" : "HPET", id);
1198 *devid = entry->devid;
1203 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1208 entry->devid = *devid;
1209 entry->cmd_line = cmd_line;
1211 list_add_tail(&entry->list, list);
1216 static int __init add_acpi_hid_device(u8 *hid, u8 *uid, u32 *devid,
1219 struct acpihid_map_entry *entry;
1220 struct list_head *list = &acpihid_map;
1222 list_for_each_entry(entry, list, list) {
1223 if (strcmp(entry->hid, hid) ||
1224 (*uid && *entry->uid && strcmp(entry->uid, uid)) ||
1228 pr_info("Command-line override for hid:%s uid:%s\n",
1230 *devid = entry->devid;
1234 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
1238 memcpy(entry->uid, uid, strlen(uid));
1239 memcpy(entry->hid, hid, strlen(hid));
1240 entry->devid = *devid;
1241 entry->cmd_line = cmd_line;
1242 entry->root_devid = (entry->devid & (~0x7));
1244 pr_info("%s, add hid:%s, uid:%s, rdevid:%d\n",
1245 entry->cmd_line ? "cmd" : "ivrs",
1246 entry->hid, entry->uid, entry->root_devid);
1248 list_add_tail(&entry->list, list);
1252 static int __init add_early_maps(void)
1256 for (i = 0; i < early_ioapic_map_size; ++i) {
1257 ret = add_special_device(IVHD_SPECIAL_IOAPIC,
1258 early_ioapic_map[i].id,
1259 &early_ioapic_map[i].devid,
1260 early_ioapic_map[i].cmd_line);
1265 for (i = 0; i < early_hpet_map_size; ++i) {
1266 ret = add_special_device(IVHD_SPECIAL_HPET,
1267 early_hpet_map[i].id,
1268 &early_hpet_map[i].devid,
1269 early_hpet_map[i].cmd_line);
1274 for (i = 0; i < early_acpihid_map_size; ++i) {
1275 ret = add_acpi_hid_device(early_acpihid_map[i].hid,
1276 early_acpihid_map[i].uid,
1277 &early_acpihid_map[i].devid,
1278 early_acpihid_map[i].cmd_line);
1287 * Takes a pointer to an AMD IOMMU entry in the ACPI table and
1288 * initializes the hardware and our data structures with it.
1290 static int __init init_iommu_from_acpi(struct amd_iommu *iommu,
1291 struct ivhd_header *h)
1294 u8 *end = p, flags = 0;
1295 u16 devid = 0, devid_start = 0, devid_to = 0, seg_id;
1296 u32 dev_i, ext_flags = 0;
1298 struct ivhd_entry *e;
1299 struct amd_iommu_pci_seg *pci_seg = iommu->pci_seg;
1304 ret = add_early_maps();
1308 amd_iommu_apply_ivrs_quirks();
1311 * First save the recommended feature enable bits from ACPI
1313 iommu->acpi_flags = h->flags;
1316 * Done. Now parse the device entries
1318 ivhd_size = get_ivhd_header_size(h);
1320 pr_err("Unsupported IVHD type %#x\n", h->type);
1330 e = (struct ivhd_entry *)p;
1331 seg_id = pci_seg->id;
1336 DUMP_printk(" DEV_ALL\t\t\tflags: %02x\n", e->flags);
1338 for (dev_i = 0; dev_i <= pci_seg->last_bdf; ++dev_i)
1339 set_dev_entry_from_acpi(iommu, dev_i, e->flags, 0);
1341 case IVHD_DEV_SELECT:
1343 DUMP_printk(" DEV_SELECT\t\t\t devid: %04x:%02x:%02x.%x "
1345 seg_id, PCI_BUS_NUM(e->devid),
1351 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1353 case IVHD_DEV_SELECT_RANGE_START:
1355 DUMP_printk(" DEV_SELECT_RANGE_START\t "
1356 "devid: %04x:%02x:%02x.%x flags: %02x\n",
1357 seg_id, PCI_BUS_NUM(e->devid),
1362 devid_start = e->devid;
1367 case IVHD_DEV_ALIAS:
1369 DUMP_printk(" DEV_ALIAS\t\t\t devid: %04x:%02x:%02x.%x "
1370 "flags: %02x devid_to: %02x:%02x.%x\n",
1371 seg_id, PCI_BUS_NUM(e->devid),
1375 PCI_BUS_NUM(e->ext >> 8),
1376 PCI_SLOT(e->ext >> 8),
1377 PCI_FUNC(e->ext >> 8));
1380 devid_to = e->ext >> 8;
1381 set_dev_entry_from_acpi(iommu, devid , e->flags, 0);
1382 set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
1383 pci_seg->alias_table[devid] = devid_to;
1385 case IVHD_DEV_ALIAS_RANGE:
1387 DUMP_printk(" DEV_ALIAS_RANGE\t\t "
1388 "devid: %04x:%02x:%02x.%x flags: %02x "
1389 "devid_to: %04x:%02x:%02x.%x\n",
1390 seg_id, PCI_BUS_NUM(e->devid),
1394 seg_id, PCI_BUS_NUM(e->ext >> 8),
1395 PCI_SLOT(e->ext >> 8),
1396 PCI_FUNC(e->ext >> 8));
1398 devid_start = e->devid;
1400 devid_to = e->ext >> 8;
1404 case IVHD_DEV_EXT_SELECT:
1406 DUMP_printk(" DEV_EXT_SELECT\t\t devid: %04x:%02x:%02x.%x "
1407 "flags: %02x ext: %08x\n",
1408 seg_id, PCI_BUS_NUM(e->devid),
1414 set_dev_entry_from_acpi(iommu, devid, e->flags,
1417 case IVHD_DEV_EXT_SELECT_RANGE:
1419 DUMP_printk(" DEV_EXT_SELECT_RANGE\t devid: "
1420 "%04x:%02x:%02x.%x flags: %02x ext: %08x\n",
1421 seg_id, PCI_BUS_NUM(e->devid),
1426 devid_start = e->devid;
1431 case IVHD_DEV_RANGE_END:
1433 DUMP_printk(" DEV_RANGE_END\t\t devid: %04x:%02x:%02x.%x\n",
1434 seg_id, PCI_BUS_NUM(e->devid),
1436 PCI_FUNC(e->devid));
1439 for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
1441 pci_seg->alias_table[dev_i] = devid_to;
1442 set_dev_entry_from_acpi(iommu,
1443 devid_to, flags, ext_flags);
1445 set_dev_entry_from_acpi(iommu, dev_i,
1449 case IVHD_DEV_SPECIAL: {
1455 handle = e->ext & 0xff;
1456 devid = PCI_SEG_DEVID_TO_SBDF(seg_id, (e->ext >> 8));
1457 type = (e->ext >> 24) & 0xff;
1459 if (type == IVHD_SPECIAL_IOAPIC)
1461 else if (type == IVHD_SPECIAL_HPET)
1466 DUMP_printk(" DEV_SPECIAL(%s[%d])\t\tdevid: %04x:%02x:%02x.%x\n",
1468 seg_id, PCI_BUS_NUM(devid),
1472 ret = add_special_device(type, handle, &devid, false);
1477 * add_special_device might update the devid in case a
1478 * command-line override is present. So call
1479 * set_dev_entry_from_acpi after add_special_device.
1481 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1485 case IVHD_DEV_ACPI_HID: {
1487 u8 hid[ACPIHID_HID_LEN];
1488 u8 uid[ACPIHID_UID_LEN];
1491 if (h->type != 0x40) {
1492 pr_err(FW_BUG "Invalid IVHD device type %#x\n",
1497 BUILD_BUG_ON(sizeof(e->ext_hid) != ACPIHID_HID_LEN - 1);
1498 memcpy(hid, &e->ext_hid, ACPIHID_HID_LEN - 1);
1499 hid[ACPIHID_HID_LEN - 1] = '\0';
1502 pr_err(FW_BUG "Invalid HID.\n");
1508 case UID_NOT_PRESENT:
1511 pr_warn(FW_BUG "Invalid UID length.\n");
1514 case UID_IS_INTEGER:
1516 sprintf(uid, "%d", e->uid);
1519 case UID_IS_CHARACTER:
1521 memcpy(uid, &e->uid, e->uidl);
1522 uid[e->uidl] = '\0';
1529 devid = PCI_SEG_DEVID_TO_SBDF(seg_id, e->devid);
1530 DUMP_printk(" DEV_ACPI_HID(%s[%s])\t\tdevid: %04x:%02x:%02x.%x\n",
1538 ret = add_acpi_hid_device(hid, uid, &devid, false);
1543 * add_special_device might update the devid in case a
1544 * command-line override is present. So call
1545 * set_dev_entry_from_acpi after add_special_device.
1547 set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
1555 p += ivhd_entry_length(p);
1561 /* Allocate PCI segment data structure */
1562 static struct amd_iommu_pci_seg *__init alloc_pci_segment(u16 id,
1563 struct acpi_table_header *ivrs_base)
1565 struct amd_iommu_pci_seg *pci_seg;
1569 * First parse ACPI tables to find the largest Bus/Dev/Func we need to
1570 * handle in this PCI segment. Upon this information the shared data
1571 * structures for the PCI segments in the system will be allocated.
1573 last_bdf = find_last_devid_acpi(ivrs_base, id);
1577 pci_seg = kzalloc(sizeof(struct amd_iommu_pci_seg), GFP_KERNEL);
1578 if (pci_seg == NULL)
1581 pci_seg->last_bdf = last_bdf;
1582 DUMP_printk("PCI segment : 0x%0x, last bdf : 0x%04x\n", id, last_bdf);
1583 pci_seg->dev_table_size = tbl_size(DEV_TABLE_ENTRY_SIZE, last_bdf);
1584 pci_seg->alias_table_size = tbl_size(ALIAS_TABLE_ENTRY_SIZE, last_bdf);
1585 pci_seg->rlookup_table_size = tbl_size(RLOOKUP_TABLE_ENTRY_SIZE, last_bdf);
1588 init_llist_head(&pci_seg->dev_data_list);
1589 INIT_LIST_HEAD(&pci_seg->unity_map);
1590 list_add_tail(&pci_seg->list, &amd_iommu_pci_seg_list);
1592 if (alloc_dev_table(pci_seg))
1594 if (alloc_alias_table(pci_seg))
1596 if (alloc_rlookup_table(pci_seg))
1602 static struct amd_iommu_pci_seg *__init get_pci_segment(u16 id,
1603 struct acpi_table_header *ivrs_base)
1605 struct amd_iommu_pci_seg *pci_seg;
1607 for_each_pci_segment(pci_seg) {
1608 if (pci_seg->id == id)
1612 return alloc_pci_segment(id, ivrs_base);
1615 static void __init free_pci_segments(void)
1617 struct amd_iommu_pci_seg *pci_seg, *next;
1619 for_each_pci_segment_safe(pci_seg, next) {
1620 list_del(&pci_seg->list);
1621 free_irq_lookup_table(pci_seg);
1622 free_rlookup_table(pci_seg);
1623 free_alias_table(pci_seg);
1624 free_dev_table(pci_seg);
1629 static void __init free_iommu_one(struct amd_iommu *iommu)
1631 free_cwwb_sem(iommu);
1632 free_command_buffer(iommu);
1633 free_event_buffer(iommu);
1634 amd_iommu_free_ppr_log(iommu);
1636 iommu_unmap_mmio_space(iommu);
1637 amd_iommu_iopf_uninit(iommu);
1640 static void __init free_iommu_all(void)
1642 struct amd_iommu *iommu, *next;
1644 for_each_iommu_safe(iommu, next) {
1645 list_del(&iommu->list);
1646 free_iommu_one(iommu);
1652 * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
1654 * BIOS should disable L2B micellaneous clock gating by setting
1655 * L2_L2B_CK_GATE_CONTROL[CKGateL2BMiscDisable](D0F2xF4_x90[2]) = 1b
1657 static void amd_iommu_erratum_746_workaround(struct amd_iommu *iommu)
1661 if ((boot_cpu_data.x86 != 0x15) ||
1662 (boot_cpu_data.x86_model < 0x10) ||
1663 (boot_cpu_data.x86_model > 0x1f))
1666 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1667 pci_read_config_dword(iommu->dev, 0xf4, &value);
1672 /* Select NB indirect register 0x90 and enable writing */
1673 pci_write_config_dword(iommu->dev, 0xf0, 0x90 | (1 << 8));
1675 pci_write_config_dword(iommu->dev, 0xf4, value | 0x4);
1676 pci_info(iommu->dev, "Applying erratum 746 workaround\n");
1678 /* Clear the enable writing bit */
1679 pci_write_config_dword(iommu->dev, 0xf0, 0x90);
1683 * Family15h Model 30h-3fh (IOMMU Mishandles ATS Write Permission)
1685 * BIOS should enable ATS write permission check by setting
1686 * L2_DEBUG_3[AtsIgnoreIWDis](D0F2xF4_x47[0]) = 1b
1688 static void amd_iommu_ats_write_check_workaround(struct amd_iommu *iommu)
1692 if ((boot_cpu_data.x86 != 0x15) ||
1693 (boot_cpu_data.x86_model < 0x30) ||
1694 (boot_cpu_data.x86_model > 0x3f))
1697 /* Test L2_DEBUG_3[AtsIgnoreIWDis] == 1 */
1698 value = iommu_read_l2(iommu, 0x47);
1703 /* Set L2_DEBUG_3[AtsIgnoreIWDis] = 1 */
1704 iommu_write_l2(iommu, 0x47, value | BIT(0));
1706 pci_info(iommu->dev, "Applying ATS write check workaround\n");
1710 * This function glues the initialization function for one IOMMU
1711 * together and also allocates the command buffer and programs the
1712 * hardware. It does NOT enable the IOMMU. This is done afterwards.
1714 static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h,
1715 struct acpi_table_header *ivrs_base)
1717 struct amd_iommu_pci_seg *pci_seg;
1719 pci_seg = get_pci_segment(h->pci_seg, ivrs_base);
1720 if (pci_seg == NULL)
1722 iommu->pci_seg = pci_seg;
1724 raw_spin_lock_init(&iommu->lock);
1725 atomic64_set(&iommu->cmd_sem_val, 0);
1727 /* Add IOMMU to internal data structures */
1728 list_add_tail(&iommu->list, &amd_iommu_list);
1729 iommu->index = amd_iommus_present++;
1731 if (unlikely(iommu->index >= MAX_IOMMUS)) {
1732 WARN(1, "System has more IOMMUs than supported by this driver\n");
1736 /* Index is fine - add IOMMU to the array */
1737 amd_iommus[iommu->index] = iommu;
1740 * Copy data from ACPI table entry to the iommu struct
1742 iommu->devid = h->devid;
1743 iommu->cap_ptr = h->cap_ptr;
1744 iommu->mmio_phys = h->mmio_phys;
1748 /* Check if IVHD EFR contains proper max banks/counters */
1749 if ((h->efr_attr != 0) &&
1750 ((h->efr_attr & (0xF << 13)) != 0) &&
1751 ((h->efr_attr & (0x3F << 17)) != 0))
1752 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1754 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1757 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1758 * GAM also requires GA mode. Therefore, we need to
1759 * check cmpxchg16b support before enabling it.
1761 if (!boot_cpu_has(X86_FEATURE_CX16) ||
1762 ((h->efr_attr & (0x1 << IOMMU_FEAT_GASUP_SHIFT)) == 0))
1763 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1767 if (h->efr_reg & (1 << 9))
1768 iommu->mmio_phys_end = MMIO_REG_END_OFFSET;
1770 iommu->mmio_phys_end = MMIO_CNTR_CONF_OFFSET;
1773 * Note: GA (128-bit IRTE) mode requires cmpxchg16b supports.
1774 * XT, GAM also requires GA mode. Therefore, we need to
1775 * check cmpxchg16b support before enabling them.
1777 if (!boot_cpu_has(X86_FEATURE_CX16) ||
1778 ((h->efr_reg & (0x1 << IOMMU_EFR_GASUP_SHIFT)) == 0)) {
1779 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY;
1783 if (h->efr_reg & BIT(IOMMU_EFR_XTSUP_SHIFT))
1784 amd_iommu_xt_mode = IRQ_REMAP_X2APIC_MODE;
1786 early_iommu_features_init(iommu, h);
1793 iommu->mmio_base = iommu_map_mmio_space(iommu->mmio_phys,
1794 iommu->mmio_phys_end);
1795 if (!iommu->mmio_base)
1798 return init_iommu_from_acpi(iommu, h);
1801 static int __init init_iommu_one_late(struct amd_iommu *iommu)
1805 if (alloc_cwwb_sem(iommu))
1808 if (alloc_command_buffer(iommu))
1811 if (alloc_event_buffer(iommu))
1814 iommu->int_enabled = false;
1816 init_translation_status(iommu);
1817 if (translation_pre_enabled(iommu) && !is_kdump_kernel()) {
1818 iommu_disable(iommu);
1819 clear_translation_pre_enabled(iommu);
1820 pr_warn("Translation was enabled for IOMMU:%d but we are not in kdump mode\n",
1823 if (amd_iommu_pre_enabled)
1824 amd_iommu_pre_enabled = translation_pre_enabled(iommu);
1826 if (amd_iommu_irq_remap) {
1827 ret = amd_iommu_create_irq_domain(iommu);
1833 * Make sure IOMMU is not considered to translate itself. The IVRS
1834 * table tells us so, but this is a lie!
1836 iommu->pci_seg->rlookup_table[iommu->devid] = NULL;
1842 * get_highest_supported_ivhd_type - Look up the appropriate IVHD type
1843 * @ivrs: Pointer to the IVRS header
1845 * This function search through all IVDB of the maximum supported IVHD
1847 static u8 get_highest_supported_ivhd_type(struct acpi_table_header *ivrs)
1849 u8 *base = (u8 *)ivrs;
1850 struct ivhd_header *ivhd = (struct ivhd_header *)
1851 (base + IVRS_HEADER_LENGTH);
1852 u8 last_type = ivhd->type;
1853 u16 devid = ivhd->devid;
1855 while (((u8 *)ivhd - base < ivrs->length) &&
1856 (ivhd->type <= ACPI_IVHD_TYPE_MAX_SUPPORTED)) {
1857 u8 *p = (u8 *) ivhd;
1859 if (ivhd->devid == devid)
1860 last_type = ivhd->type;
1861 ivhd = (struct ivhd_header *)(p + ivhd->length);
1868 * Iterates over all IOMMU entries in the ACPI table, allocates the
1869 * IOMMU structure and initializes it with init_iommu_one()
1871 static int __init init_iommu_all(struct acpi_table_header *table)
1873 u8 *p = (u8 *)table, *end = (u8 *)table;
1874 struct ivhd_header *h;
1875 struct amd_iommu *iommu;
1878 end += table->length;
1879 p += IVRS_HEADER_LENGTH;
1881 /* Phase 1: Process all IVHD blocks */
1883 h = (struct ivhd_header *)p;
1884 if (*p == amd_iommu_target_ivhd_type) {
1886 DUMP_printk("device: %04x:%02x:%02x.%01x cap: %04x "
1887 "flags: %01x info %04x\n",
1888 h->pci_seg, PCI_BUS_NUM(h->devid),
1889 PCI_SLOT(h->devid), PCI_FUNC(h->devid),
1890 h->cap_ptr, h->flags, h->info);
1891 DUMP_printk(" mmio-addr: %016llx\n",
1894 iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
1898 ret = init_iommu_one(iommu, h, table);
1907 /* Phase 2 : Early feature support check */
1910 /* Phase 3 : Enabling IOMMU features */
1911 for_each_iommu(iommu) {
1912 ret = init_iommu_one_late(iommu);
1920 static void init_iommu_perf_ctr(struct amd_iommu *iommu)
1923 struct pci_dev *pdev = iommu->dev;
1925 if (!check_feature(FEATURE_PC))
1928 amd_iommu_pc_present = true;
1930 pci_info(pdev, "IOMMU performance counters supported\n");
1932 val = readl(iommu->mmio_base + MMIO_CNTR_CONF_OFFSET);
1933 iommu->max_banks = (u8) ((val >> 12) & 0x3f);
1934 iommu->max_counters = (u8) ((val >> 7) & 0xf);
1939 static ssize_t amd_iommu_show_cap(struct device *dev,
1940 struct device_attribute *attr,
1943 struct amd_iommu *iommu = dev_to_amd_iommu(dev);
1944 return sysfs_emit(buf, "%x\n", iommu->cap);
1946 static DEVICE_ATTR(cap, S_IRUGO, amd_iommu_show_cap, NULL);
1948 static ssize_t amd_iommu_show_features(struct device *dev,
1949 struct device_attribute *attr,
1952 return sysfs_emit(buf, "%llx:%llx\n", amd_iommu_efr, amd_iommu_efr2);
1954 static DEVICE_ATTR(features, S_IRUGO, amd_iommu_show_features, NULL);
1956 static struct attribute *amd_iommu_attrs[] = {
1958 &dev_attr_features.attr,
1962 static struct attribute_group amd_iommu_group = {
1963 .name = "amd-iommu",
1964 .attrs = amd_iommu_attrs,
1967 static const struct attribute_group *amd_iommu_groups[] = {
1973 * Note: IVHD 0x11 and 0x40 also contains exact copy
1974 * of the IOMMU Extended Feature Register [MMIO Offset 0030h].
1975 * Default to EFR in IVHD since it is available sooner (i.e. before PCI init).
1977 static void __init late_iommu_features_init(struct amd_iommu *iommu)
1979 u64 features, features2;
1981 if (!(iommu->cap & (1 << IOMMU_CAP_EFR)))
1984 /* read extended feature bits */
1985 features = readq(iommu->mmio_base + MMIO_EXT_FEATURES);
1986 features2 = readq(iommu->mmio_base + MMIO_EXT_FEATURES2);
1988 if (!amd_iommu_efr) {
1989 amd_iommu_efr = features;
1990 amd_iommu_efr2 = features2;
1995 * Sanity check and warn if EFR values from
1996 * IVHD and MMIO conflict.
1998 if (features != amd_iommu_efr ||
1999 features2 != amd_iommu_efr2) {
2001 "EFR mismatch. Use IVHD EFR (%#llx : %#llx), EFR2 (%#llx : %#llx).\n",
2002 features, amd_iommu_efr,
2003 features2, amd_iommu_efr2);
2007 static int __init iommu_init_pci(struct amd_iommu *iommu)
2009 int cap_ptr = iommu->cap_ptr;
2012 iommu->dev = pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2013 PCI_BUS_NUM(iommu->devid),
2014 iommu->devid & 0xff);
2018 /* Prevent binding other PCI device drivers to IOMMU devices */
2019 iommu->dev->match_driver = false;
2021 /* ACPI _PRT won't have an IRQ for IOMMU */
2022 iommu->dev->irq_managed = 1;
2024 pci_read_config_dword(iommu->dev, cap_ptr + MMIO_CAP_HDR_OFFSET,
2027 if (!(iommu->cap & (1 << IOMMU_CAP_IOTLB)))
2028 amd_iommu_iotlb_sup = false;
2030 late_iommu_features_init(iommu);
2032 if (check_feature(FEATURE_GT)) {
2036 pasmax = amd_iommu_efr & FEATURE_PASID_MASK;
2037 pasmax >>= FEATURE_PASID_SHIFT;
2038 iommu->iommu.max_pasids = (1 << (pasmax + 1)) - 1;
2040 BUG_ON(iommu->iommu.max_pasids & ~PASID_MASK);
2042 glxval = amd_iommu_efr & FEATURE_GLXVAL_MASK;
2043 glxval >>= FEATURE_GLXVAL_SHIFT;
2045 if (amd_iommu_max_glx_val == -1)
2046 amd_iommu_max_glx_val = glxval;
2048 amd_iommu_max_glx_val = min(amd_iommu_max_glx_val, glxval);
2050 iommu_enable_gt(iommu);
2053 if (check_feature(FEATURE_PPR) && amd_iommu_alloc_ppr_log(iommu))
2056 if (iommu->cap & (1UL << IOMMU_CAP_NPCACHE)) {
2057 pr_info("Using strict mode due to virtualization\n");
2058 iommu_set_dma_strict();
2059 amd_iommu_np_cache = true;
2062 init_iommu_perf_ctr(iommu);
2064 if (amd_iommu_pgtable == AMD_IOMMU_V2) {
2065 if (!check_feature(FEATURE_GIOSUP) ||
2066 !check_feature(FEATURE_GT)) {
2067 pr_warn("Cannot enable v2 page table for DMA-API. Fallback to v1.\n");
2068 amd_iommu_pgtable = AMD_IOMMU_V1;
2072 if (is_rd890_iommu(iommu->dev)) {
2076 pci_get_domain_bus_and_slot(iommu->pci_seg->id,
2077 iommu->dev->bus->number,
2081 * Some rd890 systems may not be fully reconfigured by the
2082 * BIOS, so it's necessary for us to store this information so
2083 * it can be reprogrammed on resume
2085 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 4,
2086 &iommu->stored_addr_lo);
2087 pci_read_config_dword(iommu->dev, iommu->cap_ptr + 8,
2088 &iommu->stored_addr_hi);
2090 /* Low bit locks writes to configuration space */
2091 iommu->stored_addr_lo &= ~1;
2093 for (i = 0; i < 6; i++)
2094 for (j = 0; j < 0x12; j++)
2095 iommu->stored_l1[i][j] = iommu_read_l1(iommu, i, j);
2097 for (i = 0; i < 0x83; i++)
2098 iommu->stored_l2[i] = iommu_read_l2(iommu, i);
2101 amd_iommu_erratum_746_workaround(iommu);
2102 amd_iommu_ats_write_check_workaround(iommu);
2104 ret = iommu_device_sysfs_add(&iommu->iommu, &iommu->dev->dev,
2105 amd_iommu_groups, "ivhd%d", iommu->index);
2110 * Allocate per IOMMU IOPF queue here so that in attach device path,
2111 * PRI capable device can be added to IOPF queue
2113 if (amd_iommu_gt_ppr_supported()) {
2114 ret = amd_iommu_iopf_init(iommu);
2119 iommu_device_register(&iommu->iommu, &amd_iommu_ops, NULL);
2121 return pci_enable_device(iommu->dev);
2124 static void print_iommu_info(void)
2127 static const char * const feat_str[] = {
2128 "PreF", "PPR", "X2APIC", "NX", "GT", "[5]",
2129 "IA", "GA", "HE", "PC"
2132 if (amd_iommu_efr) {
2133 pr_info("Extended features (%#llx, %#llx):", amd_iommu_efr, amd_iommu_efr2);
2135 for (i = 0; i < ARRAY_SIZE(feat_str); ++i) {
2136 if (check_feature(1ULL << i))
2137 pr_cont(" %s", feat_str[i]);
2140 if (check_feature(FEATURE_GAM_VAPIC))
2141 pr_cont(" GA_vAPIC");
2143 if (check_feature(FEATURE_SNP))
2149 if (irq_remapping_enabled) {
2150 pr_info("Interrupt remapping enabled\n");
2151 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2152 pr_info("X2APIC enabled\n");
2154 if (amd_iommu_pgtable == AMD_IOMMU_V2) {
2155 pr_info("V2 page table enabled (Paging mode : %d level)\n",
2156 amd_iommu_gpt_level);
2160 static int __init amd_iommu_init_pci(void)
2162 struct amd_iommu *iommu;
2163 struct amd_iommu_pci_seg *pci_seg;
2166 for_each_iommu(iommu) {
2167 ret = iommu_init_pci(iommu);
2169 pr_err("IOMMU%d: Failed to initialize IOMMU Hardware (error=%d)!\n",
2173 /* Need to setup range after PCI init */
2174 iommu_set_cwwb_range(iommu);
2178 * Order is important here to make sure any unity map requirements are
2179 * fulfilled. The unity mappings are created and written to the device
2180 * table during the iommu_init_pci() call.
2182 * After that we call init_device_table_dma() to make sure any
2183 * uninitialized DTE will block DMA, and in the end we flush the caches
2184 * of all IOMMUs to make sure the changes to the device table are
2187 for_each_pci_segment(pci_seg)
2188 init_device_table_dma(pci_seg);
2190 for_each_iommu(iommu)
2191 amd_iommu_flush_all_caches(iommu);
2199 /****************************************************************************
2201 * The following functions initialize the MSI interrupts for all IOMMUs
2202 * in the system. It's a bit challenging because there could be multiple
2203 * IOMMUs per PCI BDF but we can call pci_enable_msi(x) only once per
2206 ****************************************************************************/
2208 static int iommu_setup_msi(struct amd_iommu *iommu)
2212 r = pci_enable_msi(iommu->dev);
2216 r = request_threaded_irq(iommu->dev->irq,
2217 amd_iommu_int_handler,
2218 amd_iommu_int_thread,
2223 pci_disable_msi(iommu->dev);
2234 dest_mode_logical : 1,
2241 } __attribute__ ((packed));
2244 static struct irq_chip intcapxt_controller;
2246 static int intcapxt_irqdomain_activate(struct irq_domain *domain,
2247 struct irq_data *irqd, bool reserve)
2252 static void intcapxt_irqdomain_deactivate(struct irq_domain *domain,
2253 struct irq_data *irqd)
2258 static int intcapxt_irqdomain_alloc(struct irq_domain *domain, unsigned int virq,
2259 unsigned int nr_irqs, void *arg)
2261 struct irq_alloc_info *info = arg;
2264 if (!info || info->type != X86_IRQ_ALLOC_TYPE_AMDVI)
2267 ret = irq_domain_alloc_irqs_parent(domain, virq, nr_irqs, arg);
2271 for (i = virq; i < virq + nr_irqs; i++) {
2272 struct irq_data *irqd = irq_domain_get_irq_data(domain, i);
2274 irqd->chip = &intcapxt_controller;
2275 irqd->hwirq = info->hwirq;
2276 irqd->chip_data = info->data;
2277 __irq_set_handler(i, handle_edge_irq, 0, "edge");
2283 static void intcapxt_irqdomain_free(struct irq_domain *domain, unsigned int virq,
2284 unsigned int nr_irqs)
2286 irq_domain_free_irqs_top(domain, virq, nr_irqs);
2290 static void intcapxt_unmask_irq(struct irq_data *irqd)
2292 struct amd_iommu *iommu = irqd->chip_data;
2293 struct irq_cfg *cfg = irqd_cfg(irqd);
2297 xt.dest_mode_logical = apic->dest_mode_logical;
2298 xt.vector = cfg->vector;
2299 xt.destid_0_23 = cfg->dest_apicid & GENMASK(23, 0);
2300 xt.destid_24_31 = cfg->dest_apicid >> 24;
2302 writeq(xt.capxt, iommu->mmio_base + irqd->hwirq);
2305 static void intcapxt_mask_irq(struct irq_data *irqd)
2307 struct amd_iommu *iommu = irqd->chip_data;
2309 writeq(0, iommu->mmio_base + irqd->hwirq);
2313 static int intcapxt_set_affinity(struct irq_data *irqd,
2314 const struct cpumask *mask, bool force)
2316 struct irq_data *parent = irqd->parent_data;
2319 ret = parent->chip->irq_set_affinity(parent, mask, force);
2320 if (ret < 0 || ret == IRQ_SET_MASK_OK_DONE)
2325 static int intcapxt_set_wake(struct irq_data *irqd, unsigned int on)
2327 return on ? -EOPNOTSUPP : 0;
2330 static struct irq_chip intcapxt_controller = {
2331 .name = "IOMMU-MSI",
2332 .irq_unmask = intcapxt_unmask_irq,
2333 .irq_mask = intcapxt_mask_irq,
2334 .irq_ack = irq_chip_ack_parent,
2335 .irq_retrigger = irq_chip_retrigger_hierarchy,
2336 .irq_set_affinity = intcapxt_set_affinity,
2337 .irq_set_wake = intcapxt_set_wake,
2338 .flags = IRQCHIP_MASK_ON_SUSPEND,
2341 static const struct irq_domain_ops intcapxt_domain_ops = {
2342 .alloc = intcapxt_irqdomain_alloc,
2343 .free = intcapxt_irqdomain_free,
2344 .activate = intcapxt_irqdomain_activate,
2345 .deactivate = intcapxt_irqdomain_deactivate,
2349 static struct irq_domain *iommu_irqdomain;
2351 static struct irq_domain *iommu_get_irqdomain(void)
2353 struct fwnode_handle *fn;
2355 /* No need for locking here (yet) as the init is single-threaded */
2356 if (iommu_irqdomain)
2357 return iommu_irqdomain;
2359 fn = irq_domain_alloc_named_fwnode("AMD-Vi-MSI");
2363 iommu_irqdomain = irq_domain_create_hierarchy(x86_vector_domain, 0, 0,
2364 fn, &intcapxt_domain_ops,
2366 if (!iommu_irqdomain)
2367 irq_domain_free_fwnode(fn);
2369 return iommu_irqdomain;
2372 static int __iommu_setup_intcapxt(struct amd_iommu *iommu, const char *devname,
2373 int hwirq, irq_handler_t thread_fn)
2375 struct irq_domain *domain;
2376 struct irq_alloc_info info;
2378 int node = dev_to_node(&iommu->dev->dev);
2380 domain = iommu_get_irqdomain();
2384 init_irq_alloc_info(&info, NULL);
2385 info.type = X86_IRQ_ALLOC_TYPE_AMDVI;
2389 irq = irq_domain_alloc_irqs(domain, 1, node, &info);
2391 irq_domain_remove(domain);
2395 ret = request_threaded_irq(irq, amd_iommu_int_handler,
2396 thread_fn, 0, devname, iommu);
2398 irq_domain_free_irqs(irq, 1);
2399 irq_domain_remove(domain);
2406 static int iommu_setup_intcapxt(struct amd_iommu *iommu)
2410 snprintf(iommu->evt_irq_name, sizeof(iommu->evt_irq_name),
2411 "AMD-Vi%d-Evt", iommu->index);
2412 ret = __iommu_setup_intcapxt(iommu, iommu->evt_irq_name,
2413 MMIO_INTCAPXT_EVT_OFFSET,
2414 amd_iommu_int_thread_evtlog);
2418 snprintf(iommu->ppr_irq_name, sizeof(iommu->ppr_irq_name),
2419 "AMD-Vi%d-PPR", iommu->index);
2420 ret = __iommu_setup_intcapxt(iommu, iommu->ppr_irq_name,
2421 MMIO_INTCAPXT_PPR_OFFSET,
2422 amd_iommu_int_thread_pprlog);
2426 #ifdef CONFIG_IRQ_REMAP
2427 snprintf(iommu->ga_irq_name, sizeof(iommu->ga_irq_name),
2428 "AMD-Vi%d-GA", iommu->index);
2429 ret = __iommu_setup_intcapxt(iommu, iommu->ga_irq_name,
2430 MMIO_INTCAPXT_GALOG_OFFSET,
2431 amd_iommu_int_thread_galog);
2437 static int iommu_init_irq(struct amd_iommu *iommu)
2441 if (iommu->int_enabled)
2444 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2445 ret = iommu_setup_intcapxt(iommu);
2446 else if (iommu->dev->msi_cap)
2447 ret = iommu_setup_msi(iommu);
2454 iommu->int_enabled = true;
2457 if (amd_iommu_xt_mode == IRQ_REMAP_X2APIC_MODE)
2458 iommu_feature_enable(iommu, CONTROL_INTCAPXT_EN);
2460 iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
2465 /****************************************************************************
2467 * The next functions belong to the third pass of parsing the ACPI
2468 * table. In this last pass the memory mapping requirements are
2469 * gathered (like exclusion and unity mapping ranges).
2471 ****************************************************************************/
2473 static void __init free_unity_maps(void)
2475 struct unity_map_entry *entry, *next;
2476 struct amd_iommu_pci_seg *p, *pci_seg;
2478 for_each_pci_segment_safe(pci_seg, p) {
2479 list_for_each_entry_safe(entry, next, &pci_seg->unity_map, list) {
2480 list_del(&entry->list);
2486 /* called for unity map ACPI definition */
2487 static int __init init_unity_map_range(struct ivmd_header *m,
2488 struct acpi_table_header *ivrs_base)
2490 struct unity_map_entry *e = NULL;
2491 struct amd_iommu_pci_seg *pci_seg;
2494 pci_seg = get_pci_segment(m->pci_seg, ivrs_base);
2495 if (pci_seg == NULL)
2498 e = kzalloc(sizeof(*e), GFP_KERNEL);
2506 case ACPI_IVMD_TYPE:
2507 s = "IVMD_TYPEi\t\t\t";
2508 e->devid_start = e->devid_end = m->devid;
2510 case ACPI_IVMD_TYPE_ALL:
2511 s = "IVMD_TYPE_ALL\t\t";
2513 e->devid_end = pci_seg->last_bdf;
2515 case ACPI_IVMD_TYPE_RANGE:
2516 s = "IVMD_TYPE_RANGE\t\t";
2517 e->devid_start = m->devid;
2518 e->devid_end = m->aux;
2521 e->address_start = PAGE_ALIGN(m->range_start);
2522 e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
2523 e->prot = m->flags >> 1;
2526 * Treat per-device exclusion ranges as r/w unity-mapped regions
2527 * since some buggy BIOSes might lead to the overwritten exclusion
2528 * range (exclusion_start and exclusion_length members). This
2529 * happens when there are multiple exclusion ranges (IVMD entries)
2530 * defined in ACPI table.
2532 if (m->flags & IVMD_FLAG_EXCL_RANGE)
2533 e->prot = (IVMD_FLAG_IW | IVMD_FLAG_IR) >> 1;
2535 DUMP_printk("%s devid_start: %04x:%02x:%02x.%x devid_end: "
2536 "%04x:%02x:%02x.%x range_start: %016llx range_end: %016llx"
2537 " flags: %x\n", s, m->pci_seg,
2538 PCI_BUS_NUM(e->devid_start), PCI_SLOT(e->devid_start),
2539 PCI_FUNC(e->devid_start), m->pci_seg,
2540 PCI_BUS_NUM(e->devid_end),
2541 PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
2542 e->address_start, e->address_end, m->flags);
2544 list_add_tail(&e->list, &pci_seg->unity_map);
2549 /* iterates over all memory definitions we find in the ACPI table */
2550 static int __init init_memory_definitions(struct acpi_table_header *table)
2552 u8 *p = (u8 *)table, *end = (u8 *)table;
2553 struct ivmd_header *m;
2555 end += table->length;
2556 p += IVRS_HEADER_LENGTH;
2559 m = (struct ivmd_header *)p;
2560 if (m->flags & (IVMD_FLAG_UNITY_MAP | IVMD_FLAG_EXCL_RANGE))
2561 init_unity_map_range(m, table);
2570 * Init the device table to not allow DMA access for devices
2572 static void init_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2575 struct dev_table_entry *dev_table = pci_seg->dev_table;
2577 if (dev_table == NULL)
2580 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2581 __set_dev_entry_bit(dev_table, devid, DEV_ENTRY_VALID);
2582 if (!amd_iommu_snp_en)
2583 __set_dev_entry_bit(dev_table, devid, DEV_ENTRY_TRANSLATION);
2587 static void __init uninit_device_table_dma(struct amd_iommu_pci_seg *pci_seg)
2590 struct dev_table_entry *dev_table = pci_seg->dev_table;
2592 if (dev_table == NULL)
2595 for (devid = 0; devid <= pci_seg->last_bdf; ++devid) {
2596 dev_table[devid].data[0] = 0ULL;
2597 dev_table[devid].data[1] = 0ULL;
2601 static void init_device_table(void)
2603 struct amd_iommu_pci_seg *pci_seg;
2606 if (!amd_iommu_irq_remap)
2609 for_each_pci_segment(pci_seg) {
2610 for (devid = 0; devid <= pci_seg->last_bdf; ++devid)
2611 __set_dev_entry_bit(pci_seg->dev_table,
2612 devid, DEV_ENTRY_IRQ_TBL_EN);
2616 static void iommu_init_flags(struct amd_iommu *iommu)
2618 iommu->acpi_flags & IVHD_FLAG_HT_TUN_EN_MASK ?
2619 iommu_feature_enable(iommu, CONTROL_HT_TUN_EN) :
2620 iommu_feature_disable(iommu, CONTROL_HT_TUN_EN);
2622 iommu->acpi_flags & IVHD_FLAG_PASSPW_EN_MASK ?
2623 iommu_feature_enable(iommu, CONTROL_PASSPW_EN) :
2624 iommu_feature_disable(iommu, CONTROL_PASSPW_EN);
2626 iommu->acpi_flags & IVHD_FLAG_RESPASSPW_EN_MASK ?
2627 iommu_feature_enable(iommu, CONTROL_RESPASSPW_EN) :
2628 iommu_feature_disable(iommu, CONTROL_RESPASSPW_EN);
2630 iommu->acpi_flags & IVHD_FLAG_ISOC_EN_MASK ?
2631 iommu_feature_enable(iommu, CONTROL_ISOC_EN) :
2632 iommu_feature_disable(iommu, CONTROL_ISOC_EN);
2635 * make IOMMU memory accesses cache coherent
2637 iommu_feature_enable(iommu, CONTROL_COHERENT_EN);
2639 /* Set IOTLB invalidation timeout to 1s */
2640 iommu_set_inv_tlb_timeout(iommu, CTRL_INV_TO_1S);
2643 static void iommu_apply_resume_quirks(struct amd_iommu *iommu)
2646 u32 ioc_feature_control;
2647 struct pci_dev *pdev = iommu->root_pdev;
2649 /* RD890 BIOSes may not have completely reconfigured the iommu */
2650 if (!is_rd890_iommu(iommu->dev) || !pdev)
2654 * First, we need to ensure that the iommu is enabled. This is
2655 * controlled by a register in the northbridge
2658 /* Select Northbridge indirect register 0x75 and enable writing */
2659 pci_write_config_dword(pdev, 0x60, 0x75 | (1 << 7));
2660 pci_read_config_dword(pdev, 0x64, &ioc_feature_control);
2662 /* Enable the iommu */
2663 if (!(ioc_feature_control & 0x1))
2664 pci_write_config_dword(pdev, 0x64, ioc_feature_control | 1);
2666 /* Restore the iommu BAR */
2667 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2668 iommu->stored_addr_lo);
2669 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 8,
2670 iommu->stored_addr_hi);
2672 /* Restore the l1 indirect regs for each of the 6 l1s */
2673 for (i = 0; i < 6; i++)
2674 for (j = 0; j < 0x12; j++)
2675 iommu_write_l1(iommu, i, j, iommu->stored_l1[i][j]);
2677 /* Restore the l2 indirect regs */
2678 for (i = 0; i < 0x83; i++)
2679 iommu_write_l2(iommu, i, iommu->stored_l2[i]);
2681 /* Lock PCI setup registers */
2682 pci_write_config_dword(iommu->dev, iommu->cap_ptr + 4,
2683 iommu->stored_addr_lo | 1);
2686 static void iommu_enable_ga(struct amd_iommu *iommu)
2688 #ifdef CONFIG_IRQ_REMAP
2689 switch (amd_iommu_guest_ir) {
2690 case AMD_IOMMU_GUEST_IR_VAPIC:
2691 case AMD_IOMMU_GUEST_IR_LEGACY_GA:
2692 iommu_feature_enable(iommu, CONTROL_GA_EN);
2693 iommu->irte_ops = &irte_128_ops;
2696 iommu->irte_ops = &irte_32_ops;
2702 static void iommu_disable_irtcachedis(struct amd_iommu *iommu)
2704 iommu_feature_disable(iommu, CONTROL_IRTCACHEDIS);
2707 static void iommu_enable_irtcachedis(struct amd_iommu *iommu)
2711 if (!amd_iommu_irtcachedis)
2716 * The support for IRTCacheDis feature is dertermined by
2717 * checking if the bit is writable.
2719 iommu_feature_enable(iommu, CONTROL_IRTCACHEDIS);
2720 ctrl = readq(iommu->mmio_base + MMIO_CONTROL_OFFSET);
2721 ctrl &= (1ULL << CONTROL_IRTCACHEDIS);
2723 iommu->irtcachedis_enabled = true;
2724 pr_info("iommu%d (%#06x) : IRT cache is %s\n",
2725 iommu->index, iommu->devid,
2726 iommu->irtcachedis_enabled ? "disabled" : "enabled");
2729 static void early_enable_iommu(struct amd_iommu *iommu)
2731 iommu_disable(iommu);
2732 iommu_init_flags(iommu);
2733 iommu_set_device_table(iommu);
2734 iommu_enable_command_buffer(iommu);
2735 iommu_enable_event_buffer(iommu);
2736 iommu_set_exclusion_range(iommu);
2737 iommu_enable_ga(iommu);
2738 iommu_enable_xt(iommu);
2739 iommu_enable_irtcachedis(iommu);
2740 iommu_enable(iommu);
2741 amd_iommu_flush_all_caches(iommu);
2745 * This function finally enables all IOMMUs found in the system after
2746 * they have been initialized.
2748 * Or if in kdump kernel and IOMMUs are all pre-enabled, try to copy
2749 * the old content of device table entries. Not this case or copy failed,
2750 * just continue as normal kernel does.
2752 static void early_enable_iommus(void)
2754 struct amd_iommu *iommu;
2755 struct amd_iommu_pci_seg *pci_seg;
2757 if (!copy_device_table()) {
2759 * If come here because of failure in copying device table from old
2760 * kernel with all IOMMUs enabled, print error message and try to
2761 * free allocated old_dev_tbl_cpy.
2763 if (amd_iommu_pre_enabled)
2764 pr_err("Failed to copy DEV table from previous kernel.\n");
2766 for_each_pci_segment(pci_seg) {
2767 if (pci_seg->old_dev_tbl_cpy != NULL) {
2768 iommu_free_pages(pci_seg->old_dev_tbl_cpy,
2769 get_order(pci_seg->dev_table_size));
2770 pci_seg->old_dev_tbl_cpy = NULL;
2774 for_each_iommu(iommu) {
2775 clear_translation_pre_enabled(iommu);
2776 early_enable_iommu(iommu);
2779 pr_info("Copied DEV table from previous kernel.\n");
2781 for_each_pci_segment(pci_seg) {
2782 iommu_free_pages(pci_seg->dev_table,
2783 get_order(pci_seg->dev_table_size));
2784 pci_seg->dev_table = pci_seg->old_dev_tbl_cpy;
2787 for_each_iommu(iommu) {
2788 iommu_disable_command_buffer(iommu);
2789 iommu_disable_event_buffer(iommu);
2790 iommu_disable_irtcachedis(iommu);
2791 iommu_enable_command_buffer(iommu);
2792 iommu_enable_event_buffer(iommu);
2793 iommu_enable_ga(iommu);
2794 iommu_enable_xt(iommu);
2795 iommu_enable_irtcachedis(iommu);
2796 iommu_set_device_table(iommu);
2797 amd_iommu_flush_all_caches(iommu);
2802 static void enable_iommus_ppr(void)
2804 struct amd_iommu *iommu;
2806 if (!amd_iommu_gt_ppr_supported())
2809 for_each_iommu(iommu)
2810 amd_iommu_enable_ppr_log(iommu);
2813 static void enable_iommus_vapic(void)
2815 #ifdef CONFIG_IRQ_REMAP
2817 struct amd_iommu *iommu;
2819 for_each_iommu(iommu) {
2821 * Disable GALog if already running. It could have been enabled
2822 * in the previous boot before kdump.
2824 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2825 if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2828 iommu_feature_disable(iommu, CONTROL_GALOG_EN);
2829 iommu_feature_disable(iommu, CONTROL_GAINT_EN);
2832 * Need to set and poll check the GALOGRun bit to zero before
2833 * we can set/ modify GA Log registers safely.
2835 for (i = 0; i < MMIO_STATUS_TIMEOUT; ++i) {
2836 status = readl(iommu->mmio_base + MMIO_STATUS_OFFSET);
2837 if (!(status & MMIO_STATUS_GALOG_RUN_MASK))
2842 if (WARN_ON(i >= MMIO_STATUS_TIMEOUT))
2846 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir) &&
2847 !check_feature(FEATURE_GAM_VAPIC)) {
2848 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2852 if (amd_iommu_snp_en &&
2853 !FEATURE_SNPAVICSUP_GAM(amd_iommu_efr2)) {
2854 pr_warn("Force to disable Virtual APIC due to SNP\n");
2855 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
2859 /* Enabling GAM and SNPAVIC support */
2860 for_each_iommu(iommu) {
2861 if (iommu_init_ga_log(iommu) ||
2862 iommu_ga_log_enable(iommu))
2865 iommu_feature_enable(iommu, CONTROL_GAM_EN);
2866 if (amd_iommu_snp_en)
2867 iommu_feature_enable(iommu, CONTROL_SNPAVIC_EN);
2870 amd_iommu_irq_ops.capability |= (1 << IRQ_POSTING_CAP);
2871 pr_info("Virtual APIC enabled\n");
2875 static void enable_iommus(void)
2877 early_enable_iommus();
2880 static void disable_iommus(void)
2882 struct amd_iommu *iommu;
2884 for_each_iommu(iommu)
2885 iommu_disable(iommu);
2887 #ifdef CONFIG_IRQ_REMAP
2888 if (AMD_IOMMU_GUEST_IR_VAPIC(amd_iommu_guest_ir))
2889 amd_iommu_irq_ops.capability &= ~(1 << IRQ_POSTING_CAP);
2894 * Suspend/Resume support
2895 * disable suspend until real resume implemented
2898 static void amd_iommu_resume(void)
2900 struct amd_iommu *iommu;
2902 for_each_iommu(iommu)
2903 iommu_apply_resume_quirks(iommu);
2905 /* re-load the hardware */
2908 amd_iommu_enable_interrupts();
2911 static int amd_iommu_suspend(void)
2913 /* disable IOMMUs to go out of the way for BIOS */
2919 static struct syscore_ops amd_iommu_syscore_ops = {
2920 .suspend = amd_iommu_suspend,
2921 .resume = amd_iommu_resume,
2924 static void __init free_iommu_resources(void)
2926 kmem_cache_destroy(amd_iommu_irq_cache);
2927 amd_iommu_irq_cache = NULL;
2930 free_pci_segments();
2933 /* SB IOAPIC is always on this device in AMD systems */
2934 #define IOAPIC_SB_DEVID ((0x00 << 8) | PCI_DEVFN(0x14, 0))
2936 static bool __init check_ioapic_information(void)
2938 const char *fw_bug = FW_BUG;
2939 bool ret, has_sb_ioapic;
2942 has_sb_ioapic = false;
2946 * If we have map overrides on the kernel command line the
2947 * messages in this function might not describe firmware bugs
2948 * anymore - so be careful
2953 for (idx = 0; idx < nr_ioapics; idx++) {
2954 int devid, id = mpc_ioapic_id(idx);
2956 devid = get_ioapic_devid(id);
2958 pr_err("%s: IOAPIC[%d] not in IVRS table\n",
2961 } else if (devid == IOAPIC_SB_DEVID) {
2962 has_sb_ioapic = true;
2967 if (!has_sb_ioapic) {
2969 * We expect the SB IOAPIC to be listed in the IVRS
2970 * table. The system timer is connected to the SB IOAPIC
2971 * and if we don't have it in the list the system will
2972 * panic at boot time. This situation usually happens
2973 * when the BIOS is buggy and provides us the wrong
2974 * device id for the IOAPIC in the system.
2976 pr_err("%s: No southbridge IOAPIC found\n", fw_bug);
2980 pr_err("Disabling interrupt remapping\n");
2985 static void __init free_dma_resources(void)
2987 iommu_free_pages(amd_iommu_pd_alloc_bitmap,
2988 get_order(MAX_DOMAIN_ID / 8));
2989 amd_iommu_pd_alloc_bitmap = NULL;
2994 static void __init ivinfo_init(void *ivrs)
2996 amd_iommu_ivinfo = *((u32 *)(ivrs + IOMMU_IVINFO_OFFSET));
3000 * This is the hardware init function for AMD IOMMU in the system.
3001 * This function is called either from amd_iommu_init or from the interrupt
3002 * remapping setup code.
3004 * This function basically parses the ACPI table for AMD IOMMU (IVRS)
3007 * 1 pass) Discover the most comprehensive IVHD type to use.
3009 * 2 pass) Find the highest PCI device id the driver has to handle.
3010 * Upon this information the size of the data structures is
3011 * determined that needs to be allocated.
3013 * 3 pass) Initialize the data structures just allocated with the
3014 * information in the ACPI table about available AMD IOMMUs
3015 * in the system. It also maps the PCI devices in the
3016 * system to specific IOMMUs
3018 * 4 pass) After the basic data structures are allocated and
3019 * initialized we update them with information about memory
3020 * remapping requirements parsed out of the ACPI table in
3023 * After everything is set up the IOMMUs are enabled and the necessary
3024 * hotplug and suspend notifiers are registered.
3026 static int __init early_amd_iommu_init(void)
3028 struct acpi_table_header *ivrs_base;
3029 int remap_cache_sz, ret;
3032 if (!amd_iommu_detected)
3035 status = acpi_get_table("IVRS", 0, &ivrs_base);
3036 if (status == AE_NOT_FOUND)
3038 else if (ACPI_FAILURE(status)) {
3039 const char *err = acpi_format_exception(status);
3040 pr_err("IVRS table error: %s\n", err);
3045 * Validate checksum here so we don't need to do it when
3046 * we actually parse the table
3048 ret = check_ivrs_checksum(ivrs_base);
3052 ivinfo_init(ivrs_base);
3054 amd_iommu_target_ivhd_type = get_highest_supported_ivhd_type(ivrs_base);
3055 DUMP_printk("Using IVHD type %#x\n", amd_iommu_target_ivhd_type);
3057 /* Device table - directly used by all IOMMUs */
3060 amd_iommu_pd_alloc_bitmap = iommu_alloc_pages(GFP_KERNEL,
3061 get_order(MAX_DOMAIN_ID / 8));
3062 if (amd_iommu_pd_alloc_bitmap == NULL)
3066 * never allocate domain 0 because its used as the non-allocated and
3067 * error value placeholder
3069 __set_bit(0, amd_iommu_pd_alloc_bitmap);
3072 * now the data structures are allocated and basically initialized
3073 * start the real acpi table scan
3075 ret = init_iommu_all(ivrs_base);
3079 /* 5 level guest page table */
3080 if (cpu_feature_enabled(X86_FEATURE_LA57) &&
3081 check_feature_gpt_level() == GUEST_PGTABLE_5_LEVEL)
3082 amd_iommu_gpt_level = PAGE_MODE_5_LEVEL;
3084 /* Disable any previously enabled IOMMUs */
3085 if (!is_kdump_kernel() || amd_iommu_disabled)
3088 if (amd_iommu_irq_remap)
3089 amd_iommu_irq_remap = check_ioapic_information();
3091 if (amd_iommu_irq_remap) {
3092 struct amd_iommu_pci_seg *pci_seg;
3094 * Interrupt remapping enabled, create kmem_cache for the
3098 if (!AMD_IOMMU_GUEST_IR_GA(amd_iommu_guest_ir))
3099 remap_cache_sz = MAX_IRQS_PER_TABLE * sizeof(u32);
3101 remap_cache_sz = MAX_IRQS_PER_TABLE * (sizeof(u64) * 2);
3102 amd_iommu_irq_cache = kmem_cache_create("irq_remap_cache",
3104 DTE_INTTAB_ALIGNMENT,
3106 if (!amd_iommu_irq_cache)
3109 for_each_pci_segment(pci_seg) {
3110 if (alloc_irq_lookup_table(pci_seg))
3115 ret = init_memory_definitions(ivrs_base);
3119 /* init the device table */
3120 init_device_table();
3123 /* Don't leak any ACPI memory */
3124 acpi_put_table(ivrs_base);
3129 static int amd_iommu_enable_interrupts(void)
3131 struct amd_iommu *iommu;
3134 for_each_iommu(iommu) {
3135 ret = iommu_init_irq(iommu);
3141 * Interrupt handler is ready to process interrupts. Enable
3142 * PPR and GA log interrupt for all IOMMUs.
3144 enable_iommus_vapic();
3145 enable_iommus_ppr();
3151 static bool __init detect_ivrs(void)
3153 struct acpi_table_header *ivrs_base;
3157 status = acpi_get_table("IVRS", 0, &ivrs_base);
3158 if (status == AE_NOT_FOUND)
3160 else if (ACPI_FAILURE(status)) {
3161 const char *err = acpi_format_exception(status);
3162 pr_err("IVRS table error: %s\n", err);
3166 acpi_put_table(ivrs_base);
3168 if (amd_iommu_force_enable)
3171 /* Don't use IOMMU if there is Stoney Ridge graphics */
3172 for (i = 0; i < 32; i++) {
3175 pci_id = read_pci_config(0, i, 0, 0);
3176 if ((pci_id & 0xffff) == 0x1002 && (pci_id >> 16) == 0x98e4) {
3177 pr_info("Disable IOMMU on Stoney Ridge\n");
3183 /* Make sure ACS will be enabled during PCI probe */
3189 static void iommu_snp_enable(void)
3191 #ifdef CONFIG_KVM_AMD_SEV
3192 if (!cc_platform_has(CC_ATTR_HOST_SEV_SNP))
3195 * The SNP support requires that IOMMU must be enabled, and is
3196 * configured with V1 page table (DTE[Mode] = 0 is not supported).
3198 if (no_iommu || iommu_default_passthrough()) {
3199 pr_warn("SNP: IOMMU disabled or configured in passthrough mode, SNP cannot be supported.\n");
3203 if (amd_iommu_pgtable != AMD_IOMMU_V1) {
3204 pr_warn("SNP: IOMMU is configured with V2 page table mode, SNP cannot be supported.\n");
3208 amd_iommu_snp_en = check_feature(FEATURE_SNP);
3209 if (!amd_iommu_snp_en) {
3210 pr_warn("SNP: IOMMU SNP feature not enabled, SNP cannot be supported.\n");
3214 pr_info("IOMMU SNP support enabled.\n");
3218 cc_platform_clear(CC_ATTR_HOST_SEV_SNP);
3222 /****************************************************************************
3224 * AMD IOMMU Initialization State Machine
3226 ****************************************************************************/
3228 static int __init state_next(void)
3232 switch (init_state) {
3233 case IOMMU_START_STATE:
3234 if (!detect_ivrs()) {
3235 init_state = IOMMU_NOT_FOUND;
3238 init_state = IOMMU_IVRS_DETECTED;
3241 case IOMMU_IVRS_DETECTED:
3242 if (amd_iommu_disabled) {
3243 init_state = IOMMU_CMDLINE_DISABLED;
3246 ret = early_amd_iommu_init();
3247 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_ACPI_FINISHED;
3250 case IOMMU_ACPI_FINISHED:
3251 early_enable_iommus();
3252 x86_platform.iommu_shutdown = disable_iommus;
3253 init_state = IOMMU_ENABLED;
3256 register_syscore_ops(&amd_iommu_syscore_ops);
3258 ret = amd_iommu_init_pci();
3259 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_PCI_INIT;
3261 case IOMMU_PCI_INIT:
3262 ret = amd_iommu_enable_interrupts();
3263 init_state = ret ? IOMMU_INIT_ERROR : IOMMU_INTERRUPTS_EN;
3265 case IOMMU_INTERRUPTS_EN:
3266 init_state = IOMMU_INITIALIZED;
3268 case IOMMU_INITIALIZED:
3271 case IOMMU_NOT_FOUND:
3272 case IOMMU_INIT_ERROR:
3273 case IOMMU_CMDLINE_DISABLED:
3274 /* Error states => do nothing */
3283 free_dma_resources();
3284 if (!irq_remapping_enabled) {
3286 free_iommu_resources();
3288 struct amd_iommu *iommu;
3289 struct amd_iommu_pci_seg *pci_seg;
3291 for_each_pci_segment(pci_seg)
3292 uninit_device_table_dma(pci_seg);
3294 for_each_iommu(iommu)
3295 amd_iommu_flush_all_caches(iommu);
3301 static int __init iommu_go_to_state(enum iommu_init_state state)
3305 while (init_state != state) {
3306 if (init_state == IOMMU_NOT_FOUND ||
3307 init_state == IOMMU_INIT_ERROR ||
3308 init_state == IOMMU_CMDLINE_DISABLED)
3316 #ifdef CONFIG_IRQ_REMAP
3317 int __init amd_iommu_prepare(void)
3321 amd_iommu_irq_remap = true;
3323 ret = iommu_go_to_state(IOMMU_ACPI_FINISHED);
3325 amd_iommu_irq_remap = false;
3329 return amd_iommu_irq_remap ? 0 : -ENODEV;
3332 int __init amd_iommu_enable(void)
3336 ret = iommu_go_to_state(IOMMU_ENABLED);
3340 irq_remapping_enabled = 1;
3341 return amd_iommu_xt_mode;
3344 void amd_iommu_disable(void)
3346 amd_iommu_suspend();
3349 int amd_iommu_reenable(int mode)
3356 int __init amd_iommu_enable_faulting(unsigned int cpu)
3358 /* We enable MSI later when PCI is initialized */
3364 * This is the core init function for AMD IOMMU hardware in the system.
3365 * This function is called from the generic x86 DMA layer initialization
3368 static int __init amd_iommu_init(void)
3370 struct amd_iommu *iommu;
3373 ret = iommu_go_to_state(IOMMU_INITIALIZED);
3374 #ifdef CONFIG_GART_IOMMU
3375 if (ret && list_empty(&amd_iommu_list)) {
3377 * We failed to initialize the AMD IOMMU - try fallback
3378 * to GART if possible.
3384 for_each_iommu(iommu)
3385 amd_iommu_debugfs_setup(iommu);
3390 static bool amd_iommu_sme_check(void)
3392 if (!cc_platform_has(CC_ATTR_HOST_MEM_ENCRYPT) ||
3393 (boot_cpu_data.x86 != 0x17))
3396 /* For Fam17h, a specific level of support is required */
3397 if (boot_cpu_data.microcode >= 0x08001205)
3400 if ((boot_cpu_data.microcode >= 0x08001126) &&
3401 (boot_cpu_data.microcode <= 0x080011ff))
3404 pr_notice("IOMMU not currently supported when SME is active\n");
3409 /****************************************************************************
3411 * Early detect code. This code runs at IOMMU detection time in the DMA
3412 * layer. It just looks if there is an IVRS ACPI table to detect AMD
3415 ****************************************************************************/
3416 int __init amd_iommu_detect(void)
3420 if (no_iommu || (iommu_detected && !gart_iommu_aperture))
3423 if (!amd_iommu_sme_check())
3426 ret = iommu_go_to_state(IOMMU_IVRS_DETECTED);
3430 amd_iommu_detected = true;
3432 x86_init.iommu.iommu_init = amd_iommu_init;
3437 /****************************************************************************
3439 * Parsing functions for the AMD IOMMU specific kernel command line
3442 ****************************************************************************/
3444 static int __init parse_amd_iommu_dump(char *str)
3446 amd_iommu_dump = true;
3451 static int __init parse_amd_iommu_intr(char *str)
3453 for (; *str; ++str) {
3454 if (strncmp(str, "legacy", 6) == 0) {
3455 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_LEGACY_GA;
3458 if (strncmp(str, "vapic", 5) == 0) {
3459 amd_iommu_guest_ir = AMD_IOMMU_GUEST_IR_VAPIC;
3466 static int __init parse_amd_iommu_options(char *str)
3472 if (strncmp(str, "fullflush", 9) == 0) {
3473 pr_warn("amd_iommu=fullflush deprecated; use iommu.strict=1 instead\n");
3474 iommu_set_dma_strict();
3475 } else if (strncmp(str, "force_enable", 12) == 0) {
3476 amd_iommu_force_enable = true;
3477 } else if (strncmp(str, "off", 3) == 0) {
3478 amd_iommu_disabled = true;
3479 } else if (strncmp(str, "force_isolation", 15) == 0) {
3480 amd_iommu_force_isolation = true;
3481 } else if (strncmp(str, "pgtbl_v1", 8) == 0) {
3482 amd_iommu_pgtable = AMD_IOMMU_V1;
3483 } else if (strncmp(str, "pgtbl_v2", 8) == 0) {
3484 amd_iommu_pgtable = AMD_IOMMU_V2;
3485 } else if (strncmp(str, "irtcachedis", 11) == 0) {
3486 amd_iommu_irtcachedis = true;
3488 pr_notice("Unknown option - '%s'\n", str);
3491 str += strcspn(str, ",");
3499 static int __init parse_ivrs_ioapic(char *str)
3501 u32 seg = 0, bus, dev, fn;
3505 if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3506 sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3509 if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3510 sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3511 pr_warn("ivrs_ioapic%s option format deprecated; use ivrs_ioapic=%d@%04x:%02x:%02x.%d instead\n",
3512 str, id, seg, bus, dev, fn);
3516 pr_err("Invalid command line: ivrs_ioapic%s\n", str);
3520 if (early_ioapic_map_size == EARLY_MAP_SIZE) {
3521 pr_err("Early IOAPIC map overflow - ignoring ivrs_ioapic%s\n",
3526 devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3528 cmdline_maps = true;
3529 i = early_ioapic_map_size++;
3530 early_ioapic_map[i].id = id;
3531 early_ioapic_map[i].devid = devid;
3532 early_ioapic_map[i].cmd_line = true;
3537 static int __init parse_ivrs_hpet(char *str)
3539 u32 seg = 0, bus, dev, fn;
3543 if (sscanf(str, "=%d@%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3544 sscanf(str, "=%d@%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5)
3547 if (sscanf(str, "[%d]=%x:%x.%x", &id, &bus, &dev, &fn) == 4 ||
3548 sscanf(str, "[%d]=%x:%x:%x.%x", &id, &seg, &bus, &dev, &fn) == 5) {
3549 pr_warn("ivrs_hpet%s option format deprecated; use ivrs_hpet=%d@%04x:%02x:%02x.%d instead\n",
3550 str, id, seg, bus, dev, fn);
3554 pr_err("Invalid command line: ivrs_hpet%s\n", str);
3558 if (early_hpet_map_size == EARLY_MAP_SIZE) {
3559 pr_err("Early HPET map overflow - ignoring ivrs_hpet%s\n",
3564 devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3566 cmdline_maps = true;
3567 i = early_hpet_map_size++;
3568 early_hpet_map[i].id = id;
3569 early_hpet_map[i].devid = devid;
3570 early_hpet_map[i].cmd_line = true;
3575 #define ACPIID_LEN (ACPIHID_UID_LEN + ACPIHID_HID_LEN)
3577 static int __init parse_ivrs_acpihid(char *str)
3579 u32 seg = 0, bus, dev, fn;
3580 char *hid, *uid, *p, *addr;
3581 char acpiid[ACPIID_LEN] = {0};
3584 addr = strchr(str, '@');
3586 addr = strchr(str, '=');
3592 if (strlen(addr) > ACPIID_LEN)
3595 if (sscanf(str, "[%x:%x.%x]=%s", &bus, &dev, &fn, acpiid) == 4 ||
3596 sscanf(str, "[%x:%x:%x.%x]=%s", &seg, &bus, &dev, &fn, acpiid) == 5) {
3597 pr_warn("ivrs_acpihid%s option format deprecated; use ivrs_acpihid=%s@%04x:%02x:%02x.%d instead\n",
3598 str, acpiid, seg, bus, dev, fn);
3604 /* We have the '@', make it the terminator to get just the acpiid */
3607 if (strlen(str) > ACPIID_LEN + 1)
3610 if (sscanf(str, "=%s", acpiid) != 1)
3613 if (sscanf(addr, "%x:%x.%x", &bus, &dev, &fn) == 3 ||
3614 sscanf(addr, "%x:%x:%x.%x", &seg, &bus, &dev, &fn) == 4)
3618 pr_err("Invalid command line: ivrs_acpihid%s\n", str);
3623 hid = strsep(&p, ":");
3626 if (!hid || !(*hid) || !uid) {
3627 pr_err("Invalid command line: hid or uid\n");
3632 * Ignore leading zeroes after ':', so e.g., AMDI0095:00
3633 * will match AMDI0095:0 in the second strcmp in acpi_dev_hid_uid_match
3635 while (*uid == '0' && *(uid + 1))
3638 i = early_acpihid_map_size++;
3639 memcpy(early_acpihid_map[i].hid, hid, strlen(hid));
3640 memcpy(early_acpihid_map[i].uid, uid, strlen(uid));
3641 early_acpihid_map[i].devid = IVRS_GET_SBDF_ID(seg, bus, dev, fn);
3642 early_acpihid_map[i].cmd_line = true;
3647 __setup("amd_iommu_dump", parse_amd_iommu_dump);
3648 __setup("amd_iommu=", parse_amd_iommu_options);
3649 __setup("amd_iommu_intr=", parse_amd_iommu_intr);
3650 __setup("ivrs_ioapic", parse_ivrs_ioapic);
3651 __setup("ivrs_hpet", parse_ivrs_hpet);
3652 __setup("ivrs_acpihid", parse_ivrs_acpihid);
3654 bool amd_iommu_pasid_supported(void)
3656 /* CPU page table size should match IOMMU guest page table size */
3657 if (cpu_feature_enabled(X86_FEATURE_LA57) &&
3658 amd_iommu_gpt_level != PAGE_MODE_5_LEVEL)
3662 * Since DTE[Mode]=0 is prohibited on SNP-enabled system
3663 * (i.e. EFR[SNPSup]=1), IOMMUv2 page table cannot be used without
3664 * setting up IOMMUv1 page table.
3666 return amd_iommu_gt_ppr_supported() && !amd_iommu_snp_en;
3669 struct amd_iommu *get_amd_iommu(unsigned int idx)
3672 struct amd_iommu *iommu;
3674 for_each_iommu(iommu)
3680 /****************************************************************************
3682 * IOMMU EFR Performance Counter support functionality. This code allows
3683 * access to the IOMMU PC functionality.
3685 ****************************************************************************/
3687 u8 amd_iommu_pc_get_max_banks(unsigned int idx)
3689 struct amd_iommu *iommu = get_amd_iommu(idx);
3692 return iommu->max_banks;
3697 bool amd_iommu_pc_supported(void)
3699 return amd_iommu_pc_present;
3702 u8 amd_iommu_pc_get_max_counters(unsigned int idx)
3704 struct amd_iommu *iommu = get_amd_iommu(idx);
3707 return iommu->max_counters;
3712 static int iommu_pc_get_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr,
3713 u8 fxn, u64 *value, bool is_write)
3718 /* Make sure the IOMMU PC resource is available */
3719 if (!amd_iommu_pc_present)
3722 /* Check for valid iommu and pc register indexing */
3723 if (WARN_ON(!iommu || (fxn > 0x28) || (fxn & 7)))
3726 offset = (u32)(((0x40 | bank) << 12) | (cntr << 8) | fxn);
3728 /* Limit the offset to the hw defined mmio region aperture */
3729 max_offset_lim = (u32)(((0x40 | iommu->max_banks) << 12) |
3730 (iommu->max_counters << 8) | 0x28);
3731 if ((offset < MMIO_CNTR_REG_OFFSET) ||
3732 (offset > max_offset_lim))
3736 u64 val = *value & GENMASK_ULL(47, 0);
3738 writel((u32)val, iommu->mmio_base + offset);
3739 writel((val >> 32), iommu->mmio_base + offset + 4);
3741 *value = readl(iommu->mmio_base + offset + 4);
3743 *value |= readl(iommu->mmio_base + offset);
3744 *value &= GENMASK_ULL(47, 0);
3750 int amd_iommu_pc_get_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3755 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, false);
3758 int amd_iommu_pc_set_reg(struct amd_iommu *iommu, u8 bank, u8 cntr, u8 fxn, u64 *value)
3763 return iommu_pc_get_set_reg(iommu, bank, cntr, fxn, value, true);
3766 #ifdef CONFIG_KVM_AMD_SEV
3767 static int iommu_page_make_shared(void *page)
3769 unsigned long paddr, pfn;
3771 paddr = iommu_virt_to_phys(page);
3772 /* Cbit maybe set in the paddr */
3773 pfn = __sme_clr(paddr) >> PAGE_SHIFT;
3775 if (!(pfn % PTRS_PER_PMD)) {
3779 ret = snp_lookup_rmpentry(pfn, &assigned, &level);
3781 pr_warn("IOMMU PFN %lx RMP lookup failed, ret %d\n", pfn, ret);
3786 pr_warn("IOMMU PFN %lx not assigned in RMP table\n", pfn);
3790 if (level > PG_LEVEL_4K) {
3795 pr_warn("PSMASH failed for IOMMU PFN %lx huge RMP entry, ret: %d, level: %d\n",
3802 return rmp_make_shared(pfn, PG_LEVEL_4K);
3805 static int iommu_make_shared(void *va, size_t size)
3813 for (page = va; page < (va + size); page += PAGE_SIZE) {
3814 ret = iommu_page_make_shared(page);
3822 int amd_iommu_snp_disable(void)
3824 struct amd_iommu *iommu;
3827 if (!amd_iommu_snp_en)
3830 for_each_iommu(iommu) {
3831 ret = iommu_make_shared(iommu->evt_buf, EVT_BUFFER_SIZE);
3835 ret = iommu_make_shared(iommu->ppr_log, PPR_LOG_SIZE);
3839 ret = iommu_make_shared((void *)iommu->cmd_sem, PAGE_SIZE);
3846 EXPORT_SYMBOL_GPL(amd_iommu_snp_disable);