Hi Jan,
2014-08-12 15:34 GMT+08:00 Jan Kiszka <jan.kis...@web.de>:
> On 2014-08-11 09:04, Le Tan wrote:
>> Add support for emulating Intel IOMMU according to the VT-d specification for
>> the q35 chipset machine. Implement the logics for DMAR (DMA remapping)
>> without
>> PASID support. The emulation supports register-based invalidation and primary
>> fault logging.
>
> Some arbitrary comments below (means, I didn't read every line and
> likely missed some things). In general, this looks and works pretty good!
>
>>
>> Signed-off-by: Le Tan <tamlokv...@gmail.com>
>> ---
>> hw/i386/Makefile.objs | 1 +
>> hw/i386/intel_iommu.c | 1345
>> ++++++++++++++++++++++++++++++++++++++++
>> hw/i386/intel_iommu_internal.h | 345 +++++++++++
>> include/hw/i386/intel_iommu.h | 90 +++
>> 4 files changed, 1781 insertions(+)
>> create mode 100644 hw/i386/intel_iommu.c
>> create mode 100644 hw/i386/intel_iommu_internal.h
>> create mode 100644 include/hw/i386/intel_iommu.h
>>
>> diff --git a/hw/i386/Makefile.objs b/hw/i386/Makefile.objs
>> index 48014ab..6936111 100644
>> --- a/hw/i386/Makefile.objs
>> +++ b/hw/i386/Makefile.objs
>> @@ -2,6 +2,7 @@ obj-$(CONFIG_KVM) += kvm/
>> obj-y += multiboot.o smbios.o
>> obj-y += pc.o pc_piix.o pc_q35.o
>> obj-y += pc_sysfw.o
>> +obj-y += intel_iommu.o
>> obj-$(CONFIG_XEN) += ../xenpv/ xen/
>>
>> obj-y += kvmvapic.o
>> diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
>> new file mode 100644
>> index 0000000..b3a4f78
>> --- /dev/null
>> +++ b/hw/i386/intel_iommu.c
>> @@ -0,0 +1,1345 @@
>> +/*
>> + * QEMU emulation of an Intel IOMMU (VT-d)
>> + * (DMA Remapping device)
>> + *
>> + * Copyright (C) 2013 Knut Omang, Oracle <knut.om...@oracle.com>
>> + * Copyright (C) 2014 Le Tan, <tamlokv...@gmail.com>
>> + *
>> + * This program is free software; you can redistribute it and/or modify
>> + * it under the terms of the GNU General Public License as published by
>> + * the Free Software Foundation; either version 2 of the License, or
>> + * (at your option) any later version.
>> +
>> + * This program is distributed in the hope that it will be useful,
>> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
>> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
>> + * GNU General Public License for more details.
>> +
>> + * You should have received a copy of the GNU General Public License along
>> + * with this program; if not, see <http://www.gnu.org/licenses/>.
>> + */
>> +
>> +#include "hw/sysbus.h"
>> +#include "exec/address-spaces.h"
>> +#include "intel_iommu_internal.h"
>> +
>> +
>> +/*#define DEBUG_INTEL_IOMMU*/
>> +#ifdef DEBUG_INTEL_IOMMU
>> +enum {
>> + DEBUG_GENERAL, DEBUG_CSR, DEBUG_INV, DEBUG_MMU, DEBUG_FLOG,
>> +};
>> +#define VTD_DBGBIT(x) (1 << DEBUG_##x)
>> +static int vtd_dbgflags = VTD_DBGBIT(GENERAL) | VTD_DBGBIT(CSR) |
>> + VTD_DBGBIT(FLOG);
>> +
>> +#define VTD_DPRINTF(what, fmt, ...) do { \
>> + if (vtd_dbgflags & VTD_DBGBIT(what)) { \
>> + fprintf(stderr, "(vtd)%s: " fmt "\n", __func__, \
>> + ## __VA_ARGS__); } \
>> + } while (0)
>> +#else
>> +#define VTD_DPRINTF(what, fmt, ...) do {} while (0)
>> +#endif
>> +
>> +static inline void define_quad(IntelIOMMUState *s, hwaddr addr, uint64_t
>> val,
>> + uint64_t wmask, uint64_t w1cmask)
>
> In general, don't declare functions inline needlessly. It makes sense
> for trivial ones you export via a header, but even a 2- or 3-liner like
> this can be more efficient as stand-alone function. Anything bigger
> definitely does not deserve that tag.
>
> Inline is just a hint to the compiler anyway, so you can perfectly leave
> it out for any almost-trivial static function.
OK, I will delete most of them except those just have one "return" line.
>> +{
>> + stq_le_p(&s->csr[addr], val);
>> + stq_le_p(&s->wmask[addr], wmask);
>> + stq_le_p(&s->w1cmask[addr], w1cmask);
>> +}
>> +
>> +static inline void define_quad_wo(IntelIOMMUState *s, hwaddr addr,
>> + uint64_t mask)
>> +{
>> + stq_le_p(&s->womask[addr], mask);
>> +}
>> +
>> +static inline void define_long(IntelIOMMUState *s, hwaddr addr, uint32_t
>> val,
>> + uint32_t wmask, uint32_t w1cmask)
>> +{
>> + stl_le_p(&s->csr[addr], val);
>> + stl_le_p(&s->wmask[addr], wmask);
>> + stl_le_p(&s->w1cmask[addr], w1cmask);
>> +}
>> +
>> +static inline void define_long_wo(IntelIOMMUState *s, hwaddr addr,
>> + uint32_t mask)
>> +{
>> + stl_le_p(&s->womask[addr], mask);
>> +}
>> +
>> +/* "External" get/set operations */
>> +static inline void set_quad(IntelIOMMUState *s, hwaddr addr, uint64_t val)
>> +{
>> + uint64_t oldval = ldq_le_p(&s->csr[addr]);
>> + uint64_t wmask = ldq_le_p(&s->wmask[addr]);
>> + uint64_t w1cmask = ldq_le_p(&s->w1cmask[addr]);
>> + stq_le_p(&s->csr[addr],
>> + ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
>> +}
>> +
>> +static inline void set_long(IntelIOMMUState *s, hwaddr addr, uint32_t val)
>> +{
>> + uint32_t oldval = ldl_le_p(&s->csr[addr]);
>> + uint32_t wmask = ldl_le_p(&s->wmask[addr]);
>> + uint32_t w1cmask = ldl_le_p(&s->w1cmask[addr]);
>> + stl_le_p(&s->csr[addr],
>> + ((oldval & ~wmask) | (val & wmask)) & ~(w1cmask & val));
>> +}
>> +
>> +static inline uint64_t get_quad(IntelIOMMUState *s, hwaddr addr)
>> +{
>> + uint64_t val = ldq_le_p(&s->csr[addr]);
>> + uint64_t womask = ldq_le_p(&s->womask[addr]);
>> + return val & ~womask;
>> +}
>> +
>> +
>> +static inline uint32_t get_long(IntelIOMMUState *s, hwaddr addr)
>> +{
>> + uint32_t val = ldl_le_p(&s->csr[addr]);
>> + uint32_t womask = ldl_le_p(&s->womask[addr]);
>> + return val & ~womask;
>> +}
>> +
>> +/* "Internal" get/set operations */
>> +static inline uint64_t get_quad_raw(IntelIOMMUState *s, hwaddr addr)
>> +{
>> + return ldq_le_p(&s->csr[addr]);
>> +}
>> +
>> +static inline uint32_t get_long_raw(IntelIOMMUState *s, hwaddr addr)
>> +{
>> + return ldl_le_p(&s->csr[addr]);
>> +}
>> +
>> +static inline void set_quad_raw(IntelIOMMUState *s, hwaddr addr, uint64_t
>> val)
>> +{
>> + stq_le_p(&s->csr[addr], val);
>> +}
>> +
>> +static inline uint32_t set_clear_mask_long(IntelIOMMUState *s, hwaddr addr,
>> + uint32_t clear, uint32_t mask)
>> +{
>> + uint32_t new_val = (ldl_le_p(&s->csr[addr]) & ~clear) | mask;
>> + stl_le_p(&s->csr[addr], new_val);
>> + return new_val;
>> +}
>> +
>> +static inline uint64_t set_clear_mask_quad(IntelIOMMUState *s, hwaddr addr,
>> + uint64_t clear, uint64_t mask)
>> +{
>> + uint64_t new_val = (ldq_le_p(&s->csr[addr]) & ~clear) | mask;
>> + stq_le_p(&s->csr[addr], new_val);
>> + return new_val;
>> +}
>> +
>> +/* Given the reg addr of both the message data and address, generate an
>> + * interrupt via MSI.
>> + */
>> +static void vtd_generate_interrupt(IntelIOMMUState *s, hwaddr mesg_addr_reg,
>> + hwaddr mesg_data_reg)
>> +{
>> + hwaddr addr;
>> + uint32_t data;
>> +
>> + assert(mesg_data_reg < DMAR_REG_SIZE);
>> + assert(mesg_addr_reg < DMAR_REG_SIZE);
>> +
>> + addr = get_long_raw(s, mesg_addr_reg);
>> + data = get_long_raw(s, mesg_data_reg);
>> +
>> + VTD_DPRINTF(FLOG, "msi: addr 0x%"PRIx64 " data 0x%"PRIx32, addr, data);
>> + stl_le_phys(&address_space_memory, addr, data);
>> +}
>> +
>> +/* Generate a fault event to software via MSI if conditions are met.
>> + * Notice that the value of FSTS_REG being passed to it should be the one
>> + * before any update.
>> + */
>> +static void vtd_generate_fault_event(IntelIOMMUState *s, uint32_t pre_fsts)
>> +{
>> + /* Check if there are any previously reported interrupt conditions */
>> + if (pre_fsts & VTD_FSTS_PPF || pre_fsts & VTD_FSTS_PFO ||
>> + pre_fsts & VTD_FSTS_IQE) {
>> + VTD_DPRINTF(FLOG, "there are previous interrupt conditions "
>> + "to be serviced by software, fault event is not
>> generated "
>> + "(FSTS_REG 0x%"PRIx32 ")", pre_fsts);
>> + return;
>> + }
>> + set_clear_mask_long(s, DMAR_FECTL_REG, 0, VTD_FECTL_IP);
>> + if (get_long_raw(s, DMAR_FECTL_REG) & VTD_FECTL_IM) {
>> + /* Interrupt Mask */
>> + VTD_DPRINTF(FLOG, "Interrupt Mask set, fault event is not
>> generated");
>> + } else {
>> + /* generate interrupt */
>> + vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
>> + set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
>> + }
>> +}
>> +
>> +/* Check if the Fault (F) field of the Fault Recording Register referenced
>> by
>> + * @index is Set.
>> + */
>> +static inline bool is_frcd_set(IntelIOMMUState *s, uint16_t index)
>> +{
>> + /* Each reg is 128-bit */
>> + hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
>> + addr += 8; /* Access the high 64-bit half */
>> +
>> + assert(index < DMAR_FRCD_REG_NR);
>> +
>> + return get_quad_raw(s, addr) & VTD_FRCD_F;
>> +}
>> +
>> +/* Update the PPF field of Fault Status Register.
>> + * Should be called whenever change the F field of any fault recording
>> + * registers.
>> + */
>> +static inline void update_fsts_ppf(IntelIOMMUState *s)
>> +{
>> + uint32_t i;
>> + uint32_t ppf_mask = 0;
>> +
>> + for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
>> + if (is_frcd_set(s, i)) {
>> + ppf_mask = VTD_FSTS_PPF;
>> + break;
>> + }
>> + }
>> + set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_PPF, ppf_mask);
>> + VTD_DPRINTF(FLOG, "set PPF of FSTS_REG to %d", ppf_mask ? 1 : 0);
>> +}
>> +
>> +static inline void set_frcd_and_update_ppf(IntelIOMMUState *s, uint16_t
>> index)
>> +{
>> + /* Each reg is 128-bit */
>> + hwaddr addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
>> + addr += 8; /* Access the high 64-bit half */
>> +
>> + assert(index < DMAR_FRCD_REG_NR);
>> +
>> + set_clear_mask_quad(s, addr, 0, VTD_FRCD_F);
>> + update_fsts_ppf(s);
>> +}
>> +
>> +/* Must not update F field now, should be done later */
>> +static void record_frcd(IntelIOMMUState *s, uint16_t index, uint16_t
>> source_id,
>> + hwaddr addr, VTDFaultReason fault, bool is_write)
>> +{
>> + uint64_t hi = 0, lo;
>> + hwaddr frcd_reg_addr = DMAR_FRCD_REG_OFFSET + (((uint64_t)index) << 4);
>> +
>> + assert(index < DMAR_FRCD_REG_NR);
>> +
>> + lo = VTD_FRCD_FI(addr);
>> + hi = VTD_FRCD_SID(source_id) | VTD_FRCD_FR(fault);
>> + if (!is_write) {
>> + hi |= VTD_FRCD_T;
>> + }
>> +
>> + set_quad_raw(s, frcd_reg_addr, lo);
>> + set_quad_raw(s, frcd_reg_addr + 8, hi);
>> + VTD_DPRINTF(FLOG, "record to FRCD_REG #%"PRIu16 ": hi 0x%"PRIx64
>> + ", lo 0x%"PRIx64, index, hi, lo);
>> +}
>> +
>> +/* Try to collapse multiple pending faults from the same requester */
>> +static inline bool try_collapse_fault(IntelIOMMUState *s, uint16_t
>> source_id)
>> +{
>> + uint32_t i;
>> + uint64_t frcd_reg;
>> + hwaddr addr = DMAR_FRCD_REG_OFFSET + 8; /* The high 64-bit half */
>> +
>> + for (i = 0; i < DMAR_FRCD_REG_NR; i++) {
>> + frcd_reg = get_quad_raw(s, addr);
>> + VTD_DPRINTF(FLOG, "frcd_reg #%d 0x%"PRIx64, i, frcd_reg);
>> + if ((frcd_reg & VTD_FRCD_F) &&
>> + ((frcd_reg & VTD_FRCD_SID_MASK) == source_id)) {
>> + return true;
>> + }
>> + addr += 16; /* 128-bit for each */
>> + }
>> +
>> + return false;
>> +}
>> +
>> +/* Log and report an DMAR (address translation) fault to software */
>> +static void vtd_report_dmar_fault(IntelIOMMUState *s, uint16_t source_id,
>> + hwaddr addr, VTDFaultReason fault,
>> + bool is_write)
>> +{
>> + uint32_t fsts_reg = get_long_raw(s, DMAR_FSTS_REG);
>> +
>> + assert(fault < VTD_FR_MAX);
>> +
>> + if (fault == VTD_FR_RESERVED_ERR) {
>> + /* This is not a normal fault reason case. Drop it. */
>> + return;
>> + }
>> +
>> + VTD_DPRINTF(FLOG, "sid 0x%"PRIx16 ", fault %d, addr 0x%"PRIx64
>> + ", is_write %d", source_id, fault, addr, is_write);
>> +
>> + /* Check PFO field in FSTS_REG */
>> + if (fsts_reg & VTD_FSTS_PFO) {
>> + VTD_DPRINTF(FLOG, "new fault is not recorded due to "
>> + "Primary Fault Overflow");
>> + return;
>> + }
>> +
>> + /* Compression of multiple faults from the same requester */
>> + if (try_collapse_fault(s, source_id)) {
>> + VTD_DPRINTF(FLOG, "new fault is not recorded due to "
>> + "compression of faults");
>> + return;
>> + }
>> +
>> + /* Check next_frcd_reg to see whether it is overflow now */
>> + if (is_frcd_set(s, s->next_frcd_reg)) {
>> + VTD_DPRINTF(FLOG, "Primary Fault Overflow and "
>> + "new fault is not recorded, set PFO field");
>> + set_clear_mask_long(s, DMAR_FSTS_REG, 0, VTD_FSTS_PFO);
>> + return;
>> + }
>> +
>> + record_frcd(s, s->next_frcd_reg, source_id, addr, fault, is_write);
>> +
>> + if (fsts_reg & VTD_FSTS_PPF) {
>> + /* There are already one or more pending faults */
>> + VTD_DPRINTF(FLOG, "there are pending faults already, "
>> + "fault event is not generated");
>> + set_frcd_and_update_ppf(s, s->next_frcd_reg);
>> + s->next_frcd_reg++;
>> + if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
>> + s->next_frcd_reg = 0;
>> + }
>> + } else {
>> + set_clear_mask_long(s, DMAR_FSTS_REG, VTD_FSTS_FRI_MASK,
>> + VTD_FSTS_FRI(s->next_frcd_reg));
>> + set_frcd_and_update_ppf(s, s->next_frcd_reg); /* It will also set
>> PPF */
>> + s->next_frcd_reg++;
>> + if (s->next_frcd_reg == DMAR_FRCD_REG_NR) {
>> + s->next_frcd_reg = 0;
>> + }
>> +
>> + /* This case actually cause the PPF to be Set.
>> + * So generate fault event (interrupt).
>> + */
>> + vtd_generate_fault_event(s, fsts_reg);
>> + }
>> +}
>> +
>> +static inline bool root_entry_present(VTDRootEntry *root)
>> +{
>> + return root->val & VTD_ROOT_ENTRY_P;
>> +}
>> +
>> +static int get_root_entry(IntelIOMMUState *s, uint32_t index, VTDRootEntry
>> *re)
>> +{
>> + dma_addr_t addr;
>> +
>> + assert(index < VTD_ROOT_ENTRY_NR);
>> +
>> + addr = s->root + index * sizeof(*re);
>> +
>> + if (dma_memory_read(&address_space_memory, addr, re, sizeof(*re))) {
>> + VTD_DPRINTF(GENERAL, "error: fail to access root-entry at 0x%"PRIx64
>> + " + %"PRIu32, s->root, index);
>> + re->val = 0;
>> + return -VTD_FR_ROOT_TABLE_INV;
>> + }
>> +
>> + re->val = le64_to_cpu(re->val);
>> + return VTD_FR_RESERVED;
>
> This looks a bit weird, here and elsewhere: VTD_FR_RESERVED is a
> reserved error code in the VT-d specification, and it's 0. OK, but here
> the meaning of returning 0 is actually "everything went well". So either
> provide a constant that documents this or simply use 0 consistently to
> declare the absence of errors.
I think VTD_FR_RESERVED is reserved and VT-d will not use it, and it
happens to be zero, so I just use it to represent the absence of
errors. So maybe I just simply return 0 here.
>> +}
>> +
>> +static inline bool context_entry_present(VTDContextEntry *context)
>> +{
>> + return context->lo & VTD_CONTEXT_ENTRY_P;
>> +}
>> +
>> +static int get_context_entry_from_root(VTDRootEntry *root, uint32_t index,
>> + VTDContextEntry *ce)
>> +{
>> + dma_addr_t addr;
>> +
>> + if (!root_entry_present(root)) {
>> + ce->lo = 0;
>> + ce->hi = 0;
>> + VTD_DPRINTF(GENERAL, "error: root-entry is not present");
>> + return -VTD_FR_ROOT_ENTRY_P;
>> + }
>> +
>> + assert(index < VTD_CONTEXT_ENTRY_NR);
>> +
>> + addr = (root->val & VTD_ROOT_ENTRY_CTP) + index * sizeof(*ce);
>> +
>> + if (dma_memory_read(&address_space_memory, addr, ce, sizeof(*ce))) {
>> + VTD_DPRINTF(GENERAL, "error: fail to access context-entry at
>> 0x%"PRIx64
>> + " + %"PRIu32,
>> + (uint64_t)(root->val & VTD_ROOT_ENTRY_CTP), index);
>> + ce->lo = 0;
>> + ce->hi = 0;
>> + return -VTD_FR_CONTEXT_TABLE_INV;
>> + }
>> +
>> + ce->lo = le64_to_cpu(ce->lo);
>> + ce->hi = le64_to_cpu(ce->hi);
>> + return VTD_FR_RESERVED;
>> +}
>> +
>> +static inline dma_addr_t get_slpt_base_from_context(VTDContextEntry *ce)
>> +{
>> + return ce->lo & VTD_CONTEXT_ENTRY_SLPTPTR;
>> +}
>> +
>> +/* The shift of an addr for a certain level of paging structure */
>> +static inline uint32_t slpt_level_shift(uint32_t level)
>> +{
>> + return VTD_PAGE_SHIFT_4K + (level - 1) * VTD_SL_LEVEL_BITS;
>> +}
>> +
>> +static inline uint64_t get_slpte_addr(uint64_t slpte)
>> +{
>> + return slpte & VTD_SL_PT_BASE_ADDR_MASK;
>> +}
>> +
>> +/* Whether the pte indicates the address of the page frame */
>> +static inline bool is_last_slpte(uint64_t slpte, uint32_t level)
>> +{
>> + return level == VTD_SL_PT_LEVEL || (slpte & VTD_SL_PT_PAGE_SIZE_MASK);
>> +}
>> +
>> +/* Get the content of a spte located in @base_addr[@index] */
>> +static inline uint64_t get_slpte(dma_addr_t base_addr, uint32_t index)
>> +{
>> + uint64_t slpte;
>> +
>> + assert(index < VTD_SL_PT_ENTRY_NR);
>> +
>> + if (dma_memory_read(&address_space_memory,
>> + base_addr + index * sizeof(slpte), &slpte,
>> + sizeof(slpte))) {
>> + slpte = (uint64_t)-1;
>> + return slpte;
>> + }
>> +
>> + slpte = le64_to_cpu(slpte);
>> + return slpte;
>> +}
>> +
>> +/* Given a gpa and the level of paging structure, return the offset of
>> current
>> + * level.
>> + */
>> +static inline uint32_t gpa_level_offset(uint64_t gpa, uint32_t level)
>> +{
>> + return (gpa >> slpt_level_shift(level)) & ((1ULL << VTD_SL_LEVEL_BITS)
>> - 1);
>> +}
>> +
>> +/* Check Capability Register to see if the @level of page-table is
>> supported */
>> +static inline bool is_level_supported(IntelIOMMUState *s, uint32_t level)
>> +{
>> + return VTD_CAP_SAGAW_MASK & s->cap &
>> + (1ULL << (level - 2 + VTD_CAP_SAGAW_SHIFT));
>> +}
>> +
>> +/* Get the page-table level that hardware should use for the second-level
>> + * page-table walk from the Address Width field of context-entry.
>> + */
>> +static inline uint32_t get_level_from_context_entry(VTDContextEntry *ce)
>> +{
>> + return 2 + (ce->hi & VTD_CONTEXT_ENTRY_AW);
>> +}
>> +
>> +static inline uint32_t get_agaw_from_context_entry(VTDContextEntry *ce)
>> +{
>> + return 30 + (ce->hi & VTD_CONTEXT_ENTRY_AW) * 9;
>> +}
>> +
>> +static const uint64_t paging_entry_rsvd_field[] = {
>> + [0] = ~0ULL,
>> + /* For not large page */
>> + [1] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM),
>> + [2] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM),
>> + [3] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM),
>> + [4] = 0x880ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM),
>> + /* For large page */
>> + [5] = 0x800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM),
>> + [6] = 0x1ff800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM),
>> + [7] = 0x3ffff800ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM),
>> + [8] = 0x880ULL | ~(VTD_HAW_MASK | VTD_SL_IGN_COM),
>> +};
>> +
>> +static inline bool slpte_nonzero_rsvd(uint64_t slpte, uint32_t level)
>> +{
>> + if (slpte & VTD_SL_PT_PAGE_SIZE_MASK) {
>> + /* Maybe large page */
>> + return slpte & paging_entry_rsvd_field[level + 4];
>> + } else {
>> + return slpte & paging_entry_rsvd_field[level];
>> + }
>> +}
>> +
>> +/* Given the @gpa, get relevant @slptep. @slpte_level will be the last level
>> + * of the translation, can be used for deciding the size of large page.
>> + * @slptep and @slpte_level will not be touched if error happens.
>> + */
>> +static int gpa_to_slpte(VTDContextEntry *ce, uint64_t gpa, bool is_write,
>> + uint64_t *slptep, uint32_t *slpte_level)
>> +{
>> + dma_addr_t addr = get_slpt_base_from_context(ce);
>> + uint32_t level = get_level_from_context_entry(ce);
>> + uint32_t offset;
>> + uint64_t slpte;
>> + uint32_t ce_agaw = get_agaw_from_context_entry(ce);
>> + uint64_t access_right_check;
>> +
>> + /* Check if @gpa is above 2^X-1, where X is the minimum of MGAW in
>> CAP_REG
>> + * and AW in context-entry.
>> + */
>> + if (gpa & ~((1ULL << MIN(ce_agaw, VTD_MGAW)) - 1)) {
>> + VTD_DPRINTF(GENERAL, "error: gpa 0x%"PRIx64 " exceeds limits", gpa);
>> + return -VTD_FR_ADDR_BEYOND_MGAW;
>> + }
>> +
>> + /* FIXME: what is the Atomics request here? */
>> + access_right_check = is_write ? VTD_SL_W : VTD_SL_R;
>> +
>> + while (true) {
>> + offset = gpa_level_offset(gpa, level);
>> + slpte = get_slpte(addr, offset);
>> +
>> + if (slpte == (uint64_t)-1) {
>> + VTD_DPRINTF(GENERAL, "error: fail to access second-level paging
>> "
>> + "entry at level %"PRIu32 " for gpa 0x%"PRIx64,
>> + level, gpa);
>> + if (level == get_level_from_context_entry(ce)) {
>> + /* Invalid programming of context-entry */
>> + return -VTD_FR_CONTEXT_ENTRY_INV;
>> + } else {
>> + return -VTD_FR_PAGING_ENTRY_INV;
>> + }
>> + }
>> + if (!(slpte & access_right_check)) {
>> + VTD_DPRINTF(GENERAL, "error: lack of %s permission for "
>> + "gpa 0x%"PRIx64 " slpte 0x%"PRIx64,
>> + (is_write ? "write" : "read"), gpa, slpte);
>> + return is_write ? -VTD_FR_WRITE : -VTD_FR_READ;
>> + }
>> + if (slpte_nonzero_rsvd(slpte, level)) {
>> + VTD_DPRINTF(GENERAL, "error: non-zero reserved field in second "
>> + "level paging entry level %"PRIu32 " slpte
>> 0x%"PRIx64,
>> + level, slpte);
>> + return -VTD_FR_PAGING_ENTRY_RSVD;
>> + }
>> +
>> + if (is_last_slpte(slpte, level)) {
>> + *slptep = slpte;
>> + *slpte_level = level;
>> + return VTD_FR_RESERVED;
>> + }
>> + addr = get_slpte_addr(slpte);
>> + level--;
>> + }
>> +}
>> +
>> +/* Map a device to its corresponding domain (context-entry). @ce will be set
>> + * to Zero if error happens while accessing the context-entry.
>> + */
>> +static inline int dev_to_context_entry(IntelIOMMUState *s, int bus_num,
>> + int devfn, VTDContextEntry *ce)
>> +{
>> + VTDRootEntry re;
>> + int ret_fr;
>> +
>> + assert(0 <= bus_num && bus_num < VTD_PCI_BUS_MAX);
>> + assert(0 <= devfn && devfn < VTD_PCI_SLOT_MAX * VTD_PCI_FUNC_MAX);
>
> Use the proper types for bus_num and devfn, and your can get rid of
> these assertions: uint8_t. I know that the PCI layer improperly uses int
> for them in many places, but you don't need to copy this.
OK. I will translate them to uint8_t in iommu_translate().
>> +
>> + ret_fr = get_root_entry(s, bus_num, &re);
>> + if (ret_fr) {
>> + ce->hi = 0;
>> + ce->lo = 0;
>
> That's a bit too defensive programming: The context entry is simply
> invalid when such a function returns an error, no? You can document that
> in the function description.
Ah, yes, it is needless. I made a mistake before. I thought I need to
evaluate the FPD field in context entry, so to avoid using arbitrary
value, I set it to 0 when error happens. But it is needless because
is_qualified_fault(ret_fr) will fail when the context entry is
undefined (it is surly caused by some not qualified fault). Thanks!
>> + return ret_fr;
>> + }
>> +
>> + if (!root_entry_present(&re)) {
>> + VTD_DPRINTF(GENERAL, "error: root-entry #%d is not present",
>> bus_num);
>> + ce->hi = 0;
>> + ce->lo = 0;
>> + return -VTD_FR_ROOT_ENTRY_P;
>> + } else if (re.rsvd || (re.val & VTD_ROOT_ENTRY_RSVD)) {
>> + VTD_DPRINTF(GENERAL, "error: non-zero reserved field in root-entry "
>> + "hi 0x%"PRIx64 " lo 0x%"PRIx64, re.rsvd, re.val);
>> + ce->hi = 0;
>> + ce->lo = 0;
>> + return -VTD_FR_ROOT_ENTRY_RSVD;
>> + }
>> +
>> + ret_fr = get_context_entry_from_root(&re, devfn, ce);
>> + if (ret_fr) {
>> + return ret_fr;
>> + }
>> +
>> + if (!context_entry_present(ce)) {
>> + VTD_DPRINTF(GENERAL,
>> + "error: context-entry #%d(bus #%d) is not present",
>> devfn,
>> + bus_num);
>> + return -VTD_FR_CONTEXT_ENTRY_P;
>> + } else if ((ce->hi & VTD_CONTEXT_ENTRY_RSVD_HI) ||
>> + (ce->lo & VTD_CONTEXT_ENTRY_RSVD_LO)) {
>> + VTD_DPRINTF(GENERAL,
>> + "error: non-zero reserved field in context-entry "
>> + "hi 0x%"PRIx64 " lo 0x%"PRIx64, ce->hi, ce->lo);
>> + return -VTD_FR_CONTEXT_ENTRY_RSVD;
>> + }
>> +
>> + /* Check if the programming of context-entry is valid */
>> + if (!is_level_supported(s, get_level_from_context_entry(ce))) {
>> + VTD_DPRINTF(GENERAL, "error: unsupported Address Width value in "
>> + "context-entry hi 0x%"PRIx64 " lo 0x%"PRIx64,
>> + ce->hi, ce->lo);
>> + return -VTD_FR_CONTEXT_ENTRY_INV;
>> + } else if (ce->lo & VTD_CONTEXT_ENTRY_TT) {
>> + VTD_DPRINTF(GENERAL, "error: unsupported Translation Type in "
>> + "context-entry hi 0x%"PRIx64 " lo 0x%"PRIx64,
>> + ce->hi, ce->lo);
>> + return -VTD_FR_CONTEXT_ENTRY_INV;
>> + }
>> +
>> + return VTD_FR_RESERVED;
>> +}
>> +
>> +static inline uint16_t make_source_id(int bus_num, int devfn)
>> +{
>> + return ((bus_num & 0xffUL) << 8) | (devfn & 0xffUL);
>> +}
>> +
>> +static const bool qualified_faults[] = {
>> + [VTD_FR_RESERVED] = false,
>> + [VTD_FR_ROOT_ENTRY_P] = false,
>> + [VTD_FR_CONTEXT_ENTRY_P] = true,
>> + [VTD_FR_CONTEXT_ENTRY_INV] = true,
>> + [VTD_FR_ADDR_BEYOND_MGAW] = true,
>> + [VTD_FR_WRITE] = true,
>> + [VTD_FR_READ] = true,
>> + [VTD_FR_PAGING_ENTRY_INV] = true,
>> + [VTD_FR_ROOT_TABLE_INV] = false,
>> + [VTD_FR_CONTEXT_TABLE_INV] = false,
>> + [VTD_FR_ROOT_ENTRY_RSVD] = false,
>> + [VTD_FR_PAGING_ENTRY_RSVD] = true,
>> + [VTD_FR_CONTEXT_ENTRY_TT] = true,
>> + [VTD_FR_RESERVED_ERR] = false,
>> + [VTD_FR_MAX] = false,
>> +};
>> +
>> +/* To see if a fault condition is "qualified", which is reported to software
>> + * only if the FPD field in the context-entry used to process the faulting
>> + * request is 0.
>> + */
>> +static inline bool is_qualified_fault(VTDFaultReason fault)
>> +{
>> + return qualified_faults[fault];
>> +}
>> +
>> +static inline bool is_interrupt_addr(hwaddr addr)
>> +{
>> + return VTD_INTERRUPT_ADDR_FIRST <= addr && addr <=
>> VTD_INTERRUPT_ADDR_LAST;
>> +}
>> +
>> +/* Map dev to context-entry then do a paging-structures walk to do a iommu
>> + * translation.
>> + * @bus_num: The bus number
>> + * @devfn: The devfn, which is the combined of device and function number
>> + * @is_write: The access is a write operation
>> + * @entry: IOMMUTLBEntry that contain the addr to be translated and result
>> + */
>> +static void iommu_translate(IntelIOMMUState *s, int bus_num, int devfn,
>> + hwaddr addr, bool is_write, IOMMUTLBEntry
>> *entry)
>> +{
>> + VTDContextEntry ce;
>> + uint64_t slpte;
>> + uint32_t level;
>> + uint64_t page_mask;
>> + uint16_t source_id = make_source_id(bus_num, devfn);
>> + int ret_fr;
>> + bool is_fpd_set = false;
>> +
>> + /* Check if the request is in interrupt address range */
>> + if (is_interrupt_addr(addr)) {
>> + if (is_write) {
>> + /* FIXME: since we don't know the length of the access here, we
>> + * treat Non-DWORD length write requests without PASID as
>> + * interrupt requests, too. Withoud interrupt remapping support,
>> + * we just use 1:1 mapping.
>> + */
>> + VTD_DPRINTF(MMU, "write request to interrupt address "
>> + "gpa 0x%"PRIx64, addr);
>> + entry->iova = addr & VTD_PAGE_MASK_4K;
>> + entry->translated_addr = addr & VTD_PAGE_MASK_4K;
>> + entry->addr_mask = ~VTD_PAGE_MASK_4K;
>> + entry->perm = IOMMU_WO;
>> + return;
>> + } else {
>> + VTD_DPRINTF(GENERAL, "error: read request from interrupt
>> address "
>> + "gpa 0x%"PRIx64, addr);
>> + vtd_report_dmar_fault(s, source_id, addr, VTD_FR_READ,
>> is_write);
>> + return;
>> + }
>> + }
>> +
>> + ret_fr = dev_to_context_entry(s, bus_num, devfn, &ce);
>> + is_fpd_set = ce.lo & VTD_CONTEXT_ENTRY_FPD;
>> + if (ret_fr) {
>> + ret_fr = -ret_fr;
>> + if (is_fpd_set && is_qualified_fault(ret_fr)) {
>> + VTD_DPRINTF(FLOG, "fault processing is disabled for DMA
>> requests "
>> + "through this context-entry (with FPD Set)");
>> + } else {
>> + vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write);
>> + }
>> + return;
>> + }
>> +
>> + ret_fr = gpa_to_slpte(&ce, addr, is_write, &slpte, &level);
>> + if (ret_fr) {
>> + ret_fr = -ret_fr;
>> + if (is_fpd_set && is_qualified_fault(ret_fr)) {
>> + VTD_DPRINTF(FLOG, "fault processing is disabled for DMA
>> requests "
>> + "through this context-entry (with FPD Set)");
>> + } else {
>> + vtd_report_dmar_fault(s, source_id, addr, ret_fr, is_write);
>> + }
>> + return;
>> + }
>> +
>> + if (level == VTD_SL_PT_LEVEL) {
>> + /* 4-KB page */
>> + page_mask = VTD_PAGE_MASK_4K;
>> + } else if (level == VTD_SL_PDP_LEVEL) {
>> + /* 1-GB page */
>> + page_mask = VTD_PAGE_MASK_1G;
>> + } else {
>> + /* 2-MB page */
>> + page_mask = VTD_PAGE_MASK_2M;
>> + }
>
> You don't declare 1G and 2M pages as supported in caps.sllps, do you?
> I'm wondering if we should have some device property for intel-iommu
> that enables all available features, even if our emulated chipset never
> supported them (I guess, Q35 had no support - my younger QM57 does not
> have as well). Then you could do "-global intel-iommu.full_featured=on"
> and have all those nice things available.
Oh, I forgot to report this in the Capability Register. Maybe I can
add the device property later.
>> +
>> + entry->iova = addr & page_mask;
>> + entry->translated_addr = get_slpte_addr(slpte) & page_mask;
>> + entry->addr_mask = ~page_mask;
>> + entry->perm = slpte & VTD_SL_RW_MASK;
>> +}
>> +
>> +static void vtd_root_table_setup(IntelIOMMUState *s)
>> +{
>> + s->root = get_quad_raw(s, DMAR_RTADDR_REG);
>> + s->root_extended = s->root & VTD_RTADDR_RTT;
>> + s->root &= VTD_RTADDR_ADDR_MASK;
>> +
>> + VTD_DPRINTF(CSR, "root_table addr 0x%"PRIx64 " %s", s->root,
>> + (s->root_extended ? "(extended)" : ""));
>> +}
>> +
>> +/* Context-cache invalidation
>> + * Returns the Context Actual Invalidation Granularity.
>> + * @val: the content of the CCMD_REG
>> + */
>> +static uint64_t vtd_context_cache_invalidate(IntelIOMMUState *s, uint64_t
>> val)
>> +{
>> + uint64_t caig;
>> + uint64_t type = val & VTD_CCMD_CIRG_MASK;
>> +
>> + switch (type) {
>> + case VTD_CCMD_GLOBAL_INVL:
>> + VTD_DPRINTF(INV, "Global invalidation request");
>> + caig = VTD_CCMD_GLOBAL_INVL_A;
>> + break;
>> +
>> + case VTD_CCMD_DOMAIN_INVL:
>> + VTD_DPRINTF(INV, "Domain-selective invalidation request");
>> + caig = VTD_CCMD_DOMAIN_INVL_A;
>> + break;
>> +
>> + case VTD_CCMD_DEVICE_INVL:
>> + VTD_DPRINTF(INV, "Domain-selective invalidation request");
>> + caig = VTD_CCMD_DEVICE_INVL_A;
>> + break;
>> +
>> + default:
>> + VTD_DPRINTF(GENERAL,
>> + "error: wrong context-cache invalidation granularity");
>> + caig = 0;
>> + }
>> +
>> + return caig;
>> +}
>> +
>> +/* Flush IOTLB
>> + * Returns the IOTLB Actual Invalidation Granularity.
>> + * @val: the content of the IOTLB_REG
>> + */
>> +static uint64_t vtd_iotlb_flush(IntelIOMMUState *s, uint64_t val)
>> +{
>> + uint64_t iaig;
>> + uint64_t type = val & VTD_TLB_FLUSH_GRANU_MASK;
>> +
>> + switch (type) {
>> + case VTD_TLB_GLOBAL_FLUSH:
>> + VTD_DPRINTF(INV, "Global IOTLB flush");
>> + iaig = VTD_TLB_GLOBAL_FLUSH_A;
>> + break;
>> +
>> + case VTD_TLB_DSI_FLUSH:
>> + VTD_DPRINTF(INV, "Domain-selective IOTLB flush");
>> + iaig = VTD_TLB_DSI_FLUSH_A;
>> + break;
>> +
>> + case VTD_TLB_PSI_FLUSH:
>> + VTD_DPRINTF(INV, "Page-selective-within-domain IOTLB flush");
>> + iaig = VTD_TLB_PSI_FLUSH_A;
>> + break;
>> +
>> + default:
>> + VTD_DPRINTF(GENERAL, "error: wrong iotlb flush granularity");
>> + iaig = 0;
>> + }
>> +
>> + return iaig;
>> +}
>> +
>> +/* Set Root Table Pointer */
>> +static void handle_gcmd_srtp(IntelIOMMUState *s)
>> +{
>> + VTD_DPRINTF(CSR, "set Root Table Pointer");
>> +
>> + vtd_root_table_setup(s);
>> + /* Ok - report back to driver */
>> + set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_RTPS);
>> +}
>> +
>> +/* Handle Translation Enable/Disable */
>> +static void handle_gcmd_te(IntelIOMMUState *s, bool en)
>> +{
>> + VTD_DPRINTF(CSR, "Translation Enable %s", (en ? "on" : "off"));
>> +
>> + if (en) {
>> + s->dmar_enabled = true;
>> + /* Ok - report back to driver */
>> + set_clear_mask_long(s, DMAR_GSTS_REG, 0, VTD_GSTS_TES);
>> + } else {
>> + s->dmar_enabled = false;
>> +
>> + /* Clear the index of Fault Recording Register */
>> + s->next_frcd_reg = 0;
>> + /* Ok - report back to driver */
>> + set_clear_mask_long(s, DMAR_GSTS_REG, VTD_GSTS_TES, 0);
>> + }
>> +}
>> +
>> +/* Handle write to Global Command Register */
>> +static void handle_gcmd_write(IntelIOMMUState *s)
>> +{
>> + uint32_t status = get_long_raw(s, DMAR_GSTS_REG);
>> + uint32_t val = get_long_raw(s, DMAR_GCMD_REG);
>> + uint32_t changed = status ^ val;
>> +
>> + VTD_DPRINTF(CSR, "value 0x%"PRIx32 " status 0x%"PRIx32, val, status);
>> + if (changed & VTD_GCMD_TE) {
>> + /* Translation enable/disable */
>> + handle_gcmd_te(s, val & VTD_GCMD_TE);
>> + }
>> + if (val & VTD_GCMD_SRTP) {
>> + /* Set/update the root-table pointer */
>> + handle_gcmd_srtp(s);
>> + }
>> +}
>> +
>> +/* Handle write to Context Command Register */
>> +static void handle_ccmd_write(IntelIOMMUState *s)
>> +{
>> + uint64_t ret;
>> + uint64_t val = get_quad_raw(s, DMAR_CCMD_REG);
>> +
>> + /* Context-cache invalidation request */
>> + if (val & VTD_CCMD_ICC) {
>> + ret = vtd_context_cache_invalidate(s, val);
>> +
>> + /* Invalidation completed. Change something to show */
>> + set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_ICC, 0ULL);
>> + ret = set_clear_mask_quad(s, DMAR_CCMD_REG, VTD_CCMD_CAIG_MASK,
>> ret);
>> + VTD_DPRINTF(INV, "CCMD_REG write-back val: 0x%"PRIx64, ret);
>> + }
>> +}
>> +
>> +/* Handle write to IOTLB Invalidation Register */
>> +static void handle_iotlb_write(IntelIOMMUState *s)
>> +{
>> + uint64_t ret;
>> + uint64_t val = get_quad_raw(s, DMAR_IOTLB_REG);
>> +
>> + /* IOTLB invalidation request */
>> + if (val & VTD_TLB_IVT) {
>> + ret = vtd_iotlb_flush(s, val);
>> +
>> + /* Invalidation completed. Change something to show */
>> + set_clear_mask_quad(s, DMAR_IOTLB_REG, VTD_TLB_IVT, 0ULL);
>> + ret = set_clear_mask_quad(s, DMAR_IOTLB_REG,
>> + VTD_TLB_FLUSH_GRANU_MASK_A, ret);
>> + VTD_DPRINTF(INV, "IOTLB_REG write-back val: 0x%"PRIx64, ret);
>> + }
>> +}
>> +
>> +static inline void handle_fsts_write(IntelIOMMUState *s)
>> +{
>> + uint32_t fsts_reg = get_long_raw(s, DMAR_FSTS_REG);
>> + uint32_t fectl_reg = get_long_raw(s, DMAR_FECTL_REG);
>> + uint32_t status_fields = VTD_FSTS_PFO | VTD_FSTS_PPF | VTD_FSTS_IQE;
>> +
>> + if ((fectl_reg & VTD_FECTL_IP) && !(fsts_reg & status_fields)) {
>> + set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
>> + VTD_DPRINTF(FLOG, "all pending interrupt conditions serviced, clear
>> "
>> + "IP field of FECTL_REG");
>> + }
>> +}
>> +
>> +static inline void handle_fectl_write(IntelIOMMUState *s)
>> +{
>> + uint32_t fectl_reg;
>> + /* When software clears the IM field, check the IP field. But do we
>> + * need to compare the old value and the new value to conclude that
>> + * software clears the IM field? Or just check if the IM field is zero?
>> + */
>> + fectl_reg = get_long_raw(s, DMAR_FECTL_REG);
>> + if ((fectl_reg & VTD_FECTL_IP) && !(fectl_reg & VTD_FECTL_IM)) {
>> + vtd_generate_interrupt(s, DMAR_FEADDR_REG, DMAR_FEDATA_REG);
>> + set_clear_mask_long(s, DMAR_FECTL_REG, VTD_FECTL_IP, 0);
>> + VTD_DPRINTF(FLOG, "IM field is cleared, generate "
>> + "fault event interrupt");
>> + }
>> +}
>> +
>> +static uint64_t vtd_mem_read(void *opaque, hwaddr addr, unsigned size)
>> +{
>> + IntelIOMMUState *s = opaque;
>> + uint64_t val;
>> +
>> + if (addr + size > DMAR_REG_SIZE) {
>> + VTD_DPRINTF(GENERAL, "error: addr outside region: max 0x%"PRIx64
>> + ", got 0x%"PRIx64 " %d",
>> + (uint64_t)DMAR_REG_SIZE, addr, size);
>> + return (uint64_t)-1;
>> + }
>> +
>> + assert(size == 4 || size == 8);
>
> You already declare in the ops that you only support 4 and 8 byte
> accesses, no?
Yes, I added this for safe because I didn't know the mechanism of QEMU
well. I will remove them soon. :)
>> +
>> + switch (addr) {
>> + /* Root Table Address Register, 64-bit */
>> + case DMAR_RTADDR_REG:
>> + if (size == 4) {
>> + val = s->root & ((1ULL << 32) - 1);
>> + } else {
>> + val = s->root;
>> + }
>> + break;
>> +
>> + case DMAR_RTADDR_REG_HI:
>> + assert(size == 4);
>> + val = s->root >> 32;
>> + break;
>> +
>> + default:
>> + if (size == 4) {
>> + val = get_long(s, addr);
>> + } else {
>> + val = get_quad(s, addr);
>> + }
>> + }
>> +
>> + VTD_DPRINTF(CSR, "addr 0x%"PRIx64 " size %d val 0x%"PRIx64,
>> + addr, size, val);
>> + return val;
>> +}
>> +
>> +static void vtd_mem_write(void *opaque, hwaddr addr,
>> + uint64_t val, unsigned size)
>> +{
>> + IntelIOMMUState *s = opaque;
>> +
>> + if (addr + size > DMAR_REG_SIZE) {
>> + VTD_DPRINTF(GENERAL, "error: addr outside region: max 0x%"PRIx64
>> + ", got 0x%"PRIx64 " %d",
>> + (uint64_t)DMAR_REG_SIZE, addr, size);
>> + return;
>> + }
>> +
>> + assert(size == 4 || size == 8);
>> +
>> + switch (addr) {
>> + /* Global Command Register, 32-bit */
>> + case DMAR_GCMD_REG:
>> + VTD_DPRINTF(CSR, "DMAR_GCMD_REG write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + set_long(s, addr, val);
>> + handle_gcmd_write(s);
>> + break;
>> +
>> + /* Context Command Register, 64-bit */
>> + case DMAR_CCMD_REG:
>> + VTD_DPRINTF(CSR, "DMAR_CCMD_REG write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + if (size == 4) {
>> + set_long(s, addr, val);
>> + } else {
>> + set_quad(s, addr, val);
>> + handle_ccmd_write(s);
>> + }
>> + break;
>> +
>> + case DMAR_CCMD_REG_HI:
>> + VTD_DPRINTF(CSR, "DMAR_CCMD_REG_HI write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + handle_ccmd_write(s);
>> + break;
>> +
>> +
>> + /* IOTLB Invalidation Register, 64-bit */
>> + case DMAR_IOTLB_REG:
>> + VTD_DPRINTF(INV, "DMAR_IOTLB_REG write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + if (size == 4) {
>> + set_long(s, addr, val);
>> + } else {
>> + set_quad(s, addr, val);
>> + handle_iotlb_write(s);
>> + }
>> + break;
>> +
>> + case DMAR_IOTLB_REG_HI:
>> + VTD_DPRINTF(INV, "DMAR_IOTLB_REG_HI write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + handle_iotlb_write(s);
>> + break;
>> +
>> + /* Fault Status Register, 32-bit */
>> + case DMAR_FSTS_REG:
>> + VTD_DPRINTF(FLOG, "DMAR_FSTS_REG write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + handle_fsts_write(s);
>> + break;
>> +
>> + /* Fault Event Control Register, 32-bit */
>> + case DMAR_FECTL_REG:
>> + VTD_DPRINTF(FLOG, "DMAR_FECTL_REG write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + handle_fectl_write(s);
>> + break;
>> +
>> + /* Fault Event Data Register, 32-bit */
>> + case DMAR_FEDATA_REG:
>> + VTD_DPRINTF(FLOG, "DMAR_FEDATA_REG write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + break;
>> +
>> + /* Fault Event Address Register, 32-bit */
>> + case DMAR_FEADDR_REG:
>> + VTD_DPRINTF(FLOG, "DMAR_FEADDR_REG write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + break;
>> +
>> + /* Fault Event Upper Address Register, 32-bit */
>> + case DMAR_FEUADDR_REG:
>> + VTD_DPRINTF(FLOG, "DMAR_FEUADDR_REG write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + break;
>> +
>> + /* Protected Memory Enable Register, 32-bit */
>> + case DMAR_PMEN_REG:
>> + VTD_DPRINTF(CSR, "DMAR_PMEN_REG write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + break;
>> +
>> +
>> + /* Root Table Address Register, 64-bit */
>> + case DMAR_RTADDR_REG:
>> + VTD_DPRINTF(CSR, "DMAR_RTADDR_REG write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + if (size == 4) {
>> + set_long(s, addr, val);
>> + } else {
>> + set_quad(s, addr, val);
>> + }
>> + break;
>> +
>> + case DMAR_RTADDR_REG_HI:
>> + VTD_DPRINTF(CSR, "DMAR_RTADDR_REG_HI write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + break;
>> +
>> + /* Fault Recording Registers, 128-bit */
>> + case DMAR_FRCD_REG_0_0:
>> + VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_0 write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + if (size == 4) {
>> + set_long(s, addr, val);
>> + } else {
>> + set_quad(s, addr, val);
>> + }
>> + break;
>> +
>> + case DMAR_FRCD_REG_0_1:
>> + VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_1 write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + break;
>> +
>> + case DMAR_FRCD_REG_0_2:
>> + VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_2 write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + if (size == 4) {
>> + set_long(s, addr, val);
>> + } else {
>> + set_quad(s, addr, val);
>> + /* May clear bit 127 (Fault), update PPF */
>> + update_fsts_ppf(s);
>> + }
>> + break;
>> +
>> + case DMAR_FRCD_REG_0_3:
>> + VTD_DPRINTF(FLOG, "DMAR_FRCD_REG_0_3 write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + assert(size == 4);
>> + set_long(s, addr, val);
>> + /* May clear bit 127 (Fault), update PPF */
>> + update_fsts_ppf(s);
>> + break;
>> +
>> + default:
>> + VTD_DPRINTF(GENERAL, "error: unhandled reg write addr 0x%"PRIx64
>> + ", size %d, val 0x%"PRIx64, addr, size, val);
>> + if (size == 4) {
>> + set_long(s, addr, val);
>> + } else {
>> + set_quad(s, addr, val);
>> + }
>> + }
>> +
>> +}
>> +
>> +static IOMMUTLBEntry vtd_iommu_translate(MemoryRegion *iommu, hwaddr addr,
>> + bool is_write)
>> +{
>> + VTDAddressSpace *vtd_as = container_of(iommu, VTDAddressSpace, iommu);
>> + IntelIOMMUState *s = vtd_as->iommu_state;
>> + int bus_num = vtd_as->bus_num;
>> + int devfn = vtd_as->devfn;
>> + IOMMUTLBEntry ret = {
>> + .target_as = &address_space_memory,
>> + .iova = addr,
>> + .translated_addr = 0,
>> + .addr_mask = ~(hwaddr)0,
>> + .perm = IOMMU_NONE,
>> + };
>> +
>> + if (!s->dmar_enabled) {
>> + /* DMAR disabled, passthrough, use 4k-page*/
>> + ret.iova = addr & VTD_PAGE_MASK_4K;
>> + ret.translated_addr = addr & VTD_PAGE_MASK_4K;
>> + ret.addr_mask = ~VTD_PAGE_MASK_4K;
>> + ret.perm = IOMMU_RW;
>> + return ret;
>> + }
>> +
>> + iommu_translate(s, bus_num, devfn, addr, is_write, &ret);
>> +
>> + VTD_DPRINTF(MMU,
>> + "bus %d slot %d func %d devfn %d gpa %"PRIx64 " hpa
>> %"PRIx64,
>> + bus_num, VTD_PCI_SLOT(devfn), VTD_PCI_FUNC(devfn), devfn,
>> addr,
>> + ret.translated_addr);
>> + return ret;
>> +}
>> +
>> +static const VMStateDescription vtd_vmstate = {
>> + .name = "iommu_intel",
>> + .version_id = 1,
>> + .minimum_version_id = 1,
>> + .minimum_version_id_old = 1,
>> + .fields = (VMStateField[]) {
>> + VMSTATE_UINT8_ARRAY(csr, IntelIOMMUState, DMAR_REG_SIZE),
>> + VMSTATE_END_OF_LIST()
>> + }
>> +};
>
> Did you test migration? I suppose not. :)
>
> Background: you mirror several register states into IntelIOMMUState
> fields, I guess to make them more handle to use. However, those need to
> be updated on vmload. And there are surely more internal states that
> have to be migrated as well, e.g. the currently active root pointer.
>
> I would suggest to either review and fix this or leave migration support
> out for now (".unmigratable = 1").
Not tested yet. I declare some of them for the efficiency. For
example, the dmar_enabled variable. We need to decide whether the dmar
is enabled or not in iommu_translate(). I think it will be faster to
just read the dmar_enabled variable than using ldl_le_p().
So maybe for now I just leave migration support out and come back to
fix it someday.
>> +
>> +static const MemoryRegionOps vtd_mem_ops = {
>> + .read = vtd_mem_read,
>> + .write = vtd_mem_write,
>> + .endianness = DEVICE_LITTLE_ENDIAN,
>> + .impl = {
>> + .min_access_size = 4,
>> + .max_access_size = 8,
>> + },
>> + .valid = {
>> + .min_access_size = 4,
>> + .max_access_size = 8,
>> + },
>> +};
>> +
>> +static Property iommu_properties[] = {
>> + DEFINE_PROP_UINT32("version", IntelIOMMUState, version, 0),
>> + DEFINE_PROP_END_OF_LIST(),
>> +};
>> +
>> +/* Do the real initialization. It will also be called when reset, so pay
>> + * attention when adding new initialization stuff.
>> + */
>> +static void do_vtd_init(IntelIOMMUState *s)
>> +{
>> + memset(s->csr, 0, DMAR_REG_SIZE);
>> + memset(s->wmask, 0, DMAR_REG_SIZE);
>> + memset(s->w1cmask, 0, DMAR_REG_SIZE);
>> + memset(s->womask, 0, DMAR_REG_SIZE);
>> +
>> + s->iommu_ops.translate = vtd_iommu_translate;
>> + s->root = 0;
>> + s->root_extended = false;
>> + s->dmar_enabled = false;
>> + s->iq_head = 0;
>> + s->iq_tail = 0;
>> + s->iq = 0;
>> + s->iq_size = 0;
>> + s->qi_enabled = false;
>> + s->iq_last_desc_type = VTD_INV_DESC_NONE;
>> + s->next_frcd_reg = 0;
>> +
>> + /* b.0:2 = 6: Number of domains supported: 64K using 16 bit ids
>> + * b.3 = 0: Advanced fault logging not supported
>> + * b.4 = 0: Required write buffer flushing not supported
>> + * b.5 = 0: Protected low memory region not supported
>> + * b.6 = 0: Protected high memory region not supported
>> + * b.8:12 = 2: SAGAW(Supported Adjusted Guest Address Widths), 39-bit,
>> + * 3-level page-table
>> + * b.16:21 = 38: MGAW(Maximum Guest Address Width) = 39
>> + * b.22 = 0: ZLR(Zero Length Read) zero length DMA read requests
>> + * to write-only pages not supported
>> + * b.24:33 = 34: FRO(Fault-recording Register offset)
>> + * b.54 = 0: DWD(Write Draining), draining of write requests not
>> supported
>> + * b.55 = 0: DRD(Read Draining), draining of read requests not supported
>> + */
>
> I think this level of documentation is a bit overkill. You already
> document the register layout implicitly by defining the constants.
> Applies elsewhere, too.
Get it.
>
>> + s->cap = VTD_CAP_FRO | VTD_CAP_NFR | VTD_CAP_ND | VTD_CAP_MGAW |
>> + VTD_CAP_SAGAW;
>> +
>> + /* b.1 = 0: QI(Queued Invalidation support) not supported
>> + * b.2 = 0: DT(Device-TLB support) not supported
>> + * b.3 = 0: IR(Interrupt Remapping support) not supported
>> + * b.4 = 0: EIM(Extended Interrupt Mode) not supported
>> + * b.8:17 = 15: IRO(IOTLB Register Offset)
>> + * b.20:23 = 0: MHMV(Maximum Handle Mask Value) not valid
>> + */
>> + s->ecap = VTD_ECAP_IRO;
>> +
>> + /* Define registers with default values and bit semantics */
>> + define_long(s, DMAR_VER_REG, 0x10UL, 0, 0); /* set MAX = 1, RO */
>> + define_quad(s, DMAR_CAP_REG, s->cap, 0, 0);
>> + define_quad(s, DMAR_ECAP_REG, s->ecap, 0, 0);
>> + define_long(s, DMAR_GCMD_REG, 0, 0xff800000UL, 0);
>> + define_long_wo(s, DMAR_GCMD_REG, 0xff800000UL);
>> + define_long(s, DMAR_GSTS_REG, 0, 0, 0); /* All bits RO, default 0 */
>> + define_quad(s, DMAR_RTADDR_REG, 0, 0xfffffffffffff000ULL, 0);
>> + define_quad(s, DMAR_CCMD_REG, 0, 0xe0000003ffffffffULL, 0);
>> + define_quad_wo(s, DMAR_CCMD_REG, 0x3ffff0000ULL);
>> +
>> + /* Advanced Fault Logging not supported */
>> + define_long(s, DMAR_FSTS_REG, 0, 0, 0x11UL);
>> + define_long(s, DMAR_FECTL_REG, 0x80000000UL, 0x80000000UL, 0);
>> + define_long(s, DMAR_FEDATA_REG, 0, 0x0000ffffUL, 0); /* 15:0 RW */
>> + define_long(s, DMAR_FEADDR_REG, 0, 0xfffffffcUL, 0); /* 31:2 RW */
>> +
>> + /* Treated as RsvdZ when EIM in ECAP_REG is not supported
>> + * define_long(s, DMAR_FEUADDR_REG, 0, 0xffffffffUL, 0);
>> + */
>> + define_long(s, DMAR_FEUADDR_REG, 0, 0, 0);
>> +
>> + /* Treated as RO for implementations that PLMR and PHMR fields reported
>> + * as Clear in the CAP_REG.
>> + * define_long(s, DMAR_PMEN_REG, 0, 0x80000000UL, 0);
>> + */
>> + define_long(s, DMAR_PMEN_REG, 0, 0, 0);
>> +
>> + /* IOTLB registers */
>> + define_quad(s, DMAR_IOTLB_REG, 0, 0Xb003ffff00000000ULL, 0);
>> + define_quad(s, DMAR_IVA_REG, 0, 0xfffffffffffff07fULL, 0);
>> + define_quad_wo(s, DMAR_IVA_REG, 0xfffffffffffff07fULL);
>> +
>> + /* Fault Recording Registers, 128-bit */
>> + define_quad(s, DMAR_FRCD_REG_0_0, 0, 0, 0);
>> + define_quad(s, DMAR_FRCD_REG_0_2, 0, 0, 0x8000000000000000ULL);
>> +}
>> +
>> +/* Reset function of QOM
>> + * Should not reset address_spaces when reset
>
> What does "should not" mean here? Is it an open todo?
Because after the realize of VT-d, devices will call
pci_device_iommu_address_space() to get the address space, and then
qemu will call the vtd_reset(). What's more, when L1 resets,
vtd_reset() will be called twice, but devices won't try to get the
address space again. They just use the original one. So I think we
should not reset the address_spaces[] in vtd_reset(). And anyway,
there seems to be nothing in VTDAddressSpace need to be reset.
>> + */
>> +static void vtd_reset(DeviceState *dev)
>> +{
>> + IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
>> +
>> + VTD_DPRINTF(GENERAL, "");
>> + do_vtd_init(s);
>> +}
>> +
>> +/* Initialization function of QOM */
>> +static void vtd_realize(DeviceState *dev, Error **errp)
>> +{
>> + IntelIOMMUState *s = INTEL_IOMMU_DEVICE(dev);
>> +
>> + VTD_DPRINTF(GENERAL, "");
>> + memset(s->address_spaces, 0, sizeof(s->address_spaces));
>> + memory_region_init_io(&s->csrmem, OBJECT(s), &vtd_mem_ops, s,
>> + "intel_iommu", DMAR_REG_SIZE);
>> + sysbus_init_mmio(SYS_BUS_DEVICE(s), &s->csrmem);
>> + do_vtd_init(s);
>> +}
>> +
>> +static void vtd_class_init(ObjectClass *klass, void *data)
>> +{
>> + DeviceClass *dc = DEVICE_CLASS(klass);
>> +
>> + dc->reset = vtd_reset;
>> + dc->realize = vtd_realize;
>> + dc->vmsd = &vtd_vmstate;
>> + dc->props = iommu_properties;
>> +}
>> +
>> +static const TypeInfo vtd_info = {
>> + .name = TYPE_INTEL_IOMMU_DEVICE,
>> + .parent = TYPE_SYS_BUS_DEVICE,
>> + .instance_size = sizeof(IntelIOMMUState),
>> + .class_init = vtd_class_init,
>> +};
>> +
>> +static void vtd_register_types(void)
>> +{
>> + VTD_DPRINTF(GENERAL, "");
>> + type_register_static(&vtd_info);
>> +}
>> +
>> +type_init(vtd_register_types)
>> diff --git a/hw/i386/intel_iommu_internal.h b/hw/i386/intel_iommu_internal.h
>> new file mode 100644
>> index 0000000..7bc679a
>> --- /dev/null
>> +++ b/hw/i386/intel_iommu_internal.h
>> @@ -0,0 +1,345 @@
>> +/*
>> + * QEMU emulation of an Intel IOMMU (VT-d)
>> + * (DMA Remapping device)
>> + *
>> + * Copyright (C) 2013 Knut Omang, Oracle <knut.om...@oracle.com>
>> + * Copyright (C) 2014 Le Tan, <tamlokv...@gmail.com>
>> + *
>> + * This program is free software; you can redistribute it and/or modify
>> + * it under the terms of the GNU General Public License as published by
>> + * the Free Software Foundation; either version 2 of the License, or
>> + * (at your option) any later version.
>> +
>> + * This program is distributed in the hope that it will be useful,
>> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
>> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
>> + * GNU General Public License for more details.
>> +
>> + * You should have received a copy of the GNU General Public License along
>> + * with this program; if not, see <http://www.gnu.org/licenses/>.
>> + *
>> + * Lots of defines copied from kernel/include/linux/intel-iommu.h:
>> + * Copyright (C) 2006-2008 Intel Corporation
>> + * Author: Ashok Raj <ashok....@intel.com>
>> + * Author: Anil S Keshavamurthy <anil.s.keshavamur...@intel.com>
>> + *
>> + */
>> +
>> +#ifndef HW_I386_INTEL_IOMMU_INTERNAL_H
>> +#define HW_I386_INTEL_IOMMU_INTERNAL_H
>> +#include "hw/i386/intel_iommu.h"
>> +
>> +/*
>> + * Intel IOMMU register specification
>> + */
>> +#define DMAR_VER_REG 0x0 /* Arch version supported by this IOMMU */
>> +#define DMAR_CAP_REG 0x8 /* Hardware supported capabilities */
>> +#define DMAR_CAP_REG_HI 0xc /* High 32-bit of DMAR_CAP_REG */
>> +#define DMAR_ECAP_REG 0x10 /* Extended capabilities supported */
>> +#define DMAR_ECAP_REG_HI 0X14
>> +#define DMAR_GCMD_REG 0x18 /* Global command register */
>> +#define DMAR_GSTS_REG 0x1c /* Global status register */
>> +#define DMAR_RTADDR_REG 0x20 /* Root entry table */
>> +#define DMAR_RTADDR_REG_HI 0X24
>> +#define DMAR_CCMD_REG 0x28 /* Context command reg */
>> +#define DMAR_CCMD_REG_HI 0x2c
>> +#define DMAR_FSTS_REG 0x34 /* Fault Status register */
>> +#define DMAR_FECTL_REG 0x38 /* Fault control register */
>> +#define DMAR_FEDATA_REG 0x3c /* Fault event interrupt data register */
>> +#define DMAR_FEADDR_REG 0x40 /* Fault event interrupt addr register */
>> +#define DMAR_FEUADDR_REG 0x44 /* Upper address register */
>> +#define DMAR_AFLOG_REG 0x58 /* Advanced Fault control */
>> +#define DMAR_AFLOG_REG_HI 0X5c
>> +#define DMAR_PMEN_REG 0x64 /* Enable Protected Memory Region */
>> +#define DMAR_PLMBASE_REG 0x68 /* PMRR Low addr */
>> +#define DMAR_PLMLIMIT_REG 0x6c /* PMRR low limit */
>> +#define DMAR_PHMBASE_REG 0x70 /* pmrr high base addr */
>> +#define DMAR_PHMBASE_REG_HI 0X74
>> +#define DMAR_PHMLIMIT_REG 0x78 /* pmrr high limit */
>> +#define DMAR_PHMLIMIT_REG_HI 0x7c
>> +#define DMAR_IQH_REG 0x80 /* Invalidation queue head register */
>> +#define DMAR_IQH_REG_HI 0X84
>> +#define DMAR_IQT_REG 0x88 /* Invalidation queue tail register */
>> +#define DMAR_IQT_REG_HI 0X8c
>> +#define DMAR_IQ_SHIFT 4 /* Invalidation queue head/tail shift */
>> +#define DMAR_IQA_REG 0x90 /* Invalidation queue addr register */
>> +#define DMAR_IQA_REG_HI 0x94
>> +#define DMAR_ICS_REG 0x9c /* Invalidation complete status register */
>> +#define DMAR_IRTA_REG 0xb8 /* Interrupt remapping table addr register
>> */
>> +#define DMAR_IRTA_REG_HI 0xbc
>
> Please align all those constants:
>
> #define CONSTANT 0x1234
> #define CONSTANT_WITH_LONGER_NAME 0x5678
>
What does align mean here? Like this?
#define CONSTANT 0x1234
#define CONSTANT_WITH_LONGER_NAME 0x5678
>> +
>> +#define DMAR_IECTL_REG 0xa0 /* Invalidation event control register */
>> +#define DMAR_IEDATA_REG 0xa4 /* Invalidation event data register */
>> +#define DMAR_IEADDR_REG 0xa8 /* Invalidation event address register */
>> +#define DMAR_IEUADDR_REG 0xac /* Invalidation event address register */
>> +#define DMAR_PQH_REG 0xc0 /* Page request queue head register */
>> +#define DMAR_PQH_REG_HI 0xc4
>> +#define DMAR_PQT_REG 0xc8 /* Page request queue tail register*/
>> +#define DMAR_PQT_REG_HI 0xcc
>> +#define DMAR_PQA_REG 0xd0 /* Page request queue address register */
>> +#define DMAR_PQA_REG_HI 0xd4
>> +#define DMAR_PRS_REG 0xdc /* Page request status register */
>> +#define DMAR_PECTL_REG 0xe0 /* Page request event control register */
>> +#define DMAR_PEDATA_REG 0xe4 /* Page request event data register */
>> +#define DMAR_PEADDR_REG 0xe8 /* Page request event address register */
>> +#define DMAR_PEUADDR_REG 0xec /* Page event upper address register */
>> +#define DMAR_MTRRCAP_REG 0x100 /* MTRR capability register */
>> +#define DMAR_MTRRCAP_REG_HI 0x104
>> +#define DMAR_MTRRDEF_REG 0x108 /* MTRR default type register */
>> +#define DMAR_MTRRDEF_REG_HI 0x10c
>> +
>> +/* IOTLB */
>> +#define DMAR_IOTLB_REG_OFFSET 0xf0 /* Offset to the IOTLB registers */
>> +#define DMAR_IVA_REG DMAR_IOTLB_REG_OFFSET /* Invalidate Address Register
>> */
>> +#define DMAR_IVA_REG_HI (DMAR_IVA_REG + 4)
>> +/* IOTLB Invalidate Register */
>> +#define DMAR_IOTLB_REG (DMAR_IOTLB_REG_OFFSET + 0x8)
>> +#define DMAR_IOTLB_REG_HI (DMAR_IOTLB_REG + 4)
>> +
>> +/* FRCD */
>> +#define DMAR_FRCD_REG_OFFSET 0x220 /* Offset to the Fault Recording
>> Registers */
>> +/* NOTICE: If you change the DMAR_FRCD_REG_NR, please remember to change the
>> + * DMAR_REG_SIZE in include/hw/i386/intel_iommu.h.
>> + * #define DMAR_REG_SIZE (DMAR_FRCD_REG_OFFSET + 16 * DMAR_FRCD_REG_NR)
>> + */
>> +#define DMAR_FRCD_REG_NR 1ULL /* Num of Fault Recording Registers */
>> +
>> +#define DMAR_FRCD_REG_0_0 0x220 /* The 0th Fault Recording Register */
>> +#define DMAR_FRCD_REG_0_1 0x224
>> +#define DMAR_FRCD_REG_0_2 0x228
>> +#define DMAR_FRCD_REG_0_3 0x22c
>> +
>> +/* Interrupt Address Range */
>> +#define VTD_INTERRUPT_ADDR_FIRST 0xfee00000ULL
>> +#define VTD_INTERRUPT_ADDR_LAST 0xfeefffffULL
>> +
>> +/* IOTLB_REG */
>> +#define VTD_TLB_GLOBAL_FLUSH (1ULL << 60) /* Global invalidation */
>> +#define VTD_TLB_DSI_FLUSH (2ULL << 60) /* Domain-selective invalidation */
>> +#define VTD_TLB_PSI_FLUSH (3ULL << 60) /* Page-selective invalidation */
>> +#define VTD_TLB_FLUSH_GRANU_MASK (3ULL << 60)
>> +#define VTD_TLB_GLOBAL_FLUSH_A (1ULL << 57)
>> +#define VTD_TLB_DSI_FLUSH_A (2ULL << 57)
>> +#define VTD_TLB_PSI_FLUSH_A (3ULL << 57)
>> +#define VTD_TLB_FLUSH_GRANU_MASK_A (3ULL << 57)
>> +#define VTD_TLB_IVT (1ULL << 63)
>> +
>> +/* GCMD_REG */
>> +#define VTD_GCMD_TE (1UL << 31)
>> +#define VTD_GCMD_SRTP (1UL << 30)
>> +#define VTD_GCMD_SFL (1UL << 29)
>> +#define VTD_GCMD_EAFL (1UL << 28)
>> +#define VTD_GCMD_WBF (1UL << 27)
>> +#define VTD_GCMD_QIE (1UL << 26)
>> +#define VTD_GCMD_IRE (1UL << 25)
>> +#define VTD_GCMD_SIRTP (1UL << 24)
>> +#define VTD_GCMD_CFI (1UL << 23)
>> +
>> +/* GSTS_REG */
>> +#define VTD_GSTS_TES (1UL << 31)
>> +#define VTD_GSTS_RTPS (1UL << 30)
>> +#define VTD_GSTS_FLS (1UL << 29)
>> +#define VTD_GSTS_AFLS (1UL << 28)
>> +#define VTD_GSTS_WBFS (1UL << 27)
>> +#define VTD_GSTS_QIES (1UL << 26)
>> +#define VTD_GSTS_IRES (1UL << 25)
>> +#define VTD_GSTS_IRTPS (1UL << 24)
>> +#define VTD_GSTS_CFIS (1UL << 23)
>> +
>> +/* CCMD_REG */
>> +#define VTD_CCMD_ICC (1ULL << 63)
>> +#define VTD_CCMD_GLOBAL_INVL (1ULL << 61)
>> +#define VTD_CCMD_DOMAIN_INVL (2ULL << 61)
>> +#define VTD_CCMD_DEVICE_INVL (3ULL << 61)
>> +#define VTD_CCMD_CIRG_MASK (3ULL << 61)
>> +#define VTD_CCMD_GLOBAL_INVL_A (1ULL << 59)
>> +#define VTD_CCMD_DOMAIN_INVL_A (2ULL << 59)
>> +#define VTD_CCMD_DEVICE_INVL_A (3ULL << 59)
>> +#define VTD_CCMD_CAIG_MASK (3ULL << 59)
>> +
>> +/* RTADDR_REG */
>> +#define VTD_RTADDR_RTT (1ULL << 11)
>> +#define VTD_RTADDR_ADDR_MASK (VTD_HAW_MASK ^ 0xfffULL)
>> +
>> +/* ECAP_REG */
>> +#define VTD_ECAP_IRO (DMAR_IOTLB_REG_OFFSET << 4) /* (offset >> 4) << 8 */
>> +#define VTD_ECAP_QI (1ULL << 1)
>> +
>> +/* CAP_REG */
>> +#define VTD_CAP_FRO (DMAR_FRCD_REG_OFFSET << 20) /* (offset >> 4) << 24 */
>> +#define VTD_CAP_NFR ((DMAR_FRCD_REG_NR - 1) << 40)
>> +#define VTD_DOMAIN_ID_SHIFT 16 /* 16-bit domain id for 64K domains */
>> +#define VTD_CAP_ND (((VTD_DOMAIN_ID_SHIFT - 4) / 2) & 7ULL)
>> +#define VTD_MGAW 39 /* Maximum Guest Address Width */
>> +#define VTD_CAP_MGAW (((VTD_MGAW - 1) & 0x3fULL) << 16)
>> +
>> +/* Supported Adjusted Guest Address Widths */
>> +#define VTD_CAP_SAGAW_SHIFT (8)
>> +#define VTD_CAP_SAGAW_MASK (0x1fULL << VTD_CAP_SAGAW_SHIFT)
>> + /* 39-bit AGAW, 3-level page-table */
>> +#define VTD_CAP_SAGAW_39bit (0x2ULL << VTD_CAP_SAGAW_SHIFT)
>> + /* 48-bit AGAW, 4-level page-table */
>> +#define VTD_CAP_SAGAW_48bit (0x4ULL << VTD_CAP_SAGAW_SHIFT)
>> +#define VTD_CAP_SAGAW VTD_CAP_SAGAW_39bit
>> +
>> +/* IQT_REG */
>> +#define VTD_IQT_QT(val) (((val) >> 4) & 0x7fffULL)
>> +
>> +/* IQA_REG */
>> +#define VTD_IQA_IQA_MASK (VTD_HAW_MASK ^ 0xfffULL)
>> +#define VTD_IQA_QS (0x7ULL)
>> +
>> +/* IQH_REG */
>> +#define VTD_IQH_QH_SHIFT (4)
>> +#define VTD_IQH_QH_MASK (0x7fff0ULL)
>
> No need for braces around plain values (i.e. when there are no
> operators), here and elsewhere.
Get it.
>> +
>> +/* ICS_REG */
>> +#define VTD_ICS_IWC (1UL)
>> +
>> +/* IECTL_REG */
>> +#define VTD_IECTL_IM (1UL << 31)
>> +#define VTD_IECTL_IP (1UL << 30)
>> +
>> +/* FSTS_REG */
>> +#define VTD_FSTS_FRI_MASK (0xff00)
>> +#define VTD_FSTS_FRI(val) ((((uint32_t)(val)) << 8) & VTD_FSTS_FRI_MASK)
>> +#define VTD_FSTS_IQE (1UL << 4)
>> +#define VTD_FSTS_PPF (1UL << 1)
>> +#define VTD_FSTS_PFO (1UL)
>> +
>> +/* FECTL_REG */
>> +#define VTD_FECTL_IM (1UL << 31)
>> +#define VTD_FECTL_IP (1UL << 30)
>> +
>> +/* Fault Recording Register */
>> +/* For the high 64-bit of 128-bit */
>> +#define VTD_FRCD_F (1ULL << 63)
>> +#define VTD_FRCD_T (1ULL << 62)
>> +#define VTD_FRCD_FR(val) (((val) & 0xffULL) << 32)
>> +#define VTD_FRCD_SID_MASK 0xffffULL
>> +#define VTD_FRCD_SID(val) ((val) & VTD_FRCD_SID_MASK)
>> +/* For the low 64-bit of 128-bit */
>> +#define VTD_FRCD_FI(val) ((val) & (((1ULL << VTD_MGAW) - 1) ^ 0xfffULL))
>> +
>> +/* DMA Remapping Fault Conditions */
>> +typedef enum VTDFaultReason {
>> + /* Reserved for Advanced Fault logging. We use this to represent the
>> case
>> + * with no fault event.
>> + */
>> + VTD_FR_RESERVED = 0,
>> + VTD_FR_ROOT_ENTRY_P = 1, /* The Present(P) field of root-entry is 0 */
>> + VTD_FR_CONTEXT_ENTRY_P, /* The Present(P) field of context-entry is 0 */
>> + VTD_FR_CONTEXT_ENTRY_INV, /* Invalid programming of a context-entry */
>> + VTD_FR_ADDR_BEYOND_MGAW, /* Input-address above (2^x-1) */
>> + VTD_FR_WRITE, /* No write permission */
>> + VTD_FR_READ, /* No read permission */
>> + /* Fail to access a second-level paging entry (not SL_PML4E) */
>> + VTD_FR_PAGING_ENTRY_INV,
>> + VTD_FR_ROOT_TABLE_INV, /* Fail to access a root-entry */
>> + VTD_FR_CONTEXT_TABLE_INV, /* Fail to access a context-entry */
>> + /* Non-zero reserved field in a present root-entry */
>> + VTD_FR_ROOT_ENTRY_RSVD,
>> + /* Non-zero reserved field in a present context-entry */
>> + VTD_FR_CONTEXT_ENTRY_RSVD,
>> + /* Non-zero reserved field in a second-level paging entry with at lease
>> one
>> + * Read(R) and Write(W) or Execute(E) field is Set.
>> + */
>> + VTD_FR_PAGING_ENTRY_RSVD,
>> + /* Translation request or translated request explicitly blocked dut to
>> the
>> + * programming of the Translation Type (T) field in the present
>> + * context-entry.
>> + */
>> + VTD_FR_CONTEXT_ENTRY_TT,
>> + /* This is not a normal fault reason. We use this to indicate some
>> faults
>> + * that are not referenced by the VT-d specification.
>> + * Fault event with such reason should not be recorded.
>> + */
>> + VTD_FR_RESERVED_ERR,
>> + /* Guard */
>> + VTD_FR_MAX,
>> +} VTDFaultReason;
>> +
>> +
>> +/* Masks for Queued Invalidation Descriptor */
>> +#define VTD_INV_DESC_TYPE (0xf)
>> +#define VTD_INV_DESC_CC (0x1) /* Context-cache Invalidate Descriptor */
>> +#define VTD_INV_DESC_IOTLB (0x2)
>> +#define VTD_INV_DESC_WAIT (0x5) /* Invalidation Wait Descriptor */
>> +#define VTD_INV_DESC_NONE (0) /* Not an Invalidate Descriptor */
>> +
>> +
>> +/* Pagesize of VTD paging structures, including root and context tables */
>> +#define VTD_PAGE_SHIFT (12)
>> +#define VTD_PAGE_SIZE (1ULL << VTD_PAGE_SHIFT)
>> +
>> +#define VTD_PAGE_SHIFT_4K (12)
>> +#define VTD_PAGE_MASK_4K (~((1ULL << VTD_PAGE_SHIFT_4K) - 1))
>> +#define VTD_PAGE_SHIFT_2M (21)
>> +#define VTD_PAGE_MASK_2M (~((1ULL << VTD_PAGE_SHIFT_2M) - 1))
>> +#define VTD_PAGE_SHIFT_1G (30)
>> +#define VTD_PAGE_MASK_1G (~((1ULL << VTD_PAGE_SHIFT_1G) - 1))
>> +
>> +/* Root-Entry
>> + * 0: Present
>> + * 1-11: Reserved
>> + * 12-63: Context-table Pointer
>> + * 64-127: Reserved
>> + */
>> +struct VTDRootEntry {
>> + uint64_t val;
>> + uint64_t rsvd;
>> +};
>> +typedef struct VTDRootEntry VTDRootEntry;
>> +
>> +/* Masks for struct VTDRootEntry */
>> +#define VTD_ROOT_ENTRY_P (1ULL << 0)
>> +#define VTD_ROOT_ENTRY_CTP (~0xfffULL)
>> +
>> +#define VTD_ROOT_ENTRY_NR (VTD_PAGE_SIZE / sizeof(VTDRootEntry))
>> +#define VTD_ROOT_ENTRY_RSVD (0xffeULL | ~VTD_HAW_MASK)
>> +
>> +/* Context-Entry */
>> +struct VTDContextEntry {
>> + uint64_t lo;
>> + uint64_t hi;
>> +};
>> +typedef struct VTDContextEntry VTDContextEntry;
>> +
>> +/* Masks for struct VTDContextEntry */
>> +/* lo */
>> +#define VTD_CONTEXT_ENTRY_P (1ULL << 0)
>> +#define VTD_CONTEXT_ENTRY_FPD (1ULL << 1) /* Fault Processing Disable */
>> +#define VTD_CONTEXT_ENTRY_TT (3ULL << 2) /* Translation Type */
>> +#define VTD_CONTEXT_TT_MULTI_LEVEL (0)
>> +#define VTD_CONTEXT_TT_DEV_IOTLB (1)
>> +#define VTD_CONTEXT_TT_PASS_THROUGH (2)
>> +/* Second Level Page Translation Pointer*/
>> +#define VTD_CONTEXT_ENTRY_SLPTPTR (~0xfffULL)
>> +#define VTD_CONTEXT_ENTRY_RSVD_LO (0xff0ULL | ~VTD_HAW_MASK)
>> +/* hi */
>> +#define VTD_CONTEXT_ENTRY_AW (7ULL) /* Adjusted guest-address-width */
>> +#define VTD_CONTEXT_ENTRY_DID (0xffffULL << 8) /* Domain Identifier */
>> +#define VTD_CONTEXT_ENTRY_RSVD_HI (0xffffffffff000080ULL)
>> +
>> +#define VTD_CONTEXT_ENTRY_NR (VTD_PAGE_SIZE / sizeof(VTDContextEntry))
>> +
>> +
>> +/* Paging Structure common */
>> +#define VTD_SL_PT_PAGE_SIZE_MASK (1ULL << 7)
>> +#define VTD_SL_LEVEL_BITS 9 /* Bits to decide the offset for each level
>> */
>> +
>> +/* Second Level Paging Structure */
>> +#define VTD_SL_PML4_LEVEL 4
>> +#define VTD_SL_PDP_LEVEL 3
>> +#define VTD_SL_PD_LEVEL 2
>> +#define VTD_SL_PT_LEVEL 1
>> +#define VTD_SL_PT_ENTRY_NR 512
>> +
>> +/* Masks for Second Level Paging Entry */
>> +#define VTD_SL_RW_MASK (3ULL)
>> +#define VTD_SL_R (1ULL)
>> +#define VTD_SL_W (1ULL << 1)
>> +#define VTD_SL_PT_BASE_ADDR_MASK (~(VTD_PAGE_SIZE - 1) & VTD_HAW_MASK)
>> +#define VTD_SL_IGN_COM (0xbff0000000000000ULL)
>> +
>> +#endif
>> diff --git a/include/hw/i386/intel_iommu.h b/include/hw/i386/intel_iommu.h
>> new file mode 100644
>> index 0000000..6601e62
>> --- /dev/null
>> +++ b/include/hw/i386/intel_iommu.h
>> @@ -0,0 +1,90 @@
>> +/*
>> + * QEMU emulation of an Intel IOMMU (VT-d)
>> + * (DMA Remapping device)
>> + *
>> + * Copyright (C) 2013 Knut Omang, Oracle <knut.om...@oracle.com>
>> + * Copyright (C) 2014 Le Tan, <tamlokv...@gmail.com>
>> + *
>> + * This program is free software; you can redistribute it and/or modify
>> + * it under the terms of the GNU General Public License as published by
>> + * the Free Software Foundation; either version 2 of the License, or
>> + * (at your option) any later version.
>> +
>> + * This program is distributed in the hope that it will be useful,
>> + * but WITHOUT ANY WARRANTY; without even the implied warranty of
>> + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
>> + * GNU General Public License for more details.
>> +
>> + * You should have received a copy of the GNU General Public License along
>> + * with this program; if not, see <http://www.gnu.org/licenses/>.
>> + */
>> +
>> +#ifndef INTEL_IOMMU_H
>> +#define INTEL_IOMMU_H
>> +#include "hw/qdev.h"
>> +#include "sysemu/dma.h"
>> +
>> +#define TYPE_INTEL_IOMMU_DEVICE "intel-iommu"
>> +#define INTEL_IOMMU_DEVICE(obj) \
>> + OBJECT_CHECK(IntelIOMMUState, (obj), TYPE_INTEL_IOMMU_DEVICE)
>> +
>> +/* DMAR Hardware Unit Definition address (IOMMU unit) */
>> +#define Q35_HOST_BRIDGE_IOMMU_ADDR 0xfed90000ULL
>> +
>> +#define VTD_PCI_BUS_MAX 256
>> +#define VTD_PCI_SLOT_MAX 32
>> +#define VTD_PCI_FUNC_MAX 8
>> +#define VTD_PCI_SLOT(devfn) (((devfn) >> 3) & 0x1f)
>> +#define VTD_PCI_FUNC(devfn) ((devfn) & 0x07)
>> +
>> +#define DMAR_REG_SIZE 0x230
>> +
>> +/* FIXME: do not know how to decide the haw */
>
> Nothing to fix IMHO. Just state that this definition is arbitrary, just
> large enough to cover all currently expected guest RAM sizes.
OK.
>> +#define VTD_HOST_ADDRESS_WIDTH 39
>> +#define VTD_HAW_MASK ((1ULL << VTD_HOST_ADDRESS_WIDTH) - 1)
>> +
>> +typedef struct IntelIOMMUState IntelIOMMUState;
>> +typedef struct VTDAddressSpace VTDAddressSpace;
>> +
>> +struct VTDAddressSpace {
>> + int bus_num;
>> + int devfn;
>> + AddressSpace as;
>> + MemoryRegion iommu;
>> + IntelIOMMUState *iommu_state;
>> +};
>> +
>> +/* The iommu (DMAR) device state struct */
>> +struct IntelIOMMUState {
>> + SysBusDevice busdev;
>> + MemoryRegion csrmem;
>> + uint8_t csr[DMAR_REG_SIZE]; /* register values */
>> + uint8_t wmask[DMAR_REG_SIZE]; /* R/W bytes */
>> + uint8_t w1cmask[DMAR_REG_SIZE]; /* RW1C(Write 1 to Clear) bytes */
>> + uint8_t womask[DMAR_REG_SIZE]; /* WO (write only - read returns 0) */
>> + uint32_t version;
>> +
>> + dma_addr_t root; /* Current root table pointer */
>> + bool root_extended; /* Type of root table (extended or not) */
>> + bool dmar_enabled; /* Set if DMA remapping is enabled */
>> +
>> + uint16_t iq_head; /* Current invalidation queue head */
>> + uint16_t iq_tail; /* Current invalidation queue tail */
>> + dma_addr_t iq; /* Current invalidation queue (IQ) pointer */
>> + uint16_t iq_size; /* IQ Size in number of entries */
>> + bool qi_enabled; /* Set if the QI is enabled */
>> + uint8_t iq_last_desc_type; /* The type of last completed descriptor */
>> +
>> + /* The index of the Fault Recording Register to be used next.
>> + * Wraps around from N-1 to 0, where N is the number of FRCD_REG.
>> + */
>> + uint16_t next_frcd_reg;
>> +
>> + uint64_t cap; /* The value of Capability Register */
>> + uint64_t ecap; /* The value of Extended Capability Register */
>> +
>> + MemoryRegionIOMMUOps iommu_ops;
>> + VTDAddressSpace **address_spaces[VTD_PCI_BUS_MAX];
>> +};
>> +
>> +#endif
>>
>
> Very nice job!
Thanks very much for your review! :)
Le
> Jan
>
