On Fri, Feb 04, 2011 at 01:32:57AM +0200, Eduard - Gabriel Munteanu wrote:
> This introduces emulation for the AMD IOMMU, described in "AMD I/O
> Virtualization Technology (IOMMU) Specification".
>
> Signed-off-by: Eduard - Gabriel Munteanu <eduard.munteanu(a)linux360.ro>
> ---
> Makefile.target | 2 +-
> hw/amd_iommu.c | 694 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
> hw/pc.c | 2 +
> hw/pci_ids.h | 2 +
> hw/pci_regs.h | 1 +
> 5 files changed, 700 insertions(+), 1 deletions(-)
> create mode 100644 hw/amd_iommu.c
>
> diff --git a/Makefile.target b/Makefile.target
> index e5817ab..4b650bd 100644
> --- a/Makefile.target
> +++ b/Makefile.target
> @@ -218,7 +218,7 @@ obj-i386-y += cirrus_vga.o apic.o ioapic.o piix_pci.o
> obj-i386-y += vmmouse.o vmport.o hpet.o applesmc.o
> obj-i386-y += device-hotplug.o pci-hotplug.o smbios.o wdt_ib700.o
> obj-i386-y += debugcon.o multiboot.o
> -obj-i386-y += pc_piix.o dma_rw.o
> +obj-i386-y += pc_piix.o dma_rw.o amd_iommu.o
> obj-i386-$(CONFIG_SPICE) += qxl.o qxl-logger.o qxl-render.o
>
> # shared objects
> diff --git a/hw/amd_iommu.c b/hw/amd_iommu.c
> new file mode 100644
> index 0000000..6c6346a
> --- /dev/null
> +++ b/hw/amd_iommu.c
> @@ -0,0 +1,694 @@
> +/*
> + * AMD IOMMU emulation
> + *
> + * Copyright (c) 2011 Eduard - Gabriel Munteanu
> + *
> + * Permission is hereby granted, free of charge, to any person obtaining a copy
> + * of this software and associated documentation files (the "Software"), to deal
> + * in the Software without restriction, including without limitation the rights
> + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
> + * copies of the Software, and to permit persons to whom the Software is
> + * furnished to do so, subject to the following conditions:
> + *
> + * The above copyright notice and this permission notice shall be included in
> + * all copies or substantial portions of the Software.
> + *
> + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
> + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
> + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
> + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
> + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
> + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
> + * THE SOFTWARE.
> + */
> +
> +#include "pc.h"
> +#include "hw.h"
> +#include "pci.h"
> +#include "qlist.h"
> +#include "dma_rw.h"
> +
> +/* Capability registers */
> +#define CAPAB_HEADER 0x00
> +#define CAPAB_REV_TYPE 0x02
> +#define CAPAB_FLAGS 0x03
> +#define CAPAB_BAR_LOW 0x04
> +#define CAPAB_BAR_HIGH 0x08
> +#define CAPAB_RANGE 0x0C
> +#define CAPAB_MISC 0x10
> +
> +#define CAPAB_SIZE 0x14
> +#define CAPAB_REG_SIZE 0x04
> +
> +/* Capability header data */
> +#define CAPAB_FLAG_IOTLBSUP (1 << 0)
> +#define CAPAB_FLAG_HTTUNNEL (1 << 1)
> +#define CAPAB_FLAG_NPCACHE (1 << 2)
> +#define CAPAB_INIT_REV (1 << 3)
> +#define CAPAB_INIT_TYPE 3
> +#define CAPAB_INIT_REV_TYPE (CAPAB_REV | CAPAB_TYPE)
> +#define CAPAB_INIT_FLAGS (CAPAB_FLAG_NPCACHE | CAPAB_FLAG_HTTUNNEL)
> +#define CAPAB_INIT_MISC (64 << 15) | (48 << 8)
> +#define CAPAB_BAR_MASK ~((1UL << 14) - 1)
> +
> +/* MMIO registers */
> +#define MMIO_DEVICE_TABLE 0x0000
> +#define MMIO_COMMAND_BASE 0x0008
> +#define MMIO_EVENT_BASE 0x0010
> +#define MMIO_CONTROL 0x0018
> +#define MMIO_EXCL_BASE 0x0020
> +#define MMIO_EXCL_LIMIT 0x0028
> +#define MMIO_COMMAND_HEAD 0x2000
> +#define MMIO_COMMAND_TAIL 0x2008
> +#define MMIO_EVENT_HEAD 0x2010
> +#define MMIO_EVENT_TAIL 0x2018
> +#define MMIO_STATUS 0x2020
> +
> +#define MMIO_SIZE 0x4000
> +
> +#define MMIO_DEVTAB_SIZE_MASK ((1ULL << 12) - 1)
> +#define MMIO_DEVTAB_BASE_MASK (((1ULL << 52) - 1) & ~MMIO_DEVTAB_SIZE_MASK)
> +#define MMIO_DEVTAB_ENTRY_SIZE 32
> +#define MMIO_DEVTAB_SIZE_UNIT 4096
> +
> +#define MMIO_CMDBUF_SIZE_BYTE (MMIO_COMMAND_BASE + 7)
> +#define MMIO_CMDBUF_SIZE_MASK 0x0F
> +#define MMIO_CMDBUF_BASE_MASK MMIO_DEVTAB_BASE_MASK
> +#define MMIO_CMDBUF_DEFAULT_SIZE 8
> +#define MMIO_CMDBUF_HEAD_MASK (((1ULL << 19) - 1) & ~0x0F)
> +#define MMIO_CMDBUF_TAIL_MASK MMIO_EVTLOG_HEAD_MASK
> +
> +#define MMIO_EVTLOG_SIZE_BYTE (MMIO_EVENT_BASE + 7)
> +#define MMIO_EVTLOG_SIZE_MASK MMIO_CMDBUF_SIZE_MASK
> +#define MMIO_EVTLOG_BASE_MASK MMIO_CMDBUF_BASE_MASK
> +#define MMIO_EVTLOG_DEFAULT_SIZE MMIO_CMDBUF_DEFAULT_SIZE
> +#define MMIO_EVTLOG_HEAD_MASK (((1ULL << 19) - 1) & ~0x0F)
> +#define MMIO_EVTLOG_TAIL_MASK MMIO_EVTLOG_HEAD_MASK
> +
> +#define MMIO_EXCL_BASE_MASK MMIO_DEVTAB_BASE_MASK
> +#define MMIO_EXCL_ENABLED_MASK (1ULL << 0)
> +#define MMIO_EXCL_ALLOW_MASK (1ULL << 1)
> +#define MMIO_EXCL_LIMIT_MASK MMIO_DEVTAB_BASE_MASK
> +#define MMIO_EXCL_LIMIT_LOW 0xFFF
> +
> +#define MMIO_CONTROL_IOMMUEN (1ULL << 0)
> +#define MMIO_CONTROL_HTTUNEN (1ULL << 1)
> +#define MMIO_CONTROL_EVENTLOGEN (1ULL << 2)
> +#define MMIO_CONTROL_EVENTINTEN (1ULL << 3)
> +#define MMIO_CONTROL_COMWAITINTEN (1ULL << 4)
> +#define MMIO_CONTROL_CMDBUFEN (1ULL << 12)
> +
> +#define MMIO_STATUS_EVTLOG_OF (1ULL << 0)
> +#define MMIO_STATUS_EVTLOG_INTR (1ULL << 1)
> +#define MMIO_STATUS_COMWAIT_INTR (1ULL << 2)
> +#define MMIO_STATUS_EVTLOG_RUN (1ULL << 3)
> +#define MMIO_STATUS_CMDBUF_RUN (1ULL << 4)
> +
> +#define CMDBUF_ID_BYTE 0x07
> +#define CMDBUF_ID_RSHIFT 4
> +#define CMDBUF_ENTRY_SIZE 0x10
> +
> +#define CMD_COMPLETION_WAIT 0x01
> +#define CMD_INVAL_DEVTAB_ENTRY 0x02
> +#define CMD_INVAL_IOMMU_PAGES 0x03
> +#define CMD_INVAL_IOTLB_PAGES 0x04
> +#define CMD_INVAL_INTR_TABLE 0x05
> +
> +#define DEVTAB_ENTRY_SIZE 32
> +
> +/* Device table entry bits 0:63 */
> +#define DEV_VALID (1ULL << 0)
> +#define DEV_TRANSLATION_VALID (1ULL << 1)
> +#define DEV_MODE_MASK 0x7
> +#define DEV_MODE_RSHIFT 9
> +#define DEV_PT_ROOT_MASK 0xFFFFFFFFFF000
> +#define DEV_PT_ROOT_RSHIFT 12
> +#define DEV_PERM_SHIFT 61
> +#define DEV_PERM_READ (1ULL << 61)
> +#define DEV_PERM_WRITE (1ULL << 62)
> +
> +/* Device table entry bits 64:127 */
> +#define DEV_DOMAIN_ID_MASK ((1ULL << 16) - 1)
> +#define DEV_IOTLB_SUPPORT (1ULL << 17)
> +#define DEV_SUPPRESS_PF (1ULL << 18)
> +#define DEV_SUPPRESS_ALL_PF (1ULL << 19)
> +#define DEV_IOCTL_MASK ~3
> +#define DEV_IOCTL_RSHIFT 20
> +#define DEV_IOCTL_DENY 0
> +#define DEV_IOCTL_PASSTHROUGH 1
> +#define DEV_IOCTL_TRANSLATE 2
> +#define DEV_CACHE (1ULL << 37)
> +#define DEV_SNOOP_DISABLE (1ULL << 38)
> +#define DEV_EXCL (1ULL << 39)
> +
> +/* Event codes and flags, as stored in the info field */
> +#define EVENT_ILLEGAL_DEVTAB_ENTRY (0x1U << 24)
> +#define EVENT_IOPF (0x2U << 24)
> +#define EVENT_IOPF_I (1U << 3)
> +#define EVENT_IOPF_PR (1U << 4)
> +#define EVENT_IOPF_RW (1U << 5)
> +#define EVENT_IOPF_PE (1U << 6)
> +#define EVENT_IOPF_RZ (1U << 7)
> +#define EVENT_IOPF_TR (1U << 8)
> +#define EVENT_DEV_TAB_HW_ERROR (0x3U << 24)
> +#define EVENT_PAGE_TAB_HW_ERROR (0x4U << 24)
> +#define EVENT_ILLEGAL_COMMAND_ERROR (0x5U << 24)
> +#define EVENT_COMMAND_HW_ERROR (0x6U << 24)
> +#define EVENT_IOTLB_INV_TIMEOUT (0x7U << 24)
> +#define EVENT_INVALID_DEV_REQUEST (0x8U << 24)
> +
> +#define EVENT_LEN 16
> +
> +#define IOMMU_PERM_READ (1 << 0)
> +#define IOMMU_PERM_WRITE (1 << 1)
> +#define IOMMU_PERM_RW (IOMMU_PERM_READ | IOMMU_PERM_WRITE)
> +
> +typedef struct AMDIOMMUState {
> + PCIDevice dev;
> +
> + int capab_offset;
> + unsigned char *capab;
> +
> + int mmio_index;
> + target_phys_addr_t mmio_addr;
> + unsigned char *mmio_buf;
> + int mmio_enabled;
> +
> + int enabled;
> + int ats_enabled;
> +
> + target_phys_addr_t devtab;
> + size_t devtab_len;
> +
> + target_phys_addr_t cmdbuf;
> + int cmdbuf_enabled;
> + size_t cmdbuf_len;
> + size_t cmdbuf_head;
> + size_t cmdbuf_tail;
> + int completion_wait_intr;
> +
> + target_phys_addr_t evtlog;
> + int evtlog_enabled;
> + int evtlog_intr;
> + target_phys_addr_t evtlog_len;
> + target_phys_addr_t evtlog_head;
> + target_phys_addr_t evtlog_tail;
> +
> + target_phys_addr_t excl_base;
> + target_phys_addr_t excl_limit;
> + int excl_enabled;
> + int excl_allow;
> +} AMDIOMMUState;
> +
> +typedef struct AMDIOMMUEvent {
> + uint16_t devfn;
> + uint16_t reserved;
> + uint16_t domid;
> + uint16_t info;
> + uint64_t addr;
> +} __attribute__((packed)) AMDIOMMUEvent;
> +
> +static void amd_iommu_completion_wait(AMDIOMMUState *st,
> + uint8_t *cmd)
> +{
> + uint64_t addr;
> +
> + if (cmd[0] & 1) {
> + addr = le64_to_cpu(*(uint64_t *) cmd) & 0xFFFFFFFFFFFF8;
> + cpu_physical_memory_write(addr, cmd + 8, 8);
> + }
> +
> + if (cmd[0] & 2)
> + st->mmio_buf[MMIO_STATUS] |= MMIO_STATUS_COMWAIT_INTR;
> +}
> +
> +static void amd_iommu_invalidate_iotlb(AMDIOMMUState *st,
> + uint8_t *cmd)
> +{
> + PCIDevice *dev;
> + PCIBus *bus = st->dev.bus;
> + int bus_num = pci_bus_num(bus);
> + int devfn = *(uint16_t *) cmd;
> +
> + dev = pci_find_device(bus, bus_num, devfn);
> + if (dev) {
> + dma_invalidate_memory_range(&dev->dma, 0, -1);
> + }
> +}
> +
> +static void amd_iommu_cmdbuf_exec(AMDIOMMUState *st)
> +{
> + uint8_t cmd[16];
> + int type;
> +
> + cpu_physical_memory_read(st->cmdbuf + st->cmdbuf_head, cmd, 16);
> + type = cmd[CMDBUF_ID_BYTE] >> CMDBUF_ID_RSHIFT;
> + switch (type) {
> + case CMD_COMPLETION_WAIT:
> + amd_iommu_completion_wait(st, cmd);
> + break;
> + case CMD_INVAL_DEVTAB_ENTRY:
> + break;
> + case CMD_INVAL_IOMMU_PAGES:
> + break;
> + case CMD_INVAL_IOTLB_PAGES:
> + amd_iommu_invalidate_iotlb(st, cmd);
> + break;
> + case CMD_INVAL_INTR_TABLE:
> + break;
> + default:
> + break;
> + }
> +}
> +
> +static void amd_iommu_cmdbuf_run(AMDIOMMUState *st)
> +{
> + if (!st->cmdbuf_enabled) {
> + return;
> + }
> +
> + /* Check if there's work to do. */
> + while (st->cmdbuf_head != st->cmdbuf_tail) {
> + /* Wrap head pointer. */
> + if (st->cmdbuf_head >= st->cmdbuf_len * CMDBUF_ENTRY_SIZE) {
> + st->cmdbuf_head = 0;
> + }
> +
> + amd_iommu_cmdbuf_exec(st);
> +
> + /* Increment head pointer. */
> + st->cmdbuf_head += CMDBUF_ENTRY_SIZE;
> + }
> +
> + *((uint64_t *) (st->mmio_buf + MMIO_COMMAND_HEAD)) = cpu_to_le64(st->cmdbuf_head);
> +}
> +
> +static uint32_t amd_iommu_mmio_buf_read(AMDIOMMUState *st,
> + size_t offset,
> + size_t size)
> +{
> + ssize_t i;
> + uint32_t ret;
> +
> + if (!size) {
> + return 0;
> + }
> +
> + ret = st->mmio_buf[offset + size - 1];
> + for (i = size - 2; i >= 0; i--) {
> + ret <<= 8;
> + ret |= st->mmio_buf[offset + i];
> + }
> +
> + return ret;
> +}
> +
> +static void amd_iommu_mmio_buf_write(AMDIOMMUState *st,
> + size_t offset,
> + size_t size,
> + uint32_t val)
> +{
> + size_t i;
> +
> + for (i = 0; i < size; i++) {
> + st->mmio_buf[offset + i] = val & 0xFF;
> + val >>= 8;
> + }
> +}
The above seem to do something like LE/BE conversion?
If yes, it's better to use the appropriate macros for this.
To support unaligned access, memcpy the data to an
aligned buffer.
> +
> +static void amd_iommu_update_mmio(AMDIOMMUState *st,
> + target_phys_addr_t addr)
> +{
> + size_t reg = addr & ~0x07;
> + uint64_t *base = (uint64_t *) &st->mmio_buf[reg];
> + uint64_t val = le64_to_cpu(*base);
> +
> + switch (reg) {
> + case MMIO_CONTROL:
> + st->enabled = !!(val & MMIO_CONTROL_IOMMUEN);
> + st->ats_enabled = !!(val & MMIO_CONTROL_HTTUNEN);
> + st->evtlog_enabled = st->enabled &&
> + !!(val & MMIO_CONTROL_EVENTLOGEN);
> + st->evtlog_intr = !!(val & MMIO_CONTROL_EVENTINTEN);
> + st->completion_wait_intr = !!(val & MMIO_CONTROL_COMWAITINTEN);
> + st->cmdbuf_enabled = st->enabled &&
> + !!(val & MMIO_CONTROL_CMDBUFEN);
> +
> + /* Update status flags depending on the control register. */
> + if (st->cmdbuf_enabled) {
> + st->mmio_buf[MMIO_STATUS] |= MMIO_STATUS_CMDBUF_RUN;
> + } else {
> + st->mmio_buf[MMIO_STATUS] &= ~MMIO_STATUS_CMDBUF_RUN;
> + }
> + if (st->evtlog_enabled) {
> + st->mmio_buf[MMIO_STATUS] |= MMIO_STATUS_EVTLOG_RUN;
> + } else {
> + st->mmio_buf[MMIO_STATUS] &= ~MMIO_STATUS_EVTLOG_RUN;
> + }
> +
> + amd_iommu_cmdbuf_run(st);
> + break;
> + case MMIO_DEVICE_TABLE:
> + st->devtab = (target_phys_addr_t) (val & MMIO_DEVTAB_BASE_MASK);
> + st->devtab_len = ((val & MMIO_DEVTAB_SIZE_MASK) + 1) *
> + (MMIO_DEVTAB_SIZE_UNIT / MMIO_DEVTAB_ENTRY_SIZE);
> + break;
> + case MMIO_COMMAND_BASE:
> + st->cmdbuf = (target_phys_addr_t) (val & MMIO_CMDBUF_BASE_MASK);
> + st->cmdbuf_len = 1UL << (st->mmio_buf[MMIO_CMDBUF_SIZE_BYTE] &
> + MMIO_CMDBUF_SIZE_MASK);
> +
> + /* We must reset the head and tail pointers. */
> + st->cmdbuf_head = st->cmdbuf_tail = 0;
> + memset(st->mmio_buf + MMIO_COMMAND_HEAD, 0, 8);
> + memset(st->mmio_buf + MMIO_COMMAND_TAIL, 0, 8);
> + break;
> + case MMIO_COMMAND_HEAD:
> + st->cmdbuf_head = val & MMIO_CMDBUF_HEAD_MASK;
> + amd_iommu_cmdbuf_run(st);
> + break;
> + case MMIO_COMMAND_TAIL:
> + st->cmdbuf_tail = val & MMIO_CMDBUF_TAIL_MASK;
> + amd_iommu_cmdbuf_run(st);
> + break;
> + case MMIO_EVENT_BASE:
> + st->evtlog = (target_phys_addr_t) (val & MMIO_EVTLOG_BASE_MASK);
> + st->evtlog_len = 1UL << (st->mmio_buf[MMIO_EVTLOG_SIZE_BYTE] &
> + MMIO_EVTLOG_SIZE_MASK);
> + break;
> + case MMIO_EVENT_HEAD:
> + st->evtlog_head = val & MMIO_EVTLOG_HEAD_MASK;
> + break;
> + case MMIO_EVENT_TAIL:
> + st->evtlog_tail = val & MMIO_EVTLOG_TAIL_MASK;
> + break;
> + case MMIO_EXCL_BASE:
> + st->excl_base = (target_phys_addr_t) (val & MMIO_EXCL_BASE_MASK);
> + st->excl_enabled = val & MMIO_EXCL_ENABLED_MASK;
> + st->excl_allow = val & MMIO_EXCL_ALLOW_MASK;
> + break;
> + case MMIO_EXCL_LIMIT:
> + st->excl_limit = (target_phys_addr_t) ((val & MMIO_EXCL_LIMIT_MASK) |
> + MMIO_EXCL_LIMIT_LOW);
> + break;
> + default:
> + break;
> + }
> +}
> +
> +static uint32_t amd_iommu_mmio_readb(void *opaque, target_phys_addr_t addr)
> +{
> + AMDIOMMUState *st = opaque;
> +
> + return amd_iommu_mmio_buf_read(st, addr, 1);
> +}
> +
> +static uint32_t amd_iommu_mmio_readw(void *opaque, target_phys_addr_t addr)
> +{
> + AMDIOMMUState *st = opaque;
> +
> + return amd_iommu_mmio_buf_read(st, addr, 2);
> +}
> +
> +static uint32_t amd_iommu_mmio_readl(void *opaque, target_phys_addr_t addr)
> +{
> + AMDIOMMUState *st = opaque;
> +
> + return amd_iommu_mmio_buf_read(st, addr, 4);
> +}
> +
> +static void amd_iommu_mmio_writeb(void *opaque,
> + target_phys_addr_t addr,
> + uint32_t val)
> +{
> + AMDIOMMUState *st = opaque;
> +
> + amd_iommu_mmio_buf_write(st, addr, 1, val);
> + amd_iommu_update_mmio(st, addr);
> +}
> +
> +static void amd_iommu_mmio_writew(void *opaque,
> + target_phys_addr_t addr,
> + uint32_t val)
> +{
> + AMDIOMMUState *st = opaque;
> +
> + amd_iommu_mmio_buf_write(st, addr, 2, val);
> + amd_iommu_update_mmio(st, addr);
> +}
> +
> +static void amd_iommu_mmio_writel(void *opaque,
> + target_phys_addr_t addr,
> + uint32_t val)
> +{
> + AMDIOMMUState *st = opaque;
> +
> + amd_iommu_mmio_buf_write(st, addr, 4, val);
> + amd_iommu_update_mmio(st, addr);
> +}
> +
> +static CPUReadMemoryFunc * const amd_iommu_mmio_read[] = {
> + amd_iommu_mmio_readb,
> + amd_iommu_mmio_readw,
> + amd_iommu_mmio_readl,
> +};
> +
> +static CPUWriteMemoryFunc * const amd_iommu_mmio_write[] = {
> + amd_iommu_mmio_writeb,
> + amd_iommu_mmio_writew,
> + amd_iommu_mmio_writel,
> +};
> +
> +static void amd_iommu_enable_mmio(AMDIOMMUState *st)
> +{
> + target_phys_addr_t addr;
> + uint8_t *capab_wmask = st->dev.wmask + st->capab_offset;
> +
> + st->mmio_index = cpu_register_io_memory(amd_iommu_mmio_read,
> + amd_iommu_mmio_write,
> + st, DEVICE_LITTLE_ENDIAN);
> + if (st->mmio_index < 0) {
> + return;
> + }
> +
> + addr = le64_to_cpu(*(uint64_t *) &st->capab[CAPAB_BAR_LOW]) & CAPAB_BAR_MASK;
remove space before &.
> + cpu_register_physical_memory(addr, MMIO_SIZE, st->mmio_index);
> +
> + st->mmio_addr = addr;
> + st->mmio_enabled = 1;
> +
> + /* Further changes to the capability are prohibited. */
> + memset(capab_wmask + CAPAB_BAR_LOW, 0x00, CAPAB_REG_SIZE);
> + memset(capab_wmask + CAPAB_BAR_HIGH, 0x00, CAPAB_REG_SIZE);
> +}
> +
> +static void amd_iommu_write_capab(PCIDevice *dev,
> + uint32_t addr, uint32_t val, int len)
> +{
> + AMDIOMMUState *st = DO_UPCAST(AMDIOMMUState, dev, dev);
> +
> + pci_default_write_config(dev, addr, val, len);
> +
> + if (!st->mmio_enabled && st->capab[CAPAB_BAR_LOW] & 0x1) {
> + amd_iommu_enable_mmio(st);
> + }
> +}
> +
> +static void amd_iommu_reset(DeviceState *dev)
> +{
> + AMDIOMMUState *st = DO_UPCAST(AMDIOMMUState, dev.qdev, dev);
> + unsigned char *capab = st->capab;
> + uint8_t *capab_wmask = st->dev.wmask + st->capab_offset;
> +
> + st->enabled = 0;
> + st->ats_enabled = 0;
> + st->mmio_enabled = 0;
> +
> + capab[CAPAB_REV_TYPE] = CAPAB_REV_TYPE;
> + capab[CAPAB_FLAGS] = CAPAB_FLAGS;
> + capab[CAPAB_BAR_LOW] = 0;
> + capab[CAPAB_BAR_HIGH] = 0;
> + capab[CAPAB_RANGE] = 0;
> + *((uint32_t *) &capab[CAPAB_MISC]) = cpu_to_le32(CAPAB_INIT_MISC);
> +
> + /* Changes to the capability are allowed after system reset. */
> + memset(capab_wmask + CAPAB_BAR_LOW, 0xFF, CAPAB_REG_SIZE);
> + memset(capab_wmask + CAPAB_BAR_HIGH, 0xFF, CAPAB_REG_SIZE);
> +
> + memset(st->mmio_buf, 0, MMIO_SIZE);
> + st->mmio_buf[MMIO_CMDBUF_SIZE_BYTE] = MMIO_CMDBUF_DEFAULT_SIZE;
> + st->mmio_buf[MMIO_EVTLOG_SIZE_BYTE] = MMIO_EVTLOG_DEFAULT_SIZE;
> +}
> +
> +static void amd_iommu_log_event(AMDIOMMUState *st, AMDIOMMUEvent *evt)
> +{
> + if (!st->evtlog_enabled ||
> + (st->mmio_buf[MMIO_STATUS] | MMIO_STATUS_EVTLOG_OF)) {
> + return;
> + }
> +
> + if (st->evtlog_tail >= st->evtlog_len) {
> + st->mmio_buf[MMIO_STATUS] |= MMIO_STATUS_EVTLOG_OF;
> + }
> +
> + cpu_physical_memory_write(st->evtlog + st->evtlog_tail,
> + (uint8_t *) evt, EVENT_LEN);
> +
> + st->evtlog_tail += EVENT_LEN;
> + st->mmio_buf[MMIO_STATUS] |= MMIO_STATUS_EVTLOG_INTR;
> +}
> +
> +static void amd_iommu_page_fault(AMDIOMMUState *st,
> + int devfn,
> + unsigned domid,
> + target_phys_addr_t addr,
> + int present,
> + int is_write)
> +{
> + AMDIOMMUEvent evt;
> + unsigned info;
> +
> + evt.devfn = cpu_to_le16(devfn);
> + evt.reserved = 0;
> + evt.domid = cpu_to_le16(domid);
> + evt.addr = cpu_to_le64(addr);
> +
> + info = EVENT_IOPF;
> + if (present) {
> + info |= EVENT_IOPF_PR;
> + }
> + if (is_write) {
> + info |= EVENT_IOPF_RW;
> + }
> + evt.info = cpu_to_le16(info);
> +
> + amd_iommu_log_event(st, &evt);
> +}
> +
> +static inline uint64_t amd_iommu_get_perms(uint64_t entry)
> +{
> + return (entry & (DEV_PERM_READ | DEV_PERM_WRITE)) >> DEV_PERM_SHIFT;
> +}
> +
> +static inline AMDIOMMUState *amd_iommu_dma_to_state(DMADevice *dev)
> +{
> + PCIDevice *pci_dev = DO_UPCAST(PCIDevice, qdev, dev->mmu->iommu);
> +
> + return DO_UPCAST(AMDIOMMUState, dev, pci_dev);
> +}
> +
> +static int amd_iommu_translate(DMADevice *dev,
> + dma_addr_t addr,
> + dma_addr_t *paddr,
> + dma_addr_t *len,
> + int is_write)
> +{
> + PCIDevice *pci_dev = container_of(dev, PCIDevice, dma);
> + PCIDevice *iommu_dev = DO_UPCAST(PCIDevice, qdev, dev->mmu->iommu);
> + AMDIOMMUState *st = DO_UPCAST(AMDIOMMUState, dev, iommu_dev);
> + int devfn, present;
> + target_phys_addr_t entry_addr, pte_addr;
> + uint64_t entry[4], pte, page_offset, pte_perms;
> + unsigned level, domid;
> + unsigned perms;
> +
> + if (!st->enabled) {
> + goto no_translation;
> + }
> +
> + /*
> + * It's okay to check for either read or write permissions
> + * even for memory maps, since we don't support R/W maps.
> + */
> + perms = is_write ? IOMMU_PERM_WRITE : IOMMU_PERM_READ;
> +
> + /* Get device table entry. */
> + devfn = pci_dev->devfn;
> + entry_addr = st->devtab + devfn * DEVTAB_ENTRY_SIZE;
> + cpu_physical_memory_read(entry_addr, (uint8_t *) entry, 32);
> +
> + pte = entry[0];
> + if (!(pte & DEV_VALID) || !(pte & DEV_TRANSLATION_VALID)) {
> + goto no_translation;
> + }
> + domid = entry[1] & DEV_DOMAIN_ID_MASK;
> + level = (pte >> DEV_MODE_RSHIFT) & DEV_MODE_MASK;
> + while (level > 0) {
> + /*
> + * Check permissions: the bitwise
> + * implication perms -> entry_perms must be true.
> + */
> + pte_perms = amd_iommu_get_perms(pte);
> + present = pte & 1;
> + if (!present || perms != (perms & pte_perms)) {
> + amd_iommu_page_fault(st, devfn, domid, addr,
> + present, !!(perms & IOMMU_PERM_WRITE));
> + return -EPERM;
> + }
> +
> + /* Go to the next lower level. */
> + pte_addr = pte & DEV_PT_ROOT_MASK;
> + pte_addr += ((addr >> (3 + 9 * level)) & 0x1FF) << 3;
> + pte = ldq_phys(pte_addr);
> + level = (pte >> DEV_MODE_RSHIFT) & DEV_MODE_MASK;
> + }
> + page_offset = addr & 4095;
> + *paddr = (pte & DEV_PT_ROOT_MASK) + page_offset;
> + *len = 4096 - page_offset;
> +
> + return 0;
> +
> +no_translation:
> + *paddr = addr;
> + *len = -1;
> + return 0;
The spec seems to specify that errors
should signal target abort (in pci config space).
Is the iommu ever an extress device?
if yes this could interact with AER.
> +}
> +
> +static int amd_iommu_pci_initfn(PCIDevice *dev)
> +{
> + AMDIOMMUState *st = DO_UPCAST(AMDIOMMUState, dev, dev);
> +
> + pci_config_set_vendor_id(st->dev.config, PCI_VENDOR_ID_AMD);
> + pci_config_set_device_id(st->dev.config, PCI_DEVICE_ID_AMD_IOMMU);
> + pci_config_set_class(st->dev.config, PCI_CLASS_SYSTEM_IOMMU);
> +
> + /* Secure Device capability */
> + st->capab_offset = pci_add_capability(&st->dev,
> + PCI_CAP_ID_SEC, 0, CAPAB_SIZE);
> + st->capab = st->dev.config + st->capab_offset;
> + dev->config_write = amd_iommu_write_capab;
> +
> + /* Allocate backing space for the MMIO registers. */
> + st->mmio_buf = qemu_malloc(MMIO_SIZE);
> +
> + pci_register_iommu(dev, amd_iommu_translate);
> +
> + return 0;
> +}
> +
> +static const VMStateDescription vmstate_amd_iommu = {
> + .name = "amd-iommu",
> + .version_id = 1,
> + .minimum_version_id = 1,
> + .minimum_version_id_old = 1,
> + .fields = (VMStateField []) {
> + VMSTATE_PCI_DEVICE(dev, AMDIOMMUState),
> + VMSTATE_END_OF_LIST()
> + }
> +};
> +
> +static PCIDeviceInfo amd_iommu_pci_info = {
> + .qdev.name = "amd-iommu",
> + .qdev.desc = "AMD IOMMU",
> + .qdev.size = sizeof(AMDIOMMUState),
> + .qdev.reset = amd_iommu_reset,
> + .qdev.vmsd = &vmstate_amd_iommu,
> + .init = amd_iommu_pci_initfn,
> +};
> +
> +static void amd_iommu_register(void)
> +{
> + pci_qdev_register(&amd_iommu_pci_info);
> +}
> +
> +device_init(amd_iommu_register);
> diff --git a/hw/pc.c b/hw/pc.c
> index fface7d..9f51e95 100644
> --- a/hw/pc.c
> +++ b/hw/pc.c
> @@ -1163,6 +1163,8 @@ void pc_pci_device_init(PCIBus *pci_bus)
> int max_bus;
> int bus;
>
> + pci_create_simple(pci_bus, -1, "amd-iommu");
> +
> max_bus = drive_get_max_bus(IF_SCSI);
> for (bus = 0; bus <= max_bus; bus++) {
> pci_create_simple(pci_bus, -1, "lsi53c895a");
Will this affect the default pc?
If yes it's probably not a good idea.
I'd prefer new devices used qdev exclusively.
> diff --git a/hw/pci_ids.h b/hw/pci_ids.h
> index ea3418c..5dbe281 100644
> --- a/hw/pci_ids.h
> +++ b/hw/pci_ids.h
> @@ -27,6 +27,7 @@
>
> #define PCI_CLASS_MEMORY_RAM 0x0500
>
> +#define PCI_CLASS_SYSTEM_IOMMU 0x0806
> #define PCI_CLASS_SYSTEM_OTHER 0x0880
>
> #define PCI_CLASS_SERIAL_USB 0x0c03
> @@ -57,6 +58,7 @@
>
> #define PCI_VENDOR_ID_AMD 0x1022
> #define PCI_DEVICE_ID_AMD_LANCE 0x2000
> +#define PCI_DEVICE_ID_AMD_IOMMU 0x0000 /* FIXME */
ID 0?
Also, out ids file is a copy from linux. Add the id there
or put it in your .c file.
>
> #define PCI_VENDOR_ID_TI 0x104c
>
> diff --git a/hw/pci_regs.h b/hw/pci_regs.h
> index dd0bed4..3d098aa 100644
> --- a/hw/pci_regs.h
> +++ b/hw/pci_regs.h
> @@ -209,6 +209,7 @@
> #define PCI_CAP_ID_SHPC 0x0C /* PCI Standard Hot-Plug Controller */
> #define PCI_CAP_ID_SSVID 0x0D /* Bridge subsystem vendor/device ID */
> #define PCI_CAP_ID_AGP3 0x0E /* AGP Target PCI-PCI bridge */
> +#define PCI_CAP_ID_SEC 0x0F /* Secure Device (AMD IOMMU) */
Our pci_regs.h is a copy of the one in linux. Either add the id there,
or put it in your .c file.
> #define PCI_CAP_ID_EXP 0x10 /* PCI Express */
> #define PCI_CAP_ID_MSIX 0x11 /* MSI-X */
> #define PCI_CAP_ID_AF 0x13 /* PCI Advanced Features */
> --
> 1.7.3.4
