From: David <davidkiar...@gmail.com> Introduce basic AMD IOMMU emulation in Qemu. IOMMU implements event logging and host translation which should allow nested PCI passthrough.It also implemented a very basic IOTLB implementation
Signed-off-by: David Kiarie <davidkiar...@gmail.com> --- hw/i386/Makefile.objs | 1 + hw/i386/amd_iommu.c | 1266 +++++++++++++++++++++++++++++++++++++++++++++++++ hw/i386/amd_iommu.h | 363 ++++++++++++++ 3 files changed, 1630 insertions(+) create mode 100644 hw/i386/amd_iommu.c create mode 100644 hw/i386/amd_iommu.h diff --git a/hw/i386/Makefile.objs b/hw/i386/Makefile.objs index c250deb..85fa90f 100644 --- a/hw/i386/Makefile.objs +++ b/hw/i386/Makefile.objs @@ -3,6 +3,7 @@ obj-y += multiboot.o obj-y += pc.o pc_piix.o pc_q35.o obj-y += pc_sysfw.o obj-y += intel_iommu.o +obj-y += amd_iommu.o obj-$(CONFIG_XEN) += ../xenpv/ xen/ obj-y += kvmvapic.o diff --git a/hw/i386/amd_iommu.c b/hw/i386/amd_iommu.c new file mode 100644 index 0000000..092f64e --- /dev/null +++ b/hw/i386/amd_iommu.c @@ -0,0 +1,1266 @@ +/* + * QEMU emulation of AMD IOMMU (AMD-Vi) + * + * Copyright (C) 2011 Eduard - Gabriel Munteanu + * Copyright (C) 2015 David Kiarie, <davidkiar...@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/>. + * + * Cache implementation inspired by hw/i386/intel_iommu.c + * + */ +#include "hw/i386/amd_iommu.h" + +#define DEBUG_AMD_IOMMU +#ifdef DEBUG_AMD_IOMMU +enum { + DEBUG_GENERAL, DEBUG_CAPAB, DEBUG_MMIO, DEBUG_ELOG, + DEBUG_CACHE, DEBUG_COMMAND, DEBUG_MMU +}; + +#define IOMMU_DBGBIT(x) (1 << DEBUG_##x) +static int iommu_dbgflags = IOMMU_DBGBIT(GENERAL) | IOMMU_DBGBIT(MMU) | IOMMU_DBGBIT(MMIO); +//| IOMMU_DBGBIT(CAPAB) | IOMMU_DBGBIT(ELOG) | IOMMU_DBGBIT(CACHE) | IOMMU_DBGBIT(COMMAND); + +#define IOMMU_DPRINTF(what, fmt, ...) do { \ + if (iommu_dbgflags & IOMMU_DBGBIT(what)) { \ + fprintf(stderr, "(amd-iommu)%s: " fmt "\n", __func__, \ + ## __VA_ARGS__); } \ + } while (0) +#else +#define IOMMU_DPRINTF(what, fmt, ...) do {} while (0) +#endif + +/* helper functions - FIXME - provide for reading one byte */ +static uint16_t amd_iommu_readw(AMDIOMMUState *s, hwaddr addr) +{ + return lduw_le_p(&s->mmior[addr]); +} + +static uint32_t amd_iommu_readl(AMDIOMMUState *s, hwaddr addr) +{ + return ldl_le_p(&s->mmior[addr]); +} + +static uint64_t amd_iommu_readq(AMDIOMMUState *s, hwaddr addr) +{ + return ldq_le_p(&s->mmior[addr]); +} + +static void amd_iommu_writew(AMDIOMMUState *s, hwaddr addr, uint16_t val) +{ + stw_le_p(&s->mmior[addr], val); +} + +static void amd_iommu_writel(AMDIOMMUState *s, hwaddr addr, uint32_t val) +{ + stl_le_p(&s->mmior[addr], val); +} + +static void amd_iommu_writeq(AMDIOMMUState *s, hwaddr addr, uint64_t val) +{ + stq_le_p(&s->mmior[addr], val); +} + +static void amd_iommu_log_event(AMDIOMMUState *s, uint16_t *evt) +{ + /* event logging not enabled */ + if(!s->evtlog_enabled || *(uint64_t*)&s->mmior[MMIO_STATUS] + | MMIO_STATUS_EVTLOG_OF){ + return; + } + + /* event log buffer full */ + if(s->evtlog_tail >= s->evtlog_len) { + *(uint64_t*)&s->mmior[MMIO_STATUS] |= MMIO_STATUS_EVTLOG_OF; + /* generate interrupt */ + } + + if(dma_memory_write(&address_space_memory, s->evtlog_len + s->evtlog_tail, + &evt, EVENT_LEN)){ + IOMMU_DPRINTF(ELOG, "error: fail to write at address 0x%"PRIx64 + " + offset 0x%"PRIx32, s->evtlog, s->evtlog_tail); + } + + s->evtlog_tail += EVENT_LEN; + *(uint64_t*)&s->mmior[MMIO_STATUS] |= MMIO_STATUS_EVTLOG_INTR; +} + +/* log an error encountered page-walking + * + * @addr: virtual address in translation request + */ +static void amd_iommu_page_fault(AMDIOMMUState *s, uint16_t devid, + dma_addr_t addr, uint8_t info) +{ + IOMMU_DPRINTF(ELOG, ""); + + uint16_t evt[8]; + uint8_t status; + + info |= EVENT_IOPF_I; + + /* encode information */ + *(uint16_t*)&evt[0] = devid; + *(uint16_t*)&evt[3] = info; + *(uint64_t*)&evt[4] = cpu_to_be64(addr); + + /* log a page fault */ + amd_iommu_log_event(s, evt); + + /* Abort the translation */ + status = pci_get_word(s->dev.config + PCI_STATUS); + pci_set_word(s->dev.config + PCI_STATUS, + status | PCI_STATUS_SIG_TARGET_ABORT); +} +/* + * log a master abort accessing device table + * @devtab : address of device table entry + * @info : error flags + */ +static void amd_iommu_log_devtab_error(AMDIOMMUState *s, uint16_t devid, + dma_addr_t devtab, uint8_t info) +{ + + IOMMU_DPRINTF(ELOG, ""); + + uint16_t evt[8]; + uint8_t status; + + info |= EVENT_DEV_TAB_HW_ERROR; + + /* encode information */ + *(uint16_t*)&evt[0] = devid; + *(uint8_t*)&evt[3] = info; + *(uint64_t*)&evt[4] = cpu_to_be64(devtab); + + amd_iommu_log_event(s, evt); + + /* Abort the translation */ + status = pci_get_word(s->dev.config + PCI_STATUS); + pci_set_word(s->dev.config + PCI_STATUS, + status | PCI_STATUS_SIG_TARGET_ABORT); +} + +/* log a master abort encountered during a page-walk + * @addr : address that couldn't be accessed + */ +static void amd_iommu_log_pagetab_error(AMDIOMMUState *s, uint16_t devid, + dma_addr_t addr, uint16_t info) +{ + IOMMU_DPRINTF(ELOG, ""); + + uint16_t evt[8]; + uint8_t status; + + info |= EVENT_PAGE_TAB_HW_ERROR; + + /* encode information */ + *(uint16_t*)&evt[0] = devid; + *(uint8_t*)&evt[3] = info; + *(uint64_t*)&evt[4] = (cpu_to_be64(addr) >> 3); + + amd_iommu_log_event(s, evt); + + /* Abort the translation */ + status = pci_get_word(s->dev.config + PCI_STATUS); + pci_set_word(s->dev.config + PCI_STATUS, + status | PCI_STATUS_SIG_TARGET_ABORT); + +} + +/* log an event trying to access command buffer + * @addr : address that couldn't be accessed + */ +static void amd_iommu_log_command_error(AMDIOMMUState *s, dma_addr_t addr) +{ + IOMMU_DPRINTF(ELOG, ""); + + uint16_t evt[8]; + + /* encode information */ + *(uint8_t*)&evt[3] = (uint8_t)EVENT_COMMAND_HW_ERROR; + *(uint64_t*)&evt[4] = (cpu_to_be64(addr) >> 3); + + amd_iommu_log_event(s, evt); + + /* Abort the translation */ + uint8_t status = pci_get_word(s->dev.config + PCI_STATUS); + pci_set_word(s->dev.config + PCI_STATUS, + status | PCI_STATUS_SIG_TARGET_ABORT); +} + +/* log an illegal comand event + * @addr : address of illegal command + */ +static void amd_iommu_log_illegalcom_error(AMDIOMMUState *s, uint8_t info, + dma_addr_t addr) +{ + IOMMU_DPRINTF(ELOG, ""); + + uint16_t evt[8]; + + /* encode information */ + *(uint8_t*)&evt[3] = (uint8_t)EVENT_ILLEGAL_COMMAND_ERROR; + *(uint64_t*)&evt[4] = (cpu_to_be64(addr) >> 3); + + amd_iommu_log_event(s, evt); +} + +/* log an error accessing device table + * + * @devid : device owning the table entry + * @devtab : address of device table entry + * @info : error flags + */ +static void amd_iommu_log_illegaldevtab_error(AMDIOMMUState *s, uint16_t devid, + dma_addr_t addr, uint16_t info) +{ + IOMMU_DPRINTF(ELOG, ""); + + uint16_t evt[8]; + + info |= EVENT_ILLEGAL_DEVTAB_ENTRY; + + *(uint16_t*)&evt[0] = devid; + *(uint8_t*)&evt[3] = info; + *(uint64_t*)&evt[4] = (cpu_to_be64(addr) >> 3); + + amd_iommu_log_event(s, evt); +} + +static gboolean amd_iommu_uint64_equal(gconstpointer v1, gconstpointer v2) +{ + return *((const uint64_t *)v1) == *((const uint64_t *)v2); +} + +static guint amd_iommu_uint64_hash(gconstpointer v) +{ + return (guint)*(const uint64_t *)v; +} + +static IOMMUIOTLBEntry* amd_iommu_iotlb_lookup(AMDIOMMUState *s, hwaddr addr, + uint64_t devid) +{ + uint64_t key = (addr >> IOMMU_PAGE_SHIFT_4K) | + ((uint64_t)(devid) << IOMMU_DEVID_SHIFT); + + return g_hash_table_lookup(s->iotlb, &key); +} + +static void amd_iommu_iotlb_reset(AMDIOMMUState *s) +{ + g_hash_table_remove_all(s->iotlb); +} + +static gboolean amd_iommu_iotlb_remove_by_devid(gpointer key, gpointer value, + gpointer user_data) +{ + IOMMUIOTLBEntry *entry = (IOMMUIOTLBEntry *)value; + uint16_t devid = *(uint16_t*)user_data; + return entry->devid == devid; +} + +static void amd_iommu_iotlb_remove_page(AMDIOMMUState *s, hwaddr addr, + uint64_t devid) +{ + uint64_t key = (addr >> IOMMU_PAGE_SHIFT_4K) | + ((uint64_t)(devid) << IOMMU_DEVID_SHIFT); + + g_hash_table_remove(s->iotlb, &key); +} + +static void amd_iommu_invalidate_iotlb(AMDIOMMUState *s, uint64_t *cmd) +{ + /* extract device id */ + uint16_t devid = cmd[2] & INVAL_DEV_ID_MASK; + /* if invalidation of more than one page requested */ + if(INVAL_ALL(cmd[3])){ + g_hash_table_foreach_remove(s->iotlb, amd_iommu_iotlb_remove_by_devid, + &devid); + } else { + hwaddr addr = (hwaddr)(cmd[3] & INVAL_ADDR_MASK); + amd_iommu_iotlb_remove_page(s, addr, devid); + } +} + +static void amd_iommu_update_iotlb(AMDIOMMUState *s, uint16_t devid, + uint64_t gpa, uint64_t spa, uint64_t perms) +{ + IOMMUIOTLBEntry *entry = g_malloc(sizeof(*entry)); + uint64_t *key = g_malloc(sizeof(key)); + uint64_t gfn = spa >> IOMMU_PAGE_SHIFT_4K; + + IOMMU_DPRINTF(CACHE, " update iotlb devid: %02x:%02x.%x gpa 0x%"PRIx64 + "hpa 0x%"PRIx64, PCI_BUS_NUM(devid), PCI_SLOT(devid), + PCI_FUNC(devid), gpa, spa); + + entry->gfn = gfn; + entry->perms = perms; + *key = gfn | ((uint64_t)(devid)); + g_hash_table_replace(s->iotlb, key, entry); +} + +/* execute a completion wait command */ +static void amd_iommu_completion_wait(AMDIOMMUState *s, uint8_t *cmd) +{ + unsigned int addr; + + /* completion store */ + if(cmd[0] & COM_COMPLETION_STORE_MASK){ + addr = le64_to_cpu(*(uint64_t*)cmd) & COM_STORE_ADDRESS_MASK; + if(dma_memory_write(&address_space_memory, addr, cmd + 8, 8)){ + IOMMU_DPRINTF(ELOG, "error: fail to write at address 0%x"PRIx64, + addr); + } + + } + + /* set completion interrupt */ + if (cmd[0] & COM_COMPLETION_INTR){ + s->mmior[MMIO_STATUS] |= MMIO_STATUS_COMWAIT_INTR; + } +} + +/* not honouring reserved bits is regarded as an illegal command */ +static void amd_iommu_cmdbuf_exec(AMDIOMMUState *s) +{ + unsigned type; + uint8_t cmd[IOMMU_COMMAND_SIZE]; + + IOMMU_DPRINTF(COMMAND, ""); + memset(cmd, 0, IOMMU_COMMAND_SIZE); + + if(dma_memory_read(&address_space_memory, s->cmdbuf + s->cmdbuf_head, cmd, + IOMMU_COMMAND_SIZE)){ + IOMMU_DPRINTF(COMMAND, "error: fail to access memory at 0x%"PRIx64 + " + %"PRIu8, s->cmdbuf, s->cmdbuf_head); + amd_iommu_log_command_error(s, s->cmdbuf + s->cmdbuf_head); + return; + } + + type = cmd[CMDBUF_ID_BYTE] >> CMDBUF_ID_RSHIFT; + switch(type){ + case CMD_COMPLETION_WAIT: + if(*(uint32_t*)&cmd[1] & COMPLETION_WAIT_RSVD){ + amd_iommu_log_illegalcom_error(s, type, + s->cmdbuf + s->cmdbuf_head); + return; + } + /* pretend to wait for command execution to complete */ + IOMMU_DPRINTF(COMMAND, "completion wait requested"); + amd_iommu_completion_wait(s, cmd); + break; + + case CMD_INVAL_DEVTAB_ENTRY: + /* This command should invalidate internal caches of which there isn't */ + if(*(uint64_t*)&cmd[0] & CMD_INVAL_DEV_RSVD || *(uint64_t*)&cmd[2]){ + amd_iommu_log_illegalcom_error(s, type, + s->cmdbuf + s->cmdbuf_head); + return; + } + + IOMMU_DPRINTF(COMMAND, "IOTLB entries for device --- invalidated"); + break; + + case CMD_INVAL_IOMMU_PAGES: + if(*(uint64_t*)&cmd[0] & INVAL_IOMMU_PAGES_RSVD + || *(uint32_t*)&cmd[2] & 0x00000ff0){ + amd_iommu_log_illegalcom_error(s, type, + s->cmdbuf + s->cmdbuf_head); + return; + } + + IOMMU_DPRINTF(COMMAND, "IOMMU pages invalidated"); + break; + + case CMD_INVAL_IOTLB_PAGES: + if(*(uint32_t*)&cmd[2] & INVAL_IOTLB_PAGES_RSVD){ + amd_iommu_log_illegalcom_error(s, type, + s->cmdbuf + s->cmdbuf_head); + return; + } + + amd_iommu_invalidate_iotlb(s, (uint64_t*)cmd); + IOMMU_DPRINTF(COMMAND, "IOTLB pages invalidated"); + break; + + case CMD_INVAL_INTR_TABLE: + if((*(uint64_t*)&cmd[0] & INVAL_INTR_TABLE_RSVD) + || *(uint64_t*)&cmd[2]){ + amd_iommu_log_illegalcom_error(s, type, + s->cmdbuf + s->cmdbuf_head); + return; + } + + IOMMU_DPRINTF(COMMAND, "interrupt table invalidated"); + break; + + case CMD_PREFETCH_IOMMU_PAGES: + if((*(uint64_t*)&cmd[0] & PRF_IOMMU_PAGES_RSVD) + || (*(uint32_t*)&cmd[3] & 0x00000fd4)){ + amd_iommu_log_illegalcom_error(s, type, + s->cmdbuf + s->cmdbuf_head); + return; + } + + IOMMU_DPRINTF(COMMAND, "Pre-fetch of IOMMU pages requested"); + break; + + case CMD_COMPLETE_PPR_REQUEST: + if((*(uint64_t*)&cmd[0] & COMPLETE_PPR_RQ_RSVD) + || *(uint64_t*)&cmd[3] & 0xffff000000000000){ + amd_iommu_log_illegalcom_error(s, type, + s->cmdbuf + s->cmdbuf_head); + return; + } + + IOMMU_DPRINTF(COMMAND, "Execution of PPR queue requested"); + break; + + case CMD_INVAL_IOMMU_ALL: + if((*(uint64_t*)&cmd[0] & INVAL_IOMMU_ALL_RSVD) + || *(uint64_t*)&cmd[2]){ + amd_iommu_log_illegalcom_error(s, type, + s->cmdbuf + s->cmdbuf_head); + return; + } + + amd_iommu_iotlb_reset(s); + IOMMU_DPRINTF(COMMAND, "Invalidation of all IOMMU cache requested"); + break; + + default: + IOMMU_DPRINTF(COMMAND, "unhandled command %d", type); + /* log illegal command */ + amd_iommu_log_illegalcom_error(s, type, + s->cmdbuf + s->cmdbuf_head); + break; + } + +} + +static void amd_iommu_cmdbuf_run(AMDIOMMUState *s) +{ + IOMMU_DPRINTF(COMMAND, ""); + + uint64_t *mmio_cmdbuf_head = (uint64_t*)s->mmior + MMIO_COMMAND_HEAD; + + if(!s->cmdbuf_enabled){ + IOMMU_DPRINTF(COMMAND, "Command buffer not enabled"); + return; + } + + while(s->cmdbuf_head != s->cmdbuf_tail) { + /* check if there is work to do. */ + IOMMU_DPRINTF(COMMAND, "COMMAND BUFFER head at %x COMMAND BUFFER " + "tail at %x", s->cmdbuf_head, s->cmdbuf_tail); + amd_iommu_cmdbuf_exec(s); + s->cmdbuf_head += CMDBUF_ENTRY_SIZE; + + /* wrap head pointer */ + if (s->cmdbuf_head >= s->cmdbuf_len * CMDBUF_ENTRY_SIZE) { + s->cmdbuf_head = 0; + } + } + + *mmio_cmdbuf_head = cpu_to_le64(s->cmdbuf_head); +} + +/* System Software might never read from some of this fields but anyways */ +static uint64_t amd_iommu_mmio_read(void *opaque, hwaddr addr, unsigned size) +{ + AMDIOMMUState *s = opaque; + + uint64_t val = -1; + if(addr + size > MMIO_SIZE) { + IOMMU_DPRINTF(MMIO, "error: addr outside region: max 0x%"PRIX64 + ", got 0x%"PRIx64 " %d", (uint64_t)MMIO_SIZE, addr, size); + return (uint64_t) - 1; + } + + if(size == 2){ + val = amd_iommu_readw(s, addr); + } else if(size == 4){ + val = amd_iommu_readl(s, addr); + } else if(size == 8){ + val = amd_iommu_readq(s, addr); + } + + switch(addr & ~0x07){ + case MMIO_DEVICE_TABLE: + IOMMU_DPRINTF(MMIO, "MMIO_DEVICE_TABLE read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_COMMAND_BASE: + IOMMU_DPRINTF(MMIO, "MMIO_COMMAND_BASE read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_EVENT_BASE: + IOMMU_DPRINTF(MMIO, "MMIO_EVENT_BASE read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_CONTROL: + IOMMU_DPRINTF(MMIO, "MMIO_CONTROL read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_EXCL_BASE: + IOMMU_DPRINTF(MMIO, "MMIO_EXCL_BASE read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_EXCL_LIMIT: + IOMMU_DPRINTF(MMIO, "MMIO_EXCL_LIMIT read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_COMMAND_HEAD: + IOMMU_DPRINTF(MMIO, "MMIO_COMMAND_HEAD read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_COMMAND_TAIL: + IOMMU_DPRINTF(MMIO, "MMIO_COMMAND_TAIL read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_EVENT_HEAD: + IOMMU_DPRINTF(MMIO, "MMIO_EVENT_HEAD read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_EVENT_TAIL: + IOMMU_DPRINTF(MMIO, "MMIO_EVENT_TAIL read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_STATUS: + IOMMU_DPRINTF(MMIO, "MMIO_STATUS read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + case MMIO_EXT_FEATURES: + IOMMU_DPRINTF(MMIO, "MMIO_EXT_FEATURES read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + break; + + default: + IOMMU_DPRINTF(MMIO, "UNHANDLED MMIO read addr 0x%"PRIx64 + ", size %d offset 0x%"PRIx64, addr, size, + addr & ~0x07); + } + + return val; +} + +static void iommu_handle_control_write(AMDIOMMUState *s) +{ + /* + * read whatever is already written in case + * software is writing in chucks less than 8 bytes + */ + unsigned long control = amd_iommu_readq(s, MMIO_CONTROL); + s->enabled = !!(control & MMIO_CONTROL_IOMMUEN); + + s->ats_enabled = !!(control & MMIO_CONTROL_HTTUNEN); + s->evtlog_enabled = s->enabled && !!(control & MMIO_CONTROL_EVENTLOGEN); + + s->evtlog_intr = !!(control & MMIO_CONTROL_EVENTINTEN); + s->completion_wait_intr = !!(control & MMIO_CONTROL_COMWAITINTEN); + s->cmdbuf_enabled = s->enabled && !!(control & MMIO_CONTROL_CMDBUFLEN); + + /* update the flags depending on the control register */ + if(s->cmdbuf_enabled) { + (*(uint64_t*)&s->mmior[MMIO_STATUS]) |= MMIO_STATUS_CMDBUF_RUN; + } else { + (*(uint64_t*)&s->mmior[MMIO_STATUS]) &= ~MMIO_STATUS_CMDBUF_RUN; + } + if (s->evtlog_enabled) { + (*(uint64_t*)&s->mmior[MMIO_STATUS]) |= MMIO_STATUS_EVTLOG_RUN; + } else { + (*(uint64_t*)&s->mmior[MMIO_STATUS]) &= ~MMIO_STATUS_EVTLOG_RUN; + } + + IOMMU_DPRINTF(MMIO, "MMIO_STATUS state 0x%"PRIx64, control); + + amd_iommu_cmdbuf_run(s); +} + +static inline void iommu_handle_devtab_write(AMDIOMMUState *s) +{ + uint64_t device_base = amd_iommu_readq(s, MMIO_DEVICE_TABLE); + s->devtab = (dma_addr_t)(device_base & MMIO_CMDBUF_BASE_MASK); + + /* set device table length */ + s->devtab_len = ((device_base & MMIO_DEVTAB_SIZE_MASK) + 1 * + (MMIO_DEVTAB_SIZE_UNIT / MMIO_DEVTAB_ENTRY_SIZE)); +} + +static inline void iommu_handle_cmdhead_write(AMDIOMMUState *s) +{ + uint64_t cmdbuf = amd_iommu_readq(s, MMIO_COMMAND_HEAD); + s->cmdbuf_head = cmdbuf & MMIO_CMDBUF_HEAD_MASK; +} + +static inline void iommu_handle_cmdbase_write(AMDIOMMUState *s) +{ + uint64_t command_base = amd_iommu_readq(s, MMIO_COMMAND_BASE); + s->cmdbuf = (dma_addr_t)(command_base & MMIO_CMDBUF_BASE_MASK); + s->cmdbuf_len = 1UL << (s->mmior[MMIO_CMDBUF_SIZE_BYTE] + & MMIO_CMDBUF_SIZE_MASK); + s->cmdbuf_head = s->cmdbuf_tail = 0; +} + +static inline void iommu_handle_cmdtail_write(AMDIOMMUState *s) +{ + uint64_t val = amd_iommu_readq(s, MMIO_COMMAND_TAIL); + s->cmdbuf_tail = val & MMIO_CMDBUF_TAIL_MASK; +} + +static inline void iommu_handle_excllim_write(AMDIOMMUState *s) +{ + uint64_t val = amd_iommu_readq(s, MMIO_EXCL_LIMIT); + s->excl_limit = (val & MMIO_EXCL_LIMIT_MASK ) | MMIO_EXCL_LIMIT_LOW; +} + +static inline void iommu_handle_evtbase_write(AMDIOMMUState *s) +{ + uint64_t val = amd_iommu_readq(s, MMIO_EVENT_BASE); + s->evtlog = val & MMIO_EVTLOG_BASE_MASK; + s->evtlog_len = 1UL << (*(uint64_t*)&s->mmior[MMIO_EVTLOG_SIZE_BYTE] + & MMIO_EVTLOG_SIZE_MASK); +} + +static inline void iommu_handle_evttail_write(AMDIOMMUState *s) +{ + uint64_t val = amd_iommu_readq(s, MMIO_EVENT_TAIL); + s->evtlog_tail = val & MMIO_EVTLOG_TAIL_MASK; +} + +static inline void iommu_handle_evthead_write(AMDIOMMUState *s) +{ + uint64_t val = amd_iommu_readq(s, MMIO_EVENT_HEAD); + s->evtlog_head = val & MMIO_EVTLOG_HEAD_MASK; +} + +static inline void iommu_handle_pprbase_write(AMDIOMMUState *s) +{ + uint64_t val = amd_iommu_readq(s, MMIO_PPR_BASE); + s->ppr_log = val & MMIO_PPRLOG_BASE_MASK; + s->pprlog_len = 1UL << (*(uint64_t*)&s->mmior[MMIO_PPRLOG_SIZE_BYTE] + & MMIO_PPRLOG_SIZE_MASK); +} + +static inline void iommu_handle_pprhead_write(AMDIOMMUState *s) +{ + uint64_t val = amd_iommu_readq(s, MMIO_PPR_HEAD); + s->pprlog_head = val & MMIO_PPRLOG_HEAD_MASK; +} + +static inline void iommu_handle_pprtail_write(AMDIOMMUState *s) +{ + uint64_t val = amd_iommu_readq(s, MMIO_PPR_TAIL); + s->pprlog_tail = val & MMIO_PPRLOG_TAIL_MASK; +} + +/* FIXME: something might go wrong if System Software writes in chunks + * of one byte but linux writes in chunks of 4 bytes so currently it + * works correctly with linux but will definitely be busted if software + * reads/writes 8 bytes + */ +static void amd_iommu_mmio_write(void *opaque, hwaddr addr, uint64_t val, + unsigned size) +{ + + AMDIOMMUState *s = opaque; + unsigned long offset = addr & 0x07; + + if(addr + size > MMIO_SIZE) { + IOMMU_DPRINTF(MMIO, "error: addr outside region: max 0x%"PRIx64 + ", got 0x%"PRIx64 " %d",(uint64_t)MMIO_SIZE, addr, size); + return; + } + + switch(addr & ~0x07){ + case MMIO_CONTROL: + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + + IOMMU_DPRINTF(MMIO, "MMIO_CONTROL write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + iommu_handle_control_write(s); + break; + + case MMIO_DEVICE_TABLE: + IOMMU_DPRINTF(MMIO, "MMIO_DEVICE_TABLE write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + + /* set device table address + * This also suffers from inability to tell whether software + * is done writing + */ + if(offset || (size == 8)){ + iommu_handle_devtab_write(s); + } + break; + + case MMIO_COMMAND_HEAD: + IOMMU_DPRINTF(MMIO, "MMIO_COMMAND_HEAD write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + + iommu_handle_cmdhead_write(s); + amd_iommu_cmdbuf_run(s); + break; + + case MMIO_COMMAND_BASE: + IOMMU_DPRINTF(MMIO, "MMIO_COMMAND_BASE write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + + /* FIXME - make sure System Software has finished writing incase + * it writes in chucks less than 8 bytes in a robust way.As for + * now, this hacks works for the linux driver + */ + if(offset || (size == 8)){ + iommu_handle_cmdbase_write(s); + } + break; + + case MMIO_COMMAND_TAIL: + IOMMU_DPRINTF(MMIO, "MMIO_COMMAND_TAIL write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + iommu_handle_cmdtail_write(s); + break; + + case MMIO_EVENT_BASE: + IOMMU_DPRINTF(MMIO, "MMIO_EVENT_BASE write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + iommu_handle_evtbase_write(s); + break; + + case MMIO_EVENT_HEAD: + IOMMU_DPRINTF(MMIO, "MMIO_EVENT_HEAD write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + iommu_handle_evthead_write(s); + break; + + case MMIO_EVENT_TAIL: + IOMMU_DPRINTF(MMIO, "MMIO_EVENT_TAIL write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + iommu_handle_evttail_write(s); + break; + + case MMIO_EXCL_LIMIT: + IOMMU_DPRINTF(MMIO, "MMIO_EXCL_LIMIT write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + iommu_handle_excllim_write(s); + break; + + /* PPR log base - unused for now */ + case MMIO_PPR_BASE: + IOMMU_DPRINTF(MMIO, "MMIO_PPR_BASE write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + iommu_handle_pprbase_write(s); + break; + /* PPR log head - also unused for now */ + case MMIO_PPR_HEAD: + IOMMU_DPRINTF(MMIO, "MMIO_PPR_HEAD write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + iommu_handle_pprhead_write(s); + break; + /* PPR log tail - unused for now */ + case MMIO_PPR_TAIL: + IOMMU_DPRINTF(MMIO, "MMIO_PPR_TAIL write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + if(size == 2){ + amd_iommu_writew(s, addr, val); + } else if(size == 4){ + amd_iommu_writel(s, addr, val); + } else if(size == 8){ + amd_iommu_writeq(s, addr, val); + } + iommu_handle_pprtail_write(s); + break; + + default: + IOMMU_DPRINTF(MMIO, "UNHANDLED MMIO write addr 0x%"PRIx64 + ", size %d, val 0x%"PRIx64 ", offset 0x%"PRIx64, + addr, size, val, offset); + ; + } + +} + +static inline uint64_t amd_iommu_get_perms(uint64_t entry) +{ + return (entry & (DEV_PERM_READ | DEV_PERM_WRITE )) >> DEV_PERM_SHIFT; +} + +AddressSpace *bridge_host_amd_iommu(PCIBus *bus, void *opaque, int devfn) +{ + AMDIOMMUState *s = opaque; + AMDIOMMUAddressSpace **iommu_as; + int bus_num = pci_bus_num(bus); + + /* just in case */ + assert(0 <= bus_num && bus_num <= PCI_BUS_MAX); + assert(0 <= devfn && devfn <= PCI_DEVFN_MAX); + + iommu_as = s->address_spaces[bus_num]; + + /* allocate memory during the first run */ + if(!iommu_as) { + iommu_as = g_malloc0(sizeof(AMDIOMMUAddressSpace*) * PCI_DEVFN_MAX); + s->address_spaces[bus_num] = iommu_as; + } + + /* set up IOMMU region */ + if(!iommu_as[devfn]){ + iommu_as[devfn] = g_malloc0(sizeof(AMDIOMMUAddressSpace)); + iommu_as[devfn]->bus_num = (uint8_t)bus_num; + iommu_as[devfn]->devfn = (uint8_t)devfn; + iommu_as[devfn]->iommu_state = s; + + memory_region_init_iommu(&iommu_as[devfn]->iommu, OBJECT(s), + &s->iommu_ops, "amd-iommu", UINT64_MAX); + address_space_init(&iommu_as[devfn]->as, &iommu_as[devfn]->iommu, + "amd-iommu"); + } + return &iommu_as[devfn]->as; +} + +/* validate a page table entry */ +static bool amd_iommu_validate_dte(AMDIOMMUState *s, uint16_t devid, uint64_t *dte) +{ + if((dte[0] & DTE_LOWER_QUAD_RESERVED) || (dte[1] & DTE_MIDDLE_QUAD_RESERVED) + || (dte[2] & DTE_UPPER_QUAD_RESERVED) || dte[3]){ + amd_iommu_log_illegaldevtab_error(s, devid, + s->devtab + devid * DEVTAB_ENTRY_SIZE, 0); + return false; + } + + return (dte[0] & DEV_VALID) && (dte[0] & DEV_TRANSLATION_VALID); +} + +/* get a device table entry given the devid */ +static bool amd_iommu_get_dte(AMDIOMMUState *s, int devid, uint64_t *entry) +{ + uint32_t offset = devid * DEVTAB_ENTRY_SIZE; + if(dma_memory_read(&address_space_memory, s->devtab + offset, entry, + DEVTAB_ENTRY_SIZE)){ + IOMMU_DPRINTF(MMU, "error: fail to access Device Entry devtab 0x%"PRIx64 + "offset 0x%"PRIx32, s->devtab, offset); + /* log ever accessing dte */ + amd_iommu_log_devtab_error(s, devid, s->devtab + offset, 0); + return false; + } + + if(!amd_iommu_validate_dte(s, devid, entry)){ + + IOMMU_DPRINTF(MMU, + "Pte entry at 0x%"PRIx64" is invalid", entry[0]); + return false; + } + + return true; +} + +/* get pte translation mode */ +static uint8_t get_pte_translation_mode(uint64_t pte) +{ + return (pte >> DEV_MODE_RSHIFT) & DEV_MODE_MASK; +} + +static int amd_iommu_page_walk(AMDIOMMUAddressSpace *as, uint64_t *dte, + IOMMUTLBEntry *ret, unsigned perms, + hwaddr addr) +{ + unsigned level, present; + uint64_t pte, pte_addr; + uint64_t pte_perms; + pte = dte[0]; + + level = get_pte_translation_mode(pte); + + if(level >= 7 || level == 0) + return -1; + + while(level > 0){ + pte_perms = amd_iommu_get_perms(pte); + present = pte & 1; + if(!present || perms != (perms & pte_perms)){ + amd_iommu_page_fault(as->iommu_state, as->devfn, addr, perms); + IOMMU_DPRINTF(MMU, "error: page fault accessing virtual addr 0x%" + PRIx64, addr); + return -1; + } + + /* go to the next lower level */ + pte_addr = pte & DEV_PT_ROOT_MASK; + pte_addr += ((addr >> ( 9 * level)) & 0xff8); + pte = ldq_phys(&address_space_memory, pte_addr); + level = (pte >> DEV_MODE_RSHIFT) & DEV_MODE_MASK; + } + + ret->iova = addr & IOMMU_PAGE_MASK_4K; + ret->translated_addr = (pte & DEV_PT_ROOT_MASK) & IOMMU_PAGE_MASK_4K; + ret->addr_mask = ~IOMMU_PAGE_MASK_4K; + ret->perm = IOMMU_RW; + return 0; +} + +/* TODO : Mark addresses as Accessed and Dirty */ +static void amd_iommu_do_translate(AMDIOMMUAddressSpace *as, hwaddr addr, + bool is_write, IOMMUTLBEntry *ret) +{ + AMDIOMMUState *s = as->iommu_state; + /* TODO: Get the correct devid instead */ + uint16_t devid = as->devfn; + IOMMUIOTLBEntry *iotlb_entry; + uint8_t err; + uint64_t entry[4]; + + /* try getting a cache entry first */ + iotlb_entry = amd_iommu_iotlb_lookup(s, addr, as->devfn); + + if(iotlb_entry){ + IOMMU_DPRINTF(CACHE, "hit iotlb devid: %02x:%02x.%x gpa 0x%"PRIx64 + "hpa 0x%"PRIx64, PCI_BUS_NUM(devid), PCI_SLOT(devid), + PCI_FUNC(devid), addr, iotlb_entry->hwaddr); + ret->iova = addr & IOMMU_PAGE_MASK_4K; + ret->translated_addr = iotlb_entry->hwaddr; + ret->addr_mask = ~IOMMU_PAGE_MASK_4K; + ret->perm = iotlb_entry->perms; + return; + } else { + if(!amd_iommu_get_dte(s, devid, entry)) + goto out; + + err = amd_iommu_page_walk(as, entry, ret, + is_write ? IOMMU_PERM_WRITE : IOMMU_PERM_READ, + addr); + if(!err){ + IOMMU_DPRINTF(MMU, "error: hardware error accessing page tables" + "while translating addr 0x%"PRIx64, addr); + amd_iommu_log_pagetab_error(s, as->devfn, addr, 0); + goto out; + } + + amd_iommu_update_iotlb(s, as->devfn, addr, ret->translated_addr, + ret->perm); + return; + } + +out: + ret->iova = addr; + ret->translated_addr = addr & IOMMU_PAGE_MASK_4K; + ret->addr_mask = ~IOMMU_PAGE_MASK_4K; + ret->perm = IOMMU_RW; + return; +} + +static IOMMUTLBEntry amd_iommu_translate(MemoryRegion *iommu, hwaddr addr, + bool is_write) +{ + + IOMMU_DPRINTF(GENERAL, ""); + + AMDIOMMUAddressSpace *as = container_of(iommu, AMDIOMMUAddressSpace, iommu); + AMDIOMMUState *s = as->iommu_state; + + IOMMUTLBEntry ret = { + .target_as = &address_space_memory, + .iova = addr, + .translated_addr = 0, + .addr_mask = ~(hwaddr)0, + .perm = IOMMU_NONE, + }; + + if(!s->enabled){ + /* IOMMU disabled - corresponds to iommu=off not + * failure to provide any parameter + */ + ret.iova = addr & IOMMU_PAGE_MASK_4K; + ret.translated_addr = addr & IOMMU_PAGE_MASK_4K; + ret.addr_mask = ~IOMMU_PAGE_MASK_4K; + ret.perm = IOMMU_RW; + return ret; + } + + amd_iommu_do_translate(as, addr, is_write, &ret); + + IOMMU_DPRINTF(MMU, "devid: %02x:%02x.%x gpa 0x%"PRIx64 " hpa 0x%"PRIx64, + as->bus_num, PCI_SLOT(as->devfn), PCI_FUNC(as->devfn), addr, + ret.translated_addr); + + return ret; +} + +static const MemoryRegionOps mmio_mem_ops = { + .read = amd_iommu_mmio_read, + .write = amd_iommu_mmio_write, + .endianness = DEVICE_LITTLE_ENDIAN, + .valid = { + .min_access_size = 1, + .max_access_size = 8, + .unaligned = false, + }, + + .valid = { + .min_access_size = 8, + .max_access_size = 8, + }, +}; + +static void amd_iommu_set_misc_capab(uint32_t host_va, uint32_t host_pa, + uint32_t guest_va, uint32_t *reg) +{ + *reg |= MAX_VA_ADDR | MAX_PH_ADDR | MAX_GVA_ADDR; +} + +static void amd_iommu_init(AMDIOMMUState *s) +{ + uint32_t capab_misc = 0; + + amd_iommu_iotlb_reset(s); + + s->iommu_ops.translate = amd_iommu_translate; + + s->devtab_len = 0; + s->cmdbuf_len = 0; + s->cmdbuf_head = 0; + s->cmdbuf_tail = 0; + s->evtlog_head = 0; + s->evtlog_tail = 0; + s->excl_enabled = false; + s->excl_allow = false; + s->mmio_enabled = false; + s->enabled = false; + s->ats_enabled = false; + s->cmdbuf_enabled = false; + + /* reset MMIO */ + memset(s->mmior, 0, MMIO_SIZE); + amd_iommu_writeq(s, MMIO_EXT_FEATURES, EXT_FEATURES); + amd_iommu_set_misc_capab(MAX_PH_ADDR, MAX_VA_ADDR, MAX_GVA_ADDR, + &capab_misc); + + /* reset device ident */ + pci_config_set_vendor_id(s->dev.config, PCI_VENDOR_ID_AMD); + pci_config_set_device_id(s->dev.config, PCI_DEVICE_ID_RD890_IOMMU); + pci_config_set_prog_interface(s->dev.config, 00); + pci_config_set_class(s->dev.config, 0x0806); + + /* add msi and hypertransport capabilities */ + pci_add_capability(&s->dev, PCI_CAP_ID_MSI, 0, CAPAB_REG_SIZE); + pci_add_capability(&s->dev, PCI_CAP_ID_HT, 0, CAPAB_REG_SIZE); + + /* reset IOMMU specific capabilities */ + pci_set_long(s->dev.config + s->capab_offset, CAPAB_FEATURES); + pci_set_long(s->dev.config + s->capab_offset + CAPAB_BAR_LOW, + s->mmio.addr & ~(0xffff0000)); + pci_set_long(s->dev.config + s->capab_offset + CAPAB_BAR_HIGH, + (s->mmio.addr & ~(0xffff)) >> 16); + pci_set_long(s->dev.config + s->capab_offset + CAPAB_RANGE, 0xff000000); + pci_set_long(s->dev.config + s->capab_offset + CAPAB_MISC, capab_misc); +} + +/* I honestly don't know how to reserve MMIO */ +static void amd_iommu_mmio_map(AMDIOMMUState *s, hwaddr addr) +{ + if(s->mmio.addr == addr){ + return; + } + +// if(s->mmio.addr != (hwaddr)-1) { +// memory_region_del_subregion(get_system_memory(), &s->mmio); +// } + + s->mmio.addr = addr; + memory_region_add_subregion(get_system_memory(), addr, &s->mmio); + +} + +static void amd_iommu_reset(DeviceState *dev) +{ + AMDIOMMUState *s = AMD_IOMMU_DEVICE(dev); + + amd_iommu_init(s); +} + +static void amd_iommu_realize(PCIDevice *dev, Error **error) +{ + AMDIOMMUState *s = container_of(dev, AMDIOMMUState, dev); + + s->capab_offset = pci_add_capability(dev, PCI_CAP_ID_SEC, 0, CAPAB_SIZE); + + /* set up MMIO */ + memory_region_init_io(&s->mmio, OBJECT(s), &mmio_mem_ops, s, "amd-iommu", + MMIO_SIZE); + amd_iommu_mmio_map(s, BUS_AMD_IOMMU_ADDR); + + s->iotlb = g_hash_table_new_full(amd_iommu_uint64_hash, + amd_iommu_uint64_equal, g_free, g_free); + + amd_iommu_init(s); +} + +static const VMStateDescription vmstate_amd_iommu = { + .name = "amd-iommu", + .fields = (VMStateField[]) { + VMSTATE_PCI_DEVICE(dev, AMDIOMMUState), + VMSTATE_END_OF_LIST() + } +}; + +static Property amd_iommu_properties[] = { + DEFINE_PROP_UINT32("version", AMDIOMMUState, version, 2), + DEFINE_PROP_END_OF_LIST(), +}; + +static void amd_iommu_uninit(PCIDevice *dev) +{ + AMDIOMMUState *s = container_of(dev, AMDIOMMUState, dev); + + amd_iommu_iotlb_reset(s); +} + +static void amd_iommu_class_init(ObjectClass *klass, void* data) +{ + DeviceClass *dc = DEVICE_CLASS(klass); + PCIDeviceClass *k = PCI_DEVICE_CLASS(klass); + + k->realize = amd_iommu_realize; + k->exit = amd_iommu_uninit; + + dc->reset = amd_iommu_reset; + dc->vmsd = &vmstate_amd_iommu; + dc->props = amd_iommu_properties; +} + +static const TypeInfo amd_iommu = { + .name = TYPE_AMD_IOMMU_DEVICE, + .parent = TYPE_PCI_DEVICE, + .instance_size = sizeof(AMDIOMMUState), + .class_init = amd_iommu_class_init +}; + +static void amd_iommu_register_types(void) +{ + type_register_static(&amd_iommu); +} + +type_init(amd_iommu_register_types); diff --git a/hw/i386/amd_iommu.h b/hw/i386/amd_iommu.h new file mode 100644 index 0000000..c67ec4a --- /dev/null +++ b/hw/i386/amd_iommu.h @@ -0,0 +1,363 @@ +/* + * QEMU emulation of an AMD IOMMU (AMD-Vi) + * + * Copyright (C) 2011 Eduard - Gabriel Munteanu + * Copyright (C) 2015 David Kiarie, <davidkiar...@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 AMD_IOMMU_H_ +#define AMD_IOMMU_H_ + +#include "hw/hw.h" +#include "hw/pci/pci.h" +#include "hw/sysbus.h" +#include "sysemu/dma.h" +#include "qemu/osdep.h" +#include "qemu/event_notifier.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_MISC1 0x14 + +#define CAPAB_SIZE 0x18 +#define CAPAB_REG_SIZE 0x04 + +/* Capability header data */ +#define CAPAB_FLAT_EXT (1 << 28) +#define CAPAB_EFR_SUP (1 << 27) +#define CAPAB_FLAG_NPCACHE (1 << 26) +#define CAPAB_FLAG_HTTUNNEL (1 << 25) +#define CAPAB_FLAG_IOTLBSUP (1 << 24) +#define CAPAB_INIT_REV (1 << 19) +#define CAPAB_INIT_TYPE (3 << 16) +#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_EXT_FEATURES 0x0030 +#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_PPR_BASE 0x0038 +#define MMIO_PPR_HEAD 0x2030 +#define MMIO_PPR_TAIL 0x2038 + +#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 + +/* some of this are similar but just for readability */ +#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_PPRLOG_SIZE_BYTE (MMIO_EVENT_BASE + 7) +#define MMIO_PPRLOG_HEAD_MASK MMIO_EVTLOG_HEAD_MASK +#define MMIO_PPRLOG_TAIL_MASK MMIO_EVTLOG_HEAD_MASK +#define MMIO_PPRLOG_BASE_MASK MMIO_EVTLOG_BASE_MASK +#define MMIO_PPRLOG_SIZE_MASK MMIO_EVTLOG_SIZE_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_CMDBUFLEN (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 CMD_PREFETCH_IOMMU_PAGES 0x06 +#define CMD_COMPLETE_PPR_REQUEST 0x07 +#define CMD_INVAL_IOMMU_ALL 0x08 + +#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_DOMID_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 << 12) +#define EVENT_IOPF (0x2U << 12) +#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 << 12) +#define EVENT_PAGE_TAB_HW_ERROR (0x4U << 12) +#define EVENT_ILLEGAL_COMMAND_ERROR (0x5U << 12) +#define EVENT_COMMAND_HW_ERROR (0x6U << 12) +#define EVENT_IOTLB_INV_TIMEOUT (0x7U << 12) +#define EVENT_INVALID_DEV_REQUEST (0x8U << 12) + +#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) + +/* AMD RD890 Chipset */ +#define PCI_DEVICE_ID_RD890_IOMMU 0x20 + +#define PCI_CAP_ID_SEC 0xf /* IOMMU capability header register */ +#define PCI_CAP_ID_MMIO_LOW 0x0 /* MMIO base address low register */ +#define PCI_CAP_ID_MMIO_HIGH 0x0 /* MMIO base address high register */ +#define PCI_CAP_ID_RANGE 0x0 /* Device range register */ +#define PCI_CAP_ID_MISC 0x0 /* miscellaneous Information register 0 */ +#define PCI_CAP_ID_MISC1 0x0 /* miscellaneous Information register 1 */ + +#define FEATURE_PREFETCH (1ULL<<0) +#define FEATURE_PPR (1ULL<<1) +#define FEATURE_NX (1ULL<<3) +#define FEATURE_GT (1ULL<<4) +#define FEATURE_IA (1ULL<<6) +#define FEATURE_GA (1ULL<<7) +#define FEATURE_HE (1ULL<<8) +#define FEATURE_PC (1ULL<<9) + +/* reserved DTE bits */ +#define DTE_LOWER_QUAD_RESERVED 0x80300000000000fc +#define DTE_MIDDLE_QUAD_RESERVED 0x0000000000000100 +#define DTE_UPPER_QUAD_RESERVED 0x08f0000000000000 + +/* IOMMU paging mode */ +#define GATS_MODE (6ULL << 12) +#define HATS_MODE (6ULL << 10) + +/* PCI SIG constants */ +#define PCI_BUS_MAX 256 +#define PCI_SLOT_MAX 32 +#define PCI_FUNC_MAX 8 +#define PCI_DEVFN_MAX 256 + +/* utility */ +#define PCI_BUS_NUM(x) (((x) >> 8) & 0xff) + +/* IOTLB */ +#define IOMMU_IOTLB_MAX_SIZE 1024 +#define IOMMU_DEVID_SHIFT 36 + +/* extended feature support */ +#define EXT_FEATURES (FEATURE_PREFETCH | FEATURE_PPR | FEATURE_NX | FEATURE_GT | FEATURE_IA | FEATURE_GA | FEATURE_HE | GATS_MODE | HATS_MODE ) + +/* capabilities header */ +#define CAPAB_FEATURES (CAPAB_FLAT_EXT | CAPAB_FLAG_NPCACHE | CAPAB_FLAG_IOTLBSUP | PCI_CAP_ID_SEC | CAPAB_INIT_TYPE | CAPAB_FLAG_HTTUNNEL | CAPAB_EFR_SUP) + +/* command constants */ +#define COM_STORE_ADDRESS_MASK 0xffffffffffff8 +#define COM_COMPLETION_STORE_MASK 0x1 +#define COM_COMPLETION_INTR 0x2 +#define COM_COMPLETION_DATA_OFF 0x8 +#define IOMMU_COMMAND_SIZE 0x10 + +/* IOMMU default address */ +#define BUS_AMD_IOMMU_ADDR 0xfeb00000 + +/* page management constants */ +#define IOMMU_PAGE_SHIFT 12 +#define IOMMU_PAGE_SIZE (1ULL << IOMMU_PAGE_SHIFT) + +#define IOMMU_PAGE_SHIFT_4K 12 +#define IOMMU_PAGE_MASK_4K (~((1ULL << IOMMU_PAGE_SHIFT_4K) - 1)) +#define IOMMU_PAGE_SHIFT_2M 21 +#define IOMMU_PAGE_MASK_2M (~((1ULL << IOMMU_PAGE_SHIFT_2M) -1)) +#define IOMMU_PAGE_SHIFT_1G 30 +#define IOMMU_PAGE_MASK_1G (~((1ULL << IOMMU_PAGE_SHIFT_1G) - 1)) + +#define PCI_SLOT(devfn) (((devfn) >> 3) & 0x1f) +#define PCI_FUNC(devfn) ((devfn) & 0x07) + +#define MAX_VA_ADDR (48UL << 5) +#define MAX_PH_ADDR (40UL << 8) +#define MAX_GVA_ADDR (48UL << 15) + +/* invalidation command device id */ +#define INVAL_DEV_ID_SHIFT 32 +#define INVAL_DEV_ID_MASK (~((1UL << INVAL_DEV_ID_SHIFT) - 1)) + +/* invalidation address */ +#define INVAL_ADDR_MASK_SHIFT 12 +#define INVAL_ADDR_MASK (~((1UL << INVAL_ADDR_MASK_SHIFT)-1)) + +/* invalidation S bit mask */ +#define INVAL_ALL(val) ((val) & (0x1)) + +/* reserved bits */ +#define COMPLETION_WAIT_RSVD 0x0ff000000 +#define CMD_INVAL_DEV_RSVD 0xffff00000fffffff +#define INVAL_IOMMU_PAGES_RSVD 0xfff000000fff0000 +#define INVAL_IOTLB_PAGES_RSVD 0x00000ff4 +#define INVAL_INTR_TABLE_RSVD 0xffff00000fffffff +#define PRF_IOMMU_PAGES_RSVD 0x00ff00000ff00000 +#define COMPLETE_PPR_RQ_RSVD 0xffff00000ff00000 +#define INVAL_IOMMU_ALL_RSVD 0x0fffffff00000000 + +#define TYPE_AMD_IOMMU_DEVICE "amd-iommu" +#define AMD_IOMMU_DEVICE(obj)\ + OBJECT_CHECK(AMDIOMMUState, (obj), TYPE_AMD_IOMMU_DEVICE) + +typedef struct AMDIOMMUState AMDIOMMUState; + +typedef struct AMDIOMMUAddressSpace{ + uint8_t bus_num; /* bus number */ + uint8_t devfn; /* device function */ + AMDIOMMUState *iommu_state; /* IOMMU - one per machine */ + MemoryRegion iommu; /* Device's iommu region */ + AddressSpace as; /* device's corresponding address space */ +}AMDIOMMUAddressSpace; + +struct AMDIOMMUState { + PCIDevice dev; /* The PCI device itself */ + + uint32_t version; + + uint32_t capab_offset; /* capability offset pointer */ + uint8_t *capab; /* capabilities registers */ + + bool enabled; /* IOMMU enabled */ + bool ats_enabled; /* address translation enabled */ + bool cmdbuf_enabled; /* command buffer enabled */ + bool evtlog_enabled; /* event log enabled */ + bool excl_enabled; + + dma_addr_t devtab; /* base address device table */ + size_t devtab_len; /* device table length */ + + dma_addr_t cmdbuf; /* command buffer base address */ + uint64_t cmdbuf_len; /* command buffer length */ + uint32_t cmdbuf_head; /* current IOMMU read position */ + uint32_t cmdbuf_tail; /* next Software write position */ + bool completion_wait_intr; + + dma_addr_t evtlog; /* base address event log */ + bool evtlog_intr; + uint32_t evtlog_len; /* event log length */ + uint32_t evtlog_head; /* current IOMMU write position */ + uint32_t evtlog_tail; /* current Software read position */ + + /* unused for now */ + dma_addr_t excl_base; /* base DVA - IOMMU exclusion range */ + dma_addr_t excl_limit; /* limit of IOMMU exclusion range */ + bool excl_allow; /* translate accesses to the exclusion range */ + bool excl_enable; /* exclusion range enabled */ + + dma_addr_t ppr_log; /* base address ppr log */ + uint32_t pprlog_len; /* ppr log len */ + uint32_t pprlog_head; /* ppr log head */ + uint32_t pprlog_tail; /* ppr log tail */ + + MemoryRegion mmio; /* MMIO region */ + uint8_t mmior[MMIO_SIZE]; /* read/write MMIO */ + uint8_t mmiow1c[MMIO_SIZE]; /* read/write 1 clear MMIO */ + bool mmio_enabled; + + /* IOMMU function */ + MemoryRegionIOMMUOps iommu_ops; + + /* for each served device */ + AMDIOMMUAddressSpace **address_spaces[PCI_BUS_MAX]; + + /* IOTLB */ + GHashTable *iotlb; +}; + +typedef struct IOMMUIOTLBEntry{ + uint64_t gfn; + uint64_t devid; + uint64_t perms; + dma_addr_t hwaddr; +} IOMMUIOTLBEntry; + +typedef struct IOMMUTLBPageInvInfo { + uint64_t devid; + uint64_t gfn; + uint64_t mask; +} IOMMUTLBPageInvInfo; + +AddressSpace *bridge_host_amd_iommu(PCIBus *bus, void *opaque, int devfn); + +#endif -- 2.1.4