include/sysemu/kvm_int.h

   1 /*
   2  * Internal definitions for a target's KVM support
   3  *
   4  * This work is licensed under the terms of the GNU GPL, version 2 or later.
   5  * See the COPYING file in the top-level directory.
   6  *
   7  */
   8
   9 #ifndef QEMU_KVM_INT_H
  10 #define QEMU_KVM_INT_H
  11
  12 #include "exec/memory.h"
  13 #include "qapi/qapi-types-common.h"
  14 #include "qemu/accel.h"
  15 #include "qemu/queue.h"
  16 #include "sysemu/kvm.h"
  17 #include "hw/boards.h"
  18 #include "hw/i386/topology.h"
  19 #include "io/channel-socket.h"
  20
  21 typedef struct KVMSlot
  22 {
  23     hwaddr start_addr;
  24     ram_addr_t memory_size;
  25     void *ram;
  26     int slot;
  27     int flags;
  28     int old_flags;
  29     /* Dirty bitmap cache for the slot */
  30     unsigned long *dirty_bmap;
  31     unsigned long dirty_bmap_size;
  32     /* Cache of the address space ID */
  33     int as_id;
  34     /* Cache of the offset in ram address space */
  35     ram_addr_t ram_start_offset;
  36     int guest_memfd;
  37     hwaddr guest_memfd_offset;
  38 } KVMSlot;
  39
  40 typedef struct KVMMemoryUpdate {
  41     QSIMPLEQ_ENTRY(KVMMemoryUpdate) next;
  42     MemoryRegionSection section;
  43 } KVMMemoryUpdate;
  44
  45 typedef struct KVMMemoryListener {
  46     MemoryListener listener;
  47     KVMSlot *slots;
  48     unsigned int nr_used_slots;
  49     int as_id;
  50     QSIMPLEQ_HEAD(, KVMMemoryUpdate) transaction_add;
  51     QSIMPLEQ_HEAD(, KVMMemoryUpdate) transaction_del;
  52 } KVMMemoryListener;
  53
  54 #define KVM_MSI_HASHTAB_SIZE    256
  55
  56 typedef struct KVMHostTopoInfo {
  57     /* Number of package on the Host */
  58     unsigned int maxpkgs;
  59     /* Number of cpus on the Host */
  60     unsigned int maxcpus;
  61     /* Number of cpus on each different package */
  62     unsigned int *pkg_cpu_count;
  63     /* Each package can have different maxticks */
  64     unsigned int *maxticks;
  65 } KVMHostTopoInfo;
  66
  67 struct KVMMsrEnergy {
  68     pid_t pid;
  69     bool enable;
  70     char *socket_path;
  71     QIOChannelSocket *sioc;
  72     QemuThread msr_thr;
  73     unsigned int guest_vcpus;
  74     unsigned int guest_vsockets;
  75     X86CPUTopoInfo guest_topo_info;
  76     KVMHostTopoInfo host_topo;
  77     const CPUArchIdList *guest_cpu_list;
  78     uint64_t *msr_value;
  79     uint64_t msr_unit;
  80     uint64_t msr_limit;
  81     uint64_t msr_info;
  82 };
  83
  84 enum KVMDirtyRingReaperState {
  85     KVM_DIRTY_RING_REAPER_NONE = 0,
  86     /* The reaper is sleeping */
  87     KVM_DIRTY_RING_REAPER_WAIT,
  88     /* The reaper is reaping for dirty pages */
  89     KVM_DIRTY_RING_REAPER_REAPING,
  90 };
  91
  92 /*
  93  * KVM reaper instance, responsible for collecting the KVM dirty bits
  94  * via the dirty ring.
  95  */
  96 struct KVMDirtyRingReaper {
  97     /* The reaper thread */
  98     QemuThread reaper_thr;
  99     volatile uint64_t reaper_iteration; /* iteration number of reaper thr */
 100     volatile enum KVMDirtyRingReaperState reaper_state; /* reap thr state */
 101 };
 102 struct KVMState
 103 {
 104     AccelState parent_obj;
 105
 106     int nr_slots;
 107     int fd;
 108     int vmfd;
 109     int coalesced_mmio;
 110     int coalesced_pio;
 111     struct kvm_coalesced_mmio_ring *coalesced_mmio_ring;
 112     bool coalesced_flush_in_progress;
 113     int vcpu_events;
 114 #ifdef TARGET_KVM_HAVE_GUEST_DEBUG
 115     QTAILQ_HEAD(, kvm_sw_breakpoint) kvm_sw_breakpoints;
 116 #endif
 117     int max_nested_state_len;
 118     int kvm_shadow_mem;
 119     bool kernel_irqchip_allowed;
 120     bool kernel_irqchip_required;
 121     OnOffAuto kernel_irqchip_split;
 122     bool sync_mmu;
 123     bool guest_state_protected;
 124     uint64_t manual_dirty_log_protect;
 125     /*
 126      * Older POSIX says that ioctl numbers are signed int, but in
 127      * practice they are not. (Newer POSIX doesn't specify ioctl
 128      * at all.) Linux, glibc and *BSD all treat ioctl numbers as
 129      * unsigned, and real-world ioctl values like KVM_GET_XSAVE have
 130      * bit 31 set, which means that passing them via an 'int' will
 131      * result in sign-extension when they get converted back to the
 132      * 'unsigned long' which the ioctl() prototype uses. Luckily Linux
 133      * always treats the argument as an unsigned 32-bit int, so any
 134      * possible sign-extension is deliberately ignored, but for
 135      * consistency we keep to the same type that glibc is using.
 136      */
 137     unsigned long irq_set_ioctl;
 138     unsigned int sigmask_len;
 139     GHashTable *gsimap;
 140 #ifdef KVM_CAP_IRQ_ROUTING
 141     struct kvm_irq_routing *irq_routes;
 142     int nr_allocated_irq_routes;
 143     unsigned long *used_gsi_bitmap;
 144     unsigned int gsi_count;
 145 #endif
 146     KVMMemoryListener memory_listener;
 147     QLIST_HEAD(, KVMParkedVcpu) kvm_parked_vcpus;
 148
 149     /* For "info mtree -f" to tell if an MR is registered in KVM */
 150     int nr_as;
 151     struct KVMAs {
 152         KVMMemoryListener *ml;
 153         AddressSpace *as;
 154     } *as;
 155     uint64_t kvm_dirty_ring_bytes;  /* Size of the per-vcpu dirty ring */
 156     uint32_t kvm_dirty_ring_size;   /* Number of dirty GFNs per ring */
 157     bool kvm_dirty_ring_with_bitmap;
 158     uint64_t kvm_eager_split_size;  /* Eager Page Splitting chunk size */
 159     struct KVMDirtyRingReaper reaper;
 160     struct KVMMsrEnergy msr_energy;
 161     NotifyVmexitOption notify_vmexit;
 162     uint32_t notify_window;
 163     uint32_t xen_version;
 164     uint32_t xen_caps;
 165     uint16_t xen_gnttab_max_frames;
 166     uint16_t xen_evtchn_max_pirq;
 167     char *device;
 168 };
 169
 170 void kvm_memory_listener_register(KVMState *s, KVMMemoryListener *kml,
 171                                   AddressSpace *as, int as_id, const char *name);
 172
 173 void kvm_set_max_memslot_size(hwaddr max_slot_size);
 174
 175 /**
 176  * kvm_hwpoison_page_add:
 177  *
 178  * Parameters:
 179  *  @ram_addr: the address in the RAM for the poisoned page
 180  *
 181  * Add a poisoned page to the list
 182  *
 183  * Return: None.
 184  */
 185 void kvm_hwpoison_page_add(ram_addr_t ram_addr);
 186 #endif