[Qemu-devel] [PATCH v2 12/23] memory: add ioeventfd support

[Avi Kivity]

As with the rest of the memory API, the caller associates an eventfd
with an address, and the memory API takes care of registering or
unregistering when the address is made visible or invisible to the
guest.

Signed-off-by: Avi Kivity <address@hidden>
---
 memory.c |  224 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
 memory.h |   45 +++++++++++++
 2 files changed, 269 insertions(+), 0 deletions(-)

diff --git a/memory.c b/memory.c
index 7dd7cac..686bbf2 100644
--- a/memory.c
+++ b/memory.c
@@ -15,6 +15,7 @@
 #include "exec-memory.h"
 #include "ioport.h"
 #include "bitops.h"
+#include "kvm.h"
 #include <assert.h>
 
 typedef struct AddrRange AddrRange;
@@ -64,6 +65,50 @@ struct CoalescedMemoryRange {
     QTAILQ_ENTRY(CoalescedMemoryRange) link;
 };
 
+struct MemoryRegionIoeventfd {
+    AddrRange addr;
+    bool match_data;
+    uint64_t data;
+    int fd;
+};
+
+static bool memory_region_ioeventfd_before(MemoryRegionIoeventfd a,
+                                           MemoryRegionIoeventfd b)
+{
+    if (a.addr.start < b.addr.start) {
+        return true;
+    } else if (a.addr.start > b.addr.start) {
+        return false;
+    } else if (a.addr.size < b.addr.size) {
+        return true;
+    } else if (a.addr.size > b.addr.size) {
+        return false;
+    } else if (a.match_data < b.match_data) {
+        return true;
+    } else  if (a.match_data > b.match_data) {
+        return false;
+    } else if (a.match_data) {
+        if (a.data < b.data) {
+            return true;
+        } else if (a.data > b.data) {
+            return false;
+        }
+    }
+    if (a.fd < b.fd) {
+        return true;
+    } else if (a.fd > b.fd) {
+        return false;
+    }
+    return false;
+}
+
+static bool memory_region_ioeventfd_equal(MemoryRegionIoeventfd a,
+                                          MemoryRegionIoeventfd b)
+{
+    return !memory_region_ioeventfd_before(a, b)
+        && !memory_region_ioeventfd_before(b, a);
+}
+
 typedef struct FlatRange FlatRange;
 typedef struct FlatView FlatView;
 
@@ -92,6 +137,8 @@ struct AddressSpace {
     const AddressSpaceOps *ops;
     MemoryRegion *root;
     FlatView current_map;
+    int ioeventfd_nb;
+    MemoryRegionIoeventfd *ioeventfds;
 };
 
 struct AddressSpaceOps {
@@ -99,6 +146,8 @@ struct AddressSpaceOps {
     void (*range_del)(AddressSpace *as, FlatRange *fr);
     void (*log_start)(AddressSpace *as, FlatRange *fr);
     void (*log_stop)(AddressSpace *as, FlatRange *fr);
+    void (*ioeventfd_add)(AddressSpace *as, MemoryRegionIoeventfd *fd);
+    void (*ioeventfd_del)(AddressSpace *as, MemoryRegionIoeventfd *fd);
 };
 
 #define FOR_EACH_FLAT_RANGE(var, view)          \
@@ -208,11 +257,35 @@ static void as_memory_log_stop(AddressSpace *as, 
FlatRange *fr)
     cpu_physical_log_stop(fr->addr.start, fr->addr.size);
 }
 
+static void as_memory_ioeventfd_add(AddressSpace *as, MemoryRegionIoeventfd 
*fd)
+{
+    int r;
+
+    assert(fd->match_data && fd->addr.size == 4);
+
+    r = kvm_set_ioeventfd_mmio_long(fd->fd, fd->addr.start, fd->data, true);
+    if (r < 0) {
+        abort();
+    }
+}
+
+static void as_memory_ioeventfd_del(AddressSpace *as, MemoryRegionIoeventfd 
*fd)
+{
+    int r;
+
+    r = kvm_set_ioeventfd_mmio_long(fd->fd, fd->addr.start, fd->data, false);
+    if (r < 0) {
+        abort();
+    }
+}
+
 static const AddressSpaceOps address_space_ops_memory = {
     .range_add = as_memory_range_add,
     .range_del = as_memory_range_del,
     .log_start = as_memory_log_start,
     .log_stop = as_memory_log_stop,
+    .ioeventfd_add = as_memory_ioeventfd_add,
+    .ioeventfd_del = as_memory_ioeventfd_del,
 };
 
 static AddressSpace address_space_memory = {
@@ -288,9 +361,33 @@ static void as_io_range_del(AddressSpace *as, FlatRange 
*fr)
     isa_unassign_ioport(fr->addr.start, fr->addr.size);
 }
 
+static void as_io_ioeventfd_add(AddressSpace *as, MemoryRegionIoeventfd *fd)
+{
+    int r;
+
+    assert(fd->match_data && fd->addr.size == 2);
+
+    r = kvm_set_ioeventfd_pio_word(fd->fd, fd->addr.start, fd->data, true);
+    if (r < 0) {
+        abort();
+    }
+}
+
+static void as_io_ioeventfd_del(AddressSpace *as, MemoryRegionIoeventfd *fd)
+{
+    int r;
+
+    r = kvm_set_ioeventfd_pio_word(fd->fd, fd->addr.start, fd->data, false);
+    if (r < 0) {
+        abort();
+    }
+}
+
 static const AddressSpaceOps address_space_ops_io = {
     .range_add = as_io_range_add,
     .range_del = as_io_range_del,
+    .ioeventfd_add = as_io_ioeventfd_add,
+    .ioeventfd_del = as_io_ioeventfd_del,
 };
 
 static AddressSpace address_space_io = {
@@ -389,6 +486,69 @@ static FlatView generate_memory_topology(MemoryRegion *mr)
     return view;
 }
 
+static void address_space_add_del_ioeventfds(AddressSpace *as,
+                                             MemoryRegionIoeventfd *fds_new,
+                                             unsigned fds_new_nb,
+                                             MemoryRegionIoeventfd *fds_old,
+                                             unsigned fds_old_nb)
+{
+    unsigned iold, inew;
+
+    /* Generate a symmetric difference of the old and new fd sets, adding
+     * and deleting as necessary.
+     */
+
+    iold = inew = 0;
+    while (iold < fds_old_nb || inew < fds_new_nb) {
+        if (iold < fds_old_nb
+            && (inew == fds_new_nb
+                || memory_region_ioeventfd_before(fds_old[iold],
+                                                  fds_new[inew]))) {
+            as->ops->ioeventfd_del(as, &fds_old[iold]);
+            ++iold;
+        } else if (inew < fds_new_nb
+                   && (iold == fds_old_nb
+                       || memory_region_ioeventfd_before(fds_new[inew],
+                                                         fds_old[iold]))) {
+            as->ops->ioeventfd_add(as, &fds_new[inew]);
+            ++inew;
+        } else {
+            ++iold;
+            ++inew;
+        }
+    }
+}
+
+static void address_space_update_ioeventfds(AddressSpace *as)
+{
+    FlatRange *fr;
+    unsigned ioeventfd_nb = 0;
+    MemoryRegionIoeventfd *ioeventfds = NULL;
+    AddrRange tmp;
+    unsigned i;
+
+    FOR_EACH_FLAT_RANGE(fr, &as->current_map) {
+        for (i = 0; i < fr->mr->ioeventfd_nb; ++i) {
+            tmp = addrrange_shift(fr->mr->ioeventfds[i].addr,
+                                  fr->addr.start - fr->offset_in_region);
+            if (addrrange_intersects(fr->addr, tmp)) {
+                ++ioeventfd_nb;
+                ioeventfds = qemu_realloc(ioeventfds,
+                                          ioeventfd_nb * sizeof(*ioeventfds));
+                ioeventfds[ioeventfd_nb-1] = fr->mr->ioeventfds[i];
+                ioeventfds[ioeventfd_nb-1].addr = tmp;
+            }
+        }
+    }
+
+    address_space_add_del_ioeventfds(as, ioeventfds, ioeventfd_nb,
+                                     as->ioeventfds, as->ioeventfd_nb);
+
+    qemu_free(as->ioeventfds);
+    as->ioeventfds = ioeventfds;
+    as->ioeventfd_nb = ioeventfd_nb;
+}
+
 static void address_space_update_topology(AddressSpace *as)
 {
     FlatView old_view = as->current_map;
@@ -441,6 +601,7 @@ static void address_space_update_topology(AddressSpace *as)
     }
     as->current_map = new_view;
     flatview_destroy(&old_view);
+    address_space_update_ioeventfds(as);
 }
 
 static void memory_region_update_topology(void)
@@ -471,6 +632,8 @@ void memory_region_init(MemoryRegion *mr,
     QTAILQ_INIT(&mr->coalesced);
     mr->name = qemu_strdup(name);
     mr->dirty_log_mask = 0;
+    mr->ioeventfd_nb = 0;
+    mr->ioeventfds = NULL;
 }
 
 static bool memory_region_access_valid(MemoryRegion *mr,
@@ -682,6 +845,7 @@ void memory_region_destroy(MemoryRegion *mr)
     assert(QTAILQ_EMPTY(&mr->subregions));
     memory_region_clear_coalescing(mr);
     qemu_free((char *)mr->name);
+    qemu_free(mr->ioeventfds);
 }
 
 uint64_t memory_region_size(MemoryRegion *mr)
@@ -803,6 +967,66 @@ void memory_region_clear_coalescing(MemoryRegion *mr)
     memory_region_update_coalesced_range(mr);
 }
 
+void memory_region_add_eventfd(MemoryRegion *mr,
+                               target_phys_addr_t addr,
+                               unsigned size,
+                               bool match_data,
+                               uint64_t data,
+                               int fd)
+{
+    MemoryRegionIoeventfd mrfd = {
+        .addr.start = addr,
+        .addr.size = size,
+        .match_data = match_data,
+        .data = data,
+        .fd = fd,
+    };
+    unsigned i;
+
+    for (i = 0; i < mr->ioeventfd_nb; ++i) {
+        if (memory_region_ioeventfd_before(mrfd, mr->ioeventfds[i])) {
+            break;
+        }
+    }
+    ++mr->ioeventfd_nb;
+    mr->ioeventfds = qemu_realloc(mr->ioeventfds,
+                                  sizeof(*mr->ioeventfds) * mr->ioeventfd_nb);
+    memmove(&mr->ioeventfds[i+1], &mr->ioeventfds[i],
+            sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb-1 - i));
+    mr->ioeventfds[i] = mrfd;
+    memory_region_update_topology();
+}
+
+void memory_region_del_eventfd(MemoryRegion *mr,
+                               target_phys_addr_t addr,
+                               unsigned size,
+                               bool match_data,
+                               uint64_t data,
+                               int fd)
+{
+    MemoryRegionIoeventfd mrfd = {
+        .addr.start = addr,
+        .addr.size = size,
+        .match_data = match_data,
+        .data = data,
+        .fd = fd,
+    };
+    unsigned i;
+
+    for (i = 0; i < mr->ioeventfd_nb; ++i) {
+        if (memory_region_ioeventfd_equal(mrfd, mr->ioeventfds[i])) {
+            break;
+        }
+    }
+    assert(i != mr->ioeventfd_nb);
+    memmove(&mr->ioeventfds[i], &mr->ioeventfds[i+1],
+            sizeof(*mr->ioeventfds) * (mr->ioeventfd_nb - (i+1)));
+    --mr->ioeventfd_nb;
+    mr->ioeventfds = qemu_realloc(mr->ioeventfds,
+                                  sizeof(*mr->ioeventfds)*mr->ioeventfd_nb + 
1);
+    memory_region_update_topology();
+}
+
 static void memory_region_add_subregion_common(MemoryRegion *mr,
                                                target_phys_addr_t offset,
                                                MemoryRegion *subregion)
diff --git a/memory.h b/memory.h
index 003c999..c280a39 100644
--- a/memory.h
+++ b/memory.h
@@ -98,6 +98,7 @@ struct MemoryRegionOps {
 };
 
 typedef struct CoalescedMemoryRange CoalescedMemoryRange;
+typedef struct MemoryRegionIoeventfd MemoryRegionIoeventfd;
 
 struct MemoryRegion {
     /* All fields are private - violators will be prosecuted */
@@ -120,6 +121,8 @@ struct MemoryRegion {
     QTAILQ_HEAD(coalesced_ranges, CoalescedMemoryRange) coalesced;
     const char *name;
     uint8_t dirty_log_mask;
+    unsigned ioeventfd_nb;
+    MemoryRegionIoeventfd *ioeventfds;
 };
 
 struct MemoryRegionPortio {
@@ -364,6 +367,48 @@ void memory_region_add_coalescing(MemoryRegion *mr,
 void memory_region_clear_coalescing(MemoryRegion *mr);
 
 /**
+ * memory_region_add_eventfd: Request an eventfd to be triggered when a word
+ *                            is written to a location.
+ *
+ * Marks a word in an IO region (initialized with memory_region_init_io())
+ * as a trigger for an eventfd event.  The I/O callback will not be called.
+ * The caller must be prepared to handle failure (hat is, take the required
+ * action if the callback _is_ called).
+ *
+ * @mr: the memory region being updated.
+ * @addr: the address within @mr that is to be monitored
+ * @size: the size of the access to trigger the eventfd
+ * @match_data: whether to match against @data, instead of just @addr
+ * @data: the data to match against the guest write
+ * @fd: the eventfd to be triggered when @addr, @size, and @data all match.
+ **/
+void memory_region_add_eventfd(MemoryRegion *mr,
+                               target_phys_addr_t addr,
+                               unsigned size,
+                               bool match_data,
+                               uint64_t data,
+                               int fd);
+
+/**
+ * memory_region_del_eventfd: Cancel and eventfd.
+ *
+ * Cancels an eventfd trigger request by a previous memory_region_add_eventfd()
+ * call.
+ *
+ * @mr: the memory region being updated.
+ * @addr: the address within @mr that is to be monitored
+ * @size: the size of the access to trigger the eventfd
+ * @match_data: whether to match against @data, instead of just @addr
+ * @data: the data to match against the guest write
+ * @fd: the eventfd to be triggered when @addr, @size, and @data all match.
+ */
+void memory_region_del_eventfd(MemoryRegion *mr,
+                               target_phys_addr_t addr,
+                               unsigned size,
+                               bool match_data,
+                               uint64_t data,
+                               int fd);
+/**
  * memory_region_add_subregion: Add a sub-region to a container.
  *
  * Adds a sub-region at @offset.  The sub-region may not overlap with other
-- 
1.7.5.3