[Qemu-devel] [PATCH 12/25] memory: iommu support

[Paolo Bonzini]

From: Avi Kivity <address@hidden>

Add a new memory region type that translates addresses it is given,
then forwards them to a target address space.  This is similar to
an alias, except that the mapping is more flexible than a linear
translation and trucation, and also less efficient since the
translation happens at runtime.

The implementation uses an AddressSpace mapping the target region to
avoid hierarchical dispatch all the way to the resolved region; only
iommu regions are looked up dynamically.

Reviewed-by: Peter Maydell <address@hidden>
Signed-off-by: Avi Kivity <address@hidden>
[Modified to put translation in address_space_translate; assume
 IOMMUs are not reachable from TCG. - Paolo]
Signed-off-by: Paolo Bonzini <address@hidden>
---
 exec.c                | 35 +++++++++++++++++++++++++--
 include/exec/memory.h | 66 +++++++++++++++++++++++++++++++++++++++++++++++----
 memory.c              | 16 +++++++++++++
 3 files changed, 111 insertions(+), 6 deletions(-)

diff --git a/exec.c b/exec.c
index 77eedd8..dfddd85 100644
--- a/exec.c
+++ b/exec.c
@@ -266,14 +266,45 @@ MemoryRegion *address_space_translate(AddressSpace *as, 
hwaddr addr,
                                       hwaddr *xlat, hwaddr *plen,
                                       bool is_write)
 {
-    return address_space_translate_internal(as, addr, xlat, plen, true)->mr;
+    IOMMUTLBEntry iotlb;
+    MemoryRegionSection *section;
+    MemoryRegion *mr;
+    hwaddr len = *plen;
+
+    for (;;) {
+        section = address_space_translate_internal(as, addr, &addr, plen, 
true);
+        mr = section->mr;
+
+        if (!mr->iommu_ops) {
+            break;
+        }
+
+        iotlb = mr->iommu_ops->translate(mr, addr);
+        addr = ((iotlb.translated_addr & ~iotlb.addr_mask)
+                | (addr & iotlb.addr_mask));
+        len = MIN(len, (addr | iotlb.addr_mask) - addr + 1);
+        if (!(iotlb.perm & (1 << is_write))) {
+            mr = &io_mem_unassigned;
+            break;
+        }
+
+        as = iotlb.target_as;
+    }
+
+    *plen = len;
+    *xlat = addr;
+    return mr;
 }
 
 MemoryRegionSection *
 address_space_translate_for_iotlb(AddressSpace *as, hwaddr addr, hwaddr *xlat,
                                   hwaddr *plen)
 {
-    return address_space_translate_internal(as, addr, xlat, plen, false);
+    MemoryRegionSection *section;
+    section = address_space_translate_internal(as, addr, xlat, plen, false);
+
+    assert(!section->mr->iommu_ops);
+    return section;
 }
 #endif
 
diff --git a/include/exec/memory.h b/include/exec/memory.h
index fddc6ad..a5e7721 100644
--- a/include/exec/memory.h
+++ b/include/exec/memory.h
@@ -53,6 +53,24 @@ struct MemoryRegionIORange {
     hwaddr offset;
 };
 
+typedef struct IOMMUTLBEntry IOMMUTLBEntry;
+
+/* See address_space_translate: bit 0 is read, bit 1 is write.  */
+typedef enum {
+    IOMMU_NONE = 0,
+    IOMMU_RO   = 1,
+    IOMMU_WO   = 2,
+    IOMMU_RW   = 3,
+} IOMMUAccessFlags;
+
+struct IOMMUTLBEntry {
+    AddressSpace    *target_as;
+    hwaddr           iova;
+    hwaddr           translated_addr;
+    hwaddr           addr_mask;  /* 0xfff = 4k translation */
+    IOMMUAccessFlags perm;
+};
+
 /*
  * Memory region callbacks
  */
@@ -115,12 +133,20 @@ struct MemoryRegionOps {
     const MemoryRegionMmio old_mmio;
 };
 
+typedef struct MemoryRegionIOMMUOps MemoryRegionIOMMUOps;
+
+struct MemoryRegionIOMMUOps {
+    /* Return a TLB entry that contains a given address. */
+    IOMMUTLBEntry (*translate)(MemoryRegion *iommu, hwaddr addr);
+};
+
 typedef struct CoalescedMemoryRange CoalescedMemoryRange;
 typedef struct MemoryRegionIoeventfd MemoryRegionIoeventfd;
 
 struct MemoryRegion {
     /* All fields are private - violators will be prosecuted */
     const MemoryRegionOps *ops;
+    const MemoryRegionIOMMUOps *iommu_ops;
     void *opaque;
     MemoryRegion *parent;
     Int128 size;
@@ -332,6 +358,24 @@ void memory_region_init_rom_device(MemoryRegion *mr,
 void memory_region_init_reservation(MemoryRegion *mr,
                                     const char *name,
                                     uint64_t size);
+
+/**
+ * memory_region_init_iommu: Initialize a memory region that translates
+ * addresses
+ *
+ * An IOMMU region translates addresses and forwards accesses to a target
+ * memory region.
+ *
+ * @mr: the #MemoryRegion to be initialized
+ * @ops: a function that translates addresses into the @target region
+ * @name: used for debugging; not visible to the user or ABI
+ * @size: size of the region.
+ */
+void memory_region_init_iommu(MemoryRegion *mr,
+                              const MemoryRegionIOMMUOps *ops,
+                              const char *name,
+                              uint64_t size);
+
 /**
  * memory_region_destroy: Destroy a memory region and reclaim all resources.
  *
@@ -371,6 +415,15 @@ static inline bool memory_region_is_romd(MemoryRegion *mr)
 }
 
 /**
+ * memory_region_is_iommu: check whether a memory region is an iommu
+ *
+ * Returns %true is a memory region is an iommu.
+ *
+ * @mr: the memory region being queried
+ */
+bool memory_region_is_iommu(MemoryRegion *mr);
+
+/**
  * memory_region_name: get a memory region's name
  *
  * Returns the string that was used to initialize the memory region.
@@ -825,7 +878,8 @@ void address_space_destroy(AddressSpace *as);
 /**
  * address_space_rw: read from or write to an address space.
  *
- * Return true if the operation hit any unassigned memory.
+ * Return true if the operation hit any unassigned memory or encountered an
+ * IOMMU fault.
  *
  * @as: #AddressSpace to be accessed
  * @addr: address within that address space
@@ -838,7 +892,8 @@ bool address_space_rw(AddressSpace *as, hwaddr addr, 
uint8_t *buf,
 /**
  * address_space_write: write to address space.
  *
- * Return true if the operation hit any unassigned memory.
+ * Return true if the operation hit any unassigned memory or encountered an
+ * IOMMU fault.
  *
  * @as: #AddressSpace to be accessed
  * @addr: address within that address space
@@ -850,7 +905,8 @@ bool address_space_write(AddressSpace *as, hwaddr addr,
 /**
  * address_space_read: read from an address space.
  *
- * Return true if the operation hit any unassigned memory.
+ * Return true if the operation hit any unassigned memory or encountered an
+ * IOMMU fault.
  *
  * @as: #AddressSpace to be accessed
  * @addr: address within that address space
@@ -875,7 +931,9 @@ MemoryRegion *address_space_translate(AddressSpace *as, 
hwaddr addr,
 /* address_space_access_valid: check for validity of accessing an address
  * space range
  *
- * Check whether memory is assigned to the given address space range.
+ * Check whether memory is assigned to the given address space range, and
+ * access is permitted by any IOMMU regions that are active for the address
+ * space.
  *
  * For now, addr and len should be aligned to a page size.  This limitation
  * will be lifted in the future.
diff --git a/memory.c b/memory.c
index 2b19745..7eb8d46 100644
--- a/memory.c
+++ b/memory.c
@@ -824,6 +824,7 @@ void memory_region_init(MemoryRegion *mr,
 {
     mr->ops = &unassigned_mem_ops;
     mr->opaque = NULL;
+    mr->iommu_ops = NULL;
     mr->parent = NULL;
     mr->size = int128_make64(size);
     if (size == UINT64_MAX) {
@@ -1063,6 +1064,16 @@ void memory_region_init_rom_device(MemoryRegion *mr,
     mr->ram_addr = qemu_ram_alloc(size, mr);
 }
 
+void memory_region_init_iommu(MemoryRegion *mr,
+                              const MemoryRegionIOMMUOps *ops,
+                              const char *name,
+                              uint64_t size)
+{
+    memory_region_init(mr, name, size);
+    mr->iommu_ops = ops,
+    mr->terminates = true;  /* then re-forwards */
+}
+
 void memory_region_init_reservation(MemoryRegion *mr,
                                     const char *name,
                                     uint64_t size)
@@ -1108,6 +1119,11 @@ bool memory_region_is_rom(MemoryRegion *mr)
     return mr->ram && mr->readonly;
 }
 
+bool memory_region_is_iommu(MemoryRegion *mr)
+{
+    return mr->iommu_ops;
+}
+
 void memory_region_set_log(MemoryRegion *mr, bool log, unsigned client)
 {
     uint8_t mask = 1 << client;
-- 
1.8.1.4