[PATCH v5 16/16] x86/hvm: track large memory mapped accesses by buffer offset

[Paul Durrant <paul.durrant@citrix.com>]

The code in hvmemul_do_io() that tracks large reads or writes, to avoid
re-issue of component I/O, is defeated by accesses across a page boundary
because it uses physical address. The code is also only relevant to memory
mapped I/O to or from a buffer.

This patch re-factors the code and moves it into hvmemul_phys_mmio_access()
where it is relevant and tracks using buffer offset rather than address.
Separate I/O emulations (of which there may be up to three per instruction)
are distinguished by linear address.

Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Cc: Keir Fraser <keir@xen.org>
Cc: Jan Beulich <jbeulich@suse.com>
Cc: Andrew Cooper <andrew.cooper3@citrix.com>
---
 xen/arch/x86/hvm/emulate.c     |  122 ++++++++++++++++++++--------------------
 xen/include/asm-x86/hvm/vcpu.h |   25 +++++---
 2 files changed, 78 insertions(+), 69 deletions(-)

diff --git a/xen/arch/x86/hvm/emulate.c b/xen/arch/x86/hvm/emulate.c
index 8bb56a2..d83d21c 100644
--- a/xen/arch/x86/hvm/emulate.c
+++ b/xen/arch/x86/hvm/emulate.c
@@ -106,29 +106,6 @@ static int hvmemul_do_io(
         return X86EMUL_UNHANDLEABLE;
     }
 
-    if ( is_mmio && !data_is_addr )
-    {
-        /* Part of a multi-cycle read or write? */
-        if ( dir == IOREQ_WRITE )
-        {
-            paddr_t pa = vio->mmio_large_write_pa;
-            unsigned int bytes = vio->mmio_large_write_bytes;
-            if ( (addr >= pa) && ((addr + size) <= (pa + bytes)) )
-                return X86EMUL_OKAY;
-        }
-        else
-        {
-            paddr_t pa = vio->mmio_large_read_pa;
-            unsigned int bytes = vio->mmio_large_read_bytes;
-            if ( (addr >= pa) && ((addr + size) <= (pa + bytes)) )
-            {
-                memcpy(p_data, &vio->mmio_large_read[addr - pa],
-                       size);
-                return X86EMUL_OKAY;
-            }
-        }
-    }
-
     switch ( vio->io_req.state )
     {
     case STATE_IOREQ_NONE:
@@ -208,33 +185,6 @@ static int hvmemul_do_io(
             memcpy(p_data, &p.data, size);
     }
 
-    if ( is_mmio && !data_is_addr )
-    {
-        /* Part of a multi-cycle read or write? */
-        if ( dir == IOREQ_WRITE )
-        {
-            paddr_t pa = vio->mmio_large_write_pa;
-            unsigned int bytes = vio->mmio_large_write_bytes;
-            if ( bytes == 0 )
-                pa = vio->mmio_large_write_pa = addr;
-            if ( addr == (pa + bytes) )
-                vio->mmio_large_write_bytes += size;
-        }
-        else
-        {
-            paddr_t pa = vio->mmio_large_read_pa;
-            unsigned int bytes = vio->mmio_large_read_bytes;
-            if ( bytes == 0 )
-                pa = vio->mmio_large_read_pa = addr;
-            if ( (addr == (pa + bytes)) &&
-                 ((bytes + size) <= sizeof(vio->mmio_large_read)) )
-            {
-                memcpy(&vio->mmio_large_read[bytes], p_data, size);
-                vio->mmio_large_read_bytes += size;
-            }
-        }
-    }
-
     return X86EMUL_OKAY;
 }
 
@@ -587,11 +537,12 @@ static int hvmemul_virtual_to_linear(
 }
 
 static int hvmemul_phys_mmio_access(
-    paddr_t gpa, unsigned int size, uint8_t dir, uint8_t *buffer)
+    struct hvm_mmio_cache *cache, paddr_t gpa, unsigned int size, uint8_t dir,
+    uint8_t *buffer, unsigned int off)
 {
     unsigned long one_rep = 1;
     unsigned int chunk;
-    int rc;
+    int rc = X86EMUL_OKAY;
 
     /* Accesses must fall within a page */
     BUG_ON((gpa & (PAGE_SIZE - 1)) + size > PAGE_SIZE);
@@ -607,14 +558,33 @@ static int hvmemul_phys_mmio_access(
 
     for ( ;; )
     {
-        rc = hvmemul_do_mmio_buffer(gpa, &one_rep, chunk, dir, 0,
-                                    buffer);
-        if ( rc != X86EMUL_OKAY )
-            break;
+        /* Have we already done this chunk? */
+        if ( off < cache->size )
+        {
+            ASSERT((off + chunk) <= cache->size);
+
+            if ( dir == IOREQ_READ )
+                memcpy(&buffer[off], &cache->buffer[off], chunk);
+            else if ( memcmp(&buffer[off], &cache->buffer[off], chunk) != 0 )
+                domain_crash(current->domain);
+        }
+        else
+        {
+            ASSERT(off == cache->size);
+
+            rc = hvmemul_do_mmio_buffer(gpa, &one_rep, chunk, dir, 0,
+                                        &buffer[off]);
+            if ( rc != X86EMUL_OKAY )
+                break;
+
+            /* Note that we have now done this chunk. */
+            memcpy(&cache->buffer[off], &buffer[off], chunk);
+            cache->size += chunk;
+        }
 
         /* Advance to the next chunk */
         gpa += chunk;
-        buffer += chunk;
+        off += chunk;
         size -= chunk;
 
         if ( size == 0 )
@@ -631,13 +601,41 @@ static int hvmemul_phys_mmio_access(
     return rc;
 }
 
+static struct hvm_mmio_cache *hvmemul_find_mmio_cache(
+    struct hvm_vcpu_io *vio, unsigned long gla, uint8_t dir)
+{
+    unsigned int i;
+    struct hvm_mmio_cache *cache;
+
+    for ( i = 0; i < vio->mmio_cache_count; i ++ )
+    {
+        cache = &vio->mmio_cache[i];
+
+        if ( gla == cache->gla &&
+             dir == cache->dir )
+            return cache;
+    }
+
+    i = vio->mmio_cache_count++;
+    BUG_ON(i == ARRAY_SIZE(vio->mmio_cache));
+
+    cache = &vio->mmio_cache[i];
+    memset(cache, 0, sizeof (*cache));
+
+    cache->gla = gla;
+    cache->dir = dir;
+
+    return cache;
+}
+
 static int hvmemul_linear_mmio_access(
-    unsigned long gla, unsigned int size, uint8_t dir, uint8_t *buffer,
+    unsigned long gla, unsigned int size, uint8_t dir, void *buffer,
     uint32_t pfec, struct hvm_emulate_ctxt *hvmemul_ctxt, bool_t known_gpfn)
 {
     struct hvm_vcpu_io *vio = &current->arch.hvm_vcpu.hvm_io;
     unsigned long page_off = gla & (PAGE_SIZE - 1);
-    unsigned int chunk;
+    struct hvm_mmio_cache *cache = hvmemul_find_mmio_cache(vio, gla, dir);
+    unsigned int chunk, buffer_off = 0;
     paddr_t gpa;
     unsigned long one_rep = 1;
     int rc;
@@ -656,12 +654,12 @@ static int hvmemul_linear_mmio_access(
 
     for ( ;; )
     {
-        rc = hvmemul_phys_mmio_access(gpa, chunk, dir, buffer);
+        rc = hvmemul_phys_mmio_access(cache, gpa, chunk, dir, buffer, buffer_off);
         if ( rc != X86EMUL_OKAY )
             break;
 
         gla += chunk;
-        buffer += chunk;
+        buffer_off += chunk;
         size -= chunk;
 
         if ( size == 0 )
@@ -1609,7 +1607,7 @@ static int _hvm_emulate_one(struct hvm_emulate_ctxt *hvmemul_ctxt,
         rc = X86EMUL_RETRY;
     if ( rc != X86EMUL_RETRY )
     {
-        vio->mmio_large_read_bytes = vio->mmio_large_write_bytes = 0;
+        vio->mmio_cache_count = 0;
         vio->mmio_insn_bytes = 0;
     }
     else
diff --git a/xen/include/asm-x86/hvm/vcpu.h b/xen/include/asm-x86/hvm/vcpu.h
index 506ce89..90dea8d 100644
--- a/xen/include/asm-x86/hvm/vcpu.h
+++ b/xen/include/asm-x86/hvm/vcpu.h
@@ -42,6 +42,17 @@ struct hvm_vcpu_asid {
     uint32_t asid;
 };
 
+/*
+ * We may read or write up to m256 as a number of device-model
+ * transactions.
+ */
+struct hvm_mmio_cache {
+    unsigned long size;
+    uint8_t buffer[32];
+    unsigned long gla;
+    uint8_t dir;
+};
+
 struct hvm_vcpu_io {
     /* I/O request in flight to device model. */
     ioreq_t                io_req;
@@ -57,13 +68,13 @@ struct hvm_vcpu_io {
     unsigned long       mmio_gva;
     unsigned long       mmio_gpfn;
 
-    /* We may read up to m256 as a number of device-model transactions. */
-    paddr_t mmio_large_read_pa;
-    uint8_t mmio_large_read[32];
-    unsigned int mmio_large_read_bytes;
-    /* We may write up to m256 as a number of device-model transactions. */
-    unsigned int mmio_large_write_bytes;
-    paddr_t mmio_large_write_pa;
+    /*
+     * We may need to handle up to 3 distinct memory accesses per
+     * instruction.
+     */
+    struct hvm_mmio_cache mmio_cache[3];
+    unsigned int mmio_cache_count;
+
     /* For retries we shouldn't re-fetch the instruction. */
     unsigned int mmio_insn_bytes;
     unsigned char mmio_insn[16];
-- 
1.7.10.4