[PATCH v4 17/17] 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 then 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     |   98 ++++++++++++++++------------------------
 xen/include/asm-x86/hvm/vcpu.h |   16 ++++---
 2 files changed, 48 insertions(+), 66 deletions(-)

diff --git a/xen/arch/x86/hvm/emulate.c b/xen/arch/x86/hvm/emulate.c
index aa68787..4424dfc 100644
--- a/xen/arch/x86/hvm/emulate.c
+++ b/xen/arch/x86/hvm/emulate.c
@@ -107,29 +107,6 @@ 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:
@@ -209,33 +186,6 @@ 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;
 }
 
@@ -601,8 +551,11 @@ static int hvmemul_virtual_to_linear(
 }
 
 static int hvmemul_phys_mmio_access(
-    paddr_t gpa, unsigned int size, uint8_t dir, uint8_t **buffer)
+    paddr_t gpa, unsigned int size, uint8_t dir, uint8_t *buffer,
+    unsigned int *off)
 {
+    struct vcpu *curr = current;
+    struct hvm_vcpu_io *vio = &curr->arch.hvm_vcpu.hvm_io;
     unsigned long one_rep = 1;
     unsigned int chunk;
     int rc;
@@ -621,14 +574,41 @@ 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 + chunk) <= vio->mmio_cache[dir].size )
+        {
+            ASSERT(*off + chunk <= vio->mmio_cache[dir].size);
+
+            if ( dir == IOREQ_READ )
+                memcpy(&buffer[*off],
+                       &vio->mmio_cache[IOREQ_READ].buffer[*off],
+                       chunk);
+            else
+            {
+                if ( memcmp(&buffer[*off],
+                            &vio->mmio_cache[IOREQ_WRITE].buffer[*off],
+                            chunk) != 0 )
+                    domain_crash(curr->domain);
+            }
+        }
+        else
+        {
+            ASSERT(*off == vio->mmio_cache[dir].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(&vio->mmio_cache[dir].buffer[*off],
+                   &buffer[*off], chunk);
+            vio->mmio_cache[dir].size += chunk;
+        }
 
         /* Advance to the next chunk */
         gpa += chunk;
-        *buffer += chunk;
+        *off += chunk;
         size -= chunk;
 
         if ( size == 0 )
@@ -651,7 +631,7 @@ static int hvmemul_linear_mmio_access(
 {
     struct hvm_vcpu_io *vio = &current->arch.hvm_vcpu.hvm_io;
     unsigned long page_off = gla & (PAGE_SIZE - 1);
-    unsigned int chunk;
+    unsigned int chunk, buffer_off = 0;
     paddr_t gpa;
     unsigned long one_rep = 1;
     int rc;
@@ -670,7 +650,7 @@ static int hvmemul_linear_mmio_access(
 
     for ( ;; )
     {
-        rc = hvmemul_phys_mmio_access(gpa, chunk, dir, &buffer);
+        rc = hvmemul_phys_mmio_access(gpa, chunk, dir, buffer, &buffer_off);
         if ( rc != X86EMUL_OKAY )
             break;
 
@@ -1625,7 +1605,7 @@ 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;
+        memset(&vio->mmio_cache, 0, sizeof(vio->mmio_cache));
         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 008c8fa..4f41c83 100644
--- a/xen/include/asm-x86/hvm/vcpu.h
+++ b/xen/include/asm-x86/hvm/vcpu.h
@@ -61,13 +61,15 @@ 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 read or write up to m256 as a number of device-model
+     * transactions.
+     */
+    struct {
+        unsigned long size;
+        uint8_t buffer[32];
+    } mmio_cache[2]; /* Indexed by ioreq type */
+
     /* For retries we shouldn't re-fetch the instruction. */
     unsigned int mmio_insn_bytes;
     unsigned char mmio_insn[16];
-- 
1.7.10.4