diff --git a/src/gpu/vk/GrVkBuffer.cpp b/src/gpu/vk/GrVkBuffer.cpp
index 54713cb..f65b15d 100644
--- a/src/gpu/vk/GrVkBuffer.cpp
+++ b/src/gpu/vk/GrVkBuffer.cpp
@@ -206,7 +206,13 @@
     SkASSERT(this->vkIsMapped());
 
     if (fDesc.fDynamic) {
-        GrVkMemory::FlushMappedAlloc(gpu, this->alloc(), fMappedSize);
+        // We currently don't use fOffset
+        SkASSERT(0 == fOffset);
+        VkDeviceSize flushOffset = this->alloc().fOffset + fOffset;
+        VkDeviceSize flushSize = gpu->vkCaps().canUseWholeSizeOnFlushMappedMemory() ? VK_WHOLE_SIZE
+                                                                                    : fMappedSize;
+
+        GrVkMemory::FlushMappedAlloc(gpu, this->alloc(), flushOffset, flushSize);
         VK_CALL(gpu, UnmapMemory(gpu->device(), this->alloc().fMemory));
         fMapPtr = nullptr;
         fMappedSize = 0;
diff --git a/src/gpu/vk/GrVkGpu.cpp b/src/gpu/vk/GrVkGpu.cpp
index e4a980a..bfd5593 100644
--- a/src/gpu/vk/GrVkGpu.cpp
+++ b/src/gpu/vk/GrVkGpu.cpp
@@ -615,7 +615,7 @@
                      height);
     }
 
-    GrVkMemory::FlushMappedAlloc(this, alloc, size);
+    GrVkMemory::FlushMappedAlloc(this, alloc, offset, size);
     GR_VK_CALL(interface, UnmapMemory(fDevice, alloc.fMemory));
 
     return true;
@@ -1169,7 +1169,7 @@
             }
         }
     }
-    GrVkMemory::FlushMappedAlloc(gpu, alloc, mapSize);
+    GrVkMemory::FlushMappedAlloc(gpu, alloc, mapOffset, mapSize);
     GR_VK_CALL(gpu->vkInterface(), UnmapMemory(gpu->device(), alloc.fMemory));
     return true;
 }
@@ -2015,7 +2015,8 @@
     // we can copy the data out of the buffer.
     this->submitCommandBuffer(kForce_SyncQueue);
     void* mappedMemory = transferBuffer->map();
-    GrVkMemory::InvalidateMappedAlloc(this, transferBuffer->alloc());
+    const GrVkAlloc& transAlloc = transferBuffer->alloc();
+    GrVkMemory::InvalidateMappedAlloc(this, transAlloc, transAlloc.fOffset, VK_WHOLE_SIZE);
 
     if (copyFromOrigin) {
         uint32_t skipRows = region.imageExtent.height - height;
diff --git a/src/gpu/vk/GrVkMemory.cpp b/src/gpu/vk/GrVkMemory.cpp
index d744a7a..e391d02 100644
--- a/src/gpu/vk/GrVkMemory.cpp
+++ b/src/gpu/vk/GrVkMemory.cpp
@@ -298,32 +298,53 @@
     return flags;
 }
 
-void GrVkMemory::FlushMappedAlloc(const GrVkGpu* gpu, const GrVkAlloc& alloc, VkDeviceSize size) {
+void GrVkMemory::FlushMappedAlloc(const GrVkGpu* gpu, const GrVkAlloc& alloc, VkDeviceSize offset,
+                                  VkDeviceSize size) {
     if (alloc.fFlags & GrVkAlloc::kNoncoherent_Flag) {
+#ifdef SK_DEBUG
+        SkASSERT(offset >= alloc.fOffset);
+        VkDeviceSize alignment = gpu->physicalDeviceProperties().limits.nonCoherentAtomSize;
+        SkASSERT(0 == (offset & (alignment-1)));
+        if (size != VK_WHOLE_SIZE) {
+            SkASSERT(size > 0);
+            SkASSERT(0 == (size & (alignment-1)) ||
+                     (offset + size) == (alloc.fOffset + alloc.fSize));
+            SkASSERT(offset + size <= alloc.fOffset + alloc.fSize);
+        }
+#endif
+
         VkMappedMemoryRange mappedMemoryRange;
         memset(&mappedMemoryRange, 0, sizeof(VkMappedMemoryRange));
         mappedMemoryRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE;
         mappedMemoryRange.memory = alloc.fMemory;
-        mappedMemoryRange.offset = alloc.fOffset;
-        if (gpu->vkCaps().canUseWholeSizeOnFlushMappedMemory()) {
-            mappedMemoryRange.size = VK_WHOLE_SIZE; // Size of what we mapped
-        } else {
-            SkASSERT(size > 0);
-            mappedMemoryRange.size = size;
-        }
+        mappedMemoryRange.offset = offset;
+        mappedMemoryRange.size = size;
         GR_VK_CALL(gpu->vkInterface(), FlushMappedMemoryRanges(gpu->device(),
                                                                1, &mappedMemoryRange));
     }
 }
 
-void GrVkMemory::InvalidateMappedAlloc(const GrVkGpu* gpu, const GrVkAlloc& alloc) {
+void GrVkMemory::InvalidateMappedAlloc(const GrVkGpu* gpu, const GrVkAlloc& alloc,
+                                       VkDeviceSize offset, VkDeviceSize size) {
     if (alloc.fFlags & GrVkAlloc::kNoncoherent_Flag) {
+#ifdef SK_DEBUG
+        SkASSERT(offset >= alloc.fOffset);
+        VkDeviceSize alignment = gpu->physicalDeviceProperties().limits.nonCoherentAtomSize;
+        SkASSERT(0 == (offset & (alignment-1)));
+        if (size != VK_WHOLE_SIZE) {
+            SkASSERT(size > 0);
+            SkASSERT(0 == (size & (alignment-1)) ||
+                     (offset + size) == (alloc.fOffset + alloc.fSize));
+            SkASSERT(offset + size <= alloc.fOffset + alloc.fSize);
+        }
+#endif
+
         VkMappedMemoryRange mappedMemoryRange;
         memset(&mappedMemoryRange, 0, sizeof(VkMappedMemoryRange));
         mappedMemoryRange.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE;
         mappedMemoryRange.memory = alloc.fMemory;
-        mappedMemoryRange.offset = alloc.fOffset;
-        mappedMemoryRange.size = VK_WHOLE_SIZE; // Size of what we mapped
+        mappedMemoryRange.offset = offset;
+        mappedMemoryRange.size = size;
         GR_VK_CALL(gpu->vkInterface(), InvalidateMappedMemoryRanges(gpu->device(),
                                                                1, &mappedMemoryRange));
     }
diff --git a/src/gpu/vk/GrVkMemory.h b/src/gpu/vk/GrVkMemory.h
index 8dd43bb..baf843e 100644
--- a/src/gpu/vk/GrVkMemory.h
+++ b/src/gpu/vk/GrVkMemory.h
@@ -38,8 +38,10 @@
 
     VkAccessFlags LayoutToSrcAccessMask(const VkImageLayout layout);
 
-    void FlushMappedAlloc(const GrVkGpu* gpu, const GrVkAlloc& alloc, VkDeviceSize size);
-    void InvalidateMappedAlloc(const GrVkGpu* gpu, const GrVkAlloc& alloc);
+    void FlushMappedAlloc(const GrVkGpu* gpu, const GrVkAlloc& alloc, VkDeviceSize offset,
+                          VkDeviceSize size);
+    void InvalidateMappedAlloc(const GrVkGpu* gpu, const GrVkAlloc& alloc, VkDeviceSize offset,
+                               VkDeviceSize size);
 }
 
 class GrVkFreeListAlloc {