diff --git a/drivers/dma-buf/dma-buf.c b/drivers/dma-buf/dma-buf.c
index 79d8b75310a8..d510b8fb5f06 100644
--- a/drivers/dma-buf/dma-buf.c
+++ b/drivers/dma-buf/dma-buf.c
@@ -688,6 +688,37 @@ void dma_buf_put(struct dma_buf *dmabuf)
 }
 EXPORT_SYMBOL_GPL(dma_buf_put);
 
+/**
+ * dma_buf_put_sync - decreases refcount of the buffer
+ * @dmabuf:	[in]	buffer to reduce refcount of
+ *
+ * Uses file's refcounting done implicitly by __fput_sync().
+ *
+ * If, as a result of this call, the refcount becomes 0, the 'release' file
+ * operation related to this fd is called. It calls &dma_buf_ops.release vfunc
+ * in turn, and frees the memory allocated for dmabuf when exported.
+ *
+ * This function is different than dma_buf_put() in the sense that it guarantees
+ * that the 'release' file operation related to this fd is called, and that the
+ * memory is released, when the refcount becomes 0. dma_buf_put() does not
+ * have the same guarantee when invoked by a kernel thread (e.g. a worker
+ * thread), and the refcount reaches 0; in that case, the buffer is added to
+ * the delayed_fput_list, and freed asynchronously.
+ *
+ * This function should not be called in atomic context, and should only be
+ * called by kernel threads. If in doubt, use dma_buf_put().
+ */
+void dma_buf_put_sync(struct dma_buf *dmabuf)
+{
+	if (WARN_ON(!dmabuf || !dmabuf->file))
+		return;
+
+	might_sleep();
+
+	dma_buf_ref_mod(to_msm_dma_buf(dmabuf), -1);
+	__fput_sync(dmabuf->file);
+}
+
 /**
  * dma_buf_attach - Add the device to dma_buf's attachments list; optionally,
  * calls attach() of dma_buf_ops to allow device-specific attach functionality
diff --git a/drivers/soc/qcom/mem-buf.c b/drivers/soc/qcom/mem-buf.c
index 2c9228f34155..92bf52b7e7ce 100644
--- a/drivers/soc/qcom/mem-buf.c
+++ b/drivers/soc/qcom/mem-buf.c
@@ -277,6 +277,12 @@ static int mem_buf_rmt_alloc_ion_mem(struct mem_buf_xfer_mem *xfer_mem)
 		xfer_mem->secure_alloc = false;
 	}
 
+	/*
+	 * If the buffer needs to be freed because of error handling, ensure
+	 * that dma_buf_put_sync() is invoked, instead of dma_buf_put(). Doing
+	 * so ensures that the memory is freed before the next allocation
+	 * request is serviced.
+	 */
 	dmabuf = ion_alloc(xfer_mem->size, heap_id, ion_flags);
 	if (IS_ERR(dmabuf)) {
 		pr_err("%s ion_alloc failure sz: 0x%x heap_id: %d flags: 0x%x rc: %d\n",
@@ -289,7 +295,7 @@ static int mem_buf_rmt_alloc_ion_mem(struct mem_buf_xfer_mem *xfer_mem)
 	if (IS_ERR(attachment)) {
 		pr_err("%s dma_buf_attach failure rc: %d\n",  __func__,
 		       PTR_ERR(attachment));
-		dma_buf_put(dmabuf);
+		dma_buf_put_sync(dmabuf);
 		return PTR_ERR(attachment);
 	}
 
@@ -298,7 +304,7 @@ static int mem_buf_rmt_alloc_ion_mem(struct mem_buf_xfer_mem *xfer_mem)
 		pr_err("%s dma_buf_map_attachment failure rc: %d\n", __func__,
 		       PTR_ERR(mem_sgt));
 		dma_buf_detach(dmabuf, attachment);
-		dma_buf_put(dmabuf);
+		dma_buf_put_sync(dmabuf);
 		return PTR_ERR(mem_sgt);
 	}
 
@@ -330,7 +336,11 @@ static void mem_buf_rmt_free_ion_mem(struct mem_buf_xfer_mem *xfer_mem)
 	pr_debug("%s: Freeing ION memory\n", __func__);
 	dma_buf_unmap_attachment(attachment, mem_sgt, DMA_BIDIRECTIONAL);
 	dma_buf_detach(dmabuf, attachment);
-	dma_buf_put(ion_mem_data->dmabuf);
+	/*
+	 * Use dma_buf_put_sync() instead of dma_buf_put() to ensure that the
+	 * memory is actually freed, before the next allocation request.
+	 */
+	dma_buf_put_sync(ion_mem_data->dmabuf);
 	pr_debug("%s: ION memory freed\n", __func__);
 }
 
@@ -2146,7 +2156,7 @@ static int mem_buf_probe(struct platform_device *pdev)
 		return ret;
 	}
 
-	mem_buf_wq = alloc_workqueue("mem_buf_wq", WQ_HIGHPRI | WQ_UNBOUND, 0);
+	mem_buf_wq = alloc_ordered_workqueue("mem_buf_wq", WQ_HIGHPRI);
 	if (!mem_buf_wq) {
 		dev_err(dev, "Unable to initialize workqueue\n");
 		return -EINVAL;
diff --git a/include/linux/dma-buf.h b/include/linux/dma-buf.h
index a7da54d72150..accce2b534e7 100644
--- a/include/linux/dma-buf.h
+++ b/include/linux/dma-buf.h
@@ -531,6 +531,7 @@ struct dma_buf *dma_buf_export(const struct dma_buf_export_info *exp_info);
 int dma_buf_fd(struct dma_buf *dmabuf, int flags);
 struct dma_buf *dma_buf_get(int fd);
 void dma_buf_put(struct dma_buf *dmabuf);
+void dma_buf_put_sync(struct dma_buf *dmabuf);
 
 struct sg_table *dma_buf_map_attachment(struct dma_buf_attachment *,
 					enum dma_data_direction);