GZIP compression offloading with QAT accelerator

This patch implement the hardware accelerator method in GZIP compression
in ZFS. When the ZFS pool is enabled GZIP compression, the compression
API will be automatically transferred to the hardware accelerator to
free up CPU resource and speed up the compression time.

* To enable Intel QAT hardware acceleration in ZOL you need to have QAT
  hardware and the driver installed:
  * QAT hardware DH8950:
  http://ark.intel.com/products/79483/Intel-QuickAssist-Adapter-8950
  * QAT driver:
  https://01.org/intel-quickassist-technology
* Start QAT driver in your system:
  service qat_service start
* Enable QAT in ZFS, e.g.:
  ./configure --with-qat=<qat-driver-path>/QAT1.6
  make
* Set GZIP compression in ZFS dataset:
  zfs set compression = gzip <dataset>
* Get QAT hardware statistics by:
  cat /proc/spl/kstat/zfs/qat
* To disable QAT in ZFS:
  insmod zfs.ko zfs_qat_disable=1

Reviewed-by: Brian Behlendorf <[email protected]>
Reviewed-by: Jinshan Xiong <[email protected]>
Signed-off-by: Weigang Li <[email protected]>
Closes #5846 

6 files changed, 656 insertions, 0 deletions

diff --git a/module/Makefile.in b/module/Makefile.in
index 1c1ea032e..093cf2286 100644
--- a/module/Makefile.in
+++ b/module/Makefile.in
@@ -11,6 +11,8 @@ INSTALL_MOD_DIR ?= extra
 ZFS_MODULE_CFLAGS += -include @SPL_OBJ@/spl_config.h
 ZFS_MODULE_CFLAGS += -include @abs_top_builddir@/zfs_config.h
 ZFS_MODULE_CFLAGS += -I@abs_top_srcdir@/include -I@SPL@/include -I@SPL@
+@CONFIG_QAT_TRUE@ZFS_MODULE_CFLAGS += -I@QAT_SRC@/include
+@CONFIG_QAT_TRUE@KBUILD_EXTRA_SYMBOLS += @QAT_SYMBOLS@
 export ZFS_MODULE_CFLAGS
 
 SUBDIR_TARGETS = icp
diff --git a/module/zfs/Makefile.in b/module/zfs/Makefile.in
index 6712b9b3c..f8d54f4dd 100644
--- a/module/zfs/Makefile.in
+++ b/module/zfs/Makefile.in
@@ -114,6 +114,7 @@ $(MODULE)-objs += zrlock.o
 $(MODULE)-objs += zvol.o
 $(MODULE)-objs += dsl_destroy.o
 $(MODULE)-objs += dsl_userhold.o
+$(MODULE)-objs += qat_compress.o
 
 $(MODULE)-$(CONFIG_X86) += vdev_raidz_math_sse2.o
 $(MODULE)-$(CONFIG_X86) += vdev_raidz_math_ssse3.o
diff --git a/module/zfs/gzip.c b/module/zfs/gzip.c
index 6e5c859fe..6c8fdd308 100644
--- a/module/zfs/gzip.c
+++ b/module/zfs/gzip.c
@@ -28,6 +28,7 @@
 
 #include <sys/debug.h>
 #include <sys/types.h>
+#include "qat_compress.h"
 
 #ifdef _KERNEL
 
@@ -56,6 +57,14 @@ gzip_compress(void *s_start, void *d_start, size_t s_len, size_t d_len, int n)
 
 	ASSERT(d_len <= s_len);
 
+	/* check if hardware accelerator can be used */
+	if (qat_use_accel(s_len)) {
+		if (qat_compress(QAT_COMPRESS, s_start,
+		    s_len, d_start, d_len, &dstlen) == CPA_STATUS_SUCCESS)
+			return ((size_t)dstlen);
+		/* if hardware compress fail, do it again with software */
+	}
+
 	if (compress_func(d_start, &dstlen, s_start, s_len, n) != Z_OK) {
 		if (d_len != s_len)
 			return (s_len);
@@ -75,6 +84,14 @@ gzip_decompress(void *s_start, void *d_start, size_t s_len, size_t d_len, int n)
 
 	ASSERT(d_len >= s_len);
 
+	/* check if hardware accelerator can be used */
+	if (qat_use_accel(d_len)) {
+		if (qat_compress(QAT_DECOMPRESS, s_start, s_len,
+		    d_start, d_len, &dstlen) == CPA_STATUS_SUCCESS)
+			return (0);
+		/* if hardware de-compress fail, do it again with software */
+	}
+
 	if (uncompress_func(d_start, &dstlen, s_start, s_len) != Z_OK)
 		return (-1);
 
diff --git a/module/zfs/qat_compress.c b/module/zfs/qat_compress.c
new file mode 100644
index 000000000..a350c0f3e
--- /dev/null
+++ b/module/zfs/qat_compress.c
@@ -0,0 +1,585 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+#if defined(_KERNEL) && defined(HAVE_QAT)
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/pagemap.h>
+#include <linux/completion.h>
+#include <sys/zfs_context.h>
+#include "qat_compress.h"
+
+/*
+ * Timeout - no response from hardware after 0.5 seconds
+ */
+#define	TIMEOUT_MS		500
+
+/*
+ * Max instances in QAT device, each instance is a channel to submit
+ * jobs to QAT hardware
+ */
+#define	MAX_INSTANCES		6
+
+/*
+ * ZLIB head and foot size
+ */
+#define	ZLIB_HEAD_SZ		2
+#define	ZLIB_FOOT_SZ		4
+
+/*
+ * The minimal and maximal buffer size, which are not restricted
+ * in the QAT hardware, but with the input buffer size between 4KB
+ * and 128KB, the hardware can provide the optimal performance.
+ */
+#define	QAT_MIN_BUF_SIZE	(4*1024)
+#define	QAT_MAX_BUF_SIZE	(128*1024)
+
+/*
+ * Used for qat kstat.
+ */
+typedef struct qat_stats {
+	/*
+	 * Number of jobs submitted to qat compression engine.
+	 */
+	kstat_named_t comp_requests;
+	/*
+	 * Total bytes sent to qat compression engine.
+	 */
+	kstat_named_t comp_total_in_bytes;
+	/*
+	 * Total bytes output from qat compression engine.
+	 */
+	kstat_named_t comp_total_out_bytes;
+	/*
+	 * Number of jobs submitted to qat de-compression engine.
+	 */
+	kstat_named_t decomp_requests;
+	/*
+	 * Total bytes sent to qat de-compression engine.
+	 */
+	kstat_named_t decomp_total_in_bytes;
+	/*
+	 * Total bytes output from qat de-compression engine.
+	 */
+	kstat_named_t decomp_total_out_bytes;
+	/*
+	 * Number of fails in qat engine.
+	 * Note: when qat fail happens, it doesn't mean a critical hardware
+	 * issue, sometimes it is because the output buffer is not big enough,
+	 * and the compression job will be transfered to gzip software again,
+	 * so the functionality of ZFS is not impacted.
+	 */
+	kstat_named_t dc_fails;
+} qat_stats_t;
+
+qat_stats_t qat_stats = {
+	{ "comp_reqests",			KSTAT_DATA_UINT64 },
+	{ "comp_total_in_bytes",		KSTAT_DATA_UINT64 },
+	{ "comp_total_out_bytes",		KSTAT_DATA_UINT64 },
+	{ "decomp_reqests",			KSTAT_DATA_UINT64 },
+	{ "decomp_total_in_bytes",		KSTAT_DATA_UINT64 },
+	{ "decomp_total_out_bytes",		KSTAT_DATA_UINT64 },
+	{ "dc_fails",				KSTAT_DATA_UINT64 },
+};
+
+static kstat_t *qat_ksp;
+static CpaInstanceHandle dc_inst_handles[MAX_INSTANCES];
+static CpaDcSessionHandle session_handles[MAX_INSTANCES];
+static CpaBufferList **buffer_array[MAX_INSTANCES];
+static Cpa16U num_inst = 0;
+static Cpa16U inst_num = 0;
+static boolean_t qat_init_done = B_FALSE;
+int zfs_qat_disable = 0;
+
+#define	QAT_STAT_INCR(stat, val) \
+	atomic_add_64(&qat_stats.stat.value.ui64, (val));
+#define	QAT_STAT_BUMP(stat) \
+	QAT_STAT_INCR(stat, 1);
+
+#define	PHYS_CONTIG_ALLOC(pp_mem_addr, size_bytes)	\
+	mem_alloc_contig((void *)(pp_mem_addr), (size_bytes))
+
+#define	PHYS_CONTIG_FREE(p_mem_addr)	\
+	mem_free_contig((void *)&(p_mem_addr))
+
+static inline struct page *
+mem_to_page(void *addr)
+{
+	if (!is_vmalloc_addr(addr))
+		return (virt_to_page(addr));
+
+	return (vmalloc_to_page(addr));
+}
+
+static void
+qat_dc_callback(void *p_callback, CpaStatus status)
+{
+	if (p_callback != NULL)
+		complete((struct completion *)p_callback);
+}
+
+static inline CpaStatus
+mem_alloc_contig(void **pp_mem_addr, Cpa32U size_bytes)
+{
+	*pp_mem_addr = kmalloc(size_bytes, GFP_KERNEL);
+	if (*pp_mem_addr == NULL)
+		return (CPA_STATUS_RESOURCE);
+	return (CPA_STATUS_SUCCESS);
+}
+
+static inline void
+mem_free_contig(void **pp_mem_addr)
+{
+	if (*pp_mem_addr != NULL) {
+		kfree(*pp_mem_addr);
+		*pp_mem_addr = NULL;
+	}
+}
+
+static void
+qat_clean(void)
+{
+	Cpa16U buff_num = 0;
+	Cpa16U num_inter_buff_lists = 0;
+	Cpa16U i = 0;
+
+	for (i = 0; i < num_inst; i++) {
+		cpaDcStopInstance(dc_inst_handles[i]);
+		PHYS_CONTIG_FREE(session_handles[i]);
+		/* free intermediate buffers  */
+		if (buffer_array[i] != NULL) {
+			cpaDcGetNumIntermediateBuffers(
+			    dc_inst_handles[i], &num_inter_buff_lists);
+			for (buff_num = 0; buff_num < num_inter_buff_lists;
+			    buff_num++) {
+				CpaBufferList *buffer_inter =
+				    buffer_array[i][buff_num];
+				if (buffer_inter->pBuffers) {
+					PHYS_CONTIG_FREE(
+					    buffer_inter->pBuffers->pData);
+					PHYS_CONTIG_FREE(
+					    buffer_inter->pBuffers);
+				}
+				PHYS_CONTIG_FREE(
+				    buffer_inter->pPrivateMetaData);
+				PHYS_CONTIG_FREE(buffer_inter);
+			}
+		}
+	}
+
+	num_inst = 0;
+	qat_init_done = B_FALSE;
+}
+
+int
+qat_init(void)
+{
+	CpaStatus status = CPA_STATUS_SUCCESS;
+	Cpa32U sess_size = 0;
+	Cpa32U ctx_size = 0;
+	Cpa16U num_inter_buff_lists = 0;
+	Cpa16U buff_num = 0;
+	Cpa32U buff_meta_size = 0;
+	CpaDcSessionSetupData sd = {0};
+	Cpa16U i;
+
+	status = cpaDcGetNumInstances(&num_inst);
+	if (status != CPA_STATUS_SUCCESS || num_inst == 0)
+		return (-1);
+
+	if (num_inst > MAX_INSTANCES)
+		num_inst = MAX_INSTANCES;
+
+	status = cpaDcGetInstances(num_inst, &dc_inst_handles[0]);
+	if (status != CPA_STATUS_SUCCESS)
+		return (-1);
+
+	for (i = 0; i < num_inst; i++) {
+		cpaDcSetAddressTranslation(dc_inst_handles[i],
+		    (void*)virt_to_phys);
+
+		status = cpaDcBufferListGetMetaSize(dc_inst_handles[i],
+		    1, &buff_meta_size);
+
+		if (status == CPA_STATUS_SUCCESS)
+			status = cpaDcGetNumIntermediateBuffers(
+			    dc_inst_handles[i], &num_inter_buff_lists);
+
+		if (status == CPA_STATUS_SUCCESS && num_inter_buff_lists != 0)
+			status = PHYS_CONTIG_ALLOC(&buffer_array[i],
+			    num_inter_buff_lists *
+			    sizeof (CpaBufferList *));
+
+		for (buff_num = 0; buff_num < num_inter_buff_lists;
+		    buff_num++) {
+			if (status == CPA_STATUS_SUCCESS)
+				status = PHYS_CONTIG_ALLOC(
+				    &buffer_array[i][buff_num],
+				    sizeof (CpaBufferList));
+
+			if (status == CPA_STATUS_SUCCESS)
+				status = PHYS_CONTIG_ALLOC(
+				    &buffer_array[i][buff_num]->
+				    pPrivateMetaData,
+				    buff_meta_size);
+
+			if (status == CPA_STATUS_SUCCESS)
+				status = PHYS_CONTIG_ALLOC(
+				    &buffer_array[i][buff_num]->pBuffers,
+				    sizeof (CpaFlatBuffer));
+
+			if (status == CPA_STATUS_SUCCESS) {
+				/*
+				 *  implementation requires an intermediate
+				 *  buffer approximately twice the size of
+				 *  output buffer, which is 2x max buffer
+				 *  size here.
+				 */
+				status = PHYS_CONTIG_ALLOC(
+				    &buffer_array[i][buff_num]->pBuffers->
+				    pData, 2 * QAT_MAX_BUF_SIZE);
+				if (status != CPA_STATUS_SUCCESS)
+					goto fail;
+
+				buffer_array[i][buff_num]->numBuffers = 1;
+				buffer_array[i][buff_num]->pBuffers->
+				    dataLenInBytes = 2 * QAT_MAX_BUF_SIZE;
+			}
+		}
+
+		status = cpaDcStartInstance(dc_inst_handles[i],
+		    num_inter_buff_lists, buffer_array[i]);
+		if (status != CPA_STATUS_SUCCESS)
+			goto fail;
+
+		sd.compLevel = CPA_DC_L1;
+		sd.compType = CPA_DC_DEFLATE;
+		sd.huffType = CPA_DC_HT_FULL_DYNAMIC;
+		sd.sessDirection = CPA_DC_DIR_COMBINED;
+		sd.sessState = CPA_DC_STATELESS;
+		sd.deflateWindowSize = 7;
+		sd.checksum = CPA_DC_ADLER32;
+		status = cpaDcGetSessionSize(dc_inst_handles[i],
+		    &sd, &sess_size, &ctx_size);
+		if (status != CPA_STATUS_SUCCESS)
+			goto fail;
+
+		PHYS_CONTIG_ALLOC(&session_handles[i], sess_size);
+		if (session_handles[i] == NULL)
+			goto fail;
+
+		status = cpaDcInitSession(dc_inst_handles[i],
+		    session_handles[i],
+		    &sd, NULL, qat_dc_callback);
+		if (status != CPA_STATUS_SUCCESS)
+			goto fail;
+	}
+
+	qat_ksp = kstat_create("zfs", 0, "qat", "misc",
+	    KSTAT_TYPE_NAMED, sizeof (qat_stats) / sizeof (kstat_named_t),
+	    KSTAT_FLAG_VIRTUAL);
+	if (qat_ksp != NULL) {
+		qat_ksp->ks_data = &qat_stats;
+		kstat_install(qat_ksp);
+	}
+
+	qat_init_done = B_TRUE;
+	return (0);
+fail:
+	qat_clean();
+	return (-1);
+}
+
+void
+qat_fini(void)
+{
+	qat_clean();
+
+	if (qat_ksp != NULL) {
+		kstat_delete(qat_ksp);
+		qat_ksp = NULL;
+	}
+}
+
+boolean_t
+qat_use_accel(size_t s_len)
+{
+	return (!zfs_qat_disable &&
+	    qat_init_done &&
+	    s_len >= QAT_MIN_BUF_SIZE &&
+	    s_len <= QAT_MAX_BUF_SIZE);
+}
+
+int
+qat_compress(qat_compress_dir_t dir, char *src, int src_len,
+    char *dst, int dst_len, size_t *c_len)
+{
+	CpaInstanceHandle dc_inst_handle;
+	CpaDcSessionHandle session_handle;
+	CpaBufferList *buf_list_src = NULL;
+	CpaBufferList *buf_list_dst = NULL;
+	CpaFlatBuffer *flat_buf_src = NULL;
+	CpaFlatBuffer *flat_buf_dst = NULL;
+	Cpa8U *buffer_meta_src = NULL;
+	Cpa8U *buffer_meta_dst = NULL;
+	Cpa32U buffer_meta_size = 0;
+	CpaDcRqResults dc_results;
+	CpaStatus status = CPA_STATUS_SUCCESS;
+	Cpa32U hdr_sz = 0;
+	Cpa32U compressed_sz;
+	Cpa32U num_src_buf = (src_len >> PAGE_SHIFT) + 1;
+	Cpa32U num_dst_buf = (dst_len >> PAGE_SHIFT) + 1;
+	Cpa32U bytes_left;
+	char *data;
+	struct page *in_page, *out_page;
+	struct page **in_pages = NULL;
+	struct page **out_pages = NULL;
+	struct completion complete;
+	size_t ret = -1;
+	Cpa16U page_num = 0;
+	Cpa16U i;
+
+	Cpa32U src_buffer_list_mem_size = sizeof (CpaBufferList) +
+	    (num_src_buf * sizeof (CpaFlatBuffer));
+	Cpa32U dst_buffer_list_mem_size = sizeof (CpaBufferList) +
+	    (num_dst_buf * sizeof (CpaFlatBuffer));
+
+	if (!is_vmalloc_addr(src) || !is_vmalloc_addr(src + src_len - 1) ||
+	    !is_vmalloc_addr(dst) || !is_vmalloc_addr(dst + dst_len - 1))
+		return (-1);
+
+	if (PHYS_CONTIG_ALLOC(&in_pages,
+	    num_src_buf * sizeof (struct page *)) != CPA_STATUS_SUCCESS)
+		goto fail;
+
+	if (PHYS_CONTIG_ALLOC(&out_pages,
+	    num_dst_buf * sizeof (struct page *)) != CPA_STATUS_SUCCESS)
+		goto fail;
+
+	i = atomic_inc_32_nv(&inst_num) % num_inst;
+	dc_inst_handle = dc_inst_handles[i];
+	session_handle = session_handles[i];
+
+	cpaDcBufferListGetMetaSize(dc_inst_handle, num_src_buf,
+	    &buffer_meta_size);
+	if (PHYS_CONTIG_ALLOC(&buffer_meta_src, buffer_meta_size) !=
+	    CPA_STATUS_SUCCESS)
+		goto fail;
+
+	cpaDcBufferListGetMetaSize(dc_inst_handle, num_dst_buf,
+	    &buffer_meta_size);
+	if (PHYS_CONTIG_ALLOC(&buffer_meta_dst, buffer_meta_size) !=
+	    CPA_STATUS_SUCCESS)
+		goto fail;
+
+	/* build source buffer list */
+	if (PHYS_CONTIG_ALLOC(&buf_list_src, src_buffer_list_mem_size) !=
+	    CPA_STATUS_SUCCESS)
+		goto fail;
+
+	flat_buf_src = (CpaFlatBuffer *)(buf_list_src + 1);
+
+	buf_list_src->pBuffers = flat_buf_src; /* always point to first one */
+
+	/* build destination buffer list */
+	if (PHYS_CONTIG_ALLOC(&buf_list_dst, dst_buffer_list_mem_size) !=
+	    CPA_STATUS_SUCCESS)
+		goto fail;
+
+	flat_buf_dst = (CpaFlatBuffer *)(buf_list_dst + 1);
+
+	buf_list_dst->pBuffers = flat_buf_dst; /* always point to first one */
+
+	buf_list_src->numBuffers = 0;
+	buf_list_src->pPrivateMetaData = buffer_meta_src;
+	bytes_left = src_len;
+	data = src;
+	page_num = 0;
+	while (bytes_left > 0) {
+		in_page = mem_to_page(data);
+		in_pages[page_num] = in_page;
+		flat_buf_src->pData = kmap(in_page);
+		flat_buf_src->dataLenInBytes =
+		    min((long)bytes_left, (long)PAGE_SIZE);
+
+		bytes_left -= flat_buf_src->dataLenInBytes;
+		data += flat_buf_src->dataLenInBytes;
+		flat_buf_src++;
+		buf_list_src->numBuffers++;
+		page_num++;
+	}
+
+	buf_list_dst->numBuffers = 0;
+	buf_list_dst->pPrivateMetaData = buffer_meta_dst;
+	bytes_left = dst_len;
+	data = dst;
+	page_num = 0;
+	while (bytes_left > 0) {
+		out_page = mem_to_page(data);
+		flat_buf_dst->pData = kmap(out_page);
+		out_pages[page_num] = out_page;
+		flat_buf_dst->dataLenInBytes =
+		    min((long)bytes_left, (long)PAGE_SIZE);
+
+		bytes_left -= flat_buf_dst->dataLenInBytes;
+		data += flat_buf_dst->dataLenInBytes;
+		flat_buf_dst++;
+		buf_list_dst->numBuffers++;
+		page_num++;
+	}
+
+	init_completion(&complete);
+
+	if (dir == QAT_COMPRESS) {
+		QAT_STAT_BUMP(comp_requests);
+		QAT_STAT_INCR(comp_total_in_bytes, src_len);
+
+		cpaDcGenerateHeader(session_handle,
+		    buf_list_dst->pBuffers, &hdr_sz);
+		buf_list_dst->pBuffers->pData += hdr_sz;
+		buf_list_dst->pBuffers->dataLenInBytes -= hdr_sz;
+		status = cpaDcCompressData(
+		    dc_inst_handle, session_handle,
+		    buf_list_src, buf_list_dst,
+		    &dc_results, CPA_DC_FLUSH_FINAL,
+		    &complete);
+		if (status != CPA_STATUS_SUCCESS) {
+			goto fail;
+		}
+
+		/* we now wait until the completion of the operation. */
+		if (!wait_for_completion_interruptible_timeout(&complete,
+		    TIMEOUT_MS)) {
+			status = CPA_STATUS_FAIL;
+			goto fail;
+		}
+
+		if (dc_results.status != CPA_STATUS_SUCCESS) {
+			status = CPA_STATUS_FAIL;
+			goto fail;
+		}
+
+		compressed_sz = dc_results.produced;
+		if (compressed_sz + hdr_sz + ZLIB_FOOT_SZ > dst_len) {
+			goto fail;
+		}
+
+		flat_buf_dst = (CpaFlatBuffer *)(buf_list_dst + 1);
+		/* move to the last page */
+		flat_buf_dst += (compressed_sz + hdr_sz) >> PAGE_SHIFT;
+
+		/* no space for gzip foot in the last page */
+		if (((compressed_sz + hdr_sz) % PAGE_SIZE)
+		    + ZLIB_FOOT_SZ > PAGE_SIZE)
+			goto fail;
+
+		flat_buf_dst->pData += (compressed_sz + hdr_sz) % PAGE_SIZE;
+		flat_buf_dst->dataLenInBytes = ZLIB_FOOT_SZ;
+
+		dc_results.produced = 0;
+		status = cpaDcGenerateFooter(session_handle,
+		    flat_buf_dst, &dc_results);
+		if (status != CPA_STATUS_SUCCESS) {
+			goto fail;
+		}
+
+		*c_len = compressed_sz + dc_results.produced + hdr_sz;
+
+		if (*c_len < PAGE_SIZE)
+			*c_len = 8 * PAGE_SIZE;
+
+		QAT_STAT_INCR(comp_total_out_bytes, *c_len);
+
+		ret = 0;
+
+	} else if (dir == QAT_DECOMPRESS) {
+		QAT_STAT_BUMP(decomp_requests);
+		QAT_STAT_INCR(decomp_total_in_bytes, src_len);
+
+		buf_list_src->pBuffers->pData += ZLIB_HEAD_SZ;
+		buf_list_src->pBuffers->dataLenInBytes -= ZLIB_HEAD_SZ;
+		status = cpaDcDecompressData(dc_inst_handle,
+		    session_handle,
+		    buf_list_src,
+		    buf_list_dst,
+		    &dc_results,
+		    CPA_DC_FLUSH_FINAL,
+		    &complete);
+
+		if (CPA_STATUS_SUCCESS != status) {
+			status = CPA_STATUS_FAIL;
+			goto fail;
+		}
+
+		/* we now wait until the completion of the operation. */
+		if (!wait_for_completion_interruptible_timeout(&complete,
+		    TIMEOUT_MS)) {
+			status = CPA_STATUS_FAIL;
+			goto fail;
+		}
+
+		if (dc_results.status != CPA_STATUS_SUCCESS) {
+			status = CPA_STATUS_FAIL;
+			goto fail;
+		}
+
+		*c_len = dc_results.produced;
+
+		QAT_STAT_INCR(decomp_total_out_bytes, *c_len);
+
+		ret = 0;
+	}
+
+fail:
+	if (status != CPA_STATUS_SUCCESS) {
+		QAT_STAT_BUMP(dc_fails);
+	}
+
+	if (in_pages) {
+		for (page_num = 0;
+		    page_num < buf_list_src->numBuffers;
+		    page_num++) {
+			kunmap(in_pages[page_num]);
+		}
+		PHYS_CONTIG_FREE(in_pages);
+	}
+
+	if (out_pages) {
+		for (page_num = 0;
+		    page_num < buf_list_dst->numBuffers;
+		    page_num++) {
+			kunmap(out_pages[page_num]);
+		}
+		PHYS_CONTIG_FREE(out_pages);
+	}
+
+	PHYS_CONTIG_FREE(buffer_meta_src);
+	PHYS_CONTIG_FREE(buffer_meta_dst);
+	PHYS_CONTIG_FREE(buf_list_src);
+	PHYS_CONTIG_FREE(buf_list_dst);
+
+	return (ret);
+}
+
+module_param(zfs_qat_disable, int, 0644);
+MODULE_PARM_DESC(zfs_qat_disable, "Disable QAT compression");
+
+#endif
diff --git a/module/zfs/qat_compress.h b/module/zfs/qat_compress.h
new file mode 100644
index 000000000..ff074646f
--- /dev/null
+++ b/module/zfs/qat_compress.h
@@ -0,0 +1,48 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+#ifndef	_SYS_QAT_COMPRESS_H
+#define	_SYS_QAT_COMPRESS_H
+
+#if defined(_KERNEL) && defined(HAVE_QAT)
+#include <sys/zio.h>
+#include "cpa.h"
+#include "dc/cpa_dc.h"
+
+typedef enum qat_compress_dir {
+	QAT_COMPRESS = 0,
+	QAT_DECOMPRESS = 1,
+} qat_compress_dir_t;
+
+extern int qat_init(void);
+extern void qat_fini(void);
+extern boolean_t qat_use_accel(size_t s_len);
+extern int qat_compress(qat_compress_dir_t dir, char *src, int src_len,
+    char *dst, int dst_len, size_t *c_len);
+#else
+#define	CPA_STATUS_SUCCESS	0
+#define	qat_init()
+#define	qat_fini()
+#define	qat_use_accel(s_len)	0
+#define	qat_compress(dir, s, sl, d, dl, cl)	0
+#endif
+
+#endif /* _SYS_QAT_COMPRESS_H */
diff --git a/module/zfs/spa_misc.c b/module/zfs/spa_misc.c
index 230c52156..ea16e69f0 100644
--- a/module/zfs/spa_misc.c
+++ b/module/zfs/spa_misc.c
@@ -55,6 +55,7 @@
 #include <sys/kstat.h>
 #include "zfs_prop.h"
 #include <sys/zfeature.h>
+#include "qat_compress.h"
 
 /*
  * SPA locking
@@ -1863,6 +1864,7 @@ spa_init(int mode)
 	zpool_feature_init();
 	spa_config_load();
 	l2arc_start();
+	qat_init();
 }
 
 void
@@ -1884,6 +1886,7 @@ spa_fini(void)
 	unique_fini();
 	refcount_fini();
 	fm_fini();
+	qat_fini();
 
 	avl_destroy(&spa_namespace_avl);
 	avl_destroy(&spa_spare_avl);