* [dpdk-dev] [PATCH] compress/qat: enable dynamic huffman encoding @ 2018-07-20 17:46 Fiona Trahe 2018-10-05 16:18 ` [dpdk-dev] [PATCH v2] " Fiona Trahe 0 siblings, 1 reply; 17+ messages in thread From: Fiona Trahe @ 2018-07-20 17:46 UTC (permalink / raw) To: dev; +Cc: pablo.de.lara.guarch, fiona.trahe, tomaszx.jozwiak Enable dynamic huffman encoding in the QAT comp PMD. Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> --- config/common_base | 1 + config/rte_config.h | 1 + doc/guides/compressdevs/features/qat.ini | 1 + doc/guides/compressdevs/qat_comp.rst | 8 +-- drivers/common/qat/qat_device.c | 3 + drivers/common/qat/qat_device.h | 6 ++ drivers/compress/qat/qat_comp.c | 39 +++++++---- drivers/compress/qat/qat_comp.h | 13 ++++ drivers/compress/qat/qat_comp_pmd.c | 109 ++++++++++++++++++++++++++++++- 9 files changed, 163 insertions(+), 18 deletions(-) diff --git a/config/common_base b/config/common_base index 6d82b91..2f88e34 100644 --- a/config/common_base +++ b/config/common_base @@ -500,6 +500,7 @@ CONFIG_RTE_LIBRTE_PMD_QAT_SYM=n # CONFIG_RTE_PMD_QAT_MAX_PCI_DEVICES=48 CONFIG_RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS=16 +CONFIG_RTE_PMD_QAT_COMP_IM_BUFFER_SIZE=65536 # # Compile PMD for virtio crypto devices diff --git a/config/rte_config.h b/config/rte_config.h index a8e4797..c93d508 100644 --- a/config/rte_config.h +++ b/config/rte_config.h @@ -90,6 +90,7 @@ /* Max. number of QuickAssist devices which can be attached */ #define RTE_PMD_QAT_MAX_PCI_DEVICES 48 #define RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS 16 +#define RTE_PMD_QAT_COMP_IM_BUFFER_SIZE 65536 /* virtio crypto defines */ #define RTE_MAX_VIRTIO_CRYPTO 32 diff --git a/doc/guides/compressdevs/features/qat.ini b/doc/guides/compressdevs/features/qat.ini index 5cd4524..6b1e7f9 100644 --- a/doc/guides/compressdevs/features/qat.ini +++ b/doc/guides/compressdevs/features/qat.ini @@ -13,3 +13,4 @@ Adler32 = Y Crc32 = Y Adler32&Crc32 = Y Fixed = Y +Dynamic = Y diff --git a/doc/guides/compressdevs/qat_comp.rst b/doc/guides/compressdevs/qat_comp.rst index 8b1270b..17e2bb4 100644 --- a/doc/guides/compressdevs/qat_comp.rst +++ b/doc/guides/compressdevs/qat_comp.rst @@ -18,11 +18,7 @@ QAT compression PMD has support for: Compression/Decompression algorithm: - * DEFLATE - -Huffman code type: - - * FIXED + * DEFLATE - using Fixed and Dynamic Huffman encoding Window size support: @@ -37,7 +33,7 @@ Limitations * Compressdev level 0, no compression, is not supported. -* Dynamic Huffman encoding is not yet supported. +* Dynamic Huffman encoding is supported up to a maximum payload size of 64Kb. Installation ------------ diff --git a/drivers/common/qat/qat_device.c b/drivers/common/qat/qat_device.c index f32d723..cbb5d79 100644 --- a/drivers/common/qat/qat_device.c +++ b/drivers/common/qat/qat_device.c @@ -7,6 +7,7 @@ #include "qat_device.h" #include "adf_transport_access_macros.h" #include "qat_sym_pmd.h" +#include "qat_comp_pmd.h" /* Hardware device information per generation */ __extension__ @@ -14,11 +15,13 @@ struct qat_gen_hw_data qat_gen_config[] = { [QAT_GEN1] = { .dev_gen = QAT_GEN1, .qp_hw_data = qat_gen1_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN1 }, [QAT_GEN2] = { .dev_gen = QAT_GEN2, .qp_hw_data = qat_gen1_qps, /* gen2 has same ring layout as gen1 */ + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN2 }, }; diff --git a/drivers/common/qat/qat_device.h b/drivers/common/qat/qat_device.h index 9599fc5..41f874b 100644 --- a/drivers/common/qat/qat_device.h +++ b/drivers/common/qat/qat_device.h @@ -16,6 +16,11 @@ #define QAT_DEV_NAME_MAX_LEN 64 +enum qat_comp_num_im_buffers { + QAT_NUM_INTERM_BUFS_GEN1 = 12, + QAT_NUM_INTERM_BUFS_GEN2 = 20 +}; + /* * This struct holds all the data about a QAT pci device * including data about all services it supports. @@ -67,6 +72,7 @@ struct qat_pci_device { struct qat_gen_hw_data { enum qat_device_gen dev_gen; const struct qat_qp_hw_data (*qp_hw_data)[ADF_MAX_QPS_ON_ANY_SERVICE]; + enum qat_comp_num_im_buffers comp_num_im_bufs_required; }; extern struct qat_gen_hw_data qat_gen_config[]; diff --git a/drivers/compress/qat/qat_comp.c b/drivers/compress/qat/qat_comp.c index cbf7614..df3180b 100644 --- a/drivers/compress/qat/qat_comp.c +++ b/drivers/compress/qat/qat_comp.c @@ -192,7 +192,7 @@ static void qat_comp_create_req_hdr(struct icp_qat_fw_comn_req_hdr *header, } static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, - const struct rte_memzone *interm_buff_mz __rte_unused, + const struct rte_memzone *interm_buff_mz, const struct rte_comp_xform *xform) { struct icp_qat_fw_comp_req *comp_req; @@ -280,10 +280,20 @@ static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, ICP_QAT_FW_SLICE_COMP); } else if (qat_xform->qat_comp_request_type == - QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { - QAT_LOG(ERR, "Dynamic huffman encoding not supported"); - return -EINVAL; + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_COMP); + + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_DRAM_WR); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + + comp_req->u1.xlt_pars.inter_buff_ptr = + interm_buff_mz->phys_addr; } #if RTE_LOG_DP_LEVEL >= RTE_LOG_DEBUG @@ -334,20 +344,27 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, (struct qat_comp_xform *)*private_xform; if (xform->type == RTE_COMP_COMPRESS) { - if (xform->compress.deflate.huffman == - RTE_COMP_HUFFMAN_DYNAMIC) { - QAT_LOG(ERR, - "QAT device doesn't support dynamic compression"); - return -ENOTSUP; - } if (xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_FIXED || ((xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_DEFAULT) - && qat->interm_buff_mz == NULL)) + && qat->interm_buff_mz == NULL)) qat_xform->qat_comp_request_type = QAT_COMP_REQUEST_FIXED_COMP_STATELESS; + else if ((xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DYNAMIC || + xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DEFAULT) && + qat->interm_buff_mz != NULL) + + qat_xform->qat_comp_request_type = + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS; + else { + QAT_LOG(ERR, + "IM buffers needed for dynamic deflate. Set size in config file"); + return -EINVAL; + } } else { qat_xform->qat_comp_request_type = QAT_COMP_REQUEST_DECOMPRESS; diff --git a/drivers/compress/qat/qat_comp.h b/drivers/compress/qat/qat_comp.h index 8d315ef..99a4462 100644 --- a/drivers/compress/qat/qat_comp.h +++ b/drivers/compress/qat/qat_comp.h @@ -15,6 +15,10 @@ #include "icp_qat_fw_comp.h" #include "icp_qat_fw_la.h" +#define QAT_64_BYTE_ALIGN_MASK (~0x3f) +#define QAT_64_BYTE_ALIGN (64) +#define QAT_NUM_BUFS_IN_IM_SGL 1 + #define ERR_CODE_QAT_COMP_WRONG_FW -99 enum qat_comp_request_type { @@ -24,6 +28,15 @@ enum qat_comp_request_type { REQ_COMP_END }; +struct array_of_ptrs { + phys_addr_t pointer[0]; +}; + +struct qat_inter_sgl { + qat_sgl_hdr; + struct qat_flat_buf buffers[QAT_NUM_BUFS_IN_IM_SGL]; +} __rte_packed __rte_cache_aligned; + struct qat_comp_sgl { qat_sgl_hdr; struct qat_flat_buf buffers[RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS]; diff --git a/drivers/compress/qat/qat_comp_pmd.c b/drivers/compress/qat/qat_comp_pmd.c index b89975f..76ae1f0 100644 --- a/drivers/compress/qat/qat_comp_pmd.c +++ b/drivers/compress/qat/qat_comp_pmd.c @@ -14,6 +14,7 @@ static const struct rte_compressdev_capabilities qat_comp_gen_capabilities[] = { RTE_COMP_FF_CRC32_ADLER32_CHECKSUM | RTE_COMP_FF_SHAREABLE_PRIV_XFORM | RTE_COMP_FF_HUFFMAN_FIXED | + RTE_COMP_FF_HUFFMAN_DYNAMIC | RTE_COMP_FF_OOP_SGL_IN_SGL_OUT | RTE_COMP_FF_OOP_SGL_IN_LB_OUT | RTE_COMP_FF_OOP_LB_IN_SGL_OUT, @@ -112,7 +113,7 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, /* store a link to the qp in the qat_pci_device */ qat_private->qat_dev->qps_in_use[QAT_SERVICE_COMPRESSION][qp_id] - = *qp_addr; + = *qp_addr; qp = (struct qat_qp *)*qp_addr; @@ -135,6 +136,91 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, return ret; } + +static const struct rte_memzone * +qat_comp_setup_inter_buffers(struct qat_comp_dev_private *comp_dev, + uint32_t buff_size) +{ + char inter_buff_mz_name[RTE_MEMZONE_NAMESIZE]; + const struct rte_memzone *memzone; + uint8_t *mz_start = NULL; + phys_addr_t mz_start_phys = 0; + struct array_of_ptrs *array_of_pointers; + int size_of_ptr_array; + uint32_t full_size; + uint32_t offset_of_sgls, offset_of_flat_buffs = 0; + int i; + int num_im_sgls = qat_gen_config[ + comp_dev->qat_dev->qat_dev_gen].comp_num_im_bufs_required; + + QAT_LOG(DEBUG, "QAT COMP device %s needs %d sgls", + comp_dev->qat_dev->name, num_im_sgls); + snprintf(inter_buff_mz_name, RTE_MEMZONE_NAMESIZE, + "%s_inter_buff", comp_dev->qat_dev->name); + memzone = rte_memzone_lookup(inter_buff_mz_name); + if (memzone != NULL) { + QAT_LOG(DEBUG, "QAT COMP im buffer memzone created already"); + return memzone; + } + + /* Create a memzone to hold intermediate buffers and associated + * meta-data needed by the firmware. The memzone contains: + * - a list of num_im_sgls physical pointers to sgls + * - the num_im_sgl sgl structures, each pointing to 2 flat buffers + * - the flat buffers: num_im_sgl * 2 + * where num_im_sgls depends on the hardware generation of the device + */ + + size_of_ptr_array = num_im_sgls * sizeof(phys_addr_t); + offset_of_sgls = (size_of_ptr_array + (~QAT_64_BYTE_ALIGN_MASK)) + & QAT_64_BYTE_ALIGN_MASK; + offset_of_flat_buffs = + offset_of_sgls + num_im_sgls * sizeof(struct qat_inter_sgl); + full_size = offset_of_flat_buffs + + num_im_sgls * buff_size * QAT_NUM_BUFS_IN_IM_SGL; + + memzone = rte_memzone_reserve_aligned(inter_buff_mz_name, full_size, + comp_dev->compressdev->data->socket_id, + RTE_MEMZONE_2MB, QAT_64_BYTE_ALIGN); + if (memzone == NULL) { + QAT_LOG(ERR, "Can't allocate intermediate buffers" + " for device %s", comp_dev->qat_dev->name); + return NULL; + } + + mz_start = (uint8_t *)memzone->addr; + mz_start_phys = memzone->phys_addr; + QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = %lx, size required %d," + "size created %ld", + inter_buff_mz_name, mz_start, mz_start_phys, + full_size, memzone->len); + + array_of_pointers = (struct array_of_ptrs *)mz_start; + for (i = 0; i < num_im_sgls; i++) { + uint32_t curr_sgl_offset = + offset_of_sgls + i * sizeof(struct qat_inter_sgl); + struct qat_inter_sgl *sgl = + (struct qat_inter_sgl *)(mz_start + curr_sgl_offset); + array_of_pointers->pointer[i] = mz_start_phys + curr_sgl_offset; + + sgl->num_bufs = QAT_NUM_BUFS_IN_IM_SGL; + sgl->num_mapped_bufs = 0; + sgl->resrvd = 0; + sgl->buffers[0].addr = mz_start_phys + offset_of_flat_buffs + + i * buff_size * QAT_NUM_BUFS_IN_IM_SGL; + sgl->buffers[0].len = buff_size; + sgl->buffers[0].resrvd = 0; + + + QAT_LOG(DEBUG, " : array_of_pointers->pointer[i]" + "= %lx", array_of_pointers->pointer[i]); + QAT_LOG(DEBUG, " : sgl[%i] = %p", i, sgl); + QAT_LOG(DEBUG, " : sgl->buffers[0].addr = %lx, len=%d", + sgl->buffers[0].addr, sgl->buffers[0].len); + } + return memzone; +} + static struct rte_mempool * qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, uint32_t num_elements) @@ -176,6 +262,12 @@ qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, static void _qat_comp_dev_config_clear(struct qat_comp_dev_private *comp_dev) { + /* Free intermediate buffers */ + if (comp_dev->interm_buff_mz) { + rte_memzone_free(comp_dev->interm_buff_mz); + comp_dev->interm_buff_mz = NULL; + } + /* Free private_xform pool */ if (comp_dev->xformpool) { /* Free internal mempool for private xforms */ @@ -197,6 +289,21 @@ qat_comp_dev_config(struct rte_compressdev *dev, return -EINVAL; } + if (RTE_PMD_QAT_COMP_IM_BUFFER_SIZE == 0) { + QAT_LOG(WARNING, + "RTE_PMD_QAT_COMP_IM_BUFFER_SIZE = 0 in config file, so" + " QAT device can't be used for Dynamic Deflate. " + "Did you really intend to do this?"); + } else { + comp_dev->interm_buff_mz = + qat_comp_setup_inter_buffers(comp_dev, + RTE_PMD_QAT_COMP_IM_BUFFER_SIZE); + if (comp_dev->interm_buff_mz == NULL) { + ret = -ENOMEM; + goto error_out; + } + } + comp_dev->xformpool = qat_comp_create_xform_pool(comp_dev, config->max_nb_priv_xforms); if (comp_dev->xformpool == NULL) { -- 2.7.4 ^ permalink raw reply [flat|nested] 17+ messages in thread
* [dpdk-dev] [PATCH v2] compress/qat: enable dynamic huffman encoding 2018-07-20 17:46 [dpdk-dev] [PATCH] compress/qat: enable dynamic huffman encoding Fiona Trahe @ 2018-10-05 16:18 ` Fiona Trahe 2018-10-09 11:01 ` Akhil Goyal 2018-10-11 17:03 ` [dpdk-dev] [PATCH v3] " Fiona Trahe 0 siblings, 2 replies; 17+ messages in thread From: Fiona Trahe @ 2018-10-05 16:18 UTC (permalink / raw) To: dev; +Cc: akhil.goyal, tomaszx.jozwiak, Fiona Trahe Enable dynamic huffman encoding in the QAT comp PMD. Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> --- v2 changes: - allocate 2 buffers per intermediate buffer sgl - Compile out trace for debugging intermediate buffers Depends on: https://patches.dpdk.org/patch/45794/ config/common_base | 1 + config/rte_config.h | 1 + doc/guides/compressdevs/features/qat.ini | 1 + doc/guides/compressdevs/qat_comp.rst | 7 +- drivers/common/qat/qat_device.c | 4 + drivers/common/qat/qat_device.h | 7 ++ drivers/compress/qat/qat_comp.c | 39 +++++++--- drivers/compress/qat/qat_comp.h | 13 ++++ drivers/compress/qat/qat_comp_pmd.c | 123 ++++++++++++++++++++++++++++++- 9 files changed, 179 insertions(+), 17 deletions(-) diff --git a/config/common_base b/config/common_base index 43c7e9a..d292a61 100644 --- a/config/common_base +++ b/config/common_base @@ -500,6 +500,7 @@ CONFIG_RTE_LIBRTE_PMD_QAT_SYM=n # CONFIG_RTE_PMD_QAT_MAX_PCI_DEVICES=48 CONFIG_RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS=16 +CONFIG_RTE_PMD_QAT_COMP_IM_BUFFER_SIZE=65536 # # Compile PMD for virtio crypto devices diff --git a/config/rte_config.h b/config/rte_config.h index 20c58df..4e501b3 100644 --- a/config/rte_config.h +++ b/config/rte_config.h @@ -94,6 +94,7 @@ /* Max. number of QuickAssist devices which can be attached */ #define RTE_PMD_QAT_MAX_PCI_DEVICES 48 #define RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS 16 +#define RTE_PMD_QAT_COMP_IM_BUFFER_SIZE 65536 /* virtio crypto defines */ #define RTE_MAX_VIRTIO_CRYPTO 32 diff --git a/doc/guides/compressdevs/features/qat.ini b/doc/guides/compressdevs/features/qat.ini index 5cd4524..6b1e7f9 100644 --- a/doc/guides/compressdevs/features/qat.ini +++ b/doc/guides/compressdevs/features/qat.ini @@ -13,3 +13,4 @@ Adler32 = Y Crc32 = Y Adler32&Crc32 = Y Fixed = Y +Dynamic = Y diff --git a/doc/guides/compressdevs/qat_comp.rst b/doc/guides/compressdevs/qat_comp.rst index 7bffbe6..aee3b99 100644 --- a/doc/guides/compressdevs/qat_comp.rst +++ b/doc/guides/compressdevs/qat_comp.rst @@ -18,11 +18,7 @@ QAT compression PMD has support for: Compression/Decompression algorithm: - * DEFLATE - -Huffman code type: - - * FIXED + * DEFLATE - using Fixed and Dynamic Huffman encoding Window size support: @@ -36,7 +32,6 @@ Limitations ----------- * Compressdev level 0, no compression, is not supported. -* Dynamic Huffman encoding is not yet supported. * Queue pairs are not thread-safe (that is, within a single queue pair, RX and TX from different lcores is not supported). * No BSD support as BSD QAT kernel driver not available. diff --git a/drivers/common/qat/qat_device.c b/drivers/common/qat/qat_device.c index b158fb9..b2db8b0 100644 --- a/drivers/common/qat/qat_device.c +++ b/drivers/common/qat/qat_device.c @@ -7,6 +7,7 @@ #include "qat_device.h" #include "adf_transport_access_macros.h" #include "qat_sym_pmd.h" +#include "qat_comp_pmd.h" /* Hardware device information per generation */ __extension__ @@ -14,15 +15,18 @@ struct qat_gen_hw_data qat_gen_config[] = { [QAT_GEN1] = { .dev_gen = QAT_GEN1, .qp_hw_data = qat_gen1_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN1 }, [QAT_GEN2] = { .dev_gen = QAT_GEN2, .qp_hw_data = qat_gen1_qps, /* gen2 has same ring layout as gen1 */ + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN2 }, [QAT_GEN3] = { .dev_gen = QAT_GEN3, .qp_hw_data = qat_gen3_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN3 }, }; diff --git a/drivers/common/qat/qat_device.h b/drivers/common/qat/qat_device.h index 3a71cd4..eb81c78 100644 --- a/drivers/common/qat/qat_device.h +++ b/drivers/common/qat/qat_device.h @@ -16,6 +16,12 @@ #define QAT_DEV_NAME_MAX_LEN 64 +enum qat_comp_num_im_buffers { + QAT_NUM_INTERM_BUFS_GEN1 = 12, + QAT_NUM_INTERM_BUFS_GEN2 = 20, + QAT_NUM_INTERM_BUFS_GEN3 = 20 +}; + /* * This struct holds all the data about a QAT pci device * including data about all services it supports. @@ -72,6 +78,7 @@ struct qat_pci_device { struct qat_gen_hw_data { enum qat_device_gen dev_gen; const struct qat_qp_hw_data (*qp_hw_data)[ADF_MAX_QPS_ON_ANY_SERVICE]; + enum qat_comp_num_im_buffers comp_num_im_bufs_required; }; extern struct qat_gen_hw_data qat_gen_config[]; diff --git a/drivers/compress/qat/qat_comp.c b/drivers/compress/qat/qat_comp.c index b9336f3..d70c594 100644 --- a/drivers/compress/qat/qat_comp.c +++ b/drivers/compress/qat/qat_comp.c @@ -192,7 +192,7 @@ static void qat_comp_create_req_hdr(struct icp_qat_fw_comn_req_hdr *header, } static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, - const struct rte_memzone *interm_buff_mz __rte_unused, + const struct rte_memzone *interm_buff_mz, const struct rte_comp_xform *xform) { struct icp_qat_fw_comp_req *comp_req; @@ -280,10 +280,20 @@ static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, ICP_QAT_FW_SLICE_COMP); } else if (qat_xform->qat_comp_request_type == - QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { - QAT_LOG(ERR, "Dynamic huffman encoding not supported"); - return -EINVAL; + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_COMP); + + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_DRAM_WR); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + + comp_req->u1.xlt_pars.inter_buff_ptr = + interm_buff_mz->phys_addr; } #if RTE_LOG_DP_LEVEL >= RTE_LOG_DEBUG @@ -334,12 +344,6 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, (struct qat_comp_xform *)*private_xform; if (xform->type == RTE_COMP_COMPRESS) { - if (xform->compress.deflate.huffman == - RTE_COMP_HUFFMAN_DYNAMIC) { - QAT_LOG(ERR, - "QAT device doesn't support dynamic compression"); - return -ENOTSUP; - } if (xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_FIXED || ((xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_DEFAULT) @@ -347,6 +351,21 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, qat_xform->qat_comp_request_type = QAT_COMP_REQUEST_FIXED_COMP_STATELESS; + else if ((xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DYNAMIC || + xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DEFAULT) && + qat->interm_buff_mz != NULL) + + qat_xform->qat_comp_request_type = + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS; + + else { + QAT_LOG(ERR, + "IM buffers needed for dynamic deflate. Set size in config file"); + return -EINVAL; + } + qat_xform->checksum_type = xform->compress.chksum; } else { diff --git a/drivers/compress/qat/qat_comp.h b/drivers/compress/qat/qat_comp.h index 8d315ef..24d0d9c 100644 --- a/drivers/compress/qat/qat_comp.h +++ b/drivers/compress/qat/qat_comp.h @@ -15,6 +15,10 @@ #include "icp_qat_fw_comp.h" #include "icp_qat_fw_la.h" +#define QAT_64_BYTE_ALIGN_MASK (~0x3f) +#define QAT_64_BYTE_ALIGN (64) +#define QAT_NUM_BUFS_IN_IM_SGL 2 + #define ERR_CODE_QAT_COMP_WRONG_FW -99 enum qat_comp_request_type { @@ -24,6 +28,15 @@ enum qat_comp_request_type { REQ_COMP_END }; +struct array_of_ptrs { + phys_addr_t pointer[0]; +}; + +struct qat_inter_sgl { + qat_sgl_hdr; + struct qat_flat_buf buffers[QAT_NUM_BUFS_IN_IM_SGL]; +} __rte_packed __rte_cache_aligned; + struct qat_comp_sgl { qat_sgl_hdr; struct qat_flat_buf buffers[RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS]; diff --git a/drivers/compress/qat/qat_comp_pmd.c b/drivers/compress/qat/qat_comp_pmd.c index 6510cca..05ffe79 100644 --- a/drivers/compress/qat/qat_comp_pmd.c +++ b/drivers/compress/qat/qat_comp_pmd.c @@ -14,6 +14,7 @@ static const struct rte_compressdev_capabilities qat_comp_gen_capabilities[] = { RTE_COMP_FF_CRC32_ADLER32_CHECKSUM | RTE_COMP_FF_SHAREABLE_PRIV_XFORM | RTE_COMP_FF_HUFFMAN_FIXED | + RTE_COMP_FF_HUFFMAN_DYNAMIC | RTE_COMP_FF_OOP_SGL_IN_SGL_OUT | RTE_COMP_FF_OOP_SGL_IN_LB_OUT | RTE_COMP_FF_OOP_LB_IN_SGL_OUT, @@ -112,7 +113,7 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, /* store a link to the qp in the qat_pci_device */ qat_private->qat_dev->qps_in_use[QAT_SERVICE_COMPRESSION][qp_id] - = *qp_addr; + = *qp_addr; qp = (struct qat_qp *)*qp_addr; @@ -135,6 +136,101 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, return ret; } + +static const struct rte_memzone * +qat_comp_setup_inter_buffers(struct qat_comp_dev_private *comp_dev, + uint32_t buff_size) +{ + char inter_buff_mz_name[RTE_MEMZONE_NAMESIZE]; + const struct rte_memzone *memzone; + uint8_t *mz_start = NULL; + phys_addr_t mz_start_phys = 0; + struct array_of_ptrs *array_of_pointers; + int size_of_ptr_array; + uint32_t full_size; + uint32_t offset_of_sgls, offset_of_flat_buffs = 0; + int i; + int num_im_sgls = qat_gen_config[ + comp_dev->qat_dev->qat_dev_gen].comp_num_im_bufs_required; + + QAT_LOG(DEBUG, "QAT COMP device %s needs %d sgls", + comp_dev->qat_dev->name, num_im_sgls); + snprintf(inter_buff_mz_name, RTE_MEMZONE_NAMESIZE, + "%s_inter_buff", comp_dev->qat_dev->name); + memzone = rte_memzone_lookup(inter_buff_mz_name); + if (memzone != NULL) { + QAT_LOG(DEBUG, "QAT COMP im buffer memzone created already"); + return memzone; + } + + /* Create a memzone to hold intermediate buffers and associated + * meta-data needed by the firmware. The memzone contains: + * - a list of num_im_sgls physical pointers to sgls + * - the num_im_sgl sgl structures, each pointing to 2 flat buffers + * - the flat buffers: num_im_sgl * 2 + * where num_im_sgls depends on the hardware generation of the device + */ + + size_of_ptr_array = num_im_sgls * sizeof(phys_addr_t); + offset_of_sgls = (size_of_ptr_array + (~QAT_64_BYTE_ALIGN_MASK)) + & QAT_64_BYTE_ALIGN_MASK; + offset_of_flat_buffs = + offset_of_sgls + num_im_sgls * sizeof(struct qat_inter_sgl); + full_size = offset_of_flat_buffs + + num_im_sgls * buff_size * QAT_NUM_BUFS_IN_IM_SGL; + + memzone = rte_memzone_reserve_aligned(inter_buff_mz_name, full_size, + comp_dev->compressdev->data->socket_id, + RTE_MEMZONE_2MB, QAT_64_BYTE_ALIGN); + if (memzone == NULL) { + QAT_LOG(ERR, "Can't allocate intermediate buffers" + " for device %s", comp_dev->qat_dev->name); + return NULL; + } + + mz_start = (uint8_t *)memzone->addr; + mz_start_phys = memzone->phys_addr; + QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = %lx, size required %d," + "size created %ld", + inter_buff_mz_name, mz_start, mz_start_phys, + full_size, memzone->len); + + array_of_pointers = (struct array_of_ptrs *)mz_start; + for (i = 0; i < num_im_sgls; i++) { + uint32_t curr_sgl_offset = + offset_of_sgls + i * sizeof(struct qat_inter_sgl); + struct qat_inter_sgl *sgl = + (struct qat_inter_sgl *)(mz_start + curr_sgl_offset); + array_of_pointers->pointer[i] = mz_start_phys + curr_sgl_offset; + + sgl->num_bufs = QAT_NUM_BUFS_IN_IM_SGL; + sgl->num_mapped_bufs = 0; + sgl->resrvd = 0; + sgl->buffers[0].addr = mz_start_phys + offset_of_flat_buffs + + ((i * QAT_NUM_BUFS_IN_IM_SGL) * buff_size); + sgl->buffers[0].len = buff_size; + sgl->buffers[0].resrvd = 0; + sgl->buffers[1].addr = mz_start_phys + offset_of_flat_buffs + + (((i * QAT_NUM_BUFS_IN_IM_SGL) + 1) * buff_size); + sgl->buffers[1].len = buff_size; + sgl->buffers[1].resrvd = 0; + +#if 1 + QAT_LOG(DEBUG, " : phys addr of sgl[%i] in array_of_pointers" + "= %lx", i, array_of_pointers->pointer[i]); + QAT_LOG(DEBUG, " : virt address of sgl[%i] = %p", i, sgl); + QAT_LOG(DEBUG, " : sgl->buffers[0].addr = %lx, len=%d", + sgl->buffers[0].addr, sgl->buffers[0].len); + QAT_LOG(DEBUG, " : sgl->buffers[1].addr = %lx, len=%d", + sgl->buffers[1].addr, sgl->buffers[1].len); +#endif + } +#if 0 + QAT_DP_HEXDUMP_LOG(DEBUG, "IM buffer memzone:", mz_start, 1504); +#endif + return memzone; +} + static struct rte_mempool * qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, uint32_t num_elements) @@ -176,6 +272,12 @@ qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, static void _qat_comp_dev_config_clear(struct qat_comp_dev_private *comp_dev) { + /* Free intermediate buffers */ + if (comp_dev->interm_buff_mz) { + rte_memzone_free(comp_dev->interm_buff_mz); + comp_dev->interm_buff_mz = NULL; + } + /* Free private_xform pool */ if (comp_dev->xformpool) { /* Free internal mempool for private xforms */ @@ -197,6 +299,21 @@ qat_comp_dev_config(struct rte_compressdev *dev, return -EINVAL; } + if (RTE_PMD_QAT_COMP_IM_BUFFER_SIZE == 0) { + QAT_LOG(WARNING, + "RTE_PMD_QAT_COMP_IM_BUFFER_SIZE = 0 in config file, so" + " QAT device can't be used for Dynamic Deflate. " + "Did you really intend to do this?"); + } else { + comp_dev->interm_buff_mz = + qat_comp_setup_inter_buffers(comp_dev, + RTE_PMD_QAT_COMP_IM_BUFFER_SIZE); + if (comp_dev->interm_buff_mz == NULL) { + ret = -ENOMEM; + goto error_out; + } + } + comp_dev->xformpool = qat_comp_create_xform_pool(comp_dev, config->max_nb_priv_xforms); if (comp_dev->xformpool == NULL) { @@ -365,6 +482,10 @@ qat_comp_dev_create(struct qat_pci_device *qat_pci_dev) QAT_LOG(ERR, "Compression PMD not supported on QAT dh895xcc"); return 0; } + if (qat_pci_dev->qat_dev_gen == QAT_GEN3) { + QAT_LOG(ERR, "Compression PMD not supported on QAT c4xxx"); + return 0; + } struct rte_compressdev_pmd_init_params init_params = { .name = "", -- 2.7.4 ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [dpdk-dev] [PATCH v2] compress/qat: enable dynamic huffman encoding 2018-10-05 16:18 ` [dpdk-dev] [PATCH v2] " Fiona Trahe @ 2018-10-09 11:01 ` Akhil Goyal 2018-10-09 14:09 ` Trahe, Fiona 2018-10-11 17:03 ` [dpdk-dev] [PATCH v3] " Fiona Trahe 1 sibling, 1 reply; 17+ messages in thread From: Akhil Goyal @ 2018-10-09 11:01 UTC (permalink / raw) To: Fiona Trahe, dev; +Cc: tomaszx.jozwiak On 10/5/2018 9:48 PM, Fiona Trahe wrote: > Enable dynamic huffman encoding in the QAT comp PMD. > > Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> > Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> > --- > v2 changes: > - allocate 2 buffers per intermediate buffer sgl > - Compile out trace for debugging intermediate buffers > > Depends on: > https://patches.dpdk.org/patch/45794/ > //snip > +#if 1 > + QAT_LOG(DEBUG, " : phys addr of sgl[%i] in array_of_pointers" > + "= %lx", i, array_of_pointers->pointer[i]); > + QAT_LOG(DEBUG, " : virt address of sgl[%i] = %p", i, sgl); > + QAT_LOG(DEBUG, " : sgl->buffers[0].addr = %lx, len=%d", > + sgl->buffers[0].addr, sgl->buffers[0].len); > + QAT_LOG(DEBUG, " : sgl->buffers[1].addr = %lx, len=%d", > + sgl->buffers[1].addr, sgl->buffers[1].len); > +#endif > + } > +#if 0 > + QAT_DP_HEXDUMP_LOG(DEBUG, "IM buffer memzone:", mz_start, 1504); > +#endif I believe #if 0 and #if 1 should be replaced with some debug macros, or else dynamic logging shall take care of that. ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [dpdk-dev] [PATCH v2] compress/qat: enable dynamic huffman encoding 2018-10-09 11:01 ` Akhil Goyal @ 2018-10-09 14:09 ` Trahe, Fiona 2018-10-12 11:39 ` Akhil Goyal 0 siblings, 1 reply; 17+ messages in thread From: Trahe, Fiona @ 2018-10-09 14:09 UTC (permalink / raw) To: Akhil Goyal, dev; +Cc: Jozwiak, TomaszX, Trahe, Fiona Hi Akhil, > -----Original Message----- > From: Akhil Goyal [mailto:akhil.goyal@nxp.com] > Sent: Tuesday, October 9, 2018 12:02 PM > To: Trahe, Fiona <fiona.trahe@intel.com>; dev@dpdk.org > Cc: Jozwiak, TomaszX <tomaszx.jozwiak@intel.com> > Subject: Re: [dpdk-dev] [PATCH v2] compress/qat: enable dynamic huffman encoding > > > > On 10/5/2018 9:48 PM, Fiona Trahe wrote: > > Enable dynamic huffman encoding in the QAT comp PMD. > > > > Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> > > Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> > > --- > > v2 changes: > > - allocate 2 buffers per intermediate buffer sgl > > - Compile out trace for debugging intermediate buffers > > > > Depends on: > > https://patches.dpdk.org/patch/45794/ > > > //snip > > +#if 1 > > + QAT_LOG(DEBUG, " : phys addr of sgl[%i] in array_of_pointers" > > + "= %lx", i, array_of_pointers->pointer[i]); > > + QAT_LOG(DEBUG, " : virt address of sgl[%i] = %p", i, sgl); > > + QAT_LOG(DEBUG, " : sgl->buffers[0].addr = %lx, len=%d", > > + sgl->buffers[0].addr, sgl->buffers[0].len); > > + QAT_LOG(DEBUG, " : sgl->buffers[1].addr = %lx, len=%d", > > + sgl->buffers[1].addr, sgl->buffers[1].len); > > +#endif > > + } > > +#if 0 > > + QAT_DP_HEXDUMP_LOG(DEBUG, "IM buffer memzone:", mz_start, 1504); > > +#endif > > I believe #if 0 and #if 1 should be replaced with some debug macros, or > else dynamic logging shall take care of that. [Fiona] These are not on the data path, so I didn't think it appropriate to wrap with the DP log level compile option. And they generate a lot of trace, so make it impossible to see the other debug trace enabled with dynamic logging using level 8/DEBUG. Really a level 9 would be useful here, but I think this is overkill. I could have removed this code, but it's handy for debugging. How about if I add a local QAT define for it? #define QAT_IM_BUFFER_DEBUG 0 #if QAT_IM_BUFFER_DEBUG ... #endif ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [dpdk-dev] [PATCH v2] compress/qat: enable dynamic huffman encoding 2018-10-09 14:09 ` Trahe, Fiona @ 2018-10-12 11:39 ` Akhil Goyal 0 siblings, 0 replies; 17+ messages in thread From: Akhil Goyal @ 2018-10-12 11:39 UTC (permalink / raw) To: Trahe, Fiona, dev; +Cc: Jozwiak, TomaszX On 10/9/2018 7:39 PM, Trahe, Fiona wrote: > Hi Akhil, > >> -----Original Message----- >> From: Akhil Goyal [mailto:akhil.goyal@nxp.com] >> Sent: Tuesday, October 9, 2018 12:02 PM >> To: Trahe, Fiona <fiona.trahe@intel.com>; dev@dpdk.org >> Cc: Jozwiak, TomaszX <tomaszx.jozwiak@intel.com> >> Subject: Re: [dpdk-dev] [PATCH v2] compress/qat: enable dynamic huffman encoding >> >> >> >> On 10/5/2018 9:48 PM, Fiona Trahe wrote: >>> Enable dynamic huffman encoding in the QAT comp PMD. >>> >>> Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> >>> Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> >>> --- >>> v2 changes: >>> - allocate 2 buffers per intermediate buffer sgl >>> - Compile out trace for debugging intermediate buffers >>> >>> Depends on: >>> https://patches.dpdk.org/patch/45794/ >>> >> //snip >>> +#if 1 >>> + QAT_LOG(DEBUG, " : phys addr of sgl[%i] in array_of_pointers" >>> + "= %lx", i, array_of_pointers->pointer[i]); >>> + QAT_LOG(DEBUG, " : virt address of sgl[%i] = %p", i, sgl); >>> + QAT_LOG(DEBUG, " : sgl->buffers[0].addr = %lx, len=%d", >>> + sgl->buffers[0].addr, sgl->buffers[0].len); >>> + QAT_LOG(DEBUG, " : sgl->buffers[1].addr = %lx, len=%d", >>> + sgl->buffers[1].addr, sgl->buffers[1].len); >>> +#endif >>> + } >>> +#if 0 >>> + QAT_DP_HEXDUMP_LOG(DEBUG, "IM buffer memzone:", mz_start, 1504); >>> +#endif >> I believe #if 0 and #if 1 should be replaced with some debug macros, or >> else dynamic logging shall take care of that. > [Fiona] These are not on the data path, so I didn't think it appropriate to wrap with the DP log level compile option. > And they generate a lot of trace, so make it impossible to see the other debug trace enabled with > dynamic logging using level 8/DEBUG. Really a level 9 would be useful here, but I think this is overkill. > I could have removed this code, but it's handy for debugging. > How about if I add a local QAT define for it? > #define QAT_IM_BUFFER_DEBUG 0 > #if QAT_IM_BUFFER_DEBUG > ... > #endif I think local define would be better. ^ permalink raw reply [flat|nested] 17+ messages in thread
* [dpdk-dev] [PATCH v3] compress/qat: enable dynamic huffman encoding 2018-10-05 16:18 ` [dpdk-dev] [PATCH v2] " Fiona Trahe 2018-10-09 11:01 ` Akhil Goyal @ 2018-10-11 17:03 ` Fiona Trahe 2018-10-15 23:16 ` [dpdk-dev] [PATCH v4] " Fiona Trahe 1 sibling, 1 reply; 17+ messages in thread From: Fiona Trahe @ 2018-10-11 17:03 UTC (permalink / raw) To: dev; +Cc: akhil.goyal, tomaszx.jozwiak, arkadiuszx.kusztal, Fiona Trahe Enable dynamic huffman encoding in the QAT comp PMD. Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> --- v3 changes: - reverted to 1 buffer per intermediate buffer sgl v2 changes: - allocate 2 buffers per intermediate buffer sgl - Compile out trace for debugging intermediate buffers config/common_base | 1 + config/rte_config.h | 1 + doc/guides/compressdevs/features/qat.ini | 1 + doc/guides/compressdevs/qat_comp.rst | 7 +- drivers/common/qat/qat_device.c | 4 + drivers/common/qat/qat_device.h | 7 ++ drivers/compress/qat/qat_comp.c | 39 +++++++--- drivers/compress/qat/qat_comp.h | 13 ++++ drivers/compress/qat/qat_comp_pmd.c | 123 ++++++++++++++++++++++++++++++- 9 files changed, 179 insertions(+), 17 deletions(-) diff --git a/config/common_base b/config/common_base index 85fad0c..9bbce31 100644 --- a/config/common_base +++ b/config/common_base @@ -500,6 +500,7 @@ CONFIG_RTE_LIBRTE_PMD_QAT_SYM=n # CONFIG_RTE_PMD_QAT_MAX_PCI_DEVICES=48 CONFIG_RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS=16 +CONFIG_RTE_PMD_QAT_COMP_IM_BUFFER_SIZE=65536 # # Compile PMD for virtio crypto devices diff --git a/config/rte_config.h b/config/rte_config.h index 20c58df..4e501b3 100644 --- a/config/rte_config.h +++ b/config/rte_config.h @@ -94,6 +94,7 @@ /* Max. number of QuickAssist devices which can be attached */ #define RTE_PMD_QAT_MAX_PCI_DEVICES 48 #define RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS 16 +#define RTE_PMD_QAT_COMP_IM_BUFFER_SIZE 65536 /* virtio crypto defines */ #define RTE_MAX_VIRTIO_CRYPTO 32 diff --git a/doc/guides/compressdevs/features/qat.ini b/doc/guides/compressdevs/features/qat.ini index 5cd4524..6b1e7f9 100644 --- a/doc/guides/compressdevs/features/qat.ini +++ b/doc/guides/compressdevs/features/qat.ini @@ -13,3 +13,4 @@ Adler32 = Y Crc32 = Y Adler32&Crc32 = Y Fixed = Y +Dynamic = Y diff --git a/doc/guides/compressdevs/qat_comp.rst b/doc/guides/compressdevs/qat_comp.rst index 7bffbe6..aee3b99 100644 --- a/doc/guides/compressdevs/qat_comp.rst +++ b/doc/guides/compressdevs/qat_comp.rst @@ -18,11 +18,7 @@ QAT compression PMD has support for: Compression/Decompression algorithm: - * DEFLATE - -Huffman code type: - - * FIXED + * DEFLATE - using Fixed and Dynamic Huffman encoding Window size support: @@ -36,7 +32,6 @@ Limitations ----------- * Compressdev level 0, no compression, is not supported. -* Dynamic Huffman encoding is not yet supported. * Queue pairs are not thread-safe (that is, within a single queue pair, RX and TX from different lcores is not supported). * No BSD support as BSD QAT kernel driver not available. diff --git a/drivers/common/qat/qat_device.c b/drivers/common/qat/qat_device.c index 6e64e22..e662e43 100644 --- a/drivers/common/qat/qat_device.c +++ b/drivers/common/qat/qat_device.c @@ -7,6 +7,7 @@ #include "qat_device.h" #include "adf_transport_access_macros.h" #include "qat_sym_pmd.h" +#include "qat_comp_pmd.h" /* Hardware device information per generation */ __extension__ @@ -14,15 +15,18 @@ struct qat_gen_hw_data qat_gen_config[] = { [QAT_GEN1] = { .dev_gen = QAT_GEN1, .qp_hw_data = qat_gen1_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN1 }, [QAT_GEN2] = { .dev_gen = QAT_GEN2, .qp_hw_data = qat_gen1_qps, /* gen2 has same ring layout as gen1 */ + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN2 }, [QAT_GEN3] = { .dev_gen = QAT_GEN3, .qp_hw_data = qat_gen3_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN3 }, }; diff --git a/drivers/common/qat/qat_device.h b/drivers/common/qat/qat_device.h index 3a71cd4..eb81c78 100644 --- a/drivers/common/qat/qat_device.h +++ b/drivers/common/qat/qat_device.h @@ -16,6 +16,12 @@ #define QAT_DEV_NAME_MAX_LEN 64 +enum qat_comp_num_im_buffers { + QAT_NUM_INTERM_BUFS_GEN1 = 12, + QAT_NUM_INTERM_BUFS_GEN2 = 20, + QAT_NUM_INTERM_BUFS_GEN3 = 20 +}; + /* * This struct holds all the data about a QAT pci device * including data about all services it supports. @@ -72,6 +78,7 @@ struct qat_pci_device { struct qat_gen_hw_data { enum qat_device_gen dev_gen; const struct qat_qp_hw_data (*qp_hw_data)[ADF_MAX_QPS_ON_ANY_SERVICE]; + enum qat_comp_num_im_buffers comp_num_im_bufs_required; }; extern struct qat_gen_hw_data qat_gen_config[]; diff --git a/drivers/compress/qat/qat_comp.c b/drivers/compress/qat/qat_comp.c index b9336f3..d70c594 100644 --- a/drivers/compress/qat/qat_comp.c +++ b/drivers/compress/qat/qat_comp.c @@ -192,7 +192,7 @@ static void qat_comp_create_req_hdr(struct icp_qat_fw_comn_req_hdr *header, } static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, - const struct rte_memzone *interm_buff_mz __rte_unused, + const struct rte_memzone *interm_buff_mz, const struct rte_comp_xform *xform) { struct icp_qat_fw_comp_req *comp_req; @@ -280,10 +280,20 @@ static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, ICP_QAT_FW_SLICE_COMP); } else if (qat_xform->qat_comp_request_type == - QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { - QAT_LOG(ERR, "Dynamic huffman encoding not supported"); - return -EINVAL; + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_COMP); + + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_DRAM_WR); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + + comp_req->u1.xlt_pars.inter_buff_ptr = + interm_buff_mz->phys_addr; } #if RTE_LOG_DP_LEVEL >= RTE_LOG_DEBUG @@ -334,12 +344,6 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, (struct qat_comp_xform *)*private_xform; if (xform->type == RTE_COMP_COMPRESS) { - if (xform->compress.deflate.huffman == - RTE_COMP_HUFFMAN_DYNAMIC) { - QAT_LOG(ERR, - "QAT device doesn't support dynamic compression"); - return -ENOTSUP; - } if (xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_FIXED || ((xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_DEFAULT) @@ -347,6 +351,21 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, qat_xform->qat_comp_request_type = QAT_COMP_REQUEST_FIXED_COMP_STATELESS; + else if ((xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DYNAMIC || + xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DEFAULT) && + qat->interm_buff_mz != NULL) + + qat_xform->qat_comp_request_type = + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS; + + else { + QAT_LOG(ERR, + "IM buffers needed for dynamic deflate. Set size in config file"); + return -EINVAL; + } + qat_xform->checksum_type = xform->compress.chksum; } else { diff --git a/drivers/compress/qat/qat_comp.h b/drivers/compress/qat/qat_comp.h index 8d315ef..99a4462 100644 --- a/drivers/compress/qat/qat_comp.h +++ b/drivers/compress/qat/qat_comp.h @@ -15,6 +15,10 @@ #include "icp_qat_fw_comp.h" #include "icp_qat_fw_la.h" +#define QAT_64_BYTE_ALIGN_MASK (~0x3f) +#define QAT_64_BYTE_ALIGN (64) +#define QAT_NUM_BUFS_IN_IM_SGL 1 + #define ERR_CODE_QAT_COMP_WRONG_FW -99 enum qat_comp_request_type { @@ -24,6 +28,15 @@ enum qat_comp_request_type { REQ_COMP_END }; +struct array_of_ptrs { + phys_addr_t pointer[0]; +}; + +struct qat_inter_sgl { + qat_sgl_hdr; + struct qat_flat_buf buffers[QAT_NUM_BUFS_IN_IM_SGL]; +} __rte_packed __rte_cache_aligned; + struct qat_comp_sgl { qat_sgl_hdr; struct qat_flat_buf buffers[RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS]; diff --git a/drivers/compress/qat/qat_comp_pmd.c b/drivers/compress/qat/qat_comp_pmd.c index 63af23a..8f34682 100644 --- a/drivers/compress/qat/qat_comp_pmd.c +++ b/drivers/compress/qat/qat_comp_pmd.c @@ -14,6 +14,7 @@ static const struct rte_compressdev_capabilities qat_comp_gen_capabilities[] = { RTE_COMP_FF_CRC32_ADLER32_CHECKSUM | RTE_COMP_FF_SHAREABLE_PRIV_XFORM | RTE_COMP_FF_HUFFMAN_FIXED | + RTE_COMP_FF_HUFFMAN_DYNAMIC | RTE_COMP_FF_OOP_SGL_IN_SGL_OUT | RTE_COMP_FF_OOP_SGL_IN_LB_OUT | RTE_COMP_FF_OOP_LB_IN_SGL_OUT, @@ -112,7 +113,7 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, /* store a link to the qp in the qat_pci_device */ qat_private->qat_dev->qps_in_use[QAT_SERVICE_COMPRESSION][qp_id] - = *qp_addr; + = *qp_addr; qp = (struct qat_qp *)*qp_addr; @@ -135,6 +136,101 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, return ret; } + +static const struct rte_memzone * +qat_comp_setup_inter_buffers(struct qat_comp_dev_private *comp_dev, + uint32_t buff_size) +{ + char inter_buff_mz_name[RTE_MEMZONE_NAMESIZE]; + const struct rte_memzone *memzone; + uint8_t *mz_start = NULL; + phys_addr_t mz_start_phys = 0; + struct array_of_ptrs *array_of_pointers; + int size_of_ptr_array; + uint32_t full_size; + uint32_t offset_of_sgls, offset_of_flat_buffs = 0; + int i; + int num_im_sgls = qat_gen_config[ + comp_dev->qat_dev->qat_dev_gen].comp_num_im_bufs_required; + + QAT_LOG(DEBUG, "QAT COMP device %s needs %d sgls", + comp_dev->qat_dev->name, num_im_sgls); + snprintf(inter_buff_mz_name, RTE_MEMZONE_NAMESIZE, + "%s_inter_buff", comp_dev->qat_dev->name); + memzone = rte_memzone_lookup(inter_buff_mz_name); + if (memzone != NULL) { + QAT_LOG(DEBUG, "QAT COMP im buffer memzone created already"); + return memzone; + } + + /* Create a memzone to hold intermediate buffers and associated + * meta-data needed by the firmware. The memzone contains: + * - a list of num_im_sgls physical pointers to sgls + * - the num_im_sgl sgl structures, each pointing to 2 flat buffers + * - the flat buffers: num_im_sgl * 2 + * where num_im_sgls depends on the hardware generation of the device + */ + + size_of_ptr_array = num_im_sgls * sizeof(phys_addr_t); + offset_of_sgls = (size_of_ptr_array + (~QAT_64_BYTE_ALIGN_MASK)) + & QAT_64_BYTE_ALIGN_MASK; + offset_of_flat_buffs = + offset_of_sgls + num_im_sgls * sizeof(struct qat_inter_sgl); + full_size = offset_of_flat_buffs + + num_im_sgls * buff_size * QAT_NUM_BUFS_IN_IM_SGL; + + memzone = rte_memzone_reserve_aligned(inter_buff_mz_name, full_size, + comp_dev->compressdev->data->socket_id, + RTE_MEMZONE_2MB, QAT_64_BYTE_ALIGN); + if (memzone == NULL) { + QAT_LOG(ERR, "Can't allocate intermediate buffers" + " for device %s", comp_dev->qat_dev->name); + return NULL; + } + + mz_start = (uint8_t *)memzone->addr; + mz_start_phys = memzone->phys_addr; + QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = %lx, size required %d," + "size created %ld", + inter_buff_mz_name, mz_start, mz_start_phys, + full_size, memzone->len); + + array_of_pointers = (struct array_of_ptrs *)mz_start; + for (i = 0; i < num_im_sgls; i++) { + uint32_t curr_sgl_offset = + offset_of_sgls + i * sizeof(struct qat_inter_sgl); + struct qat_inter_sgl *sgl = + (struct qat_inter_sgl *)(mz_start + curr_sgl_offset); + array_of_pointers->pointer[i] = mz_start_phys + curr_sgl_offset; + + sgl->num_bufs = QAT_NUM_BUFS_IN_IM_SGL; + sgl->num_mapped_bufs = 0; + sgl->resrvd = 0; + sgl->buffers[0].addr = mz_start_phys + offset_of_flat_buffs + + ((i * QAT_NUM_BUFS_IN_IM_SGL) * buff_size); + sgl->buffers[0].len = buff_size; + sgl->buffers[0].resrvd = 0; + sgl->buffers[1].addr = mz_start_phys + offset_of_flat_buffs + + (((i * QAT_NUM_BUFS_IN_IM_SGL) + 1) * buff_size); + sgl->buffers[1].len = buff_size; + sgl->buffers[1].resrvd = 0; + +#if 1 + QAT_LOG(DEBUG, " : phys addr of sgl[%i] in array_of_pointers" + "= %lx", i, array_of_pointers->pointer[i]); + QAT_LOG(DEBUG, " : virt address of sgl[%i] = %p", i, sgl); + QAT_LOG(DEBUG, " : sgl->buffers[0].addr = %lx, len=%d", + sgl->buffers[0].addr, sgl->buffers[0].len); + QAT_LOG(DEBUG, " : sgl->buffers[1].addr = %lx, len=%d", + sgl->buffers[1].addr, sgl->buffers[1].len); +#endif + } +#if 0 + QAT_DP_HEXDUMP_LOG(DEBUG, "IM buffer memzone:", mz_start, 1504); +#endif + return memzone; +} + static struct rte_mempool * qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, uint32_t num_elements) @@ -176,6 +272,12 @@ qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, static void _qat_comp_dev_config_clear(struct qat_comp_dev_private *comp_dev) { + /* Free intermediate buffers */ + if (comp_dev->interm_buff_mz) { + rte_memzone_free(comp_dev->interm_buff_mz); + comp_dev->interm_buff_mz = NULL; + } + /* Free private_xform pool */ if (comp_dev->xformpool) { /* Free internal mempool for private xforms */ @@ -197,6 +299,21 @@ qat_comp_dev_config(struct rte_compressdev *dev, return -EINVAL; } + if (RTE_PMD_QAT_COMP_IM_BUFFER_SIZE == 0) { + QAT_LOG(WARNING, + "RTE_PMD_QAT_COMP_IM_BUFFER_SIZE = 0 in config file, so" + " QAT device can't be used for Dynamic Deflate. " + "Did you really intend to do this?"); + } else { + comp_dev->interm_buff_mz = + qat_comp_setup_inter_buffers(comp_dev, + RTE_PMD_QAT_COMP_IM_BUFFER_SIZE); + if (comp_dev->interm_buff_mz == NULL) { + ret = -ENOMEM; + goto error_out; + } + } + comp_dev->xformpool = qat_comp_create_xform_pool(comp_dev, config->max_nb_priv_xforms); if (comp_dev->xformpool == NULL) { @@ -365,6 +482,10 @@ qat_comp_dev_create(struct qat_pci_device *qat_pci_dev) QAT_LOG(ERR, "Compression PMD not supported on QAT dh895xcc"); return 0; } + if (qat_pci_dev->qat_dev_gen == QAT_GEN3) { + QAT_LOG(ERR, "Compression PMD not supported on QAT c4xxx"); + return 0; + } struct rte_compressdev_pmd_init_params init_params = { .name = "", -- 2.7.4 ^ permalink raw reply [flat|nested] 17+ messages in thread
* [dpdk-dev] [PATCH v4] compress/qat: enable dynamic huffman encoding 2018-10-11 17:03 ` [dpdk-dev] [PATCH v3] " Fiona Trahe @ 2018-10-15 23:16 ` Fiona Trahe 2018-10-17 10:29 ` Kusztal, ArkadiuszX 2018-10-17 20:48 ` [dpdk-dev] [PATCH v5] " Fiona Trahe 0 siblings, 2 replies; 17+ messages in thread From: Fiona Trahe @ 2018-10-15 23:16 UTC (permalink / raw) To: dev; +Cc: akhil.goyal, tomaszx.jozwiak, arkadiuszx.kusztal, Fiona Trahe Enable dynamic huffman encoding in the QAT comp PMD. Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> --- v4 changes: - use #define for IM buffer logs, disable all by default and use param for clearer hexdump length. v3 changes: - reverted to 1 buffer per intermediate buffer sgl v2 changes: - allocate 2 buffers per intermediate buffer sgl - Compile out trace for debugging intermediate buffers config/common_base | 1 + config/rte_config.h | 1 + doc/guides/compressdevs/features/qat.ini | 1 + doc/guides/compressdevs/qat_comp.rst | 7 +- drivers/common/qat/qat_device.c | 4 + drivers/common/qat/qat_device.h | 7 ++ drivers/compress/qat/qat_comp.c | 39 +++++++--- drivers/compress/qat/qat_comp.h | 13 ++++ drivers/compress/qat/qat_comp_pmd.c | 125 ++++++++++++++++++++++++++++++- 9 files changed, 181 insertions(+), 17 deletions(-) diff --git a/config/common_base b/config/common_base index 5519d71..29c71b1 100644 --- a/config/common_base +++ b/config/common_base @@ -511,6 +511,7 @@ CONFIG_RTE_LIBRTE_PMD_QAT_SYM=n # CONFIG_RTE_PMD_QAT_MAX_PCI_DEVICES=48 CONFIG_RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS=16 +CONFIG_RTE_PMD_QAT_COMP_IM_BUFFER_SIZE=65536 # # Compile PMD for virtio crypto devices diff --git a/config/rte_config.h b/config/rte_config.h index 816e6f8..333fb0b 100644 --- a/config/rte_config.h +++ b/config/rte_config.h @@ -95,6 +95,7 @@ /* Max. number of QuickAssist devices which can be attached */ #define RTE_PMD_QAT_MAX_PCI_DEVICES 48 #define RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS 16 +#define RTE_PMD_QAT_COMP_IM_BUFFER_SIZE 65536 /* virtio crypto defines */ #define RTE_MAX_VIRTIO_CRYPTO 32 diff --git a/doc/guides/compressdevs/features/qat.ini b/doc/guides/compressdevs/features/qat.ini index 5cd4524..6b1e7f9 100644 --- a/doc/guides/compressdevs/features/qat.ini +++ b/doc/guides/compressdevs/features/qat.ini @@ -13,3 +13,4 @@ Adler32 = Y Crc32 = Y Adler32&Crc32 = Y Fixed = Y +Dynamic = Y diff --git a/doc/guides/compressdevs/qat_comp.rst b/doc/guides/compressdevs/qat_comp.rst index 7bffbe6..aee3b99 100644 --- a/doc/guides/compressdevs/qat_comp.rst +++ b/doc/guides/compressdevs/qat_comp.rst @@ -18,11 +18,7 @@ QAT compression PMD has support for: Compression/Decompression algorithm: - * DEFLATE - -Huffman code type: - - * FIXED + * DEFLATE - using Fixed and Dynamic Huffman encoding Window size support: @@ -36,7 +32,6 @@ Limitations ----------- * Compressdev level 0, no compression, is not supported. -* Dynamic Huffman encoding is not yet supported. * Queue pairs are not thread-safe (that is, within a single queue pair, RX and TX from different lcores is not supported). * No BSD support as BSD QAT kernel driver not available. diff --git a/drivers/common/qat/qat_device.c b/drivers/common/qat/qat_device.c index b158fb9..b2db8b0 100644 --- a/drivers/common/qat/qat_device.c +++ b/drivers/common/qat/qat_device.c @@ -7,6 +7,7 @@ #include "qat_device.h" #include "adf_transport_access_macros.h" #include "qat_sym_pmd.h" +#include "qat_comp_pmd.h" /* Hardware device information per generation */ __extension__ @@ -14,15 +15,18 @@ struct qat_gen_hw_data qat_gen_config[] = { [QAT_GEN1] = { .dev_gen = QAT_GEN1, .qp_hw_data = qat_gen1_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN1 }, [QAT_GEN2] = { .dev_gen = QAT_GEN2, .qp_hw_data = qat_gen1_qps, /* gen2 has same ring layout as gen1 */ + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN2 }, [QAT_GEN3] = { .dev_gen = QAT_GEN3, .qp_hw_data = qat_gen3_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN3 }, }; diff --git a/drivers/common/qat/qat_device.h b/drivers/common/qat/qat_device.h index 3a71cd4..eb81c78 100644 --- a/drivers/common/qat/qat_device.h +++ b/drivers/common/qat/qat_device.h @@ -16,6 +16,12 @@ #define QAT_DEV_NAME_MAX_LEN 64 +enum qat_comp_num_im_buffers { + QAT_NUM_INTERM_BUFS_GEN1 = 12, + QAT_NUM_INTERM_BUFS_GEN2 = 20, + QAT_NUM_INTERM_BUFS_GEN3 = 20 +}; + /* * This struct holds all the data about a QAT pci device * including data about all services it supports. @@ -72,6 +78,7 @@ struct qat_pci_device { struct qat_gen_hw_data { enum qat_device_gen dev_gen; const struct qat_qp_hw_data (*qp_hw_data)[ADF_MAX_QPS_ON_ANY_SERVICE]; + enum qat_comp_num_im_buffers comp_num_im_bufs_required; }; extern struct qat_gen_hw_data qat_gen_config[]; diff --git a/drivers/compress/qat/qat_comp.c b/drivers/compress/qat/qat_comp.c index b9336f3..d70c594 100644 --- a/drivers/compress/qat/qat_comp.c +++ b/drivers/compress/qat/qat_comp.c @@ -192,7 +192,7 @@ static void qat_comp_create_req_hdr(struct icp_qat_fw_comn_req_hdr *header, } static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, - const struct rte_memzone *interm_buff_mz __rte_unused, + const struct rte_memzone *interm_buff_mz, const struct rte_comp_xform *xform) { struct icp_qat_fw_comp_req *comp_req; @@ -280,10 +280,20 @@ static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, ICP_QAT_FW_SLICE_COMP); } else if (qat_xform->qat_comp_request_type == - QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { - QAT_LOG(ERR, "Dynamic huffman encoding not supported"); - return -EINVAL; + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_COMP); + + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_DRAM_WR); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + + comp_req->u1.xlt_pars.inter_buff_ptr = + interm_buff_mz->phys_addr; } #if RTE_LOG_DP_LEVEL >= RTE_LOG_DEBUG @@ -334,12 +344,6 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, (struct qat_comp_xform *)*private_xform; if (xform->type == RTE_COMP_COMPRESS) { - if (xform->compress.deflate.huffman == - RTE_COMP_HUFFMAN_DYNAMIC) { - QAT_LOG(ERR, - "QAT device doesn't support dynamic compression"); - return -ENOTSUP; - } if (xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_FIXED || ((xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_DEFAULT) @@ -347,6 +351,21 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, qat_xform->qat_comp_request_type = QAT_COMP_REQUEST_FIXED_COMP_STATELESS; + else if ((xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DYNAMIC || + xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DEFAULT) && + qat->interm_buff_mz != NULL) + + qat_xform->qat_comp_request_type = + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS; + + else { + QAT_LOG(ERR, + "IM buffers needed for dynamic deflate. Set size in config file"); + return -EINVAL; + } + qat_xform->checksum_type = xform->compress.chksum; } else { diff --git a/drivers/compress/qat/qat_comp.h b/drivers/compress/qat/qat_comp.h index 8d315ef..99a4462 100644 --- a/drivers/compress/qat/qat_comp.h +++ b/drivers/compress/qat/qat_comp.h @@ -15,6 +15,10 @@ #include "icp_qat_fw_comp.h" #include "icp_qat_fw_la.h" +#define QAT_64_BYTE_ALIGN_MASK (~0x3f) +#define QAT_64_BYTE_ALIGN (64) +#define QAT_NUM_BUFS_IN_IM_SGL 1 + #define ERR_CODE_QAT_COMP_WRONG_FW -99 enum qat_comp_request_type { @@ -24,6 +28,15 @@ enum qat_comp_request_type { REQ_COMP_END }; +struct array_of_ptrs { + phys_addr_t pointer[0]; +}; + +struct qat_inter_sgl { + qat_sgl_hdr; + struct qat_flat_buf buffers[QAT_NUM_BUFS_IN_IM_SGL]; +} __rte_packed __rte_cache_aligned; + struct qat_comp_sgl { qat_sgl_hdr; struct qat_flat_buf buffers[RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS]; diff --git a/drivers/compress/qat/qat_comp_pmd.c b/drivers/compress/qat/qat_comp_pmd.c index 63af23a..8ea634e 100644 --- a/drivers/compress/qat/qat_comp_pmd.c +++ b/drivers/compress/qat/qat_comp_pmd.c @@ -14,6 +14,7 @@ static const struct rte_compressdev_capabilities qat_comp_gen_capabilities[] = { RTE_COMP_FF_CRC32_ADLER32_CHECKSUM | RTE_COMP_FF_SHAREABLE_PRIV_XFORM | RTE_COMP_FF_HUFFMAN_FIXED | + RTE_COMP_FF_HUFFMAN_DYNAMIC | RTE_COMP_FF_OOP_SGL_IN_SGL_OUT | RTE_COMP_FF_OOP_SGL_IN_LB_OUT | RTE_COMP_FF_OOP_LB_IN_SGL_OUT, @@ -112,7 +113,7 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, /* store a link to the qp in the qat_pci_device */ qat_private->qat_dev->qps_in_use[QAT_SERVICE_COMPRESSION][qp_id] - = *qp_addr; + = *qp_addr; qp = (struct qat_qp *)*qp_addr; @@ -135,6 +136,103 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, return ret; } + +#define QAT_IM_BUFFER_DEBUG 0 +static const struct rte_memzone * +qat_comp_setup_inter_buffers(struct qat_comp_dev_private *comp_dev, + uint32_t buff_size) +{ + char inter_buff_mz_name[RTE_MEMZONE_NAMESIZE]; + const struct rte_memzone *memzone; + uint8_t *mz_start = NULL; + phys_addr_t mz_start_phys = 0; + struct array_of_ptrs *array_of_pointers; + int size_of_ptr_array; + uint32_t full_size; + uint32_t offset_of_sgls, offset_of_flat_buffs = 0; + int i; + int num_im_sgls = qat_gen_config[ + comp_dev->qat_dev->qat_dev_gen].comp_num_im_bufs_required; + + QAT_LOG(DEBUG, "QAT COMP device %s needs %d sgls", + comp_dev->qat_dev->name, num_im_sgls); + snprintf(inter_buff_mz_name, RTE_MEMZONE_NAMESIZE, + "%s_inter_buff", comp_dev->qat_dev->name); + memzone = rte_memzone_lookup(inter_buff_mz_name); + if (memzone != NULL) { + QAT_LOG(DEBUG, "QAT COMP im buffer memzone created already"); + return memzone; + } + + /* Create a memzone to hold intermediate buffers and associated + * meta-data needed by the firmware. The memzone contains: + * - a list of num_im_sgls physical pointers to sgls + * - the num_im_sgl sgl structures, each pointing to 2 flat buffers + * - the flat buffers: num_im_sgl * 2 + * where num_im_sgls depends on the hardware generation of the device + */ + + size_of_ptr_array = num_im_sgls * sizeof(phys_addr_t); + offset_of_sgls = (size_of_ptr_array + (~QAT_64_BYTE_ALIGN_MASK)) + & QAT_64_BYTE_ALIGN_MASK; + offset_of_flat_buffs = + offset_of_sgls + num_im_sgls * sizeof(struct qat_inter_sgl); + full_size = offset_of_flat_buffs + + num_im_sgls * buff_size * QAT_NUM_BUFS_IN_IM_SGL; + + memzone = rte_memzone_reserve_aligned(inter_buff_mz_name, full_size, + comp_dev->compressdev->data->socket_id, + RTE_MEMZONE_2MB, QAT_64_BYTE_ALIGN); + if (memzone == NULL) { + QAT_LOG(ERR, "Can't allocate intermediate buffers" + " for device %s", comp_dev->qat_dev->name); + return NULL; + } + + mz_start = (uint8_t *)memzone->addr; + mz_start_phys = memzone->phys_addr; + QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = %lx, size required %d," + "size created %ld", + inter_buff_mz_name, mz_start, mz_start_phys, + full_size, memzone->len); + + array_of_pointers = (struct array_of_ptrs *)mz_start; + for (i = 0; i < num_im_sgls; i++) { + uint32_t curr_sgl_offset = + offset_of_sgls + i * sizeof(struct qat_inter_sgl); + struct qat_inter_sgl *sgl = + (struct qat_inter_sgl *)(mz_start + curr_sgl_offset); + array_of_pointers->pointer[i] = mz_start_phys + curr_sgl_offset; + + sgl->num_bufs = QAT_NUM_BUFS_IN_IM_SGL; + sgl->num_mapped_bufs = 0; + sgl->resrvd = 0; + sgl->buffers[0].addr = mz_start_phys + offset_of_flat_buffs + + ((i * QAT_NUM_BUFS_IN_IM_SGL) * buff_size); + sgl->buffers[0].len = buff_size; + sgl->buffers[0].resrvd = 0; + sgl->buffers[1].addr = mz_start_phys + offset_of_flat_buffs + + (((i * QAT_NUM_BUFS_IN_IM_SGL) + 1) * buff_size); + sgl->buffers[1].len = buff_size; + sgl->buffers[1].resrvd = 0; + +#if QAT_IM_BUFFER_DEBUG + QAT_LOG(DEBUG, " : phys addr of sgl[%i] in array_of_pointers" + "= %lx", i, array_of_pointers->pointer[i]); + QAT_LOG(DEBUG, " : virt address of sgl[%i] = %p", i, sgl); + QAT_LOG(DEBUG, " : sgl->buffers[0].addr = %lx, len=%d", + sgl->buffers[0].addr, sgl->buffers[0].len); + QAT_LOG(DEBUG, " : sgl->buffers[1].addr = %lx, len=%d", + sgl->buffers[1].addr, sgl->buffers[1].len); +#endif + } +#if QAT_IM_BUFFER_DEBUG + QAT_DP_HEXDUMP_LOG(DEBUG, "IM buffer memzone start:", + mz_start, offset_of_flat_buffs + 32); +#endif + return memzone; +} + static struct rte_mempool * qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, uint32_t num_elements) @@ -176,6 +274,12 @@ qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, static void _qat_comp_dev_config_clear(struct qat_comp_dev_private *comp_dev) { + /* Free intermediate buffers */ + if (comp_dev->interm_buff_mz) { + rte_memzone_free(comp_dev->interm_buff_mz); + comp_dev->interm_buff_mz = NULL; + } + /* Free private_xform pool */ if (comp_dev->xformpool) { /* Free internal mempool for private xforms */ @@ -197,6 +301,21 @@ qat_comp_dev_config(struct rte_compressdev *dev, return -EINVAL; } + if (RTE_PMD_QAT_COMP_IM_BUFFER_SIZE == 0) { + QAT_LOG(WARNING, + "RTE_PMD_QAT_COMP_IM_BUFFER_SIZE = 0 in config file, so" + " QAT device can't be used for Dynamic Deflate. " + "Did you really intend to do this?"); + } else { + comp_dev->interm_buff_mz = + qat_comp_setup_inter_buffers(comp_dev, + RTE_PMD_QAT_COMP_IM_BUFFER_SIZE); + if (comp_dev->interm_buff_mz == NULL) { + ret = -ENOMEM; + goto error_out; + } + } + comp_dev->xformpool = qat_comp_create_xform_pool(comp_dev, config->max_nb_priv_xforms); if (comp_dev->xformpool == NULL) { @@ -365,6 +484,10 @@ qat_comp_dev_create(struct qat_pci_device *qat_pci_dev) QAT_LOG(ERR, "Compression PMD not supported on QAT dh895xcc"); return 0; } + if (qat_pci_dev->qat_dev_gen == QAT_GEN3) { + QAT_LOG(ERR, "Compression PMD not supported on QAT c4xxx"); + return 0; + } struct rte_compressdev_pmd_init_params init_params = { .name = "", -- 2.7.4 ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [dpdk-dev] [PATCH v4] compress/qat: enable dynamic huffman encoding 2018-10-15 23:16 ` [dpdk-dev] [PATCH v4] " Fiona Trahe @ 2018-10-17 10:29 ` Kusztal, ArkadiuszX 2018-10-17 20:48 ` [dpdk-dev] [PATCH v5] " Fiona Trahe 1 sibling, 0 replies; 17+ messages in thread From: Kusztal, ArkadiuszX @ 2018-10-17 10:29 UTC (permalink / raw) To: Trahe, Fiona, dev; +Cc: akhil.goyal, Jozwiak, TomaszX > -----Original Message----- > From: Trahe, Fiona > Sent: Tuesday, October 16, 2018 1:16 AM > To: dev@dpdk.org > Cc: akhil.goyal@nxp.com; Jozwiak, TomaszX <tomaszx.jozwiak@intel.com>; > Kusztal, ArkadiuszX <arkadiuszx.kusztal@intel.com>; Trahe, Fiona > <fiona.trahe@intel.com> > Subject: [PATCH v4] compress/qat: enable dynamic huffman encoding > > Enable dynamic huffman encoding in the QAT comp PMD. > > Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> > Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> > --- > v4 changes: > - use #define for IM buffer logs, disable all by default > and use param for clearer hexdump length. > > v3 changes: > - reverted to 1 buffer per intermediate buffer sgl > > v2 changes: > - allocate 2 buffers per intermediate buffer sgl > - Compile out trace for debugging intermediate buffers > > config/common_base | 1 + > config/rte_config.h | 1 + > doc/guides/compressdevs/features/qat.ini | 1 + > doc/guides/compressdevs/qat_comp.rst | 7 +- > drivers/common/qat/qat_device.c | 4 + > drivers/common/qat/qat_device.h | 7 ++ > drivers/compress/qat/qat_comp.c | 39 +++++++--- > drivers/compress/qat/qat_comp.h | 13 ++++ > drivers/compress/qat/qat_comp_pmd.c | 125 > ++++++++++++++++++++++++++++++- > 9 files changed, 181 insertions(+), 17 deletions(-) > > diff --git a/config/common_base b/config/common_base index > 5519d71..29c71b1 100644 > --- a/config/common_base > +++ b/config/common_base > 2.7.4 Acked-by: Arek Kusztal <arkadiuszx.kusztal@intel.com> ^ permalink raw reply [flat|nested] 17+ messages in thread
* [dpdk-dev] [PATCH v5] compress/qat: enable dynamic huffman encoding 2018-10-15 23:16 ` [dpdk-dev] [PATCH v4] " Fiona Trahe 2018-10-17 10:29 ` Kusztal, ArkadiuszX @ 2018-10-17 20:48 ` Fiona Trahe 2018-10-24 14:15 ` Akhil Goyal 2018-10-25 1:30 ` [dpdk-dev] [PATCH v6] " Fiona Trahe 1 sibling, 2 replies; 17+ messages in thread From: Fiona Trahe @ 2018-10-17 20:48 UTC (permalink / raw) To: dev; +Cc: akhil.goyal, tomaszx.jozwiak, arkadiuszx.kusztal, Fiona Trahe Enable dynamic huffman encoding in the QAT comp PMD. Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> Acked-by: Arek Kusztal <arkadiuszx.kusztal@intel.com> --- v5 changes: - update release note v4 changes: - use #define for IM buffer logs, disable all by default and use param for clearer hexdump length. v3 changes: - reverted to 1 buffer per intermediate buffer sgl v2 changes: - allocate 2 buffers per intermediate buffer sgl - Compile out trace for debugging intermediate buffers config/common_base | 1 + config/rte_config.h | 1 + doc/guides/compressdevs/features/qat.ini | 1 + doc/guides/compressdevs/qat_comp.rst | 7 +- doc/guides/rel_notes/release_18_11.rst | 5 ++ drivers/common/qat/qat_device.c | 4 + drivers/common/qat/qat_device.h | 7 ++ drivers/compress/qat/qat_comp.c | 39 +++++++--- drivers/compress/qat/qat_comp.h | 13 ++++ drivers/compress/qat/qat_comp_pmd.c | 125 ++++++++++++++++++++++++++++++- 10 files changed, 186 insertions(+), 17 deletions(-) diff --git a/config/common_base b/config/common_base index 9f45d9e..7ccf11b 100644 --- a/config/common_base +++ b/config/common_base @@ -522,6 +522,7 @@ CONFIG_RTE_LIBRTE_PMD_QAT_SYM=n # CONFIG_RTE_PMD_QAT_MAX_PCI_DEVICES=48 CONFIG_RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS=16 +CONFIG_RTE_PMD_QAT_COMP_IM_BUFFER_SIZE=65536 # # Compile PMD for virtio crypto devices diff --git a/config/rte_config.h b/config/rte_config.h index 816e6f8..333fb0b 100644 --- a/config/rte_config.h +++ b/config/rte_config.h @@ -95,6 +95,7 @@ /* Max. number of QuickAssist devices which can be attached */ #define RTE_PMD_QAT_MAX_PCI_DEVICES 48 #define RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS 16 +#define RTE_PMD_QAT_COMP_IM_BUFFER_SIZE 65536 /* virtio crypto defines */ #define RTE_MAX_VIRTIO_CRYPTO 32 diff --git a/doc/guides/compressdevs/features/qat.ini b/doc/guides/compressdevs/features/qat.ini index 5cd4524..6b1e7f9 100644 --- a/doc/guides/compressdevs/features/qat.ini +++ b/doc/guides/compressdevs/features/qat.ini @@ -13,3 +13,4 @@ Adler32 = Y Crc32 = Y Adler32&Crc32 = Y Fixed = Y +Dynamic = Y diff --git a/doc/guides/compressdevs/qat_comp.rst b/doc/guides/compressdevs/qat_comp.rst index 7bffbe6..aee3b99 100644 --- a/doc/guides/compressdevs/qat_comp.rst +++ b/doc/guides/compressdevs/qat_comp.rst @@ -18,11 +18,7 @@ QAT compression PMD has support for: Compression/Decompression algorithm: - * DEFLATE - -Huffman code type: - - * FIXED + * DEFLATE - using Fixed and Dynamic Huffman encoding Window size support: @@ -36,7 +32,6 @@ Limitations ----------- * Compressdev level 0, no compression, is not supported. -* Dynamic Huffman encoding is not yet supported. * Queue pairs are not thread-safe (that is, within a single queue pair, RX and TX from different lcores is not supported). * No BSD support as BSD QAT kernel driver not available. diff --git a/doc/guides/rel_notes/release_18_11.rst b/doc/guides/rel_notes/release_18_11.rst index af9a6f8..3031e4f 100644 --- a/doc/guides/rel_notes/release_18_11.rst +++ b/doc/guides/rel_notes/release_18_11.rst @@ -191,6 +191,11 @@ New Features Added the new caam job ring driver for NXP platforms. See the "NXP CAAM JOB RING (caam_jr)" document for more details on this new driver. +* **Added support for Dynamic Huffman Encoding to Intel QAT comp PMD.** + + The Intel QuickAssist (QAT) compression PMD has been updated with support + for Dynamic Huffman Encoding for the Deflate algorithm. + API Changes ----------- diff --git a/drivers/common/qat/qat_device.c b/drivers/common/qat/qat_device.c index 8d4df92..5df831f 100644 --- a/drivers/common/qat/qat_device.c +++ b/drivers/common/qat/qat_device.c @@ -7,6 +7,7 @@ #include "qat_device.h" #include "adf_transport_access_macros.h" #include "qat_sym_pmd.h" +#include "qat_comp_pmd.h" /* Hardware device information per generation */ __extension__ @@ -14,15 +15,18 @@ struct qat_gen_hw_data qat_gen_config[] = { [QAT_GEN1] = { .dev_gen = QAT_GEN1, .qp_hw_data = qat_gen1_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN1 }, [QAT_GEN2] = { .dev_gen = QAT_GEN2, .qp_hw_data = qat_gen1_qps, /* gen2 has same ring layout as gen1 */ + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN2 }, [QAT_GEN3] = { .dev_gen = QAT_GEN3, .qp_hw_data = qat_gen3_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN3 }, }; diff --git a/drivers/common/qat/qat_device.h b/drivers/common/qat/qat_device.h index 3a71cd4..eb81c78 100644 --- a/drivers/common/qat/qat_device.h +++ b/drivers/common/qat/qat_device.h @@ -16,6 +16,12 @@ #define QAT_DEV_NAME_MAX_LEN 64 +enum qat_comp_num_im_buffers { + QAT_NUM_INTERM_BUFS_GEN1 = 12, + QAT_NUM_INTERM_BUFS_GEN2 = 20, + QAT_NUM_INTERM_BUFS_GEN3 = 20 +}; + /* * This struct holds all the data about a QAT pci device * including data about all services it supports. @@ -72,6 +78,7 @@ struct qat_pci_device { struct qat_gen_hw_data { enum qat_device_gen dev_gen; const struct qat_qp_hw_data (*qp_hw_data)[ADF_MAX_QPS_ON_ANY_SERVICE]; + enum qat_comp_num_im_buffers comp_num_im_bufs_required; }; extern struct qat_gen_hw_data qat_gen_config[]; diff --git a/drivers/compress/qat/qat_comp.c b/drivers/compress/qat/qat_comp.c index b9336f3..d70c594 100644 --- a/drivers/compress/qat/qat_comp.c +++ b/drivers/compress/qat/qat_comp.c @@ -192,7 +192,7 @@ static void qat_comp_create_req_hdr(struct icp_qat_fw_comn_req_hdr *header, } static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, - const struct rte_memzone *interm_buff_mz __rte_unused, + const struct rte_memzone *interm_buff_mz, const struct rte_comp_xform *xform) { struct icp_qat_fw_comp_req *comp_req; @@ -280,10 +280,20 @@ static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, ICP_QAT_FW_SLICE_COMP); } else if (qat_xform->qat_comp_request_type == - QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { - QAT_LOG(ERR, "Dynamic huffman encoding not supported"); - return -EINVAL; + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_COMP); + + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_DRAM_WR); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + + comp_req->u1.xlt_pars.inter_buff_ptr = + interm_buff_mz->phys_addr; } #if RTE_LOG_DP_LEVEL >= RTE_LOG_DEBUG @@ -334,12 +344,6 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, (struct qat_comp_xform *)*private_xform; if (xform->type == RTE_COMP_COMPRESS) { - if (xform->compress.deflate.huffman == - RTE_COMP_HUFFMAN_DYNAMIC) { - QAT_LOG(ERR, - "QAT device doesn't support dynamic compression"); - return -ENOTSUP; - } if (xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_FIXED || ((xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_DEFAULT) @@ -347,6 +351,21 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, qat_xform->qat_comp_request_type = QAT_COMP_REQUEST_FIXED_COMP_STATELESS; + else if ((xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DYNAMIC || + xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DEFAULT) && + qat->interm_buff_mz != NULL) + + qat_xform->qat_comp_request_type = + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS; + + else { + QAT_LOG(ERR, + "IM buffers needed for dynamic deflate. Set size in config file"); + return -EINVAL; + } + qat_xform->checksum_type = xform->compress.chksum; } else { diff --git a/drivers/compress/qat/qat_comp.h b/drivers/compress/qat/qat_comp.h index 8d315ef..99a4462 100644 --- a/drivers/compress/qat/qat_comp.h +++ b/drivers/compress/qat/qat_comp.h @@ -15,6 +15,10 @@ #include "icp_qat_fw_comp.h" #include "icp_qat_fw_la.h" +#define QAT_64_BYTE_ALIGN_MASK (~0x3f) +#define QAT_64_BYTE_ALIGN (64) +#define QAT_NUM_BUFS_IN_IM_SGL 1 + #define ERR_CODE_QAT_COMP_WRONG_FW -99 enum qat_comp_request_type { @@ -24,6 +28,15 @@ enum qat_comp_request_type { REQ_COMP_END }; +struct array_of_ptrs { + phys_addr_t pointer[0]; +}; + +struct qat_inter_sgl { + qat_sgl_hdr; + struct qat_flat_buf buffers[QAT_NUM_BUFS_IN_IM_SGL]; +} __rte_packed __rte_cache_aligned; + struct qat_comp_sgl { qat_sgl_hdr; struct qat_flat_buf buffers[RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS]; diff --git a/drivers/compress/qat/qat_comp_pmd.c b/drivers/compress/qat/qat_comp_pmd.c index 63af23a..8ea634e 100644 --- a/drivers/compress/qat/qat_comp_pmd.c +++ b/drivers/compress/qat/qat_comp_pmd.c @@ -14,6 +14,7 @@ static const struct rte_compressdev_capabilities qat_comp_gen_capabilities[] = { RTE_COMP_FF_CRC32_ADLER32_CHECKSUM | RTE_COMP_FF_SHAREABLE_PRIV_XFORM | RTE_COMP_FF_HUFFMAN_FIXED | + RTE_COMP_FF_HUFFMAN_DYNAMIC | RTE_COMP_FF_OOP_SGL_IN_SGL_OUT | RTE_COMP_FF_OOP_SGL_IN_LB_OUT | RTE_COMP_FF_OOP_LB_IN_SGL_OUT, @@ -112,7 +113,7 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, /* store a link to the qp in the qat_pci_device */ qat_private->qat_dev->qps_in_use[QAT_SERVICE_COMPRESSION][qp_id] - = *qp_addr; + = *qp_addr; qp = (struct qat_qp *)*qp_addr; @@ -135,6 +136,103 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, return ret; } + +#define QAT_IM_BUFFER_DEBUG 0 +static const struct rte_memzone * +qat_comp_setup_inter_buffers(struct qat_comp_dev_private *comp_dev, + uint32_t buff_size) +{ + char inter_buff_mz_name[RTE_MEMZONE_NAMESIZE]; + const struct rte_memzone *memzone; + uint8_t *mz_start = NULL; + phys_addr_t mz_start_phys = 0; + struct array_of_ptrs *array_of_pointers; + int size_of_ptr_array; + uint32_t full_size; + uint32_t offset_of_sgls, offset_of_flat_buffs = 0; + int i; + int num_im_sgls = qat_gen_config[ + comp_dev->qat_dev->qat_dev_gen].comp_num_im_bufs_required; + + QAT_LOG(DEBUG, "QAT COMP device %s needs %d sgls", + comp_dev->qat_dev->name, num_im_sgls); + snprintf(inter_buff_mz_name, RTE_MEMZONE_NAMESIZE, + "%s_inter_buff", comp_dev->qat_dev->name); + memzone = rte_memzone_lookup(inter_buff_mz_name); + if (memzone != NULL) { + QAT_LOG(DEBUG, "QAT COMP im buffer memzone created already"); + return memzone; + } + + /* Create a memzone to hold intermediate buffers and associated + * meta-data needed by the firmware. The memzone contains: + * - a list of num_im_sgls physical pointers to sgls + * - the num_im_sgl sgl structures, each pointing to 2 flat buffers + * - the flat buffers: num_im_sgl * 2 + * where num_im_sgls depends on the hardware generation of the device + */ + + size_of_ptr_array = num_im_sgls * sizeof(phys_addr_t); + offset_of_sgls = (size_of_ptr_array + (~QAT_64_BYTE_ALIGN_MASK)) + & QAT_64_BYTE_ALIGN_MASK; + offset_of_flat_buffs = + offset_of_sgls + num_im_sgls * sizeof(struct qat_inter_sgl); + full_size = offset_of_flat_buffs + + num_im_sgls * buff_size * QAT_NUM_BUFS_IN_IM_SGL; + + memzone = rte_memzone_reserve_aligned(inter_buff_mz_name, full_size, + comp_dev->compressdev->data->socket_id, + RTE_MEMZONE_2MB, QAT_64_BYTE_ALIGN); + if (memzone == NULL) { + QAT_LOG(ERR, "Can't allocate intermediate buffers" + " for device %s", comp_dev->qat_dev->name); + return NULL; + } + + mz_start = (uint8_t *)memzone->addr; + mz_start_phys = memzone->phys_addr; + QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = %lx, size required %d," + "size created %ld", + inter_buff_mz_name, mz_start, mz_start_phys, + full_size, memzone->len); + + array_of_pointers = (struct array_of_ptrs *)mz_start; + for (i = 0; i < num_im_sgls; i++) { + uint32_t curr_sgl_offset = + offset_of_sgls + i * sizeof(struct qat_inter_sgl); + struct qat_inter_sgl *sgl = + (struct qat_inter_sgl *)(mz_start + curr_sgl_offset); + array_of_pointers->pointer[i] = mz_start_phys + curr_sgl_offset; + + sgl->num_bufs = QAT_NUM_BUFS_IN_IM_SGL; + sgl->num_mapped_bufs = 0; + sgl->resrvd = 0; + sgl->buffers[0].addr = mz_start_phys + offset_of_flat_buffs + + ((i * QAT_NUM_BUFS_IN_IM_SGL) * buff_size); + sgl->buffers[0].len = buff_size; + sgl->buffers[0].resrvd = 0; + sgl->buffers[1].addr = mz_start_phys + offset_of_flat_buffs + + (((i * QAT_NUM_BUFS_IN_IM_SGL) + 1) * buff_size); + sgl->buffers[1].len = buff_size; + sgl->buffers[1].resrvd = 0; + +#if QAT_IM_BUFFER_DEBUG + QAT_LOG(DEBUG, " : phys addr of sgl[%i] in array_of_pointers" + "= %lx", i, array_of_pointers->pointer[i]); + QAT_LOG(DEBUG, " : virt address of sgl[%i] = %p", i, sgl); + QAT_LOG(DEBUG, " : sgl->buffers[0].addr = %lx, len=%d", + sgl->buffers[0].addr, sgl->buffers[0].len); + QAT_LOG(DEBUG, " : sgl->buffers[1].addr = %lx, len=%d", + sgl->buffers[1].addr, sgl->buffers[1].len); +#endif + } +#if QAT_IM_BUFFER_DEBUG + QAT_DP_HEXDUMP_LOG(DEBUG, "IM buffer memzone start:", + mz_start, offset_of_flat_buffs + 32); +#endif + return memzone; +} + static struct rte_mempool * qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, uint32_t num_elements) @@ -176,6 +274,12 @@ qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, static void _qat_comp_dev_config_clear(struct qat_comp_dev_private *comp_dev) { + /* Free intermediate buffers */ + if (comp_dev->interm_buff_mz) { + rte_memzone_free(comp_dev->interm_buff_mz); + comp_dev->interm_buff_mz = NULL; + } + /* Free private_xform pool */ if (comp_dev->xformpool) { /* Free internal mempool for private xforms */ @@ -197,6 +301,21 @@ qat_comp_dev_config(struct rte_compressdev *dev, return -EINVAL; } + if (RTE_PMD_QAT_COMP_IM_BUFFER_SIZE == 0) { + QAT_LOG(WARNING, + "RTE_PMD_QAT_COMP_IM_BUFFER_SIZE = 0 in config file, so" + " QAT device can't be used for Dynamic Deflate. " + "Did you really intend to do this?"); + } else { + comp_dev->interm_buff_mz = + qat_comp_setup_inter_buffers(comp_dev, + RTE_PMD_QAT_COMP_IM_BUFFER_SIZE); + if (comp_dev->interm_buff_mz == NULL) { + ret = -ENOMEM; + goto error_out; + } + } + comp_dev->xformpool = qat_comp_create_xform_pool(comp_dev, config->max_nb_priv_xforms); if (comp_dev->xformpool == NULL) { @@ -365,6 +484,10 @@ qat_comp_dev_create(struct qat_pci_device *qat_pci_dev) QAT_LOG(ERR, "Compression PMD not supported on QAT dh895xcc"); return 0; } + if (qat_pci_dev->qat_dev_gen == QAT_GEN3) { + QAT_LOG(ERR, "Compression PMD not supported on QAT c4xxx"); + return 0; + } struct rte_compressdev_pmd_init_params init_params = { .name = "", -- 2.7.4 ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [dpdk-dev] [PATCH v5] compress/qat: enable dynamic huffman encoding 2018-10-17 20:48 ` [dpdk-dev] [PATCH v5] " Fiona Trahe @ 2018-10-24 14:15 ` Akhil Goyal 2018-10-24 21:30 ` Thomas Monjalon 2018-10-25 1:30 ` [dpdk-dev] [PATCH v6] " Fiona Trahe 1 sibling, 1 reply; 17+ messages in thread From: Akhil Goyal @ 2018-10-24 14:15 UTC (permalink / raw) To: Fiona Trahe, dev; +Cc: tomaszx.jozwiak, arkadiuszx.kusztal On 10/18/2018 2:18 AM, Fiona Trahe wrote: > Enable dynamic huffman encoding in the QAT comp PMD. > > Signed-off-by: Tomasz Jozwiak<tomaszx.jozwiak@intel.com> > Signed-off-by: Fiona Trahe<fiona.trahe@intel.com> > Acked-by: Arek Kusztal<arkadiuszx.kusztal@intel.com> > --- Applied to dpdk-next-crypto Thanks ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [dpdk-dev] [PATCH v5] compress/qat: enable dynamic huffman encoding 2018-10-24 14:15 ` Akhil Goyal @ 2018-10-24 21:30 ` Thomas Monjalon 2018-10-25 0:40 ` Trahe, Fiona 0 siblings, 1 reply; 17+ messages in thread From: Thomas Monjalon @ 2018-10-24 21:30 UTC (permalink / raw) To: Fiona Trahe; +Cc: dev, Akhil Goyal, tomaszx.jozwiak, arkadiuszx.kusztal 24/10/2018 16:15, Akhil Goyal: > > On 10/18/2018 2:18 AM, Fiona Trahe wrote: > > Enable dynamic huffman encoding in the QAT comp PMD. > > > > Signed-off-by: Tomasz Jozwiak<tomaszx.jozwiak@intel.com> > > Signed-off-by: Fiona Trahe<fiona.trahe@intel.com> > > Acked-by: Arek Kusztal<arkadiuszx.kusztal@intel.com> > > --- > Applied to dpdk-next-crypto error: format ‘%lx’ expects argument of type ‘long unsigned int’, but argument 7 has type ‘phys_addr_t’ {aka ‘long long unsigned int’} [-Werror=format=] "%s(): " fmt "\n", __func__, ## args) ^~~~~~~~ drivers/compress/qat/qat_comp_pmd.c:194:2: note: in expansion of macro ‘QAT_LOG’ QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = %lx, size required %d," One more comment about the release notes, please try to integrate compressdev features between cryptodev and eventdev. ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [dpdk-dev] [PATCH v5] compress/qat: enable dynamic huffman encoding 2018-10-24 21:30 ` Thomas Monjalon @ 2018-10-25 0:40 ` Trahe, Fiona 2018-10-25 6:45 ` Thomas Monjalon 0 siblings, 1 reply; 17+ messages in thread From: Trahe, Fiona @ 2018-10-25 0:40 UTC (permalink / raw) To: Thomas Monjalon Cc: dev, Akhil Goyal, Jozwiak, TomaszX, Kusztal, ArkadiuszX, Trahe, Fiona I'll send a v6. On which OS/compiler did this fail to compile? > -----Original Message----- > From: Thomas Monjalon [mailto:thomas@monjalon.net] > Sent: Wednesday, October 24, 2018 2:30 PM > To: Trahe, Fiona <fiona.trahe@intel.com> > Cc: dev@dpdk.org; Akhil Goyal <akhil.goyal@nxp.com>; Jozwiak, TomaszX <tomaszx.jozwiak@intel.com>; > Kusztal, ArkadiuszX <arkadiuszx.kusztal@intel.com> > Subject: Re: [dpdk-dev] [PATCH v5] compress/qat: enable dynamic huffman encoding > > 24/10/2018 16:15, Akhil Goyal: > > > > On 10/18/2018 2:18 AM, Fiona Trahe wrote: > > > Enable dynamic huffman encoding in the QAT comp PMD. > > > > > > Signed-off-by: Tomasz Jozwiak<tomaszx.jozwiak@intel.com> > > > Signed-off-by: Fiona Trahe<fiona.trahe@intel.com> > > > Acked-by: Arek Kusztal<arkadiuszx.kusztal@intel.com> > > > --- > > Applied to dpdk-next-crypto > > error: format ‘%lx’ expects argument of type ‘long unsigned int’, but argument 7 has type ‘phys_addr_t’ > {aka ‘long long unsigned int’} [-Werror=format=] > "%s(): " fmt "\n", __func__, ## args) > ^~~~~~~~ > drivers/compress/qat/qat_comp_pmd.c:194:2: note: in expansion of macro ‘QAT_LOG’ > QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = %lx, size required %d," > > > One more comment about the release notes, please try to integrate > compressdev features between cryptodev and eventdev. > ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [dpdk-dev] [PATCH v5] compress/qat: enable dynamic huffman encoding 2018-10-25 0:40 ` Trahe, Fiona @ 2018-10-25 6:45 ` Thomas Monjalon 0 siblings, 0 replies; 17+ messages in thread From: Thomas Monjalon @ 2018-10-25 6:45 UTC (permalink / raw) To: Trahe, Fiona; +Cc: dev, Akhil Goyal, Jozwiak, TomaszX, Kusztal, ArkadiuszX 25/10/2018 02:40, Trahe, Fiona: > I'll send a v6. > On which OS/compiler did this fail to compile? It fails on i686 of course. I already sent some general notes in the past, explaining how %lx is most of the time wrong. %PRIu64 is the right specifier. https://www.mail-archive.com/dev@dpdk.org/msg90644.html > From: Thomas Monjalon [mailto:thomas@monjalon.net] > > 24/10/2018 16:15, Akhil Goyal: > > > > > > On 10/18/2018 2:18 AM, Fiona Trahe wrote: > > > > Enable dynamic huffman encoding in the QAT comp PMD. > > > > > > > > Signed-off-by: Tomasz Jozwiak<tomaszx.jozwiak@intel.com> > > > > Signed-off-by: Fiona Trahe<fiona.trahe@intel.com> > > > > Acked-by: Arek Kusztal<arkadiuszx.kusztal@intel.com> > > > > --- > > > Applied to dpdk-next-crypto > > > > error: format ‘%lx’ expects argument of type ‘long unsigned int’, but argument 7 has type ‘phys_addr_t’ > > {aka ‘long long unsigned int’} [-Werror=format=] > > "%s(): " fmt "\n", __func__, ## args) > > ^~~~~~~~ > > drivers/compress/qat/qat_comp_pmd.c:194:2: note: in expansion of macro ‘QAT_LOG’ > > QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = %lx, size required %d," > > > > > > One more comment about the release notes, please try to integrate > > compressdev features between cryptodev and eventdev. > > > > ^ permalink raw reply [flat|nested] 17+ messages in thread
* [dpdk-dev] [PATCH v6] compress/qat: enable dynamic huffman encoding 2018-10-17 20:48 ` [dpdk-dev] [PATCH v5] " Fiona Trahe 2018-10-24 14:15 ` Akhil Goyal @ 2018-10-25 1:30 ` Fiona Trahe 2018-10-25 7:34 ` Akhil Goyal 2018-10-26 18:18 ` [dpdk-dev] [PATCH v7] " Fiona Trahe 1 sibling, 2 replies; 17+ messages in thread From: Fiona Trahe @ 2018-10-25 1:30 UTC (permalink / raw) To: dev; +Cc: thomas, akhil.goyal, tomaszx.jozwiak, arkadiuszx.kusztal, Fiona Trahe Enable dynamic huffman encoding in the QAT comp PMD. Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> Acked-by: Arek Kusztal <arkadiuszx.kusztal@intel.com> --- v6 changes: - replace %lx with PRIx64 to fix compiler issue - reorder entry in release notes so above eventdev items v5 changes: - update release note v4 changes: - use #define for IM buffer logs, disable all by default and use param for clearer hexdump length. v3 changes: - reverted to 1 buffer per intermediate buffer sgl v2 changes: - allocate 2 buffers per intermediate buffer sgl - Compile out trace for debugging intermediate buffers config/common_base | 1 + config/rte_config.h | 1 + doc/guides/compressdevs/features/qat.ini | 1 + doc/guides/compressdevs/qat_comp.rst | 7 +- doc/guides/rel_notes/release_18_11.rst | 5 ++ drivers/common/qat/qat_device.c | 4 + drivers/common/qat/qat_device.h | 7 ++ drivers/compress/qat/qat_comp.c | 39 +++++++--- drivers/compress/qat/qat_comp.h | 13 ++++ drivers/compress/qat/qat_comp_pmd.c | 125 ++++++++++++++++++++++++++++++- 10 files changed, 186 insertions(+), 17 deletions(-) diff --git a/config/common_base b/config/common_base index be7365e..a40c61f 100644 --- a/config/common_base +++ b/config/common_base @@ -532,6 +532,7 @@ CONFIG_RTE_LIBRTE_PMD_QAT_SYM=n # CONFIG_RTE_PMD_QAT_MAX_PCI_DEVICES=48 CONFIG_RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS=16 +CONFIG_RTE_PMD_QAT_COMP_IM_BUFFER_SIZE=65536 # # Compile PMD for virtio crypto devices diff --git a/config/rte_config.h b/config/rte_config.h index 816e6f8..333fb0b 100644 --- a/config/rte_config.h +++ b/config/rte_config.h @@ -95,6 +95,7 @@ /* Max. number of QuickAssist devices which can be attached */ #define RTE_PMD_QAT_MAX_PCI_DEVICES 48 #define RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS 16 +#define RTE_PMD_QAT_COMP_IM_BUFFER_SIZE 65536 /* virtio crypto defines */ #define RTE_MAX_VIRTIO_CRYPTO 32 diff --git a/doc/guides/compressdevs/features/qat.ini b/doc/guides/compressdevs/features/qat.ini index 5cd4524..6b1e7f9 100644 --- a/doc/guides/compressdevs/features/qat.ini +++ b/doc/guides/compressdevs/features/qat.ini @@ -13,3 +13,4 @@ Adler32 = Y Crc32 = Y Adler32&Crc32 = Y Fixed = Y +Dynamic = Y diff --git a/doc/guides/compressdevs/qat_comp.rst b/doc/guides/compressdevs/qat_comp.rst index 7bffbe6..aee3b99 100644 --- a/doc/guides/compressdevs/qat_comp.rst +++ b/doc/guides/compressdevs/qat_comp.rst @@ -18,11 +18,7 @@ QAT compression PMD has support for: Compression/Decompression algorithm: - * DEFLATE - -Huffman code type: - - * FIXED + * DEFLATE - using Fixed and Dynamic Huffman encoding Window size support: @@ -36,7 +32,6 @@ Limitations ----------- * Compressdev level 0, no compression, is not supported. -* Dynamic Huffman encoding is not yet supported. * Queue pairs are not thread-safe (that is, within a single queue pair, RX and TX from different lcores is not supported). * No BSD support as BSD QAT kernel driver not available. diff --git a/doc/guides/rel_notes/release_18_11.rst b/doc/guides/rel_notes/release_18_11.rst index 04f3745..2476cea 100644 --- a/doc/guides/rel_notes/release_18_11.rst +++ b/doc/guides/rel_notes/release_18_11.rst @@ -166,6 +166,11 @@ New Features Added the new caam job ring driver for NXP platforms. See the "NXP CAAM JOB RING (caam_jr)" document for more details on this new driver. +* **Added support for Dynamic Huffman Encoding to Intel QAT comp PMD.** + + The Intel QuickAssist (QAT) compression PMD has been updated with support + for Dynamic Huffman Encoding for the Deflate algorithm. + * **Added Event Ethernet Tx Adapter.** Added event ethernet Tx adapter library that provides configuration and diff --git a/drivers/common/qat/qat_device.c b/drivers/common/qat/qat_device.c index 8d4df92..5df831f 100644 --- a/drivers/common/qat/qat_device.c +++ b/drivers/common/qat/qat_device.c @@ -7,6 +7,7 @@ #include "qat_device.h" #include "adf_transport_access_macros.h" #include "qat_sym_pmd.h" +#include "qat_comp_pmd.h" /* Hardware device information per generation */ __extension__ @@ -14,15 +15,18 @@ struct qat_gen_hw_data qat_gen_config[] = { [QAT_GEN1] = { .dev_gen = QAT_GEN1, .qp_hw_data = qat_gen1_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN1 }, [QAT_GEN2] = { .dev_gen = QAT_GEN2, .qp_hw_data = qat_gen1_qps, /* gen2 has same ring layout as gen1 */ + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN2 }, [QAT_GEN3] = { .dev_gen = QAT_GEN3, .qp_hw_data = qat_gen3_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN3 }, }; diff --git a/drivers/common/qat/qat_device.h b/drivers/common/qat/qat_device.h index 3a71cd4..eb81c78 100644 --- a/drivers/common/qat/qat_device.h +++ b/drivers/common/qat/qat_device.h @@ -16,6 +16,12 @@ #define QAT_DEV_NAME_MAX_LEN 64 +enum qat_comp_num_im_buffers { + QAT_NUM_INTERM_BUFS_GEN1 = 12, + QAT_NUM_INTERM_BUFS_GEN2 = 20, + QAT_NUM_INTERM_BUFS_GEN3 = 20 +}; + /* * This struct holds all the data about a QAT pci device * including data about all services it supports. @@ -72,6 +78,7 @@ struct qat_pci_device { struct qat_gen_hw_data { enum qat_device_gen dev_gen; const struct qat_qp_hw_data (*qp_hw_data)[ADF_MAX_QPS_ON_ANY_SERVICE]; + enum qat_comp_num_im_buffers comp_num_im_bufs_required; }; extern struct qat_gen_hw_data qat_gen_config[]; diff --git a/drivers/compress/qat/qat_comp.c b/drivers/compress/qat/qat_comp.c index b9336f3..d70c594 100644 --- a/drivers/compress/qat/qat_comp.c +++ b/drivers/compress/qat/qat_comp.c @@ -192,7 +192,7 @@ static void qat_comp_create_req_hdr(struct icp_qat_fw_comn_req_hdr *header, } static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, - const struct rte_memzone *interm_buff_mz __rte_unused, + const struct rte_memzone *interm_buff_mz, const struct rte_comp_xform *xform) { struct icp_qat_fw_comp_req *comp_req; @@ -280,10 +280,20 @@ static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, ICP_QAT_FW_SLICE_COMP); } else if (qat_xform->qat_comp_request_type == - QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { - QAT_LOG(ERR, "Dynamic huffman encoding not supported"); - return -EINVAL; + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_COMP); + + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_DRAM_WR); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + + comp_req->u1.xlt_pars.inter_buff_ptr = + interm_buff_mz->phys_addr; } #if RTE_LOG_DP_LEVEL >= RTE_LOG_DEBUG @@ -334,12 +344,6 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, (struct qat_comp_xform *)*private_xform; if (xform->type == RTE_COMP_COMPRESS) { - if (xform->compress.deflate.huffman == - RTE_COMP_HUFFMAN_DYNAMIC) { - QAT_LOG(ERR, - "QAT device doesn't support dynamic compression"); - return -ENOTSUP; - } if (xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_FIXED || ((xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_DEFAULT) @@ -347,6 +351,21 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, qat_xform->qat_comp_request_type = QAT_COMP_REQUEST_FIXED_COMP_STATELESS; + else if ((xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DYNAMIC || + xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DEFAULT) && + qat->interm_buff_mz != NULL) + + qat_xform->qat_comp_request_type = + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS; + + else { + QAT_LOG(ERR, + "IM buffers needed for dynamic deflate. Set size in config file"); + return -EINVAL; + } + qat_xform->checksum_type = xform->compress.chksum; } else { diff --git a/drivers/compress/qat/qat_comp.h b/drivers/compress/qat/qat_comp.h index 8d315ef..99a4462 100644 --- a/drivers/compress/qat/qat_comp.h +++ b/drivers/compress/qat/qat_comp.h @@ -15,6 +15,10 @@ #include "icp_qat_fw_comp.h" #include "icp_qat_fw_la.h" +#define QAT_64_BYTE_ALIGN_MASK (~0x3f) +#define QAT_64_BYTE_ALIGN (64) +#define QAT_NUM_BUFS_IN_IM_SGL 1 + #define ERR_CODE_QAT_COMP_WRONG_FW -99 enum qat_comp_request_type { @@ -24,6 +28,15 @@ enum qat_comp_request_type { REQ_COMP_END }; +struct array_of_ptrs { + phys_addr_t pointer[0]; +}; + +struct qat_inter_sgl { + qat_sgl_hdr; + struct qat_flat_buf buffers[QAT_NUM_BUFS_IN_IM_SGL]; +} __rte_packed __rte_cache_aligned; + struct qat_comp_sgl { qat_sgl_hdr; struct qat_flat_buf buffers[RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS]; diff --git a/drivers/compress/qat/qat_comp_pmd.c b/drivers/compress/qat/qat_comp_pmd.c index 63af23a..2c6aea5 100644 --- a/drivers/compress/qat/qat_comp_pmd.c +++ b/drivers/compress/qat/qat_comp_pmd.c @@ -14,6 +14,7 @@ static const struct rte_compressdev_capabilities qat_comp_gen_capabilities[] = { RTE_COMP_FF_CRC32_ADLER32_CHECKSUM | RTE_COMP_FF_SHAREABLE_PRIV_XFORM | RTE_COMP_FF_HUFFMAN_FIXED | + RTE_COMP_FF_HUFFMAN_DYNAMIC | RTE_COMP_FF_OOP_SGL_IN_SGL_OUT | RTE_COMP_FF_OOP_SGL_IN_LB_OUT | RTE_COMP_FF_OOP_LB_IN_SGL_OUT, @@ -112,7 +113,7 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, /* store a link to the qp in the qat_pci_device */ qat_private->qat_dev->qps_in_use[QAT_SERVICE_COMPRESSION][qp_id] - = *qp_addr; + = *qp_addr; qp = (struct qat_qp *)*qp_addr; @@ -135,6 +136,103 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, return ret; } + +#define QAT_IM_BUFFER_DEBUG 0 +static const struct rte_memzone * +qat_comp_setup_inter_buffers(struct qat_comp_dev_private *comp_dev, + uint32_t buff_size) +{ + char inter_buff_mz_name[RTE_MEMZONE_NAMESIZE]; + const struct rte_memzone *memzone; + uint8_t *mz_start = NULL; + phys_addr_t mz_start_phys = 0; + struct array_of_ptrs *array_of_pointers; + int size_of_ptr_array; + uint32_t full_size; + uint32_t offset_of_sgls, offset_of_flat_buffs = 0; + int i; + int num_im_sgls = qat_gen_config[ + comp_dev->qat_dev->qat_dev_gen].comp_num_im_bufs_required; + + QAT_LOG(DEBUG, "QAT COMP device %s needs %d sgls", + comp_dev->qat_dev->name, num_im_sgls); + snprintf(inter_buff_mz_name, RTE_MEMZONE_NAMESIZE, + "%s_inter_buff", comp_dev->qat_dev->name); + memzone = rte_memzone_lookup(inter_buff_mz_name); + if (memzone != NULL) { + QAT_LOG(DEBUG, "QAT COMP im buffer memzone created already"); + return memzone; + } + + /* Create a memzone to hold intermediate buffers and associated + * meta-data needed by the firmware. The memzone contains: + * - a list of num_im_sgls physical pointers to sgls + * - the num_im_sgl sgl structures, each pointing to 2 flat buffers + * - the flat buffers: num_im_sgl * 2 + * where num_im_sgls depends on the hardware generation of the device + */ + + size_of_ptr_array = num_im_sgls * sizeof(phys_addr_t); + offset_of_sgls = (size_of_ptr_array + (~QAT_64_BYTE_ALIGN_MASK)) + & QAT_64_BYTE_ALIGN_MASK; + offset_of_flat_buffs = + offset_of_sgls + num_im_sgls * sizeof(struct qat_inter_sgl); + full_size = offset_of_flat_buffs + + num_im_sgls * buff_size * QAT_NUM_BUFS_IN_IM_SGL; + + memzone = rte_memzone_reserve_aligned(inter_buff_mz_name, full_size, + comp_dev->compressdev->data->socket_id, + RTE_MEMZONE_2MB, QAT_64_BYTE_ALIGN); + if (memzone == NULL) { + QAT_LOG(ERR, "Can't allocate intermediate buffers" + " for device %s", comp_dev->qat_dev->name); + return NULL; + } + + mz_start = (uint8_t *)memzone->addr; + mz_start_phys = memzone->phys_addr; + QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = 0x%"PRIx64 + ", size required %d, size created %ld", + inter_buff_mz_name, mz_start, mz_start_phys, + full_size, memzone->len); + + array_of_pointers = (struct array_of_ptrs *)mz_start; + for (i = 0; i < num_im_sgls; i++) { + uint32_t curr_sgl_offset = + offset_of_sgls + i * sizeof(struct qat_inter_sgl); + struct qat_inter_sgl *sgl = + (struct qat_inter_sgl *)(mz_start + curr_sgl_offset); + array_of_pointers->pointer[i] = mz_start_phys + curr_sgl_offset; + + sgl->num_bufs = QAT_NUM_BUFS_IN_IM_SGL; + sgl->num_mapped_bufs = 0; + sgl->resrvd = 0; + sgl->buffers[0].addr = mz_start_phys + offset_of_flat_buffs + + ((i * QAT_NUM_BUFS_IN_IM_SGL) * buff_size); + sgl->buffers[0].len = buff_size; + sgl->buffers[0].resrvd = 0; + sgl->buffers[1].addr = mz_start_phys + offset_of_flat_buffs + + (((i * QAT_NUM_BUFS_IN_IM_SGL) + 1) * buff_size); + sgl->buffers[1].len = buff_size; + sgl->buffers[1].resrvd = 0; + +#if QAT_IM_BUFFER_DEBUG + QAT_LOG(DEBUG, " : phys addr of sgl[%i] in array_of_pointers" + "= 0x%"PRIx64, i, array_of_pointers->pointer[i]); + QAT_LOG(DEBUG, " : virt address of sgl[%i] = %p", i, sgl); + QAT_LOG(DEBUG, " : sgl->buffers[0].addr = 0x%"PRIx64", len=%d", + sgl->buffers[0].addr, sgl->buffers[0].len); + QAT_LOG(DEBUG, " : sgl->buffers[1].addr = 0x%"PRIx64", len=%d", + sgl->buffers[1].addr, sgl->buffers[1].len); +#endif + } +#if QAT_IM_BUFFER_DEBUG + QAT_DP_HEXDUMP_LOG(DEBUG, "IM buffer memzone start:", + mz_start, offset_of_flat_buffs + 32); +#endif + return memzone; +} + static struct rte_mempool * qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, uint32_t num_elements) @@ -176,6 +274,12 @@ qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, static void _qat_comp_dev_config_clear(struct qat_comp_dev_private *comp_dev) { + /* Free intermediate buffers */ + if (comp_dev->interm_buff_mz) { + rte_memzone_free(comp_dev->interm_buff_mz); + comp_dev->interm_buff_mz = NULL; + } + /* Free private_xform pool */ if (comp_dev->xformpool) { /* Free internal mempool for private xforms */ @@ -197,6 +301,21 @@ qat_comp_dev_config(struct rte_compressdev *dev, return -EINVAL; } + if (RTE_PMD_QAT_COMP_IM_BUFFER_SIZE == 0) { + QAT_LOG(WARNING, + "RTE_PMD_QAT_COMP_IM_BUFFER_SIZE = 0 in config file, so" + " QAT device can't be used for Dynamic Deflate. " + "Did you really intend to do this?"); + } else { + comp_dev->interm_buff_mz = + qat_comp_setup_inter_buffers(comp_dev, + RTE_PMD_QAT_COMP_IM_BUFFER_SIZE); + if (comp_dev->interm_buff_mz == NULL) { + ret = -ENOMEM; + goto error_out; + } + } + comp_dev->xformpool = qat_comp_create_xform_pool(comp_dev, config->max_nb_priv_xforms); if (comp_dev->xformpool == NULL) { @@ -365,6 +484,10 @@ qat_comp_dev_create(struct qat_pci_device *qat_pci_dev) QAT_LOG(ERR, "Compression PMD not supported on QAT dh895xcc"); return 0; } + if (qat_pci_dev->qat_dev_gen == QAT_GEN3) { + QAT_LOG(ERR, "Compression PMD not supported on QAT c4xxx"); + return 0; + } struct rte_compressdev_pmd_init_params init_params = { .name = "", -- 2.7.4 ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [dpdk-dev] [PATCH v6] compress/qat: enable dynamic huffman encoding 2018-10-25 1:30 ` [dpdk-dev] [PATCH v6] " Fiona Trahe @ 2018-10-25 7:34 ` Akhil Goyal 2018-10-26 18:18 ` [dpdk-dev] [PATCH v7] " Fiona Trahe 1 sibling, 0 replies; 17+ messages in thread From: Akhil Goyal @ 2018-10-25 7:34 UTC (permalink / raw) To: Fiona Trahe, dev; +Cc: thomas, tomaszx.jozwiak, arkadiuszx.kusztal Hi Fiona, On 10/25/2018 7:00 AM, Fiona Trahe wrote: > Enable dynamic huffman encoding in the QAT comp PMD. > > Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> > Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> > Acked-by: Arek Kusztal <arkadiuszx.kusztal@intel.com> > --- The patch is still failing in i686 and armv7 **************** Testing i686 *********************** In file included from /home/b35197/dpdk/drivers/common/qat/qat_device.h:10:0, from /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.h:13, from /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.c:6: /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.c: In function âqat_comp_setup_inter_buffersâ: /home/b35197/dpdk/drivers/common/qat/qat_logs.h:13:4: error: format â%ldâ expects argument of type âlong intâ, but argument 9 has type âsize_t {aka const unsigned int}â [-Werror=format=] "%s(): " fmt "\n", __func__, ## args) ^ /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.c:194:2: note: in expansion of macro âQAT_LOGâ QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = 0x%"PRIx64 ^~~~~~~ /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.c:195:40: note: format string is defined here ", size required %d, size created %ld", ~~^ %d cc1: all warnings being treated as errors make[5]: *** [qat_comp_pmd.o] Error 1 make[5]: *** Waiting for unfinished jobs.... make[4]: *** [common/qat] Error 2 make[4]: *** Waiting for unfinished jobs.... make[3]: *** [drivers] Error 2 make[2]: *** [all] Error 2 make[1]: *** [pre_install] Error 2 make: *** [install] Error 2 Error in BUILD **************** Testing armv7 *********************** /home/b35197/dpdk/drivers/compress/qat/qat_comp.c: In function 'qat_comp_build_request': /home/b35197/dpdk/drivers/compress/qat/qat_comp.c:33:6: warning: cast increases required alignment of target type [-Wcast-align] (struct icp_qat_fw_comp_req *)out_msg; ^ /home/b35197/dpdk/drivers/compress/qat/qat_comp.c: In function 'qat_comp_process_response': /home/b35197/dpdk/drivers/compress/qat/qat_comp.c:112:4: warning: cast increases required alignment of target type [-Wcast-align] (struct icp_qat_fw_comp_resp *)resp; ^ /home/b35197/dpdk/drivers/compress/qat/qat_comp.c:146:6: warning: cast increases required alignment of target type [-Wcast-align] *((uint16_t *)(&resp_msg->comn_resp.comn_error)); ^ In file included from /home/b35197/dpdk/drivers/common/qat/qat_device.h:10:0, from /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.h:13, from /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.c:6: /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.c: In function 'qat_comp_setup_inter_buffers': /home/b35197/dpdk/drivers/common/qat/qat_logs.h:13:4: error: format '%ld' expects argument of type 'long int', but argument 9 has type 'size_t {aka const unsigned int}' [-Werror=format=] "%s(): " fmt "\n", __func__, ## args) ^ /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.c:194:2: note: in expansion of macro 'QAT_LOG' QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = 0x%"PRIx64 ^ /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.c:199:22: warning: cast increases required alignment of target type [-Wcast-align] array_of_pointers = (struct array_of_ptrs *)mz_start; ^ /home/b35197/dpdk/drivers/compress/qat/qat_comp_pmd.c:204:7: warning: cast increases required alignment of target type [-Wcast-align] (struct qat_inter_sgl *)(mz_start + curr_sgl_offset); ^ cc1: all warnings being treated as errors make[5]: *** [qat_comp_pmd.o] Error 1 make[5]: *** Waiting for unfinished jobs.... make[4]: *** [common/qat] Error 2 make[4]: *** Waiting for unfinished jobs.... make[3]: *** [drivers] Error 2 make[2]: *** [all] Error 2 make[1]: *** [pre_install] Error 2 make: *** [install] Error 2 Error in BUILD ^ permalink raw reply [flat|nested] 17+ messages in thread
* [dpdk-dev] [PATCH v7] compress/qat: enable dynamic huffman encoding 2018-10-25 1:30 ` [dpdk-dev] [PATCH v6] " Fiona Trahe 2018-10-25 7:34 ` Akhil Goyal @ 2018-10-26 18:18 ` Fiona Trahe 2018-10-27 0:06 ` Thomas Monjalon 1 sibling, 1 reply; 17+ messages in thread From: Fiona Trahe @ 2018-10-26 18:18 UTC (permalink / raw) To: dev Cc: thomas, akhil.goyal, tomaszx.jozwiak, arkadiuszx.kusztal, ferruh.yigit, Fiona Trahe Enable dynamic huffman encoding in the QAT comp PMD. Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> Acked-by: Arek Kusztal <arkadiuszx.kusztal@intel.com> --- v7 changes: - replace %ld with %zu for printing size_t to fix compiler issue - replace phys addr with rte_iova_t v6 changes: - replace %lx with PRIx64 to fix compiler issue - reorder entry in release notes so above eventdev items v5 changes: - update release note v4 changes: - use #define for IM buffer logs, disable all by default and use param for clearer hexdump length. v3 changes: - reverted to 1 buffer per intermediate buffer sgl v2 changes: - allocate 2 buffers per intermediate buffer sgl - Compile out trace for debugging intermediate buffers config/common_base | 1 + config/rte_config.h | 1 + doc/guides/compressdevs/features/qat.ini | 1 + doc/guides/compressdevs/qat_comp.rst | 7 +- doc/guides/rel_notes/release_18_11.rst | 5 ++ drivers/common/qat/qat_device.c | 4 + drivers/common/qat/qat_device.h | 7 ++ drivers/compress/qat/qat_comp.c | 39 +++++++--- drivers/compress/qat/qat_comp.h | 13 ++++ drivers/compress/qat/qat_comp_pmd.c | 125 ++++++++++++++++++++++++++++++- 10 files changed, 186 insertions(+), 17 deletions(-) diff --git a/config/common_base b/config/common_base index be7365e..a40c61f 100644 --- a/config/common_base +++ b/config/common_base @@ -532,6 +532,7 @@ CONFIG_RTE_LIBRTE_PMD_QAT_SYM=n # CONFIG_RTE_PMD_QAT_MAX_PCI_DEVICES=48 CONFIG_RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS=16 +CONFIG_RTE_PMD_QAT_COMP_IM_BUFFER_SIZE=65536 # # Compile PMD for virtio crypto devices diff --git a/config/rte_config.h b/config/rte_config.h index 816e6f8..333fb0b 100644 --- a/config/rte_config.h +++ b/config/rte_config.h @@ -95,6 +95,7 @@ /* Max. number of QuickAssist devices which can be attached */ #define RTE_PMD_QAT_MAX_PCI_DEVICES 48 #define RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS 16 +#define RTE_PMD_QAT_COMP_IM_BUFFER_SIZE 65536 /* virtio crypto defines */ #define RTE_MAX_VIRTIO_CRYPTO 32 diff --git a/doc/guides/compressdevs/features/qat.ini b/doc/guides/compressdevs/features/qat.ini index 5cd4524..6b1e7f9 100644 --- a/doc/guides/compressdevs/features/qat.ini +++ b/doc/guides/compressdevs/features/qat.ini @@ -13,3 +13,4 @@ Adler32 = Y Crc32 = Y Adler32&Crc32 = Y Fixed = Y +Dynamic = Y diff --git a/doc/guides/compressdevs/qat_comp.rst b/doc/guides/compressdevs/qat_comp.rst index 7bffbe6..aee3b99 100644 --- a/doc/guides/compressdevs/qat_comp.rst +++ b/doc/guides/compressdevs/qat_comp.rst @@ -18,11 +18,7 @@ QAT compression PMD has support for: Compression/Decompression algorithm: - * DEFLATE - -Huffman code type: - - * FIXED + * DEFLATE - using Fixed and Dynamic Huffman encoding Window size support: @@ -36,7 +32,6 @@ Limitations ----------- * Compressdev level 0, no compression, is not supported. -* Dynamic Huffman encoding is not yet supported. * Queue pairs are not thread-safe (that is, within a single queue pair, RX and TX from different lcores is not supported). * No BSD support as BSD QAT kernel driver not available. diff --git a/doc/guides/rel_notes/release_18_11.rst b/doc/guides/rel_notes/release_18_11.rst index 04f3745..2476cea 100644 --- a/doc/guides/rel_notes/release_18_11.rst +++ b/doc/guides/rel_notes/release_18_11.rst @@ -166,6 +166,11 @@ New Features Added the new caam job ring driver for NXP platforms. See the "NXP CAAM JOB RING (caam_jr)" document for more details on this new driver. +* **Added support for Dynamic Huffman Encoding to Intel QAT comp PMD.** + + The Intel QuickAssist (QAT) compression PMD has been updated with support + for Dynamic Huffman Encoding for the Deflate algorithm. + * **Added Event Ethernet Tx Adapter.** Added event ethernet Tx adapter library that provides configuration and diff --git a/drivers/common/qat/qat_device.c b/drivers/common/qat/qat_device.c index 4ff8b88..2a1cf3e 100644 --- a/drivers/common/qat/qat_device.c +++ b/drivers/common/qat/qat_device.c @@ -7,6 +7,7 @@ #include "qat_device.h" #include "adf_transport_access_macros.h" #include "qat_sym_pmd.h" +#include "qat_comp_pmd.h" /* Hardware device information per generation */ __extension__ @@ -14,15 +15,18 @@ struct qat_gen_hw_data qat_gen_config[] = { [QAT_GEN1] = { .dev_gen = QAT_GEN1, .qp_hw_data = qat_gen1_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN1 }, [QAT_GEN2] = { .dev_gen = QAT_GEN2, .qp_hw_data = qat_gen1_qps, /* gen2 has same ring layout as gen1 */ + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN2 }, [QAT_GEN3] = { .dev_gen = QAT_GEN3, .qp_hw_data = qat_gen3_qps, + .comp_num_im_bufs_required = QAT_NUM_INTERM_BUFS_GEN3 }, }; diff --git a/drivers/common/qat/qat_device.h b/drivers/common/qat/qat_device.h index 3a71cd4..eb81c78 100644 --- a/drivers/common/qat/qat_device.h +++ b/drivers/common/qat/qat_device.h @@ -16,6 +16,12 @@ #define QAT_DEV_NAME_MAX_LEN 64 +enum qat_comp_num_im_buffers { + QAT_NUM_INTERM_BUFS_GEN1 = 12, + QAT_NUM_INTERM_BUFS_GEN2 = 20, + QAT_NUM_INTERM_BUFS_GEN3 = 20 +}; + /* * This struct holds all the data about a QAT pci device * including data about all services it supports. @@ -72,6 +78,7 @@ struct qat_pci_device { struct qat_gen_hw_data { enum qat_device_gen dev_gen; const struct qat_qp_hw_data (*qp_hw_data)[ADF_MAX_QPS_ON_ANY_SERVICE]; + enum qat_comp_num_im_buffers comp_num_im_bufs_required; }; extern struct qat_gen_hw_data qat_gen_config[]; diff --git a/drivers/compress/qat/qat_comp.c b/drivers/compress/qat/qat_comp.c index b9336f3..d70c594 100644 --- a/drivers/compress/qat/qat_comp.c +++ b/drivers/compress/qat/qat_comp.c @@ -192,7 +192,7 @@ static void qat_comp_create_req_hdr(struct icp_qat_fw_comn_req_hdr *header, } static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, - const struct rte_memzone *interm_buff_mz __rte_unused, + const struct rte_memzone *interm_buff_mz, const struct rte_comp_xform *xform) { struct icp_qat_fw_comp_req *comp_req; @@ -280,10 +280,20 @@ static int qat_comp_create_templates(struct qat_comp_xform *qat_xform, ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, ICP_QAT_FW_SLICE_COMP); } else if (qat_xform->qat_comp_request_type == - QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS) { - QAT_LOG(ERR, "Dynamic huffman encoding not supported"); - return -EINVAL; + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->comp_cd_ctrl, + ICP_QAT_FW_SLICE_COMP); + + ICP_QAT_FW_COMN_NEXT_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_DRAM_WR); + ICP_QAT_FW_COMN_CURR_ID_SET(&comp_req->u2.xlt_cd_ctrl, + ICP_QAT_FW_SLICE_XLAT); + + comp_req->u1.xlt_pars.inter_buff_ptr = + interm_buff_mz->phys_addr; } #if RTE_LOG_DP_LEVEL >= RTE_LOG_DEBUG @@ -334,12 +344,6 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, (struct qat_comp_xform *)*private_xform; if (xform->type == RTE_COMP_COMPRESS) { - if (xform->compress.deflate.huffman == - RTE_COMP_HUFFMAN_DYNAMIC) { - QAT_LOG(ERR, - "QAT device doesn't support dynamic compression"); - return -ENOTSUP; - } if (xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_FIXED || ((xform->compress.deflate.huffman == RTE_COMP_HUFFMAN_DEFAULT) @@ -347,6 +351,21 @@ qat_comp_private_xform_create(struct rte_compressdev *dev, qat_xform->qat_comp_request_type = QAT_COMP_REQUEST_FIXED_COMP_STATELESS; + else if ((xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DYNAMIC || + xform->compress.deflate.huffman == + RTE_COMP_HUFFMAN_DEFAULT) && + qat->interm_buff_mz != NULL) + + qat_xform->qat_comp_request_type = + QAT_COMP_REQUEST_DYNAMIC_COMP_STATELESS; + + else { + QAT_LOG(ERR, + "IM buffers needed for dynamic deflate. Set size in config file"); + return -EINVAL; + } + qat_xform->checksum_type = xform->compress.chksum; } else { diff --git a/drivers/compress/qat/qat_comp.h b/drivers/compress/qat/qat_comp.h index 8d315ef..99a4462 100644 --- a/drivers/compress/qat/qat_comp.h +++ b/drivers/compress/qat/qat_comp.h @@ -15,6 +15,10 @@ #include "icp_qat_fw_comp.h" #include "icp_qat_fw_la.h" +#define QAT_64_BYTE_ALIGN_MASK (~0x3f) +#define QAT_64_BYTE_ALIGN (64) +#define QAT_NUM_BUFS_IN_IM_SGL 1 + #define ERR_CODE_QAT_COMP_WRONG_FW -99 enum qat_comp_request_type { @@ -24,6 +28,15 @@ enum qat_comp_request_type { REQ_COMP_END }; +struct array_of_ptrs { + phys_addr_t pointer[0]; +}; + +struct qat_inter_sgl { + qat_sgl_hdr; + struct qat_flat_buf buffers[QAT_NUM_BUFS_IN_IM_SGL]; +} __rte_packed __rte_cache_aligned; + struct qat_comp_sgl { qat_sgl_hdr; struct qat_flat_buf buffers[RTE_PMD_QAT_COMP_SGL_MAX_SEGMENTS]; diff --git a/drivers/compress/qat/qat_comp_pmd.c b/drivers/compress/qat/qat_comp_pmd.c index 63af23a..01dd736 100644 --- a/drivers/compress/qat/qat_comp_pmd.c +++ b/drivers/compress/qat/qat_comp_pmd.c @@ -14,6 +14,7 @@ static const struct rte_compressdev_capabilities qat_comp_gen_capabilities[] = { RTE_COMP_FF_CRC32_ADLER32_CHECKSUM | RTE_COMP_FF_SHAREABLE_PRIV_XFORM | RTE_COMP_FF_HUFFMAN_FIXED | + RTE_COMP_FF_HUFFMAN_DYNAMIC | RTE_COMP_FF_OOP_SGL_IN_SGL_OUT | RTE_COMP_FF_OOP_SGL_IN_LB_OUT | RTE_COMP_FF_OOP_LB_IN_SGL_OUT, @@ -112,7 +113,7 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, /* store a link to the qp in the qat_pci_device */ qat_private->qat_dev->qps_in_use[QAT_SERVICE_COMPRESSION][qp_id] - = *qp_addr; + = *qp_addr; qp = (struct qat_qp *)*qp_addr; @@ -135,6 +136,103 @@ qat_comp_qp_setup(struct rte_compressdev *dev, uint16_t qp_id, return ret; } + +#define QAT_IM_BUFFER_DEBUG 0 +static const struct rte_memzone * +qat_comp_setup_inter_buffers(struct qat_comp_dev_private *comp_dev, + uint32_t buff_size) +{ + char inter_buff_mz_name[RTE_MEMZONE_NAMESIZE]; + const struct rte_memzone *memzone; + uint8_t *mz_start = NULL; + rte_iova_t mz_start_phys = 0; + struct array_of_ptrs *array_of_pointers; + int size_of_ptr_array; + uint32_t full_size; + uint32_t offset_of_sgls, offset_of_flat_buffs = 0; + int i; + int num_im_sgls = qat_gen_config[ + comp_dev->qat_dev->qat_dev_gen].comp_num_im_bufs_required; + + QAT_LOG(DEBUG, "QAT COMP device %s needs %d sgls", + comp_dev->qat_dev->name, num_im_sgls); + snprintf(inter_buff_mz_name, RTE_MEMZONE_NAMESIZE, + "%s_inter_buff", comp_dev->qat_dev->name); + memzone = rte_memzone_lookup(inter_buff_mz_name); + if (memzone != NULL) { + QAT_LOG(DEBUG, "QAT COMP im buffer memzone created already"); + return memzone; + } + + /* Create a memzone to hold intermediate buffers and associated + * meta-data needed by the firmware. The memzone contains: + * - a list of num_im_sgls physical pointers to sgls + * - the num_im_sgl sgl structures, each pointing to 2 flat buffers + * - the flat buffers: num_im_sgl * 2 + * where num_im_sgls depends on the hardware generation of the device + */ + + size_of_ptr_array = num_im_sgls * sizeof(phys_addr_t); + offset_of_sgls = (size_of_ptr_array + (~QAT_64_BYTE_ALIGN_MASK)) + & QAT_64_BYTE_ALIGN_MASK; + offset_of_flat_buffs = + offset_of_sgls + num_im_sgls * sizeof(struct qat_inter_sgl); + full_size = offset_of_flat_buffs + + num_im_sgls * buff_size * QAT_NUM_BUFS_IN_IM_SGL; + + memzone = rte_memzone_reserve_aligned(inter_buff_mz_name, full_size, + comp_dev->compressdev->data->socket_id, + RTE_MEMZONE_2MB, QAT_64_BYTE_ALIGN); + if (memzone == NULL) { + QAT_LOG(ERR, "Can't allocate intermediate buffers" + " for device %s", comp_dev->qat_dev->name); + return NULL; + } + + mz_start = (uint8_t *)memzone->addr; + mz_start_phys = memzone->phys_addr; + QAT_LOG(DEBUG, "Memzone %s: addr = %p, phys = 0x%"PRIx64 + ", size required %d, size created %zu", + inter_buff_mz_name, mz_start, mz_start_phys, + full_size, memzone->len); + + array_of_pointers = (struct array_of_ptrs *)mz_start; + for (i = 0; i < num_im_sgls; i++) { + uint32_t curr_sgl_offset = + offset_of_sgls + i * sizeof(struct qat_inter_sgl); + struct qat_inter_sgl *sgl = + (struct qat_inter_sgl *)(mz_start + curr_sgl_offset); + array_of_pointers->pointer[i] = mz_start_phys + curr_sgl_offset; + + sgl->num_bufs = QAT_NUM_BUFS_IN_IM_SGL; + sgl->num_mapped_bufs = 0; + sgl->resrvd = 0; + sgl->buffers[0].addr = mz_start_phys + offset_of_flat_buffs + + ((i * QAT_NUM_BUFS_IN_IM_SGL) * buff_size); + sgl->buffers[0].len = buff_size; + sgl->buffers[0].resrvd = 0; + sgl->buffers[1].addr = mz_start_phys + offset_of_flat_buffs + + (((i * QAT_NUM_BUFS_IN_IM_SGL) + 1) * buff_size); + sgl->buffers[1].len = buff_size; + sgl->buffers[1].resrvd = 0; + +#if QAT_IM_BUFFER_DEBUG + QAT_LOG(DEBUG, " : phys addr of sgl[%i] in array_of_pointers" + "= 0x%"PRIx64, i, array_of_pointers->pointer[i]); + QAT_LOG(DEBUG, " : virt address of sgl[%i] = %p", i, sgl); + QAT_LOG(DEBUG, " : sgl->buffers[0].addr = 0x%"PRIx64", len=%d", + sgl->buffers[0].addr, sgl->buffers[0].len); + QAT_LOG(DEBUG, " : sgl->buffers[1].addr = 0x%"PRIx64", len=%d", + sgl->buffers[1].addr, sgl->buffers[1].len); +#endif + } +#if QAT_IM_BUFFER_DEBUG + QAT_DP_HEXDUMP_LOG(DEBUG, "IM buffer memzone start:", + mz_start, offset_of_flat_buffs + 32); +#endif + return memzone; +} + static struct rte_mempool * qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, uint32_t num_elements) @@ -176,6 +274,12 @@ qat_comp_create_xform_pool(struct qat_comp_dev_private *comp_dev, static void _qat_comp_dev_config_clear(struct qat_comp_dev_private *comp_dev) { + /* Free intermediate buffers */ + if (comp_dev->interm_buff_mz) { + rte_memzone_free(comp_dev->interm_buff_mz); + comp_dev->interm_buff_mz = NULL; + } + /* Free private_xform pool */ if (comp_dev->xformpool) { /* Free internal mempool for private xforms */ @@ -197,6 +301,21 @@ qat_comp_dev_config(struct rte_compressdev *dev, return -EINVAL; } + if (RTE_PMD_QAT_COMP_IM_BUFFER_SIZE == 0) { + QAT_LOG(WARNING, + "RTE_PMD_QAT_COMP_IM_BUFFER_SIZE = 0 in config file, so" + " QAT device can't be used for Dynamic Deflate. " + "Did you really intend to do this?"); + } else { + comp_dev->interm_buff_mz = + qat_comp_setup_inter_buffers(comp_dev, + RTE_PMD_QAT_COMP_IM_BUFFER_SIZE); + if (comp_dev->interm_buff_mz == NULL) { + ret = -ENOMEM; + goto error_out; + } + } + comp_dev->xformpool = qat_comp_create_xform_pool(comp_dev, config->max_nb_priv_xforms); if (comp_dev->xformpool == NULL) { @@ -365,6 +484,10 @@ qat_comp_dev_create(struct qat_pci_device *qat_pci_dev) QAT_LOG(ERR, "Compression PMD not supported on QAT dh895xcc"); return 0; } + if (qat_pci_dev->qat_dev_gen == QAT_GEN3) { + QAT_LOG(ERR, "Compression PMD not supported on QAT c4xxx"); + return 0; + } struct rte_compressdev_pmd_init_params init_params = { .name = "", -- 2.7.4 ^ permalink raw reply [flat|nested] 17+ messages in thread
* Re: [dpdk-dev] [PATCH v7] compress/qat: enable dynamic huffman encoding 2018-10-26 18:18 ` [dpdk-dev] [PATCH v7] " Fiona Trahe @ 2018-10-27 0:06 ` Thomas Monjalon 0 siblings, 0 replies; 17+ messages in thread From: Thomas Monjalon @ 2018-10-27 0:06 UTC (permalink / raw) To: Fiona Trahe Cc: dev, akhil.goyal, tomaszx.jozwiak, arkadiuszx.kusztal, ferruh.yigit 26/10/2018 20:18, Fiona Trahe: > Enable dynamic huffman encoding in the QAT comp PMD. > > Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com> > Signed-off-by: Fiona Trahe <fiona.trahe@intel.com> > Acked-by: Arek Kusztal <arkadiuszx.kusztal@intel.com> Applied, thanks ^ permalink raw reply [flat|nested] 17+ messages in thread
end of thread, other threads:[~2018-10-27 0:06 UTC | newest] Thread overview: 17+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2018-07-20 17:46 [dpdk-dev] [PATCH] compress/qat: enable dynamic huffman encoding Fiona Trahe 2018-10-05 16:18 ` [dpdk-dev] [PATCH v2] " Fiona Trahe 2018-10-09 11:01 ` Akhil Goyal 2018-10-09 14:09 ` Trahe, Fiona 2018-10-12 11:39 ` Akhil Goyal 2018-10-11 17:03 ` [dpdk-dev] [PATCH v3] " Fiona Trahe 2018-10-15 23:16 ` [dpdk-dev] [PATCH v4] " Fiona Trahe 2018-10-17 10:29 ` Kusztal, ArkadiuszX 2018-10-17 20:48 ` [dpdk-dev] [PATCH v5] " Fiona Trahe 2018-10-24 14:15 ` Akhil Goyal 2018-10-24 21:30 ` Thomas Monjalon 2018-10-25 0:40 ` Trahe, Fiona 2018-10-25 6:45 ` Thomas Monjalon 2018-10-25 1:30 ` [dpdk-dev] [PATCH v6] " Fiona Trahe 2018-10-25 7:34 ` Akhil Goyal 2018-10-26 18:18 ` [dpdk-dev] [PATCH v7] " Fiona Trahe 2018-10-27 0:06 ` Thomas Monjalon
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for NNTP newsgroup(s).