Spdk/include/spdk/idxd_spec.h

638 lines
14 KiB
C
Raw Normal View History

/* SPDX-License-Identifier: BSD-3-Clause
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
* Copyright (c) Intel Corporation.
* All rights reserved.
*/
/**
* \file
* IDXD specification definitions
*/
#ifndef SPDK_IDXD_SPEC_H
#define SPDK_IDXD_SPEC_H
#include "spdk/stdinc.h"
#include "spdk/assert.h"
#ifdef __cplusplus
extern "C" {
#endif
#define IDXD_MMIO_BAR 0
#define IDXD_WQ_BAR 2
#define PORTAL_SIZE 0x1000
#define WQ_TOTAL_PORTAL_SIZE (PORTAL_SIZE * 4)
#define PORTAL_STRIDE 0x40
#define PORTAL_MASK (PORTAL_SIZE - 1)
#define WQCFG_SHIFT 5
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
#define IDXD_TABLE_OFFSET_MULT 0x100
#define IDXD_CLEAR_CRC_FLAGS 0xFFFFu
#define IDXD_FLAG_FENCE (1 << 0)
#define IDXD_FLAG_COMPLETION_ADDR_VALID (1 << 2)
#define IDXD_FLAG_REQUEST_COMPLETION (1 << 3)
#define IDXD_FLAG_CACHE_CONTROL (1 << 8)
#define IDXD_FLAG_DEST_READBACK (1 << 14)
#define IDXD_FLAG_DEST_STEERING_TAG (1 << 15)
#define IDXD_FLAG_CRC_READ_CRC_SEED (1 << 16)
#define IAA_FLAG_RD_SRC2_AECS (1 << 16)
#define IAA_COMP_FLUSH_OUTPUT (1 << 1)
#define IAA_COMP_APPEND_EOB (1 << 2)
#define IAA_COMP_FLAGS (IAA_COMP_FLUSH_OUTPUT | IAA_COMP_APPEND_EOB)
#define IAA_DECOMP_ENABLE (1 << 0)
#define IAA_DECOMP_FLAGS (IAA_DECOMP_ENABLE)
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
/*
* IDXD is a family of devices, DSA and IAA.
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
*/
enum dsa_completion_status {
DSA_COMP_NONE = 0,
DSA_COMP_SUCCESS = 1,
DSA_COMP_SUCCESS_PRED = 2,
DSA_COMP_PAGE_FAULT_NOBOF = 3,
DSA_COMP_PAGE_FAULT_IR = 4,
DSA_COMP_BATCH_FAIL = 5,
DSA_COMP_BATCH_PAGE_FAULT = 6,
DSA_COMP_DR_OFFSET_NOINC = 7,
DSA_COMP_DR_OFFSET_ERANGE = 8,
DSA_COMP_DIF_ERR = 9,
DSA_COMP_BAD_OPCODE = 16,
DSA_COMP_INVALID_FLAGS = 17,
DSA_COMP_NOZERO_RESERVE = 18,
DSA_COMP_XFER_ERANGE = 19,
DSA_COMP_DESC_CNT_ERANGE = 20,
DSA_COMP_DR_ERANGE = 21,
DSA_COMP_OVERLAP_BUFFERS = 22,
DSA_COMP_DCAST_ERR = 23,
DSA_COMP_DESCLIST_ALIGN = 24,
DSA_COMP_INT_HANDLE_INVAL = 25,
DSA_COMP_CRA_XLAT = 26,
DSA_COMP_CRA_ALIGN = 27,
DSA_COMP_ADDR_ALIGN = 28,
DSA_COMP_PRIV_BAD = 29,
DSA_COMP_TRAFFIC_CLASS_CONF = 30,
DSA_COMP_PFAULT_RDBA = 31,
DSA_COMP_HW_ERR1 = 32,
DSA_COMP_HW_ERR_DRB = 33,
DSA_COMP_TRANSLATION_FAIL = 34,
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
};
enum iaa_completion_status {
IAA_COMP_NONE = 0,
IAA_COMP_SUCCESS = 1,
IAA_COMP_PAGE_FAULT_IR = 4,
IAA_COMP_OUTBUF_OVERFLOW = 5,
IAA_COMP_BAD_OPCODE = 16,
IAA_COMP_INVALID_FLAGS = 17,
IAA_COMP_NOZERO_RESERVE = 18,
IAA_COMP_INVALID_SIZE = 19,
IAA_COMP_OVERLAP_BUFFERS = 22,
IAA_COMP_INT_HANDLE_INVAL = 25,
IAA_COMP_CRA_XLAT = 32,
IAA_COMP_CRA_ALIGN = 33,
IAA_COMP_ADDR_ALIGN = 34,
IAA_COMP_PRIV_BAD = 35,
IAA_COMP_TRAFFIC_CLASS_CONF = 36,
IAA_COMP_PFAULT_RDBA = 37,
IAA_COMP_HW_ERR1 = 38,
IAA_COMP_TRANSLATION_FAIL = 39,
IAA_COMP_PRS_TIMEOUT = 40,
IAA_COMP_WATCHDOG = 41,
IAA_COMP_INVALID_COMP_FLAG = 48,
IAA_COMP_INVALID_FILTER_FLAG = 49,
IAA_COMP_INVALID_NUM_ELEMS = 50,
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
enum idxd_wq_state {
WQ_DISABLED = 0,
WQ_ENABLED = 1,
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
};
enum idxd_wq_flag {
WQ_FLAG_DEDICATED = 0,
WQ_FLAG_BOF = 1,
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
};
enum idxd_wq_type {
WQT_NONE = 0,
WQT_KERNEL = 1,
WQT_USER = 2,
WQT_MDEV = 3,
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
};
enum idxd_dev_state {
IDXD_DEVICE_STATE_DISABLED = 0,
IDXD_DEVICE_STATE_ENABLED = 1,
IDXD_DEVICE_STATE_DRAIN = 2,
IDXD_DEVICE_STATE_HALT = 3,
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
};
enum idxd_device_reset_type {
IDXD_DEVICE_RESET_SOFTWARE = 0,
IDXD_DEVICE_RESET_FLR = 1,
IDXD_DEVICE_RESET_WARM = 2,
IDXD_DEVICE_RESET_COLD = 3,
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
};
enum idxd_cmds {
IDXD_ENABLE_DEV = 1,
IDXD_DISABLE_DEV = 2,
IDXD_DRAIN_ALL = 3,
IDXD_ABORT_ALL = 4,
IDXD_RESET_DEVICE = 5,
IDXD_ENABLE_WQ = 6,
IDXD_DISABLE_WQ = 7,
IDXD_DRAIN_WQ = 8,
IDXD_ABORT_WQ = 9,
IDXD_RESET_WQ = 10,
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
};
enum idxd_cmdsts_err {
IDXD_CMDSTS_SUCCESS = 0,
IDXD_CMDSTS_INVAL_CMD = 1,
IDXD_CMDSTS_INVAL_WQIDX = 2,
IDXD_CMDSTS_HW_ERR = 3,
IDXD_CMDSTS_ERR_DEV_ENABLED = 16,
IDXD_CMDSTS_ERR_CONFIG = 17,
IDXD_CMDSTS_ERR_BUSMASTER_EN = 18,
IDXD_CMDSTS_ERR_PASID_INVAL = 19,
IDXD_CMDSTS_ERR_WQ_SIZE_ERANGE = 20,
IDXD_CMDSTS_ERR_GRP_CONFIG = 21,
IDXD_CMDSTS_ERR_GRP_CONFIG2 = 22,
IDXD_CMDSTS_ERR_GRP_CONFIG3 = 23,
IDXD_CMDSTS_ERR_GRP_CONFIG4 = 24,
IDXD_CMDSTS_ERR_DEV_NOTEN = 32,
IDXD_CMDSTS_ERR_WQ_ENABLED = 33,
IDXD_CMDSTS_ERR_WQ_SIZE = 34,
IDXD_CMDSTS_ERR_WQ_PRIOR = 35,
IDXD_CMDSTS_ERR_WQ_MODE = 36,
IDXD_CMDSTS_ERR_BOF_EN = 37,
IDXD_CMDSTS_ERR_PASID_EN = 38,
IDXD_CMDSTS_ERR_MAX_BATCH_SIZE = 39,
IDXD_CMDSTS_ERR_MAX_XFER_SIZE = 40,
IDXD_CMDSTS_ERR_DIS_DEV_EN = 49,
IDXD_CMDSTS_ERR_DEV_NOT_EN = 50,
IDXD_CMDSTS_ERR_INVAL_INT_IDX = 65,
IDXD_CMDSTS_ERR_NO_HANDLE = 66,
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
};
enum idxd_wq_hw_state {
IDXD_WQ_DEV_DISABLED = 0,
IDXD_WQ_DEV_ENABLED = 1,
IDXD_WQ_DEV_BUSY = 2,
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
};
struct idxd_hw_desc {
uint32_t pasid: 20;
uint32_t rsvd: 11;
uint32_t priv: 1;
uint32_t flags: 24;
uint32_t opcode: 8;
uint64_t completion_addr;
union {
uint64_t src_addr;
uint64_t src1_addr;
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t readback_addr;
uint64_t pattern;
uint64_t desc_list_addr;
};
union {
uint64_t dst_addr;
uint64_t readback_addr2;
uint64_t src2_addr;
uint64_t comp_pattern;
};
union {
uint32_t src1_size;
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint32_t xfer_size;
uint32_t desc_count;
};
uint16_t int_handle;
union {
uint16_t rsvd1;
uint16_t compr_flags;
uint16_t decompr_flags;
};
union {
struct {
uint64_t src2_addr;
uint32_t max_dst_size;
uint32_t src2_size;
uint32_t filter_flags;
uint32_t num_inputs;
} iaa;
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint8_t expected_res;
struct {
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t addr;
uint32_t max_size;
} delta;
uint32_t delta_rec_size;
uint64_t dest2;
struct {
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint32_t seed;
uint32_t rsvd;
uint64_t addr;
} crc32c;
struct {
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint8_t src_flags;
uint8_t rsvd1;
uint8_t flags;
uint8_t rsvd2[5];
uint32_t ref_tag_seed;
uint16_t app_tag_mask;
uint16_t app_tag_seed;
} dif_chk;
struct {
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint8_t rsvd1;
uint8_t dest_flag;
uint8_t flags;
uint8_t rsvd2[13];
uint32_t ref_tag_seed;
uint16_t app_tag_mask;
uint16_t app_tag_seed;
} dif_ins;
struct {
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint8_t src_flags;
uint8_t dest_flags;
uint8_t flags;
uint8_t rsvd[5];
uint32_t src_ref_tag_seed;
uint16_t src_app_tag_mask;
uint16_t src_app_tag_seed;
uint32_t dest_ref_tag_seed;
uint16_t dest_app_tag_mask;
uint16_t dest_app_tag_seed;
} dif_upd;
uint8_t op_specific[24];
};
} __attribute((aligned(64)));
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
SPDK_STATIC_ASSERT(sizeof(struct idxd_hw_desc) == 64, "size mismatch");
struct dsa_hw_comp_record {
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
volatile uint8_t status;
union {
uint8_t result;
uint8_t dif_status;
};
uint16_t rsvd;
uint32_t bytes_completed;
uint64_t fault_addr;
union {
uint32_t delta_rec_size;
uint32_t crc32c_val;
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
struct {
uint32_t dif_chk_ref_tag;
uint16_t dif_chk_app_tag_mask;
uint16_t dif_chk_app_tag;
};
struct {
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t rsvd;
uint32_t ref_tag;
uint16_t app_tag_mask;
uint16_t app_tag;
} dif_ins_comp;
struct {
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint32_t src_ref_tag;
uint16_t src_app_tag_mask;
uint16_t src_app_tag;
uint32_t dest_ref_tag;
uint16_t dest_app_tag_mask;
uint16_t dest_app_tag;
} dif_upd_comp;
uint8_t op_specific[16];
};
};
SPDK_STATIC_ASSERT(sizeof(struct dsa_hw_comp_record) == 32, "size mismatch");
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
struct iaa_hw_comp_record {
volatile uint8_t status;
uint8_t error_code;
uint16_t rsvd;
uint32_t bytes_completed;
uint64_t fault_addr;
uint32_t invalid_flags;
uint32_t rsvd2;
uint32_t output_size;
uint8_t output_bits;
uint8_t rsvd3;
uint16_t rsvd4;
uint64_t rsvd5[4];
};
SPDK_STATIC_ASSERT(sizeof(struct iaa_hw_comp_record) == 64, "size mismatch");
struct iaa_aecs {
uint32_t crc;
uint32_t xor_checksum;
uint32_t rsvd[5];
uint32_t num_output_accum_bits;
uint8_t output_accum[256];
uint32_t ll_sym[286];
uint32_t rsvd1;
uint32_t rsvd3;
uint32_t d_sym[30];
uint32_t pad[2];
};
SPDK_STATIC_ASSERT(sizeof(struct iaa_aecs) == 1568, "size mismatch");
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
union idxd_gencap_register {
struct {
uint64_t block_on_fault: 1;
uint64_t overlap_copy: 1;
uint64_t cache_control_mem: 1;
uint64_t cache_control_cache: 1;
uint64_t command_cap: 1;
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t rsvd: 3;
uint64_t dest_readback: 1;
uint64_t drain_readback: 1;
uint64_t rsvd2: 6;
uint64_t max_xfer_shift: 5;
uint64_t max_batch_shift: 4;
uint64_t max_ims_mult: 6;
uint64_t config_support: 1;
uint64_t rsvd3: 32;
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_gencap_register) == 8, "size mismatch");
union idxd_wqcap_register {
struct {
uint64_t total_wq_size: 16;
uint64_t num_wqs: 8;
uint64_t wqcfg_size: 4;
uint64_t rsvd: 20;
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t shared_mode: 1;
uint64_t dedicated_mode: 1;
uint64_t ats_support: 1;
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t priority: 1;
uint64_t occupancy: 1;
uint64_t occupancy_int: 1;
uint64_t rsvd1: 10;
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_wqcap_register) == 8, "size mismatch");
union idxd_groupcap_register {
struct {
uint64_t num_groups: 8;
uint64_t read_bufs: 8;
uint64_t read_bufs_ctrl: 1;
uint64_t read_bus_limit: 1;
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t rsvd: 46;
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_groupcap_register) == 8, "size mismatch");
union idxd_enginecap_register {
struct {
uint64_t num_engines: 8;
uint64_t rsvd: 56;
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_enginecap_register) == 8, "size mismatch");
struct idxd_opcap_register {
uint64_t raw[4];
};
SPDK_STATIC_ASSERT(sizeof(struct idxd_opcap_register) == 32, "size mismatch");
union idxd_offsets_register {
struct {
uint64_t grpcfg: 16;
uint64_t wqcfg: 16;
uint64_t msix_perm: 16;
uint64_t ims: 16;
uint64_t perfmon: 16;
uint64_t rsvd: 48;
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t raw[2];
};
SPDK_STATIC_ASSERT(sizeof(union idxd_offsets_register) == 16, "size mismatch");
union idxd_gencfg_register {
struct {
uint8_t global_read_buf_limit;
uint8_t reserved0 : 4;
uint8_t user_mode_int_enabled : 1;
uint8_t reserved1 : 3;
uint16_t reserved2;
};
uint32_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_gencfg_register) == 4, "size mismatch");
union idxd_genctrl_register {
struct {
uint32_t sw_err_int_enable : 1;
uint32_t halt_state_int_enable : 1;
uint32_t reserved : 30;
};
uint32_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_genctrl_register) == 4, "size mismatch");
union idxd_gensts_register {
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
struct {
uint32_t state: 2;
uint32_t reset_type: 2;
uint32_t rsvd: 28;
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint32_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_gensts_register) == 4, "size mismatch");
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
union idxd_intcause_register {
struct {
uint32_t software_err : 1;
uint32_t command_completion : 1;
uint32_t wq_occupancy_below_limit : 1;
uint32_t perfmon_counter_overflow : 1;
uint32_t halt_state : 1;
uint32_t reserved : 26;
uint32_t int_handles_revoked : 1;
};
uint32_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_intcause_register) == 4, "size mismatch");
union idxd_cmd_register {
struct {
uint32_t operand : 20;
uint32_t command_code : 5;
uint32_t reserved : 6;
uint32_t request_completion_interrupt : 1;
};
uint32_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_cmd_register) == 4, "size mismatch");
union idxd_cmdsts_register {
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
struct {
uint32_t err : 8;
uint32_t result : 16;
uint32_t rsvd: 7;
uint32_t active: 1;
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint32_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_cmdsts_register) == 4, "size mismatch");
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
union idxd_cmdcap_register {
struct {
uint32_t reserved0 : 1;
uint32_t enable_device : 1;
uint32_t disable_device : 1;
uint32_t drain_all : 1;
uint32_t abort_all : 1;
uint32_t reset_device : 1;
uint32_t enable_wq : 1;
uint32_t disable_wq : 1;
uint32_t drain_wq : 1;
uint32_t abort_wq : 1;
uint32_t reset_wq : 1;
uint32_t drain_pasid : 1;
uint32_t abort_pasid : 1;
uint32_t request_int_handle : 1;
uint32_t release_int_handle : 1;
uint32_t reserved1 : 17;
};
uint32_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_cmdcap_register) == 4, "size mismatch");
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
union idxd_swerr_register {
struct {
uint64_t valid: 1;
uint64_t overflow: 1;
uint64_t desc_valid: 1;
uint64_t wq_idx_valid: 1;
uint64_t batch: 1;
uint64_t fault_rw: 1;
uint64_t priv: 1;
uint64_t rsvd: 1;
uint64_t error: 8;
uint64_t wq_idx: 8;
uint64_t rsvd2: 8;
uint64_t operation: 8;
uint64_t pasid: 20;
uint64_t rsvd3: 4;
uint64_t batch_idx: 16;
uint64_t rsvd4: 16;
uint64_t invalid_flags: 32;
uint64_t fault_addr;
uint64_t rsvd5;
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint64_t raw[4];
};
SPDK_STATIC_ASSERT(sizeof(union idxd_swerr_register) == 32, "size mismatch");
struct idxd_registers {
uint32_t version;
uint32_t reserved0;
uint64_t reserved1;
union idxd_gencap_register gencap;
uint64_t reserved2;
union idxd_wqcap_register wqcap;
uint64_t reserved3;
union idxd_groupcap_register groupcap;
union idxd_enginecap_register enginecap;
struct idxd_opcap_register opcap;
union idxd_offsets_register offsets;
uint64_t reserved4[2];
union idxd_gencfg_register gencfg;
uint32_t reserved5;
union idxd_genctrl_register genctrl;
uint32_t reserved6;
union idxd_gensts_register gensts;
uint32_t reserved7;
union idxd_intcause_register intcause;
uint32_t reserved8;
union idxd_cmd_register cmd;
uint32_t reserved9;
union idxd_cmdsts_register cmdsts;
uint32_t reserved10;
union idxd_cmdcap_register cmdcap;
uint32_t reserved11;
uint64_t reserved12;
union idxd_swerr_register sw_err;
};
SPDK_STATIC_ASSERT(sizeof(struct idxd_registers) == 0xE0, "size mismatch");
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
union idxd_group_flags {
struct {
uint32_t tc_a : 3;
uint32_t tc_b : 3;
uint32_t reserved0 : 1;
uint32_t global_read_buffer_limit: 1;
uint32_t read_buffers_reserved : 8;
uint32_t reserved1: 4;
uint32_t read_buffers_allowed : 8;
uint32_t reserved2 : 4;
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint32_t raw;
};
SPDK_STATIC_ASSERT(sizeof(union idxd_group_flags) == 4, "size mismatch");
struct idxd_grpcfg {
uint64_t wqs[4];
uint64_t engines;
union idxd_group_flags flags;
/* This is not part of the definition, but in practice the stride in the table
* is 64 bytes. */
uint32_t reserved0;
uint64_t reserved1[2];
};
SPDK_STATIC_ASSERT(sizeof(struct idxd_grpcfg) == 64, "size mismatch");
struct idxd_grptbl {
struct idxd_grpcfg group[1];
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
};
union idxd_wqcfg {
struct {
uint16_t wq_size;
uint16_t rsvd;
uint16_t wq_thresh;
uint16_t rsvd1;
uint32_t mode: 1;
uint32_t bof: 1;
uint32_t wq_ats_disable: 1;
uint32_t rsvd2: 1;
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint32_t priority: 4;
uint32_t pasid: 20;
uint32_t pasid_en: 1;
uint32_t priv: 1;
uint32_t rsvd3: 2;
uint32_t max_xfer_shift: 5;
uint32_t max_batch_shift: 4;
uint32_t rsvd4: 23;
uint16_t occupancy_inth;
uint16_t occupancy_table_sel: 1;
uint16_t rsvd5: 15;
uint16_t occupancy_limit;
uint16_t occupancy_int_en: 1;
uint16_t rsvd6: 15;
uint16_t occupancy;
uint16_t occupancy_int: 1;
uint16_t rsvd7: 12;
uint16_t mode_support: 1;
uint16_t wq_state: 2;
uint32_t rsvd8;
};
lib/idxd: add low level idxd library Module, etc., will follow. Notes: * IDXD is an Intel silicon feature available in future Intel CPUs. Initial development is being done on a simulator. Once HW is available and the code fully tested the experimental label will be lifted. Spec can be found here: https://software.intel.com/en-us/download/intel-data-streaming-accelerator-preliminary-architecture-specification * The current implementation will only work with VFIO. * DSA has a number of engines that can be grouped based on application need such as type of memory being served or QoS. Engines are processing units and are assigned to groups. Work queues are on device structures that act as front-end groups for queueing descriptors. Full details on what is configurable & how will come in later doc patches. * There is a finite number of work queue slots that are divided amongst the number of desired work queues in some fashion (ie evenly). * SW (outside of the idxd lib) is required to manage flow control, to not over-run the work queues.This is provided in the accel plug-in module. The upper layers use public API to manage this. * Work queue submissions are done with a 64 byte atomic instruction * The design here creates a set of descriptor rings per channel that match the size of the work queues. Then, an spdk_bit_array is used to make sure we don't overrun a queue. If there are not slots available, the operation is put on a linked list to be retried later from the poller. * As we need to support any number of channels (we can't limit ourselves to the number of work queues) we need to dynamically size/resize our per channel descriptor rings based on the number of current channels. This is done from upper layers via public API into the lib. * As channels are created, the total number of work queue slots is divided across the channels evenly. Same thing when they are destroyed, remaining channels with see the ring sizes increase. This is done from upper layers via public API into the lib. * The sim has 64 total work queue entries (WQE) that get dolled out to the work queues (WQ) evenly. Signed-off-by: paul luse <paul.e.luse@intel.com> Change-Id: I899bbeda3cef3db05bea4197b8757e89dddb579d Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/1809 Community-CI: Mellanox Build Bot Reviewed-by: Jim Harris <james.r.harris@intel.com> Reviewed-by: Ben Walker <benjamin.walker@intel.com> Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com> Reviewed-by: Vitaliy Mysak <vitaliy.mysak@intel.com> Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
2020-04-10 15:29:01 +00:00
uint32_t raw[8];
};
SPDK_STATIC_ASSERT(sizeof(union idxd_wqcfg) == 32, "size mismatch");
#ifdef __cplusplus
}
#endif
#endif /* SPDK_IDXD_SPEC_H */