From patchwork Fri Aug 7 18:34:25 2015 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Tadeusz Struk X-Patchwork-Id: 6972631 X-Patchwork-Delegate: bhelgaas@google.com Return-Path: X-Original-To: patchwork-linux-pci@patchwork.kernel.org Delivered-To: patchwork-parsemail@patchwork2.web.kernel.org Received: from mail.kernel.org (mail.kernel.org [198.145.29.136]) by patchwork2.web.kernel.org (Postfix) with ESMTP id 360ABC05AC for ; Fri, 7 Aug 2015 18:36:14 +0000 (UTC) Received: from mail.kernel.org (localhost [127.0.0.1]) by mail.kernel.org (Postfix) with ESMTP id 58B5F20613 for ; Fri, 7 Aug 2015 18:36:10 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.kernel.org (Postfix) with ESMTP id 02A8C20623 for ; Fri, 7 Aug 2015 18:36:06 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1946203AbbHGSfx (ORCPT ); Fri, 7 Aug 2015 14:35:53 -0400 Received: from mga01.intel.com ([192.55.52.88]:35859 "EHLO mga01.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1946070AbbHGSfu (ORCPT ); Fri, 7 Aug 2015 14:35:50 -0400 Received: from fmsmga003.fm.intel.com ([10.253.24.29]) by fmsmga101.fm.intel.com with ESMTP; 07 Aug 2015 11:35:26 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.15,630,1432623600"; d="scan'208";a="537898775" Received: from tstruk-mobl1.jf.intel.com (HELO [127.0.1.1]) ([134.134.171.170]) by FMSMGA003.fm.intel.com with ESMTP; 07 Aug 2015 11:35:25 -0700 Subject: [PATCH 2/5] crypto: qat - Add support for SRIOV From: Tadeusz Struk To: herbert@gondor.apana.org.au Cc: linux-crypto@vger.kernel.org, bhelgaas@google.com, linux-pci@vger.kernel.org, qat-linux@intel.com, tadeusz.struk@intel.com Date: Fri, 07 Aug 2015 11:34:25 -0700 Message-ID: <20150807183425.630.32268.stgit@tstruk-mobl1> In-Reply-To: <20150807183414.630.25889.stgit@tstruk-mobl1> References: <20150807183414.630.25889.stgit@tstruk-mobl1> User-Agent: StGit/0.17.1-dirty MIME-Version: 1.0 Sender: linux-pci-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-pci@vger.kernel.org X-Spam-Status: No, score=-7.0 required=5.0 tests=BAYES_00, RCVD_IN_DNSWL_HI, RP_MATCHES_RCVD, UNPARSEABLE_RELAY autolearn=unavailable version=3.3.1 X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on mail.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP Add code that enables SRIOV on dh895xcc devices. Signed-off-by: Tadeusz Struk --- drivers/crypto/qat/qat_common/Makefile | 1 drivers/crypto/qat/qat_common/adf_accel_devices.h | 37 ++ drivers/crypto/qat/qat_common/adf_aer.c | 3 drivers/crypto/qat/qat_common/adf_cfg.c | 3 drivers/crypto/qat/qat_common/adf_cfg_common.h | 3 drivers/crypto/qat/qat_common/adf_common_drv.h | 37 ++ drivers/crypto/qat/qat_common/adf_ctl_drv.c | 6 drivers/crypto/qat/qat_common/adf_dev_mgr.c | 283 +++++++++++++- drivers/crypto/qat/qat_common/adf_init.c | 10 drivers/crypto/qat/qat_common/adf_pf2vf_msg.c | 336 +++++++++++++++++ drivers/crypto/qat/qat_common/adf_pf2vf_msg.h | 144 +++++++ drivers/crypto/qat/qat_common/adf_sriov.c | 406 ++++++++++++++++++++ drivers/crypto/qat/qat_common/qat_crypto.c | 4 .../crypto/qat/qat_dh895xcc/adf_dh895xcc_hw_data.c | 24 + .../crypto/qat/qat_dh895xcc/adf_dh895xcc_hw_data.h | 4 drivers/crypto/qat/qat_dh895xcc/adf_drv.c | 80 ++-- drivers/crypto/qat/qat_dh895xcc/adf_isr.c | 139 +++++-- 17 files changed, 1426 insertions(+), 94 deletions(-) create mode 100644 drivers/crypto/qat/qat_common/adf_pf2vf_msg.c create mode 100644 drivers/crypto/qat/qat_common/adf_pf2vf_msg.h create mode 100644 drivers/crypto/qat/qat_common/adf_sriov.c -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html diff --git a/drivers/crypto/qat/qat_common/Makefile b/drivers/crypto/qat/qat_common/Makefile index e5fe4e7..df20a9d 100644 --- a/drivers/crypto/qat/qat_common/Makefile +++ b/drivers/crypto/qat/qat_common/Makefile @@ -19,3 +19,4 @@ intel_qat-objs := adf_cfg.o \ qat_hal.o intel_qat-$(CONFIG_DEBUG_FS) += adf_transport_debug.o +intel_qat-$(CONFIG_PCI_IOV) += adf_sriov.o adf_pf2vf_msg.o diff --git a/drivers/crypto/qat/qat_common/adf_accel_devices.h b/drivers/crypto/qat/qat_common/adf_accel_devices.h index a7d3d11..ca853d5 100644 --- a/drivers/crypto/qat/qat_common/adf_accel_devices.h +++ b/drivers/crypto/qat/qat_common/adf_accel_devices.h @@ -46,13 +46,17 @@ */ #ifndef ADF_ACCEL_DEVICES_H_ #define ADF_ACCEL_DEVICES_H_ +#include #include #include #include +#include #include "adf_cfg_common.h" #define ADF_DH895XCC_DEVICE_NAME "dh895xcc" +#define ADF_DH895XCCVF_DEVICE_NAME "dh895xccvf" #define ADF_DH895XCC_PCI_DEVICE_ID 0x435 +#define ADF_DH895XCCIOV_PCI_DEVICE_ID 0x443 #define ADF_PCI_MAX_BARS 3 #define ADF_DEVICE_NAME_LENGTH 32 #define ADF_ETR_MAX_RINGS_PER_BANK 16 @@ -79,6 +83,7 @@ struct adf_bar { struct adf_accel_msix { struct msix_entry *entries; char **names; + u32 num_entries; } __packed; struct adf_accel_pci { @@ -99,6 +104,7 @@ enum dev_sku_info { DEV_SKU_2, DEV_SKU_3, DEV_SKU_4, + DEV_SKU_VF, DEV_SKU_UNKNOWN, }; @@ -113,6 +119,8 @@ static inline const char *get_sku_info(enum dev_sku_info info) return "SKU3"; case DEV_SKU_4: return "SKU4"; + case DEV_SKU_VF: + return "SKUVF"; case DEV_SKU_UNKNOWN: default: break; @@ -140,6 +148,8 @@ struct adf_hw_device_data { uint32_t (*get_etr_bar_id)(struct adf_hw_device_data *self); uint32_t (*get_num_aes)(struct adf_hw_device_data *self); uint32_t (*get_num_accels)(struct adf_hw_device_data *self); + uint32_t (*get_pf2vf_offset)(uint32_t i); + uint32_t (*get_vintmsk_offset)(uint32_t i); enum dev_sku_info (*get_sku)(struct adf_hw_device_data *self); int (*alloc_irq)(struct adf_accel_dev *accel_dev); void (*free_irq)(struct adf_accel_dev *accel_dev); @@ -151,7 +161,9 @@ struct adf_hw_device_data { void (*exit_arb)(struct adf_accel_dev *accel_dev); void (*get_arb_mapping)(struct adf_accel_dev *accel_dev, const uint32_t **cfg); + void (*disable_iov)(struct adf_accel_dev *accel_dev); void (*enable_ints)(struct adf_accel_dev *accel_dev); + int (*enable_vf2pf_comms)(struct adf_accel_dev *accel_dev); const char *fw_name; const char *fw_mmp_name; uint32_t fuses; @@ -165,6 +177,7 @@ struct adf_hw_device_data { uint8_t num_accel; uint8_t num_logical_accel; uint8_t num_engines; + uint8_t min_iov_compat_ver; } __packed; /* CSR write macro */ @@ -189,6 +202,15 @@ struct adf_fw_loader_data { const struct firmware *mmp_fw; }; +struct adf_accel_vf_info { + struct adf_accel_dev *accel_dev; + struct tasklet_struct vf2pf_bh_tasklet; + struct mutex pf2vf_lock; /* protect CSR access for PF2VF messages */ + struct ratelimit_state vf2pf_ratelimit; + u32 vf_nr; + bool init; +}; + struct adf_accel_dev { struct adf_etr_data *transport; struct adf_hw_device_data *hw_device; @@ -202,6 +224,21 @@ struct adf_accel_dev { struct list_head list; struct module *owner; struct adf_accel_pci accel_pci_dev; + union { + struct { + /* vf_info is non-zero when SR-IOV is init'ed */ + struct adf_accel_vf_info *vf_info; + } pf; + struct { + char *irq_name; + struct tasklet_struct pf2vf_bh_tasklet; + struct mutex vf2pf_lock; /* protect CSR access */ + struct completion iov_msg_completion; + uint8_t compatible; + uint8_t pf_version; + } vf; + }; + bool is_vf; uint8_t accel_id; } __packed; #endif diff --git a/drivers/crypto/qat/qat_common/adf_aer.c b/drivers/crypto/qat/qat_common/adf_aer.c index 8f34a5f..a57b419 100644 --- a/drivers/crypto/qat/qat_common/adf_aer.c +++ b/drivers/crypto/qat/qat_common/adf_aer.c @@ -91,6 +91,9 @@ static void adf_dev_restore(struct adf_accel_dev *accel_dev) dev_info(&GET_DEV(accel_dev), "Resetting device qat_dev%d\n", accel_dev->accel_id); + if (!parent) + parent = pdev; + if (!pci_wait_for_pending_transaction(pdev)) dev_info(&GET_DEV(accel_dev), "Transaction still in progress. Proceeding\n"); diff --git a/drivers/crypto/qat/qat_common/adf_cfg.c b/drivers/crypto/qat/qat_common/adf_cfg.c index 76eccb8..d087979 100644 --- a/drivers/crypto/qat/qat_common/adf_cfg.c +++ b/drivers/crypto/qat/qat_common/adf_cfg.c @@ -178,6 +178,9 @@ void adf_cfg_dev_remove(struct adf_accel_dev *accel_dev) { struct adf_cfg_device_data *dev_cfg_data = accel_dev->cfg; + if (!dev_cfg_data) + return; + down_write(&dev_cfg_data->lock); adf_cfg_section_del_all(&dev_cfg_data->sec_list); up_write(&dev_cfg_data->lock); diff --git a/drivers/crypto/qat/qat_common/adf_cfg_common.h b/drivers/crypto/qat/qat_common/adf_cfg_common.h index 88b8218..c697fb1 100644 --- a/drivers/crypto/qat/qat_common/adf_cfg_common.h +++ b/drivers/crypto/qat/qat_common/adf_cfg_common.h @@ -60,7 +60,7 @@ #define ADF_CFG_NO_DEVICE 0xFF #define ADF_CFG_AFFINITY_WHATEVER 0xFF #define MAX_DEVICE_NAME_SIZE 32 -#define ADF_MAX_DEVICES 32 +#define ADF_MAX_DEVICES (32 * 32) enum adf_cfg_val_type { ADF_DEC, @@ -71,6 +71,7 @@ enum adf_cfg_val_type { enum adf_device_type { DEV_UNKNOWN = 0, DEV_DH895XCC, + DEV_DH895XCCVF, }; struct adf_dev_status_info { diff --git a/drivers/crypto/qat/qat_common/adf_common_drv.h b/drivers/crypto/qat/qat_common/adf_common_drv.h index 7935a38..0f5f5e8 100644 --- a/drivers/crypto/qat/qat_common/adf_common_drv.h +++ b/drivers/crypto/qat/qat_common/adf_common_drv.h @@ -54,8 +54,8 @@ #include "icp_qat_hal.h" #define ADF_MAJOR_VERSION 0 -#define ADF_MINOR_VERSION 1 -#define ADF_BUILD_VERSION 4 +#define ADF_MINOR_VERSION 2 +#define ADF_BUILD_VERSION 0 #define ADF_DRV_VERSION __stringify(ADF_MAJOR_VERSION) "." \ __stringify(ADF_MINOR_VERSION) "." \ __stringify(ADF_BUILD_VERSION) @@ -95,7 +95,7 @@ struct service_hndl { static inline int get_current_node(void) { - return cpu_data(current_thread_info()->cpu).phys_proc_id; + return topology_physical_package_id(smp_processor_id()); } int adf_service_register(struct service_hndl *service); @@ -106,13 +106,23 @@ int adf_dev_start(struct adf_accel_dev *accel_dev); int adf_dev_stop(struct adf_accel_dev *accel_dev); void adf_dev_shutdown(struct adf_accel_dev *accel_dev); +void adf_enable_pf2vf_interrupts(struct adf_accel_dev *accel_dev); +void adf_disable_pf2vf_interrupts(struct adf_accel_dev *accel_dev); +int adf_iov_putmsg(struct adf_accel_dev *accel_dev, u32 msg, u8 vf_nr); +void adf_pf2vf_notify_restarting(struct adf_accel_dev *accel_dev); +int adf_enable_vf2pf_comms(struct adf_accel_dev *accel_dev); +void adf_devmgr_update_class_index(struct adf_hw_device_data *hw_data); +void adf_clean_vf_map(bool); + int adf_ctl_dev_register(void); void adf_ctl_dev_unregister(void); int adf_processes_dev_register(void); void adf_processes_dev_unregister(void); -int adf_devmgr_add_dev(struct adf_accel_dev *accel_dev); -void adf_devmgr_rm_dev(struct adf_accel_dev *accel_dev); +int adf_devmgr_add_dev(struct adf_accel_dev *accel_dev, + struct adf_accel_dev *pf); +void adf_devmgr_rm_dev(struct adf_accel_dev *accel_dev, + struct adf_accel_dev *pf); struct list_head *adf_devmgr_get_head(void); struct adf_accel_dev *adf_devmgr_get_dev_by_id(uint32_t id); struct adf_accel_dev *adf_devmgr_get_first(void); @@ -211,4 +221,21 @@ int qat_uclo_map_uof_obj(struct icp_qat_fw_loader_handle *handle, void *addr_ptr, int mem_size); void qat_uclo_wr_mimage(struct icp_qat_fw_loader_handle *handle, void *addr_ptr, int mem_size); +#if defined(CONFIG_PCI_IOV) +int adf_sriov_configure(struct pci_dev *pdev, int numvfs); +void adf_disable_sriov(struct adf_accel_dev *accel_dev); +void adf_disable_vf2pf_interrupts(struct adf_accel_dev *accel_dev, + uint32_t vf_mask); +void adf_enable_vf2pf_interrupts(struct adf_accel_dev *accel_dev, + uint32_t vf_mask); +#else +static inline int adf_sriov_configure(struct pci_dev *pdev, int numvfs) +{ + return 0; +} + +static inline void adf_disable_sriov(struct adf_accel_dev *accel_dev) +{ +} +#endif #endif diff --git a/drivers/crypto/qat/qat_common/adf_ctl_drv.c b/drivers/crypto/qat/qat_common/adf_ctl_drv.c index e056b9e..cd8a12a 100644 --- a/drivers/crypto/qat/qat_common/adf_ctl_drv.c +++ b/drivers/crypto/qat/qat_common/adf_ctl_drv.c @@ -398,10 +398,9 @@ static int adf_ctl_ioctl_get_status(struct file *fp, unsigned int cmd, } accel_dev = adf_devmgr_get_dev_by_id(dev_info.accel_id); - if (!accel_dev) { - pr_err("QAT: Device %d not found\n", dev_info.accel_id); + if (!accel_dev) return -ENODEV; - } + hw_data = accel_dev->hw_device; dev_info.state = adf_dev_started(accel_dev) ? DEV_UP : DEV_DOWN; dev_info.num_ae = hw_data->get_num_aes(hw_data); @@ -495,6 +494,7 @@ static void __exit adf_unregister_ctl_device_driver(void) adf_exit_aer(); qat_crypto_unregister(); qat_algs_exit(); + adf_clean_vf_map(false); mutex_destroy(&adf_ctl_lock); } diff --git a/drivers/crypto/qat/qat_common/adf_dev_mgr.c b/drivers/crypto/qat/qat_common/adf_dev_mgr.c index b574a82..8dfdb8f 100644 --- a/drivers/crypto/qat/qat_common/adf_dev_mgr.c +++ b/drivers/crypto/qat/qat_common/adf_dev_mgr.c @@ -50,21 +50,125 @@ #include "adf_common_drv.h" static LIST_HEAD(accel_table); +static LIST_HEAD(vfs_table); static DEFINE_MUTEX(table_lock); static uint32_t num_devices; +struct vf_id_map { + u32 bdf; + u32 id; + u32 fake_id; + bool attached; + struct list_head list; +}; + +static int adf_get_vf_id(struct adf_accel_dev *vf) +{ + return ((7 * (PCI_SLOT(accel_to_pci_dev(vf)->devfn) - 1)) + + PCI_FUNC(accel_to_pci_dev(vf)->devfn) + + (PCI_SLOT(accel_to_pci_dev(vf)->devfn) - 1)); +} + +static int adf_get_vf_num(struct adf_accel_dev *vf) +{ + return (accel_to_pci_dev(vf)->bus->number << 8) | adf_get_vf_id(vf); +} + +static struct vf_id_map *adf_find_vf(u32 bdf) +{ + struct list_head *itr; + + list_for_each(itr, &vfs_table) { + struct vf_id_map *ptr = + list_entry(itr, struct vf_id_map, list); + + if (ptr->bdf == bdf) + return ptr; + } + return NULL; +} + +static int adf_get_vf_real_id(u32 fake) +{ + struct list_head *itr; + + list_for_each(itr, &vfs_table) { + struct vf_id_map *ptr = + list_entry(itr, struct vf_id_map, list); + if (ptr->fake_id == fake) + return ptr->id; + } + return -1; +} + +/** + * adf_clean_vf_map() - Cleans VF id mapings + * + * Function cleans internal ids for virtual functions. + * @vf: flag indicating whether mappings is cleaned + * for vfs only or for vfs and pfs + */ +void adf_clean_vf_map(bool vf) +{ + struct vf_id_map *map; + struct list_head *ptr, *tmp; + + mutex_lock(&table_lock); + list_for_each_safe(ptr, tmp, &vfs_table) { + map = list_entry(ptr, struct vf_id_map, list); + if (map->bdf != -1) + num_devices--; + + if (vf && map->bdf == -1) + continue; + + list_del(ptr); + kfree(map); + } + mutex_unlock(&table_lock); +} +EXPORT_SYMBOL_GPL(adf_clean_vf_map); + +/** + * adf_devmgr_update_class_index() - Update internal index + * @hw_data: Pointer to internal device data. + * + * Function updates internal dev index for VFs + */ +void adf_devmgr_update_class_index(struct adf_hw_device_data *hw_data) +{ + struct adf_hw_device_class *class = hw_data->dev_class; + struct list_head *itr; + int i = 0; + + list_for_each(itr, &accel_table) { + struct adf_accel_dev *ptr = + list_entry(itr, struct adf_accel_dev, list); + + if (ptr->hw_device->dev_class == class) + ptr->hw_device->instance_id = i++; + + if (i == class->instances) + break; + } +} +EXPORT_SYMBOL_GPL(adf_devmgr_update_class_index); + /** * adf_devmgr_add_dev() - Add accel_dev to the acceleration framework * @accel_dev: Pointer to acceleration device. + * @pf: Corresponding PF if the accel_dev is a VF * * Function adds acceleration device to the acceleration framework. * To be used by QAT device specific drivers. * * Return: 0 on success, error code otherwise. */ -int adf_devmgr_add_dev(struct adf_accel_dev *accel_dev) +int adf_devmgr_add_dev(struct adf_accel_dev *accel_dev, + struct adf_accel_dev *pf) { struct list_head *itr; + int ret = 0; if (num_devices == ADF_MAX_DEVICES) { dev_err(&GET_DEV(accel_dev), "Only support up to %d devices\n", @@ -73,20 +177,77 @@ int adf_devmgr_add_dev(struct adf_accel_dev *accel_dev) } mutex_lock(&table_lock); - list_for_each(itr, &accel_table) { - struct adf_accel_dev *ptr = + atomic_set(&accel_dev->ref_count, 0); + + /* PF on host or VF on guest */ + if (!accel_dev->is_vf || (accel_dev->is_vf && !pf)) { + struct vf_id_map *map; + + list_for_each(itr, &accel_table) { + struct adf_accel_dev *ptr = list_entry(itr, struct adf_accel_dev, list); - if (ptr == accel_dev) { - mutex_unlock(&table_lock); - return -EEXIST; + if (ptr == accel_dev) { + ret = -EEXIST; + goto unlock; + } } + + list_add_tail(&accel_dev->list, &accel_table); + accel_dev->accel_id = num_devices++; + + map = kzalloc(sizeof(*map), GFP_KERNEL); + if (!map) { + ret = -ENOMEM; + goto unlock; + } + map->bdf = ~0; + map->id = accel_dev->accel_id; + map->fake_id = map->id; + map->attached = true; + list_add_tail(&map->list, &vfs_table); + } else if (accel_dev->is_vf && pf) { + /* VF on host */ + struct adf_accel_vf_info *vf_info; + struct vf_id_map *map; + + vf_info = pf->pf.vf_info + adf_get_vf_id(accel_dev); + + map = adf_find_vf(adf_get_vf_num(accel_dev)); + if (map) { + struct vf_id_map *next; + + accel_dev->accel_id = map->id; + list_add_tail(&accel_dev->list, &accel_table); + map->fake_id++; + map->attached = true; + next = list_next_entry(map, list); + while (next && &next->list != &vfs_table) { + next->fake_id++; + next = list_next_entry(next, list); + } + + ret = 0; + goto unlock; + } + + map = kzalloc(sizeof(*map), GFP_KERNEL); + if (!map) { + ret = -ENOMEM; + goto unlock; + } + + accel_dev->accel_id = num_devices++; + list_add_tail(&accel_dev->list, &accel_table); + map->bdf = adf_get_vf_num(accel_dev); + map->id = accel_dev->accel_id; + map->fake_id = map->id; + map->attached = true; + list_add_tail(&map->list, &vfs_table); } - atomic_set(&accel_dev->ref_count, 0); - list_add_tail(&accel_dev->list, &accel_table); - accel_dev->accel_id = num_devices++; +unlock: mutex_unlock(&table_lock); - return 0; + return ret; } EXPORT_SYMBOL_GPL(adf_devmgr_add_dev); @@ -98,17 +259,37 @@ struct list_head *adf_devmgr_get_head(void) /** * adf_devmgr_rm_dev() - Remove accel_dev from the acceleration framework. * @accel_dev: Pointer to acceleration device. + * @pf: Corresponding PF if the accel_dev is a VF * * Function removes acceleration device from the acceleration framework. * To be used by QAT device specific drivers. * * Return: void */ -void adf_devmgr_rm_dev(struct adf_accel_dev *accel_dev) +void adf_devmgr_rm_dev(struct adf_accel_dev *accel_dev, + struct adf_accel_dev *pf) { mutex_lock(&table_lock); + if (!accel_dev->is_vf || (accel_dev->is_vf && !pf)) { + num_devices--; + } else if (accel_dev->is_vf && pf) { + struct vf_id_map *map, *next; + + map = adf_find_vf(adf_get_vf_num(accel_dev)); + if (!map) { + dev_err(&GET_DEV(accel_dev), "Failed to find VF map\n"); + goto unlock; + } + map->fake_id--; + map->attached = false; + next = list_next_entry(map, list); + while (next && &next->list != &vfs_table) { + next->fake_id--; + next = list_next_entry(next, list); + } + } +unlock: list_del(&accel_dev->list); - num_devices--; mutex_unlock(&table_lock); } EXPORT_SYMBOL_GPL(adf_devmgr_rm_dev); @@ -154,17 +335,24 @@ EXPORT_SYMBOL_GPL(adf_devmgr_pci_to_accel_dev); struct adf_accel_dev *adf_devmgr_get_dev_by_id(uint32_t id) { struct list_head *itr; + int real_id; mutex_lock(&table_lock); + real_id = adf_get_vf_real_id(id); + if (real_id < 0) + goto unlock; + + id = real_id; + list_for_each(itr, &accel_table) { struct adf_accel_dev *ptr = list_entry(itr, struct adf_accel_dev, list); - if (ptr->accel_id == id) { mutex_unlock(&table_lock); return ptr; } } +unlock: mutex_unlock(&table_lock); return NULL; } @@ -180,16 +368,52 @@ int adf_devmgr_verify_id(uint32_t id) return -ENODEV; } +static int adf_get_num_dettached_vfs(void) +{ + struct list_head *itr; + int vfs = 0; + + mutex_lock(&table_lock); + list_for_each(itr, &vfs_table) { + struct vf_id_map *ptr = + list_entry(itr, struct vf_id_map, list); + if (ptr->bdf != ~0 && !ptr->attached) + vfs++; + } + mutex_unlock(&table_lock); + return vfs; +} + void adf_devmgr_get_num_dev(uint32_t *num) { - *num = num_devices; + *num = num_devices - adf_get_num_dettached_vfs(); } +/** + * adf_dev_in_use() - Check whether accel_dev is currently in use + * @accel_dev: Pointer to acceleration device. + * + * To be used by QAT device specific drivers. + * + * Return: 1 when device is in use, 0 otherwise. + */ int adf_dev_in_use(struct adf_accel_dev *accel_dev) { return atomic_read(&accel_dev->ref_count) != 0; } +EXPORT_SYMBOL_GPL(adf_dev_in_use); +/** + * adf_dev_get() - Increment accel_dev reference count + * @accel_dev: Pointer to acceleration device. + * + * Increment the accel_dev refcount and if this is the first time + * incrementing it during this period the accel_dev is in use, + * increment the module refcount too. + * To be used by QAT device specific drivers. + * + * Return: 0 when successful, EFAULT when fail to bump module refcount + */ int adf_dev_get(struct adf_accel_dev *accel_dev) { if (atomic_add_return(1, &accel_dev->ref_count) == 1) @@ -197,19 +421,50 @@ int adf_dev_get(struct adf_accel_dev *accel_dev) return -EFAULT; return 0; } +EXPORT_SYMBOL_GPL(adf_dev_get); +/** + * adf_dev_put() - Decrement accel_dev reference count + * @accel_dev: Pointer to acceleration device. + * + * Decrement the accel_dev refcount and if this is the last time + * decrementing it during this period the accel_dev is in use, + * decrement the module refcount too. + * To be used by QAT device specific drivers. + * + * Return: void + */ void adf_dev_put(struct adf_accel_dev *accel_dev) { if (atomic_sub_return(1, &accel_dev->ref_count) == 0) module_put(accel_dev->owner); } +EXPORT_SYMBOL_GPL(adf_dev_put); +/** + * adf_devmgr_in_reset() - Check whether device is in reset + * @accel_dev: Pointer to acceleration device. + * + * To be used by QAT device specific drivers. + * + * Return: 1 when the device is being reset, 0 otherwise. + */ int adf_devmgr_in_reset(struct adf_accel_dev *accel_dev) { return test_bit(ADF_STATUS_RESTARTING, &accel_dev->status); } +EXPORT_SYMBOL_GPL(adf_devmgr_in_reset); +/** + * adf_dev_started() - Check whether device has started + * @accel_dev: Pointer to acceleration device. + * + * To be used by QAT device specific drivers. + * + * Return: 1 when the device has started, 0 otherwise + */ int adf_dev_started(struct adf_accel_dev *accel_dev) { return test_bit(ADF_STATUS_STARTED, &accel_dev->status); } +EXPORT_SYMBOL_GPL(adf_dev_started); diff --git a/drivers/crypto/qat/qat_common/adf_init.c b/drivers/crypto/qat/qat_common/adf_init.c index 1aaa5fd..ac37a89 100644 --- a/drivers/crypto/qat/qat_common/adf_init.c +++ b/drivers/crypto/qat/qat_common/adf_init.c @@ -187,6 +187,7 @@ int adf_dev_init(struct adf_accel_dev *accel_dev) } hw_data->enable_error_correction(accel_dev); + hw_data->enable_vf2pf_comms(accel_dev); return 0; } @@ -235,7 +236,8 @@ int adf_dev_start(struct adf_accel_dev *accel_dev) clear_bit(ADF_STATUS_STARTING, &accel_dev->status); set_bit(ADF_STATUS_STARTED, &accel_dev->status); - if (qat_algs_register() || qat_asym_algs_register()) { + if (!list_empty(&accel_dev->crypto_list) && + (qat_algs_register() || qat_asym_algs_register())) { dev_err(&GET_DEV(accel_dev), "Failed to register crypto algs\n"); set_bit(ADF_STATUS_STARTING, &accel_dev->status); @@ -270,11 +272,12 @@ int adf_dev_stop(struct adf_accel_dev *accel_dev) clear_bit(ADF_STATUS_STARTING, &accel_dev->status); clear_bit(ADF_STATUS_STARTED, &accel_dev->status); - if (qat_algs_unregister()) + if (!list_empty(&accel_dev->crypto_list) && qat_algs_unregister()) dev_err(&GET_DEV(accel_dev), "Failed to unregister crypto algs\n"); - qat_asym_algs_unregister(); + if (!list_empty(&accel_dev->crypto_list)) + qat_asym_algs_unregister(); list_for_each(list_itr, &service_table) { service = list_entry(list_itr, struct service_hndl, list); @@ -363,6 +366,7 @@ void adf_dev_shutdown(struct adf_accel_dev *accel_dev) if (hw_data->exit_admin_comms) hw_data->exit_admin_comms(accel_dev); + hw_data->disable_iov(accel_dev); adf_cleanup_etr_data(accel_dev); } EXPORT_SYMBOL_GPL(adf_dev_shutdown); diff --git a/drivers/crypto/qat/qat_common/adf_pf2vf_msg.c b/drivers/crypto/qat/qat_common/adf_pf2vf_msg.c new file mode 100644 index 0000000..c5790cd --- /dev/null +++ b/drivers/crypto/qat/qat_common/adf_pf2vf_msg.c @@ -0,0 +1,336 @@ +/* + This file is provided under a dual BSD/GPLv2 license. When using or + redistributing this file, you may do so under either license. + + GPL LICENSE SUMMARY + Copyright(c) 2015 Intel Corporation. + This program is free software; you can redistribute it and/or modify + it under the terms of version 2 of the GNU General Public License as + published by the Free Software Foundation. + + This program is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + General Public License for more details. + + Contact Information: + qat-linux@intel.com + + BSD LICENSE + Copyright(c) 2015 Intel Corporation. + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in + the documentation and/or other materials provided with the + distribution. + * Neither the name of Intel Corporation nor the names of its + contributors may be used to endorse or promote products derived + from this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include +#include +#include +#include "adf_accel_devices.h" +#include "adf_common_drv.h" +#include "adf_pf2vf_msg.h" + +#define ADF_DH895XCC_EP_OFFSET 0x3A000 +#define ADF_DH895XCC_ERRMSK3 (ADF_DH895XCC_EP_OFFSET + 0x1C) +#define ADF_DH895XCC_ERRMSK3_VF2PF_L_MASK(vf_mask) ((vf_mask & 0xFFFF) << 9) +#define ADF_DH895XCC_ERRMSK5 (ADF_DH895XCC_EP_OFFSET + 0xDC) +#define ADF_DH895XCC_ERRMSK5_VF2PF_U_MASK(vf_mask) (vf_mask >> 16) + +/** + * adf_enable_pf2vf_interrupts() - Enable PF to VF interrupts + * @accel_dev: Pointer to acceleration device. + * + * Function enables PF to VF interrupts + */ +void adf_enable_pf2vf_interrupts(struct adf_accel_dev *accel_dev) +{ + struct adf_accel_pci *pci_info = &accel_dev->accel_pci_dev; + struct adf_hw_device_data *hw_data = accel_dev->hw_device; + void __iomem *pmisc_bar_addr = + pci_info->pci_bars[hw_data->get_misc_bar_id(hw_data)].virt_addr; + + ADF_CSR_WR(pmisc_bar_addr, hw_data->get_vintmsk_offset(0), 0x0); +} +EXPORT_SYMBOL_GPL(adf_enable_pf2vf_interrupts); + +/** + * adf_disable_pf2vf_interrupts() - Disable PF to VF interrupts + * @accel_dev: Pointer to acceleration device. + * + * Function disables PF to VF interrupts + */ +void adf_disable_pf2vf_interrupts(struct adf_accel_dev *accel_dev) +{ + struct adf_accel_pci *pci_info = &accel_dev->accel_pci_dev; + struct adf_hw_device_data *hw_data = accel_dev->hw_device; + void __iomem *pmisc_bar_addr = + pci_info->pci_bars[hw_data->get_misc_bar_id(hw_data)].virt_addr; + + ADF_CSR_WR(pmisc_bar_addr, hw_data->get_vintmsk_offset(0), 0x2); +} +EXPORT_SYMBOL_GPL(adf_disable_pf2vf_interrupts); + +void adf_enable_vf2pf_interrupts(struct adf_accel_dev *accel_dev, + u32 vf_mask) +{ + struct adf_hw_device_data *hw_data = accel_dev->hw_device; + struct adf_bar *pmisc = + &GET_BARS(accel_dev)[hw_data->get_misc_bar_id(hw_data)]; + void __iomem *pmisc_addr = pmisc->virt_addr; + u32 reg; + + /* Enable VF2PF Messaging Ints - VFs 1 through 16 per vf_mask[15:0] */ + if (vf_mask & 0xFFFF) { + reg = ADF_CSR_RD(pmisc_addr, ADF_DH895XCC_ERRMSK3); + reg &= ~ADF_DH895XCC_ERRMSK3_VF2PF_L_MASK(vf_mask); + ADF_CSR_WR(pmisc_addr, ADF_DH895XCC_ERRMSK3, reg); + } + + /* Enable VF2PF Messaging Ints - VFs 17 through 32 per vf_mask[31:16] */ + if (vf_mask >> 16) { + reg = ADF_CSR_RD(pmisc_addr, ADF_DH895XCC_ERRMSK5); + reg &= ~ADF_DH895XCC_ERRMSK5_VF2PF_U_MASK(vf_mask); + ADF_CSR_WR(pmisc_addr, ADF_DH895XCC_ERRMSK5, reg); + } +} + +/** + * adf_disable_pf2vf_interrupts() - Disable VF to PF interrupts + * @accel_dev: Pointer to acceleration device. + * + * Function disables VF to PF interrupts + */ +void adf_disable_vf2pf_interrupts(struct adf_accel_dev *accel_dev, u32 vf_mask) +{ + struct adf_hw_device_data *hw_data = accel_dev->hw_device; + struct adf_bar *pmisc = + &GET_BARS(accel_dev)[hw_data->get_misc_bar_id(hw_data)]; + void __iomem *pmisc_addr = pmisc->virt_addr; + u32 reg; + + /* Disable VF2PF interrupts for VFs 1 through 16 per vf_mask[15:0] */ + if (vf_mask & 0xFFFF) { + reg = ADF_CSR_RD(pmisc_addr, ADF_DH895XCC_ERRMSK3) | + ADF_DH895XCC_ERRMSK3_VF2PF_L_MASK(vf_mask); + ADF_CSR_WR(pmisc_addr, ADF_DH895XCC_ERRMSK3, reg); + } + + /* Disable VF2PF interrupts for VFs 17 through 32 per vf_mask[31:16] */ + if (vf_mask >> 16) { + reg = ADF_CSR_RD(pmisc_addr, ADF_DH895XCC_ERRMSK5) | + ADF_DH895XCC_ERRMSK5_VF2PF_U_MASK(vf_mask); + ADF_CSR_WR(pmisc_addr, ADF_DH895XCC_ERRMSK5, reg); + } +} +EXPORT_SYMBOL_GPL(adf_disable_vf2pf_interrupts); + +static int __adf_iov_putmsg(struct adf_accel_dev *accel_dev, u32 msg, u8 vf_nr) +{ + struct adf_accel_pci *pci_info = &accel_dev->accel_pci_dev; + struct adf_hw_device_data *hw_data = accel_dev->hw_device; + void __iomem *pmisc_bar_addr = + pci_info->pci_bars[hw_data->get_misc_bar_id(hw_data)].virt_addr; + u32 val, pf2vf_offset, count = 0; + u32 local_in_use_mask, local_in_use_pattern; + u32 remote_in_use_mask, remote_in_use_pattern; + struct mutex *lock; /* lock preventing concurrent acces of CSR */ + u32 int_bit; + int ret = 0; + + if (accel_dev->is_vf) { + pf2vf_offset = hw_data->get_pf2vf_offset(0); + lock = &accel_dev->vf.vf2pf_lock; + local_in_use_mask = ADF_VF2PF_IN_USE_BY_VF_MASK; + local_in_use_pattern = ADF_VF2PF_IN_USE_BY_VF; + remote_in_use_mask = ADF_PF2VF_IN_USE_BY_PF_MASK; + remote_in_use_pattern = ADF_PF2VF_IN_USE_BY_PF; + int_bit = ADF_VF2PF_INT; + } else { + pf2vf_offset = hw_data->get_pf2vf_offset(vf_nr); + lock = &accel_dev->pf.vf_info[vf_nr].pf2vf_lock; + local_in_use_mask = ADF_PF2VF_IN_USE_BY_PF_MASK; + local_in_use_pattern = ADF_PF2VF_IN_USE_BY_PF; + remote_in_use_mask = ADF_VF2PF_IN_USE_BY_VF_MASK; + remote_in_use_pattern = ADF_VF2PF_IN_USE_BY_VF; + int_bit = ADF_PF2VF_INT; + } + + mutex_lock(lock); + + /* Check if PF2VF CSR is in use by remote function */ + val = ADF_CSR_RD(pmisc_bar_addr, pf2vf_offset); + if ((val & remote_in_use_mask) == remote_in_use_pattern) { + dev_dbg(&GET_DEV(accel_dev), + "PF2VF CSR in use by remote function\n"); + ret = -EBUSY; + goto out; + } + + /* Attempt to get ownership of PF2VF CSR */ + msg &= ~local_in_use_mask; + msg |= local_in_use_pattern; + ADF_CSR_WR(pmisc_bar_addr, pf2vf_offset, msg); + + /* Wait in case remote func also attempting to get ownership */ + msleep(ADF_IOV_MSG_COLLISION_DETECT_DELAY); + + val = ADF_CSR_RD(pmisc_bar_addr, pf2vf_offset); + if ((val & local_in_use_mask) != local_in_use_pattern) { + dev_dbg(&GET_DEV(accel_dev), + "PF2VF CSR in use by remote - collision detected\n"); + ret = -EBUSY; + goto out; + } + + /* + * This function now owns the PV2VF CSR. The IN_USE_BY pattern must + * remain in the PF2VF CSR for all writes including ACK from remote + * until this local function relinquishes the CSR. Send the message + * by interrupting the remote. + */ + ADF_CSR_WR(pmisc_bar_addr, pf2vf_offset, msg | int_bit); + + /* Wait for confirmation from remote func it received the message */ + do { + msleep(ADF_IOV_MSG_ACK_DELAY); + val = ADF_CSR_RD(pmisc_bar_addr, pf2vf_offset); + } while ((val & int_bit) && (count++ < ADF_IOV_MSG_ACK_MAX_RETRY)); + + if (val & int_bit) { + dev_dbg(&GET_DEV(accel_dev), "ACK not received from remote\n"); + val &= ~int_bit; + ret = -EIO; + } + + /* Finished with PF2VF CSR; relinquish it and leave msg in CSR */ + ADF_CSR_WR(pmisc_bar_addr, pf2vf_offset, val & ~local_in_use_mask); +out: + mutex_unlock(lock); + return ret; +} + +/** + * adf_iov_putmsg() - send PF2VF message + * @accel_dev: Pointer to acceleration device. + * @msg: Message to send + * @vf_nr: VF number to which the message will be sent + * + * Function sends a messge from the PF to a VF + * + * Return: 0 on success, error code otherwise. + */ +int adf_iov_putmsg(struct adf_accel_dev *accel_dev, u32 msg, u8 vf_nr) +{ + u32 count = 0; + int ret; + + do { + ret = __adf_iov_putmsg(accel_dev, msg, vf_nr); + if (ret) + msleep(ADF_IOV_MSG_RETRY_DELAY); + } while (ret && (count++ < ADF_IOV_MSG_MAX_RETRIES)); + + return ret; +} +EXPORT_SYMBOL_GPL(adf_iov_putmsg); + +void adf_pf2vf_notify_restarting(struct adf_accel_dev *accel_dev) +{ + struct adf_accel_vf_info *vf; + u32 msg = (ADF_PF2VF_MSGORIGIN_SYSTEM | + (ADF_PF2VF_MSGTYPE_RESTARTING << ADF_PF2VF_MSGTYPE_SHIFT)); + int i, num_vfs = pci_num_vf(accel_to_pci_dev(accel_dev)); + + for (i = 0, vf = accel_dev->pf.vf_info; i < num_vfs; i++, vf++) { + if (vf->init && adf_iov_putmsg(accel_dev, msg, i)) + dev_err(&GET_DEV(accel_dev), + "Failed to send restarting msg to VF%d\n", i); + } +} + +static int adf_vf2pf_request_version(struct adf_accel_dev *accel_dev) +{ + unsigned long timeout = msecs_to_jiffies(ADF_IOV_MSG_RESP_TIMEOUT); + struct adf_hw_device_data *hw_data = accel_dev->hw_device; + u32 msg = 0; + int ret; + + msg = ADF_VF2PF_MSGORIGIN_SYSTEM; + msg |= ADF_VF2PF_MSGTYPE_COMPAT_VER_REQ << ADF_VF2PF_MSGTYPE_SHIFT; + msg |= ADF_PFVF_COMPATIBILITY_VERSION << ADF_VF2PF_COMPAT_VER_REQ_SHIFT; + BUILD_BUG_ON(ADF_PFVF_COMPATIBILITY_VERSION > 255); + + /* Send request from VF to PF */ + ret = adf_iov_putmsg(accel_dev, msg, 0); + if (ret) { + dev_err(&GET_DEV(accel_dev), + "Failed to send Compatibility Version Request.\n"); + return ret; + } + + /* Wait for response */ + if (!wait_for_completion_timeout(&accel_dev->vf.iov_msg_completion, + timeout)) { + dev_err(&GET_DEV(accel_dev), + "IOV request/response message timeout expired\n"); + return -EIO; + } + + /* Response from PF received, check compatibility */ + switch (accel_dev->vf.compatible) { + case ADF_PF2VF_VF_COMPATIBLE: + break; + case ADF_PF2VF_VF_COMPAT_UNKNOWN: + /* VF is newer than PF and decides whether it is compatible */ + if (accel_dev->vf.pf_version >= hw_data->min_iov_compat_ver) + break; + /* fall through */ + case ADF_PF2VF_VF_INCOMPATIBLE: + dev_err(&GET_DEV(accel_dev), + "PF (vers %d) and VF (vers %d) are not compatible\n", + accel_dev->vf.pf_version, + ADF_PFVF_COMPATIBILITY_VERSION); + return -EINVAL; + default: + dev_err(&GET_DEV(accel_dev), + "Invalid response from PF; assume not compatible\n"); + return -EINVAL; + } + return ret; +} + +/** + * adf_enable_vf2pf_comms() - Function enables communication from vf to pf + * + * @accel_dev: Pointer to acceleration device virtual function. + * + * Return: 0 on success, error code otherwise. + */ +int adf_enable_vf2pf_comms(struct adf_accel_dev *accel_dev) +{ + adf_enable_pf2vf_interrupts(accel_dev); + return adf_vf2pf_request_version(accel_dev); +} +EXPORT_SYMBOL_GPL(adf_enable_vf2pf_comms); diff --git a/drivers/crypto/qat/qat_common/adf_pf2vf_msg.h b/drivers/crypto/qat/qat_common/adf_pf2vf_msg.h new file mode 100644 index 0000000..3ceaa38 --- /dev/null +++ b/drivers/crypto/qat/qat_common/adf_pf2vf_msg.h @@ -0,0 +1,144 @@ +/* + This file is provided under a dual BSD/GPLv2 license. When using or + redistributing this file, you may do so under either license. + + GPL LICENSE SUMMARY + Copyright(c) 2015 Intel Corporation. + This program is free software; you can redistribute it and/or modify + it under the terms of version 2 of the GNU General Public License as + published by the Free Software Foundation. + + This program is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + General Public License for more details. + + Contact Information: + qat-linux@intel.com + + BSD LICENSE + Copyright(c) 2015 Intel Corporation. + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in + the documentation and/or other materials provided with the + distribution. + * Neither the name of Intel Corporation nor the names of its + contributors may be used to endorse or promote products derived + from this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ +#ifndef ADF_PF2VF_MSG_H +#define ADF_PF2VF_MSG_H + +/* + * PF<->VF Messaging + * The PF has an array of 32-bit PF2VF registers, one for each VF. The + * PF can access all these registers; each VF can access only the one + * register associated with that particular VF. + * + * The register functionally is split into two parts: + * The bottom half is for PF->VF messages. In particular when the first + * bit of this register (bit 0) gets set an interrupt will be triggered + * in the respective VF. + * The top half is for VF->PF messages. In particular when the first bit + * of this half of register (bit 16) gets set an interrupt will be triggered + * in the PF. + * + * The remaining bits within this register are available to encode messages. + * and implement a collision control mechanism to prevent concurrent use of + * the PF2VF register by both the PF and VF. + * + * 31 30 29 28 27 26 25 24 23 22 21 20 19 18 17 16 + * _______________________________________________ + * | | | | | | | | | | | | | | | | | + * +-----------------------------------------------+ + * \___________________________/ \_________/ ^ ^ + * ^ ^ | | + * | | | VF2PF Int + * | | Message Origin + * | Message Type + * Message-specific Data/Reserved + * + * 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 0 + * _______________________________________________ + * | | | | | | | | | | | | | | | | | + * +-----------------------------------------------+ + * \___________________________/ \_________/ ^ ^ + * ^ ^ | | + * | | | PF2VF Int + * | | Message Origin + * | Message Type + * Message-specific Data/Reserved + * + * Message Origin (Should always be 1) + * A legacy out-of-tree QAT driver allowed for a set of messages not supported + * by this driver; these had a Msg Origin of 0 and are ignored by this driver. + * + * When a PF or VF attempts to send a message in the lower or upper 16 bits, + * respectively, the other 16 bits are written to first with a defined + * IN_USE_BY pattern as part of a collision control scheme (see adf_iov_putmsg). + */ + +#define ADF_PFVF_COMPATIBILITY_VERSION 0x1 /* PF<->VF compat */ + +/* PF->VF messages */ +#define ADF_PF2VF_INT BIT(0) +#define ADF_PF2VF_MSGORIGIN_SYSTEM BIT(1) +#define ADF_PF2VF_MSGTYPE_MASK 0x0000003C +#define ADF_PF2VF_MSGTYPE_SHIFT 2 +#define ADF_PF2VF_MSGTYPE_RESTARTING 0x01 +#define ADF_PF2VF_MSGTYPE_VERSION_RESP 0x02 +#define ADF_PF2VF_IN_USE_BY_PF 0x6AC20000 +#define ADF_PF2VF_IN_USE_BY_PF_MASK 0xFFFE0000 + +/* PF->VF Version Response */ +#define ADF_PF2VF_VERSION_RESP_VERS_MASK 0x00003FC0 +#define ADF_PF2VF_VERSION_RESP_VERS_SHIFT 6 +#define ADF_PF2VF_VERSION_RESP_RESULT_MASK 0x0000C000 +#define ADF_PF2VF_VERSION_RESP_RESULT_SHIFT 14 +#define ADF_PF2VF_VF_COMPATIBLE 1 +#define ADF_PF2VF_VF_INCOMPATIBLE 2 +#define ADF_PF2VF_VF_COMPAT_UNKNOWN 3 + +/* VF->PF messages */ +#define ADF_VF2PF_IN_USE_BY_VF 0x00006AC2 +#define ADF_VF2PF_IN_USE_BY_VF_MASK 0x0000FFFE +#define ADF_VF2PF_INT BIT(16) +#define ADF_VF2PF_MSGORIGIN_SYSTEM BIT(17) +#define ADF_VF2PF_MSGTYPE_MASK 0x003C0000 +#define ADF_VF2PF_MSGTYPE_SHIFT 18 +#define ADF_VF2PF_MSGTYPE_INIT 0x3 +#define ADF_VF2PF_MSGTYPE_SHUTDOWN 0x4 +#define ADF_VF2PF_MSGTYPE_VERSION_REQ 0x5 +#define ADF_VF2PF_MSGTYPE_COMPAT_VER_REQ 0x6 + +/* VF->PF Compatible Version Request */ +#define ADF_VF2PF_COMPAT_VER_REQ_SHIFT 22 + +/* Collision detection */ +#define ADF_IOV_MSG_COLLISION_DETECT_DELAY 10 +#define ADF_IOV_MSG_ACK_DELAY 2 +#define ADF_IOV_MSG_ACK_MAX_RETRY 100 +#define ADF_IOV_MSG_RETRY_DELAY 5 +#define ADF_IOV_MSG_MAX_RETRIES 3 +#define ADF_IOV_MSG_RESP_TIMEOUT (ADF_IOV_MSG_ACK_DELAY * \ + ADF_IOV_MSG_ACK_MAX_RETRY + \ + ADF_IOV_MSG_COLLISION_DETECT_DELAY) +#endif /* ADF_IOV_MSG_H */ diff --git a/drivers/crypto/qat/qat_common/adf_sriov.c b/drivers/crypto/qat/qat_common/adf_sriov.c new file mode 100644 index 0000000..a766fb5 --- /dev/null +++ b/drivers/crypto/qat/qat_common/adf_sriov.c @@ -0,0 +1,406 @@ +/* + This file is provided under a dual BSD/GPLv2 license. When using or + redistributing this file, you may do so under either license. + + GPL LICENSE SUMMARY + Copyright(c) 2015 Intel Corporation. + This program is free software; you can redistribute it and/or modify + it under the terms of version 2 of the GNU General Public License as + published by the Free Software Foundation. + + This program is distributed in the hope that it will be useful, but + WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + General Public License for more details. + + Contact Information: + qat-linux@intel.com + + BSD LICENSE + Copyright(c) 2015 Intel Corporation. + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + * Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in + the documentation and/or other materials provided with the + distribution. + * Neither the name of Intel Corporation nor the names of its + contributors may be used to endorse or promote products derived + from this software without specific prior written permission. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT + OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, + SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT + LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE + OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ +#include +#include +#include +#include +#include "adf_common_drv.h" +#include "adf_cfg.h" +#include "adf_pf2vf_msg.h" + +static struct workqueue_struct *pf2vf_resp_wq; + +#define ME2FUNCTION_MAP_A_OFFSET (0x3A400 + 0x190) +#define ME2FUNCTION_MAP_A_NUM_REGS 96 + +#define ME2FUNCTION_MAP_B_OFFSET (0x3A400 + 0x310) +#define ME2FUNCTION_MAP_B_NUM_REGS 12 + +#define ME2FUNCTION_MAP_REG_SIZE 4 +#define ME2FUNCTION_MAP_VALID BIT(7) + +#define READ_CSR_ME2FUNCTION_MAP_A(pmisc_bar_addr, index) \ + ADF_CSR_RD(pmisc_bar_addr, ME2FUNCTION_MAP_A_OFFSET + \ + ME2FUNCTION_MAP_REG_SIZE * index) + +#define WRITE_CSR_ME2FUNCTION_MAP_A(pmisc_bar_addr, index, value) \ + ADF_CSR_WR(pmisc_bar_addr, ME2FUNCTION_MAP_A_OFFSET + \ + ME2FUNCTION_MAP_REG_SIZE * index, value) + +#define READ_CSR_ME2FUNCTION_MAP_B(pmisc_bar_addr, index) \ + ADF_CSR_RD(pmisc_bar_addr, ME2FUNCTION_MAP_B_OFFSET + \ + ME2FUNCTION_MAP_REG_SIZE * index) + +#define WRITE_CSR_ME2FUNCTION_MAP_B(pmisc_bar_addr, index, value) \ + ADF_CSR_WR(pmisc_bar_addr, ME2FUNCTION_MAP_B_OFFSET + \ + ME2FUNCTION_MAP_REG_SIZE * index, value) + +struct adf_pf2vf_resp_data { + struct work_struct pf2vf_resp_work; + struct adf_accel_dev *accel_dev; + u32 resp; + u8 vf_nr; +}; + +static void adf_iov_send_resp(struct work_struct *work) +{ + struct adf_pf2vf_resp_data *pf2vf_resp_data = + container_of(work, struct adf_pf2vf_resp_data, pf2vf_resp_work); + + if (adf_iov_putmsg(pf2vf_resp_data->accel_dev, pf2vf_resp_data->resp, + pf2vf_resp_data->vf_nr)) { + dev_err(&GET_DEV(pf2vf_resp_data->accel_dev), + "Failed to send response\n"); + } + + kfree(pf2vf_resp_data); +} + +static void adf_vf2pf_bh_handler(void *data) +{ + struct adf_accel_vf_info *vf_info = (struct adf_accel_vf_info *)data; + struct adf_accel_dev *accel_dev = vf_info->accel_dev; + struct adf_hw_device_data *hw_data = accel_dev->hw_device; + struct adf_bar *pmisc = + &GET_BARS(accel_dev)[hw_data->get_misc_bar_id(hw_data)]; + void __iomem *pmisc_addr = pmisc->virt_addr; + u32 msg; + + /* Read message from the VF */ + msg = ADF_CSR_RD(pmisc_addr, hw_data->get_pf2vf_offset(vf_info->vf_nr)); + + if (!(msg & ADF_VF2PF_MSGORIGIN_SYSTEM)) + /* Ignore legacy non-system (non-kernel) VF2PF messages */ + goto err; + + switch ((msg & ADF_VF2PF_MSGTYPE_MASK) >> ADF_VF2PF_MSGTYPE_SHIFT) { + case ADF_VF2PF_MSGTYPE_COMPAT_VER_REQ: + { + u8 vf_compat_ver = msg >> ADF_VF2PF_COMPAT_VER_REQ_SHIFT; + struct adf_pf2vf_resp_data *pf2vf_resp_data; + u32 resp = (ADF_PF2VF_MSGORIGIN_SYSTEM | + (ADF_PF2VF_MSGTYPE_VERSION_RESP << + ADF_PF2VF_MSGTYPE_SHIFT) | + (ADF_PFVF_COMPATIBILITY_VERSION << + ADF_PF2VF_VERSION_RESP_VERS_SHIFT)); + + dev_dbg(&GET_DEV(accel_dev), + "Compatibility Version Request from VF%d vers=%u\n", + vf_info->vf_nr + 1, vf_compat_ver); + + if (vf_compat_ver < hw_data->min_iov_compat_ver) { + dev_err(&GET_DEV(accel_dev), + "VF (vers %d) incompatible with PF (vers %d)\n", + vf_compat_ver, ADF_PFVF_COMPATIBILITY_VERSION); + resp |= ADF_PF2VF_VF_INCOMPATIBLE << + ADF_PF2VF_VERSION_RESP_RESULT_SHIFT; + } else if (vf_compat_ver > ADF_PFVF_COMPATIBILITY_VERSION) { + dev_err(&GET_DEV(accel_dev), + "VF (vers %d) compat with PF (vers %d) unkn.\n", + vf_compat_ver, ADF_PFVF_COMPATIBILITY_VERSION); + resp |= ADF_PF2VF_VF_COMPAT_UNKNOWN << + ADF_PF2VF_VERSION_RESP_RESULT_SHIFT; + } else { + dev_dbg(&GET_DEV(accel_dev), + "VF (vers %d) compatible with PF (vers %d)\n", + vf_compat_ver, ADF_PFVF_COMPATIBILITY_VERSION); + resp |= ADF_PF2VF_VF_COMPATIBLE << + ADF_PF2VF_VERSION_RESP_RESULT_SHIFT; + } + + pf2vf_resp_data = kzalloc(sizeof(*pf2vf_resp_data), GFP_ATOMIC); + if (!pf2vf_resp_data) + return; + + pf2vf_resp_data->accel_dev = accel_dev; + pf2vf_resp_data->vf_nr = vf_info->vf_nr; + pf2vf_resp_data->resp = resp; + INIT_WORK(&pf2vf_resp_data->pf2vf_resp_work, adf_iov_send_resp); + queue_work(pf2vf_resp_wq, &pf2vf_resp_data->pf2vf_resp_work); + } + break; + case ADF_VF2PF_MSGTYPE_INIT: + { + dev_dbg(&GET_DEV(accel_dev), + "Init message received from VF%d 0x%x\n", + vf_info->vf_nr + 1, msg); + vf_info->init = true; + } + break; + case ADF_VF2PF_MSGTYPE_SHUTDOWN: + { + dev_dbg(&GET_DEV(accel_dev), + "Shutdown message received from VF%d 0x%x\n", + vf_info->vf_nr + 1, msg); + vf_info->init = false; + } + break; + case ADF_VF2PF_MSGTYPE_VERSION_REQ: + dev_err(&GET_DEV(accel_dev), + "Incompatible VersionRequest received from VF%d 0x%x\n", + vf_info->vf_nr + 1, msg); + break; + default: + goto err; + } + + /* To ACK, clear the VF2PFINT bit */ + msg &= ~ADF_VF2PF_INT; + ADF_CSR_WR(pmisc_addr, hw_data->get_pf2vf_offset(vf_info->vf_nr), msg); + + /* re-enable interrupt on PF from this VF */ + adf_enable_vf2pf_interrupts(accel_dev, (1 << vf_info->vf_nr)); + return; +err: + dev_err(&GET_DEV(accel_dev), "Unknown message from VF%d (0x%x);\n", + vf_info->vf_nr + 1, msg); +} + +static int adf_enable_sriov(struct adf_accel_dev *accel_dev) +{ + struct pci_dev *pdev = accel_to_pci_dev(accel_dev); + int totalvfs = pci_sriov_get_totalvfs(pdev); + struct adf_hw_device_data *hw_data = accel_dev->hw_device; + struct adf_bar *pmisc = + &GET_BARS(accel_dev)[hw_data->get_misc_bar_id(hw_data)]; + void __iomem *pmisc_addr = pmisc->virt_addr; + struct adf_accel_vf_info *vf_info; + int i, ret; + u32 reg; + + /* Workqueue for PF2VF responses */ + pf2vf_resp_wq = create_workqueue("qat_pf2vf_resp_wq"); + if (!pf2vf_resp_wq) + return -ENOMEM; + + for (i = 0, vf_info = accel_dev->pf.vf_info; i < totalvfs; + i++, vf_info++) { + /* This ptr will be populated when VFs will be created */ + vf_info->accel_dev = accel_dev; + vf_info->vf_nr = i; + + tasklet_init(&vf_info->vf2pf_bh_tasklet, + (void *)adf_vf2pf_bh_handler, + (unsigned long)vf_info); + mutex_init(&vf_info->pf2vf_lock); + ratelimit_state_init(&vf_info->vf2pf_ratelimit, + DEFAULT_RATELIMIT_INTERVAL, + DEFAULT_RATELIMIT_BURST); + } + + /* Set Valid bits in ME Thread to PCIe Function Mapping Group A */ + for (i = 0; i < ME2FUNCTION_MAP_A_NUM_REGS; i++) { + reg = READ_CSR_ME2FUNCTION_MAP_A(pmisc_addr, i); + reg |= ME2FUNCTION_MAP_VALID; + WRITE_CSR_ME2FUNCTION_MAP_A(pmisc_addr, i, reg); + } + + /* Set Valid bits in ME Thread to PCIe Function Mapping Group B */ + for (i = 0; i < ME2FUNCTION_MAP_B_NUM_REGS; i++) { + reg = READ_CSR_ME2FUNCTION_MAP_B(pmisc_addr, i); + reg |= ME2FUNCTION_MAP_VALID; + WRITE_CSR_ME2FUNCTION_MAP_B(pmisc_addr, i, reg); + } + + /* Enable VF to PF interrupts for all VFs */ + adf_enable_vf2pf_interrupts(accel_dev, GENMASK_ULL(totalvfs - 1, 0)); + + /* + * Due to the hardware design, when SR-IOV and the ring arbiter + * are enabled all the VFs supported in hardware must be enabled in + * order for all the hardware resources (i.e. bundles) to be usable. + * When SR-IOV is enabled, each of the VFs will own one bundle. + */ + ret = pci_enable_sriov(pdev, totalvfs); + if (ret) + return ret; + + return 0; +} + +/** + * adf_disable_sriov() - Disable SRIOV for the device + * @pdev: Pointer to pci device. + * + * Function disables SRIOV for the pci device. + * + * Return: 0 on success, error code otherwise. + */ +void adf_disable_sriov(struct adf_accel_dev *accel_dev) +{ + struct adf_hw_device_data *hw_data = accel_dev->hw_device; + struct adf_bar *pmisc = + &GET_BARS(accel_dev)[hw_data->get_misc_bar_id(hw_data)]; + void __iomem *pmisc_addr = pmisc->virt_addr; + int totalvfs = pci_sriov_get_totalvfs(accel_to_pci_dev(accel_dev)); + struct adf_accel_vf_info *vf; + u32 reg; + int i; + + if (!accel_dev->pf.vf_info) + return; + + adf_pf2vf_notify_restarting(accel_dev); + + pci_disable_sriov(accel_to_pci_dev(accel_dev)); + + /* Disable VF to PF interrupts */ + adf_disable_vf2pf_interrupts(accel_dev, 0xFFFFFFFF); + + /* Clear Valid bits in ME Thread to PCIe Function Mapping Group A */ + for (i = 0; i < ME2FUNCTION_MAP_A_NUM_REGS; i++) { + reg = READ_CSR_ME2FUNCTION_MAP_A(pmisc_addr, i); + reg &= ~ME2FUNCTION_MAP_VALID; + WRITE_CSR_ME2FUNCTION_MAP_A(pmisc_addr, i, reg); + } + + /* Clear Valid bits in ME Thread to PCIe Function Mapping Group B */ + for (i = 0; i < ME2FUNCTION_MAP_B_NUM_REGS; i++) { + reg = READ_CSR_ME2FUNCTION_MAP_B(pmisc_addr, i); + reg &= ~ME2FUNCTION_MAP_VALID; + WRITE_CSR_ME2FUNCTION_MAP_B(pmisc_addr, i, reg); + } + + for (i = 0, vf = accel_dev->pf.vf_info; i < totalvfs; i++, vf++) { + tasklet_disable(&vf->vf2pf_bh_tasklet); + tasklet_kill(&vf->vf2pf_bh_tasklet); + mutex_destroy(&vf->pf2vf_lock); + } + + kfree(accel_dev->pf.vf_info); + accel_dev->pf.vf_info = NULL; + + if (pf2vf_resp_wq) { + destroy_workqueue(pf2vf_resp_wq); + pf2vf_resp_wq = NULL; + } +} +EXPORT_SYMBOL_GPL(adf_disable_sriov); + +/** + * adf_sriov_configure() - Enable SRIOV for the device + * @pdev: Pointer to pci device. + * + * Function enables SRIOV for the pci device. + * + * Return: 0 on success, error code otherwise. + */ +int adf_sriov_configure(struct pci_dev *pdev, int numvfs) +{ + struct adf_accel_dev *accel_dev = adf_devmgr_pci_to_accel_dev(pdev); + int totalvfs = pci_sriov_get_totalvfs(pdev); + unsigned long val; + int ret; + + if (!accel_dev) { + dev_err(&pdev->dev, "Failed to find accel_dev\n"); + return -EFAULT; + } + + if (!iommu_present(&pci_bus_type)) { + dev_err(&pdev->dev, + "IOMMU must be enabled for SR-IOV to work\n"); + return -EINVAL; + } + + if (accel_dev->pf.vf_info) { + dev_info(&pdev->dev, "Already enabled for this device\n"); + return -EINVAL; + } + + if (adf_dev_started(accel_dev)) { + if (adf_devmgr_in_reset(accel_dev) || + adf_dev_in_use(accel_dev)) { + dev_err(&GET_DEV(accel_dev), "Device busy\n"); + return -EBUSY; + } + + if (adf_dev_stop(accel_dev)) { + dev_err(&GET_DEV(accel_dev), + "Failed to stop qat_dev%d\n", + accel_dev->accel_id); + return -EFAULT; + } + + adf_dev_shutdown(accel_dev); + } + + if (adf_cfg_section_add(accel_dev, ADF_KERNEL_SEC)) + return -EFAULT; + val = 0; + if (adf_cfg_add_key_value_param(accel_dev, ADF_KERNEL_SEC, + ADF_NUM_CY, (void *)&val, ADF_DEC)) + return -EFAULT; + + set_bit(ADF_STATUS_CONFIGURED, &accel_dev->status); + + /* Allocate memory for VF info structs */ + accel_dev->pf.vf_info = kcalloc(totalvfs, + sizeof(struct adf_accel_vf_info), + GFP_KERNEL); + if (!accel_dev->pf.vf_info) + return -ENOMEM; + + if (adf_dev_init(accel_dev)) { + dev_err(&GET_DEV(accel_dev), "Failed to init qat_dev%d\n", + accel_dev->accel_id); + return -EFAULT; + } + + if (adf_dev_start(accel_dev)) { + dev_err(&GET_DEV(accel_dev), "Failed to start qat_dev%d\n", + accel_dev->accel_id); + return -EFAULT; + } + + ret = adf_enable_sriov(accel_dev); + if (ret) + return ret; + + return numvfs; +} +EXPORT_SYMBOL_GPL(adf_sriov_configure); diff --git a/drivers/crypto/qat/qat_common/qat_crypto.c b/drivers/crypto/qat/qat_common/qat_crypto.c index 359ae94..07c2f9f 100644 --- a/drivers/crypto/qat/qat_common/qat_crypto.c +++ b/drivers/crypto/qat/qat_common/qat_crypto.c @@ -103,9 +103,11 @@ struct qat_crypto_instance *qat_crypto_get_instance_node(int node) list_for_each(itr, adf_devmgr_get_head()) { accel_dev = list_entry(itr, struct adf_accel_dev, list); + if ((node == dev_to_node(&GET_DEV(accel_dev)) || dev_to_node(&GET_DEV(accel_dev)) < 0) && - adf_dev_started(accel_dev)) + adf_dev_started(accel_dev) && + !list_empty(&accel_dev->crypto_list)) break; accel_dev = NULL; } diff --git a/drivers/crypto/qat/qat_dh895xcc/adf_dh895xcc_hw_data.c b/drivers/crypto/qat/qat_dh895xcc/adf_dh895xcc_hw_data.c index a3b1be8..ff54257 100644 --- a/drivers/crypto/qat/qat_dh895xcc/adf_dh895xcc_hw_data.c +++ b/drivers/crypto/qat/qat_dh895xcc/adf_dh895xcc_hw_data.c @@ -45,6 +45,7 @@ OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include +#include #include #include "adf_dh895xcc_hw_data.h" #include "adf_drv.h" @@ -161,6 +162,16 @@ void adf_get_arbiter_mapping(struct adf_accel_dev *accel_dev, } } +static uint32_t get_pf2vf_offset(uint32_t i) +{ + return ADF_DH895XCC_PF2VF_OFFSET(i); +} + +static uint32_t get_vintmsk_offset(uint32_t i) +{ + return ADF_DH895XCC_VINTMSK_OFFSET(i); +} + static void adf_enable_error_correction(struct adf_accel_dev *accel_dev) { struct adf_hw_device_data *hw_device = accel_dev->hw_device; @@ -197,11 +208,17 @@ static void adf_enable_ints(struct adf_accel_dev *accel_dev) /* Enable bundle and misc interrupts */ ADF_CSR_WR(addr, ADF_DH895XCC_SMIAPF0_MASK_OFFSET, - ADF_DH895XCC_SMIA0_MASK); + accel_dev->pf.vf_info ? 0 : + GENMASK_ULL(GET_MAX_BANKS(accel_dev) - 1, 0)); ADF_CSR_WR(addr, ADF_DH895XCC_SMIAPF1_MASK_OFFSET, ADF_DH895XCC_SMIA1_MASK); } +static int adf_pf_enable_vf2pf_comms(struct adf_accel_dev *accel_dev) +{ + return 0; +} + void adf_init_hw_data_dh895xcc(struct adf_hw_device_data *hw_data) { hw_data->dev_class = &dh895xcc_class; @@ -221,17 +238,22 @@ void adf_init_hw_data_dh895xcc(struct adf_hw_device_data *hw_data) hw_data->get_num_aes = get_num_aes; hw_data->get_etr_bar_id = get_etr_bar_id; hw_data->get_misc_bar_id = get_misc_bar_id; + hw_data->get_pf2vf_offset = get_pf2vf_offset; + hw_data->get_vintmsk_offset = get_vintmsk_offset; hw_data->get_sram_bar_id = get_sram_bar_id; hw_data->get_sku = get_sku; hw_data->fw_name = ADF_DH895XCC_FW; hw_data->fw_mmp_name = ADF_DH895XCC_MMP; hw_data->init_admin_comms = adf_init_admin_comms; hw_data->exit_admin_comms = adf_exit_admin_comms; + hw_data->disable_iov = adf_disable_sriov; hw_data->send_admin_init = adf_send_admin_init; hw_data->init_arb = adf_init_arb; hw_data->exit_arb = adf_exit_arb; hw_data->get_arb_mapping = adf_get_arbiter_mapping; hw_data->enable_ints = adf_enable_ints; + hw_data->enable_vf2pf_comms = adf_pf_enable_vf2pf_comms; + hw_data->min_iov_compat_ver = ADF_PFVF_COMPATIBILITY_VERSION; } void adf_clean_hw_data_dh895xcc(struct adf_hw_device_data *hw_data) diff --git a/drivers/crypto/qat/qat_dh895xcc/adf_dh895xcc_hw_data.h b/drivers/crypto/qat/qat_dh895xcc/adf_dh895xcc_hw_data.h index d410afb..88dffb2 100644 --- a/drivers/crypto/qat/qat_dh895xcc/adf_dh895xcc_hw_data.h +++ b/drivers/crypto/qat/qat_dh895xcc/adf_dh895xcc_hw_data.h @@ -80,6 +80,10 @@ #define ADF_DH895XCC_CERRSSMSH(i) (i * 0x4000 + 0x10) #define ADF_DH895XCC_ERRSSMSH_EN BIT(3) +#define ADF_DH895XCC_ERRSOU3 (0x3A000 + 0x00C) +#define ADF_DH895XCC_ERRSOU5 (0x3A000 + 0x0D8) +#define ADF_DH895XCC_PF2VF_OFFSET(i) (0x3A000 + 0x280 + ((i) * 0x04)) +#define ADF_DH895XCC_VINTMSK_OFFSET(i) (0x3A000 + 0x200 + ((i) * 0x04)) /* FW names */ #define ADF_DH895XCC_FW "qat_895xcc.bin" #define ADF_DH895XCC_MMP "qat_mmp.bin" diff --git a/drivers/crypto/qat/qat_dh895xcc/adf_drv.c b/drivers/crypto/qat/qat_dh895xcc/adf_drv.c index 1837c32..f8dd14f 100644 --- a/drivers/crypto/qat/qat_dh895xcc/adf_drv.c +++ b/drivers/crypto/qat/qat_dh895xcc/adf_drv.c @@ -82,16 +82,21 @@ static struct pci_driver adf_driver = { .id_table = adf_pci_tbl, .name = adf_driver_name, .probe = adf_probe, - .remove = adf_remove + .remove = adf_remove, + .sriov_configure = adf_sriov_configure, }; +static void adf_cleanup_pci_dev(struct adf_accel_dev *accel_dev) +{ + pci_release_regions(accel_dev->accel_pci_dev.pci_dev); + pci_disable_device(accel_dev->accel_pci_dev.pci_dev); +} + static void adf_cleanup_accel(struct adf_accel_dev *accel_dev) { struct adf_accel_pci *accel_pci_dev = &accel_dev->accel_pci_dev; int i; - adf_dev_shutdown(accel_dev); - for (i = 0; i < ADF_PCI_MAX_BARS; i++) { struct adf_bar *bar = &accel_pci_dev->pci_bars[i]; @@ -108,13 +113,11 @@ static void adf_cleanup_accel(struct adf_accel_dev *accel_dev) break; } kfree(accel_dev->hw_device); + accel_dev->hw_device = NULL; } adf_cfg_dev_remove(accel_dev); debugfs_remove(accel_dev->debugfs_dir); - adf_devmgr_rm_dev(accel_dev); - pci_release_regions(accel_pci_dev->pci_dev); - pci_disable_device(accel_pci_dev->pci_dev); - kfree(accel_dev); + adf_devmgr_rm_dev(accel_dev, NULL); } static int adf_dev_configure(struct adf_accel_dev *accel_dev) @@ -205,7 +208,7 @@ static int adf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) struct adf_hw_device_data *hw_data; char name[ADF_DEVICE_NAME_LENGTH]; unsigned int i, bar_nr; - int ret; + int ret, bar_mask; switch (ent->device) { case ADF_DH895XCC_PCI_DEVICE_ID: @@ -229,10 +232,12 @@ static int adf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) return -ENOMEM; INIT_LIST_HEAD(&accel_dev->crypto_list); + accel_pci_dev = &accel_dev->accel_pci_dev; + accel_pci_dev->pci_dev = pdev; /* Add accel device to accel table. * This should be called before adf_cleanup_accel is called */ - if (adf_devmgr_add_dev(accel_dev)) { + if (adf_devmgr_add_dev(accel_dev, NULL)) { dev_err(&pdev->dev, "Failed to add new accelerator device.\n"); kfree(accel_dev); return -EFAULT; @@ -255,7 +260,6 @@ static int adf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) default: return -ENODEV; } - accel_pci_dev = &accel_dev->accel_pci_dev; pci_read_config_byte(pdev, PCI_REVISION_ID, &accel_pci_dev->revid); pci_read_config_dword(pdev, ADF_DH895XCC_FUSECTL_OFFSET, &hw_data->fuses); @@ -264,7 +268,6 @@ static int adf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) hw_data->accel_mask = hw_data->get_accel_mask(hw_data->fuses); hw_data->ae_mask = hw_data->get_ae_mask(hw_data->fuses); accel_pci_dev->sku = hw_data->get_sku(hw_data); - accel_pci_dev->pci_dev = pdev; /* If the device has no acceleration engines then ignore it. */ if (!hw_data->accel_mask || !hw_data->ae_mask || ((~hw_data->ae_mask) & 0x01)) { @@ -274,11 +277,14 @@ static int adf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) } /* Create dev top level debugfs entry */ - snprintf(name, sizeof(name), "%s%s_dev%d", ADF_DEVICE_NAME_PREFIX, - hw_data->dev_class->name, hw_data->instance_id); + snprintf(name, sizeof(name), "%s%s_%02x:%02d.%02d", + ADF_DEVICE_NAME_PREFIX, hw_data->dev_class->name, + pdev->bus->number, PCI_SLOT(pdev->devfn), + PCI_FUNC(pdev->devfn)); + accel_dev->debugfs_dir = debugfs_create_dir(name, NULL); if (!accel_dev->debugfs_dir) { - dev_err(&pdev->dev, "Could not create debugfs dir\n"); + dev_err(&pdev->dev, "Could not create debugfs dir %s\n", name); ret = -EINVAL; goto out_err; } @@ -301,7 +307,7 @@ static int adf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) if ((pci_set_dma_mask(pdev, DMA_BIT_MASK(32)))) { dev_err(&pdev->dev, "No usable DMA configuration\n"); ret = -EFAULT; - goto out_err; + goto out_err_disable; } else { pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)); } @@ -312,7 +318,7 @@ static int adf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) if (pci_request_regions(pdev, adf_driver_name)) { ret = -EFAULT; - goto out_err; + goto out_err_disable; } /* Read accelerator capabilities mask */ @@ -320,19 +326,21 @@ static int adf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) &hw_data->accel_capabilities_mask); /* Find and map all the device's BARS */ - for (i = 0; i < ADF_PCI_MAX_BARS; i++) { - struct adf_bar *bar = &accel_pci_dev->pci_bars[i]; + i = 0; + bar_mask = pci_select_bars(pdev, IORESOURCE_MEM); + for_each_set_bit(bar_nr, (const unsigned long *)&bar_mask, + ADF_PCI_MAX_BARS * 2) { + struct adf_bar *bar = &accel_pci_dev->pci_bars[i++]; - bar_nr = i * 2; bar->base_addr = pci_resource_start(pdev, bar_nr); if (!bar->base_addr) break; bar->size = pci_resource_len(pdev, bar_nr); bar->virt_addr = pci_iomap(accel_pci_dev->pci_dev, bar_nr, 0); if (!bar->virt_addr) { - dev_err(&pdev->dev, "Failed to map BAR %d\n", i); + dev_err(&pdev->dev, "Failed to map BAR %d\n", bar_nr); ret = -EFAULT; - goto out_err; + goto out_err_free_reg; } } pci_set_master(pdev); @@ -340,32 +348,40 @@ static int adf_probe(struct pci_dev *pdev, const struct pci_device_id *ent) if (adf_enable_aer(accel_dev, &adf_driver)) { dev_err(&pdev->dev, "Failed to enable aer\n"); ret = -EFAULT; - goto out_err; + goto out_err_free_reg; } if (pci_save_state(pdev)) { dev_err(&pdev->dev, "Failed to save pci state\n"); ret = -ENOMEM; - goto out_err; + goto out_err_free_reg; } ret = adf_dev_configure(accel_dev); if (ret) - goto out_err; + goto out_err_free_reg; ret = adf_dev_init(accel_dev); if (ret) - goto out_err; + goto out_err_dev_shutdown; ret = adf_dev_start(accel_dev); - if (ret) { - adf_dev_stop(accel_dev); - goto out_err; - } + if (ret) + goto out_err_dev_stop; - return 0; + return ret; + +out_err_dev_stop: + adf_dev_stop(accel_dev); +out_err_dev_shutdown: + adf_dev_shutdown(accel_dev); +out_err_free_reg: + pci_release_regions(accel_pci_dev->pci_dev); +out_err_disable: + pci_disable_device(accel_pci_dev->pci_dev); out_err: adf_cleanup_accel(accel_dev); + kfree(accel_dev); return ret; } @@ -379,8 +395,12 @@ static void adf_remove(struct pci_dev *pdev) } if (adf_dev_stop(accel_dev)) dev_err(&GET_DEV(accel_dev), "Failed to stop QAT accel dev\n"); + + adf_dev_shutdown(accel_dev); adf_disable_aer(accel_dev); adf_cleanup_accel(accel_dev); + adf_cleanup_pci_dev(accel_dev); + kfree(accel_dev); } static int __init adfdrv_init(void) diff --git a/drivers/crypto/qat/qat_dh895xcc/adf_isr.c b/drivers/crypto/qat/qat_dh895xcc/adf_isr.c index 0d03c10..5570f78 100644 --- a/drivers/crypto/qat/qat_dh895xcc/adf_isr.c +++ b/drivers/crypto/qat/qat_dh895xcc/adf_isr.c @@ -59,21 +59,30 @@ #include #include #include "adf_drv.h" +#include "adf_dh895xcc_hw_data.h" static int adf_enable_msix(struct adf_accel_dev *accel_dev) { struct adf_accel_pci *pci_dev_info = &accel_dev->accel_pci_dev; struct adf_hw_device_data *hw_data = accel_dev->hw_device; - uint32_t msix_num_entries = hw_data->num_banks + 1; - int i; - - for (i = 0; i < msix_num_entries; i++) - pci_dev_info->msix_entries.entries[i].entry = i; + u32 msix_num_entries = 1; + + /* If SR-IOV is disabled, add entries for each bank */ + if (!accel_dev->pf.vf_info) { + int i; + + msix_num_entries += hw_data->num_banks; + for (i = 0; i < msix_num_entries; i++) + pci_dev_info->msix_entries.entries[i].entry = i; + } else { + pci_dev_info->msix_entries.entries[0].entry = + hw_data->num_banks; + } if (pci_enable_msix_exact(pci_dev_info->pci_dev, pci_dev_info->msix_entries.entries, msix_num_entries)) { - dev_err(&GET_DEV(accel_dev), "Failed to enable MSIX IRQ\n"); + dev_err(&GET_DEV(accel_dev), "Failed to enable MSI-X IRQ(s)\n"); return -EFAULT; } return 0; @@ -97,9 +106,58 @@ static irqreturn_t adf_msix_isr_ae(int irq, void *dev_ptr) { struct adf_accel_dev *accel_dev = dev_ptr; - dev_info(&GET_DEV(accel_dev), "qat_dev%d spurious AE interrupt\n", - accel_dev->accel_id); - return IRQ_HANDLED; +#ifdef CONFIG_PCI_IOV + /* If SR-IOV is enabled (vf_info is non-NULL), check for VF->PF ints */ + if (accel_dev->pf.vf_info) { + void __iomem *pmisc_bar_addr = + (&GET_BARS(accel_dev)[ADF_DH895XCC_PMISC_BAR])->virt_addr; + u32 vf_mask; + + /* Get the interrupt sources triggered by VFs */ + vf_mask = ((ADF_CSR_RD(pmisc_bar_addr, ADF_DH895XCC_ERRSOU5) & + 0x0000FFFF) << 16) | + ((ADF_CSR_RD(pmisc_bar_addr, ADF_DH895XCC_ERRSOU3) & + 0x01FFFE00) >> 9); + + if (vf_mask) { + struct adf_accel_vf_info *vf_info; + bool irq_handled = false; + int i; + + /* Disable VF2PF interrupts for VFs with pending ints */ + adf_disable_vf2pf_interrupts(accel_dev, vf_mask); + + /* + * Schedule tasklets to handle VF2PF interrupt BHs + * unless the VF is malicious and is attempting to + * flood the host OS with VF2PF interrupts. + */ + for_each_set_bit(i, (const unsigned long *)&vf_mask, + (sizeof(vf_mask) * BITS_PER_BYTE)) { + vf_info = accel_dev->pf.vf_info + i; + + if (!__ratelimit(&vf_info->vf2pf_ratelimit)) { + dev_info(&GET_DEV(accel_dev), + "Too many ints from VF%d\n", + vf_info->vf_nr + 1); + continue; + } + + /* Tasklet will re-enable ints from this VF */ + tasklet_hi_schedule(&vf_info->vf2pf_bh_tasklet); + irq_handled = true; + } + + if (irq_handled) + return IRQ_HANDLED; + } + } +#endif /* CONFIG_PCI_IOV */ + + dev_dbg(&GET_DEV(accel_dev), "qat_dev%d spurious AE interrupt\n", + accel_dev->accel_id); + + return IRQ_NONE; } static int adf_request_irqs(struct adf_accel_dev *accel_dev) @@ -108,28 +166,32 @@ static int adf_request_irqs(struct adf_accel_dev *accel_dev) struct adf_hw_device_data *hw_data = accel_dev->hw_device; struct msix_entry *msixe = pci_dev_info->msix_entries.entries; struct adf_etr_data *etr_data = accel_dev->transport; - int ret, i; + int ret, i = 0; char *name; - /* Request msix irq for all banks */ - for (i = 0; i < hw_data->num_banks; i++) { - struct adf_etr_bank_data *bank = &etr_data->banks[i]; - unsigned int cpu, cpus = num_online_cpus(); - - name = *(pci_dev_info->msix_entries.names + i); - snprintf(name, ADF_MAX_MSIX_VECTOR_NAME, - "qat%d-bundle%d", accel_dev->accel_id, i); - ret = request_irq(msixe[i].vector, - adf_msix_isr_bundle, 0, name, bank); - if (ret) { - dev_err(&GET_DEV(accel_dev), - "failed to enable irq %d for %s\n", - msixe[i].vector, name); - return ret; + /* Request msix irq for all banks unless SR-IOV enabled */ + if (!accel_dev->pf.vf_info) { + for (i = 0; i < hw_data->num_banks; i++) { + struct adf_etr_bank_data *bank = &etr_data->banks[i]; + unsigned int cpu, cpus = num_online_cpus(); + + name = *(pci_dev_info->msix_entries.names + i); + snprintf(name, ADF_MAX_MSIX_VECTOR_NAME, + "qat%d-bundle%d", accel_dev->accel_id, i); + ret = request_irq(msixe[i].vector, + adf_msix_isr_bundle, 0, name, bank); + if (ret) { + dev_err(&GET_DEV(accel_dev), + "failed to enable irq %d for %s\n", + msixe[i].vector, name); + return ret; + } + + cpu = ((accel_dev->accel_id * hw_data->num_banks) + + i) % cpus; + irq_set_affinity_hint(msixe[i].vector, + get_cpu_mask(cpu)); } - - cpu = ((accel_dev->accel_id * hw_data->num_banks) + i) % cpus; - irq_set_affinity_hint(msixe[i].vector, get_cpu_mask(cpu)); } /* Request msix irq for AE */ @@ -152,11 +214,13 @@ static void adf_free_irqs(struct adf_accel_dev *accel_dev) struct adf_hw_device_data *hw_data = accel_dev->hw_device; struct msix_entry *msixe = pci_dev_info->msix_entries.entries; struct adf_etr_data *etr_data = accel_dev->transport; - int i; + int i = 0; - for (i = 0; i < hw_data->num_banks; i++) { - irq_set_affinity_hint(msixe[i].vector, NULL); - free_irq(msixe[i].vector, &etr_data->banks[i]); + if (pci_dev_info->msix_entries.num_entries > 1) { + for (i = 0; i < hw_data->num_banks; i++) { + irq_set_affinity_hint(msixe[i].vector, NULL); + free_irq(msixe[i].vector, &etr_data->banks[i]); + } } irq_set_affinity_hint(msixe[i].vector, NULL); free_irq(msixe[i].vector, accel_dev); @@ -168,7 +232,11 @@ static int adf_isr_alloc_msix_entry_table(struct adf_accel_dev *accel_dev) char **names; struct msix_entry *entries; struct adf_hw_device_data *hw_data = accel_dev->hw_device; - uint32_t msix_num_entries = hw_data->num_banks + 1; + u32 msix_num_entries = 1; + + /* If SR-IOV is disabled (vf_info is NULL), add entries for each bank */ + if (!accel_dev->pf.vf_info) + msix_num_entries += hw_data->num_banks; entries = kzalloc_node(msix_num_entries * sizeof(*entries), GFP_KERNEL, dev_to_node(&GET_DEV(accel_dev))); @@ -185,6 +253,7 @@ static int adf_isr_alloc_msix_entry_table(struct adf_accel_dev *accel_dev) if (!(*(names + i))) goto err; } + accel_dev->accel_pci_dev.msix_entries.num_entries = msix_num_entries; accel_dev->accel_pci_dev.msix_entries.entries = entries; accel_dev->accel_pci_dev.msix_entries.names = names; return 0; @@ -198,13 +267,11 @@ err: static void adf_isr_free_msix_entry_table(struct adf_accel_dev *accel_dev) { - struct adf_hw_device_data *hw_data = accel_dev->hw_device; - uint32_t msix_num_entries = hw_data->num_banks + 1; char **names = accel_dev->accel_pci_dev.msix_entries.names; int i; kfree(accel_dev->accel_pci_dev.msix_entries.entries); - for (i = 0; i < msix_num_entries; i++) + for (i = 0; i < accel_dev->accel_pci_dev.msix_entries.num_entries; i++) kfree(*(names + i)); kfree(names); }