From patchwork Fri Mar 5 13:38:34 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Nithin Dabilpuram X-Patchwork-Id: 88535 X-Patchwork-Delegate: jerinj@marvell.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 490AAA0547; Fri, 5 Mar 2021 14:41:03 +0100 (CET) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id AFDAD22A379; Fri, 5 Mar 2021 14:39:58 +0100 (CET) Received: from mx0b-0016f401.pphosted.com (mx0b-0016f401.pphosted.com [67.231.156.173]) by mails.dpdk.org (Postfix) with ESMTP id 5FF2A22A379 for ; Fri, 5 Mar 2021 14:39:57 +0100 (CET) Received: from pps.filterd (m0045851.ppops.net [127.0.0.1]) by mx0b-0016f401.pphosted.com (8.16.0.43/8.16.0.43) with SMTP id 125Ddmfh008314 for ; Fri, 5 Mar 2021 05:39:56 -0800 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=marvell.com; h=from : to : cc : subject : date : message-id : in-reply-to : references : mime-version : content-type; s=pfpt0220; bh=IKaAQpTWS0ADbpjRa5hb3vQ7jWTZrKMfbtMO5+2E9I8=; b=Sy3w5oE5Z0NrxwcQ57/x8kBa+PdrOnvB1NQg4JaTip2r80gke/yYVmNFRjrwCZgUGNde 7LkPFvyLahPkZq2G/I0VfR4NAYKVuXNZBmw16heb3CN/SfuaBehq+LWAvs+20QnhqHEc eFKmz2b3ToxerOvYgk3lwflW/ug2oGVwU6ppBe5Ez4xpZ4Q0KIMg7ac7atb5/BttZIHI zuQLSWM5XglNZegrnEuIrPnuggZktYQtsbipIH7hN+62Ao9z0dA9T7443OjaKL7hPMr5 6V8IQRjjdyOZaU1+60nku+8P3eEu2BJ/l7eaz5SIFO+oYK7DfYNT4D2tZAf+V4SXnfBJ qQ== Received: from dc5-exch02.marvell.com ([199.233.59.182]) by mx0b-0016f401.pphosted.com with ESMTP id 370p7p0db9-1 (version=TLSv1.2 cipher=ECDHE-RSA-AES256-SHA384 bits=256 verify=NOT) for ; Fri, 05 Mar 2021 05:39:56 -0800 Received: from SC-EXCH01.marvell.com (10.93.176.81) by DC5-EXCH02.marvell.com (10.69.176.39) with Microsoft SMTP Server (TLS) id 15.0.1497.2; Fri, 5 Mar 2021 05:39:54 -0800 Received: from DC5-EXCH01.marvell.com (10.69.176.38) by SC-EXCH01.marvell.com (10.93.176.81) with Microsoft SMTP Server (TLS) id 15.0.1497.2; Fri, 5 Mar 2021 05:39:53 -0800 Received: from maili.marvell.com (10.69.176.80) by DC5-EXCH01.marvell.com (10.69.176.38) with Microsoft SMTP Server id 15.0.1497.2 via Frontend Transport; Fri, 5 Mar 2021 05:39:53 -0800 Received: from hyd1588t430.marvell.com (unknown [10.29.52.204]) by maili.marvell.com (Postfix) with ESMTP id 7597F3F7040; Fri, 5 Mar 2021 05:39:51 -0800 (PST) From: Nithin Dabilpuram To: CC: , , , , , , Date: Fri, 5 Mar 2021 19:08:34 +0530 Message-ID: <20210305133918.8005-9-ndabilpuram@marvell.com> X-Mailer: git-send-email 2.8.4 In-Reply-To: <20210305133918.8005-1-ndabilpuram@marvell.com> References: <20210305133918.8005-1-ndabilpuram@marvell.com> MIME-Version: 1.0 X-Proofpoint-Virus-Version: vendor=fsecure engine=2.50.10434:6.0.369, 18.0.761 definitions=2021-03-05_08:2021-03-03, 2021-03-05 signatures=0 Subject: [dpdk-dev] [PATCH 08/52] common/cnxk: add VF support to base device class X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" From: Jerin Jacob Add VF specific handling such as BAR4 setup, forwarding VF mbox messages to AF and vice-versa, VF FLR handling etc. Signed-off-by: Jerin Jacob --- drivers/common/cnxk/roc_dev.c | 857 ++++++++++++++++++++++++++++++++++++- drivers/common/cnxk/roc_dev_priv.h | 42 ++ 2 files changed, 879 insertions(+), 20 deletions(-) diff --git a/drivers/common/cnxk/roc_dev.c b/drivers/common/cnxk/roc_dev.c index 4ef15df..1fe1371 100644 --- a/drivers/common/cnxk/roc_dev.c +++ b/drivers/common/cnxk/roc_dev.c @@ -17,6 +17,337 @@ /* Single Root I/O Virtualization */ #define ROC_PCI_SRIOV_TOTAL_VF 0x0e /* Total VFs */ +static void * +mbox_mem_map(off_t off, size_t size) +{ + void *va = MAP_FAILED; + int mem_fd; + + if (size <= 0 || !off) { + plt_err("Invalid mbox area off 0x%lx size %lu", off, size); + goto error; + } + + mem_fd = open("/dev/mem", O_RDWR); + if (mem_fd < 0) + goto error; + + va = plt_mmap(NULL, size, PLT_PROT_READ | PLT_PROT_WRITE, + PLT_MAP_SHARED, mem_fd, off); + close(mem_fd); + + if (va == MAP_FAILED) + plt_err("Failed to mmap sz=0x%zx, fd=%d, off=%jd", size, mem_fd, + (intmax_t)off); +error: + return va; +} + +static void +mbox_mem_unmap(void *va, size_t size) +{ + if (va) + munmap(va, size); +} + +static int +pf_af_sync_msg(struct dev *dev, struct mbox_msghdr **rsp) +{ + uint32_t timeout = 0, sleep = 1; + struct mbox *mbox = dev->mbox; + struct mbox_dev *mdev = &mbox->dev[0]; + + volatile uint64_t int_status; + struct mbox_msghdr *msghdr; + uint64_t off; + int rc = 0; + + /* We need to disable PF interrupts. We are in timer interrupt */ + plt_write64(~0ull, dev->bar2 + RVU_PF_INT_ENA_W1C); + + /* Send message */ + mbox_msg_send(mbox, 0); + + do { + plt_delay_ms(sleep); + timeout += sleep; + if (timeout >= mbox->rsp_tmo) { + plt_err("Message timeout: %dms", mbox->rsp_tmo); + rc = -EIO; + break; + } + int_status = plt_read64(dev->bar2 + RVU_PF_INT); + } while ((int_status & 0x1) != 0x1); + + /* Clear */ + plt_write64(int_status, dev->bar2 + RVU_PF_INT); + + /* Enable interrupts */ + plt_write64(~0ull, dev->bar2 + RVU_PF_INT_ENA_W1S); + + if (rc == 0) { + /* Get message */ + off = mbox->rx_start + + PLT_ALIGN(sizeof(struct mbox_hdr), MBOX_MSG_ALIGN); + msghdr = (struct mbox_msghdr *)((uintptr_t)mdev->mbase + off); + if (rsp) + *rsp = msghdr; + rc = msghdr->rc; + } + + return rc; +} + +static int +af_pf_wait_msg(struct dev *dev, uint16_t vf, int num_msg) +{ + uint32_t timeout = 0, sleep = 1; + struct mbox *mbox = dev->mbox; + struct mbox_dev *mdev = &mbox->dev[0]; + volatile uint64_t int_status; + struct mbox_hdr *req_hdr; + struct mbox_msghdr *msg; + struct mbox_msghdr *rsp; + uint64_t offset; + size_t size; + int i; + + /* We need to disable PF interrupts. We are in timer interrupt */ + plt_write64(~0ull, dev->bar2 + RVU_PF_INT_ENA_W1C); + + /* Send message */ + mbox_msg_send(mbox, 0); + + do { + plt_delay_ms(sleep); + timeout++; + if (timeout >= mbox->rsp_tmo) { + plt_err("Routed messages %d timeout: %dms", num_msg, + mbox->rsp_tmo); + break; + } + int_status = plt_read64(dev->bar2 + RVU_PF_INT); + } while ((int_status & 0x1) != 0x1); + + /* Clear */ + plt_write64(~0ull, dev->bar2 + RVU_PF_INT); + + /* Enable interrupts */ + plt_write64(~0ull, dev->bar2 + RVU_PF_INT_ENA_W1S); + + plt_spinlock_lock(&mdev->mbox_lock); + + req_hdr = (struct mbox_hdr *)((uintptr_t)mdev->mbase + mbox->rx_start); + if (req_hdr->num_msgs != num_msg) + plt_err("Routed messages: %d received: %d", num_msg, + req_hdr->num_msgs); + + /* Get messages from mbox */ + offset = mbox->rx_start + + PLT_ALIGN(sizeof(struct mbox_hdr), MBOX_MSG_ALIGN); + for (i = 0; i < req_hdr->num_msgs; i++) { + msg = (struct mbox_msghdr *)((uintptr_t)mdev->mbase + offset); + size = mbox->rx_start + msg->next_msgoff - offset; + + /* Reserve PF/VF mbox message */ + size = PLT_ALIGN(size, MBOX_MSG_ALIGN); + rsp = mbox_alloc_msg(&dev->mbox_vfpf, vf, size); + mbox_rsp_init(msg->id, rsp); + + /* Copy message from AF<->PF mbox to PF<->VF mbox */ + mbox_memcpy((uint8_t *)rsp + sizeof(struct mbox_msghdr), + (uint8_t *)msg + sizeof(struct mbox_msghdr), + size - sizeof(struct mbox_msghdr)); + + /* Set status and sender pf_func data */ + rsp->rc = msg->rc; + rsp->pcifunc = msg->pcifunc; + + offset = mbox->rx_start + msg->next_msgoff; + } + plt_spinlock_unlock(&mdev->mbox_lock); + + return req_hdr->num_msgs; +} + +static int +vf_pf_process_msgs(struct dev *dev, uint16_t vf) +{ + struct mbox *mbox = &dev->mbox_vfpf; + struct mbox_dev *mdev = &mbox->dev[vf]; + struct mbox_hdr *req_hdr; + struct mbox_msghdr *msg; + int offset, routed = 0; + size_t size; + uint16_t i; + + req_hdr = (struct mbox_hdr *)((uintptr_t)mdev->mbase + mbox->rx_start); + if (!req_hdr->num_msgs) + return 0; + + offset = mbox->rx_start + PLT_ALIGN(sizeof(*req_hdr), MBOX_MSG_ALIGN); + + for (i = 0; i < req_hdr->num_msgs; i++) { + msg = (struct mbox_msghdr *)((uintptr_t)mdev->mbase + offset); + size = mbox->rx_start + msg->next_msgoff - offset; + + /* RVU_PF_FUNC_S */ + msg->pcifunc = dev_pf_func(dev->pf, vf); + + if (msg->id == MBOX_MSG_READY) { + struct ready_msg_rsp *rsp; + uint16_t max_bits = sizeof(dev->active_vfs[0]) * 8; + + /* Handle READY message in PF */ + dev->active_vfs[vf / max_bits] |= + BIT_ULL(vf % max_bits); + rsp = (struct ready_msg_rsp *)mbox_alloc_msg( + mbox, vf, sizeof(*rsp)); + mbox_rsp_init(msg->id, rsp); + + /* PF/VF function ID */ + rsp->hdr.pcifunc = msg->pcifunc; + rsp->hdr.rc = 0; + } else { + struct mbox_msghdr *af_req; + /* Reserve AF/PF mbox message */ + size = PLT_ALIGN(size, MBOX_MSG_ALIGN); + af_req = mbox_alloc_msg(dev->mbox, 0, size); + if (af_req == NULL) + return -ENOSPC; + mbox_req_init(msg->id, af_req); + + /* Copy message from VF<->PF mbox to PF<->AF mbox */ + mbox_memcpy((uint8_t *)af_req + + sizeof(struct mbox_msghdr), + (uint8_t *)msg + sizeof(struct mbox_msghdr), + size - sizeof(struct mbox_msghdr)); + af_req->pcifunc = msg->pcifunc; + routed++; + } + offset = mbox->rx_start + msg->next_msgoff; + } + + if (routed > 0) { + plt_base_dbg("pf:%d routed %d messages from vf:%d to AF", + dev->pf, routed, vf); + af_pf_wait_msg(dev, vf, routed); + mbox_reset(dev->mbox, 0); + } + + /* Send mbox responses to VF */ + if (mdev->num_msgs) { + plt_base_dbg("pf:%d reply %d messages to vf:%d", dev->pf, + mdev->num_msgs, vf); + mbox_msg_send(mbox, vf); + } + + return i; +} + +static int +vf_pf_process_up_msgs(struct dev *dev, uint16_t vf) +{ + struct mbox *mbox = &dev->mbox_vfpf_up; + struct mbox_dev *mdev = &mbox->dev[vf]; + struct mbox_hdr *req_hdr; + struct mbox_msghdr *msg; + int msgs_acked = 0; + int offset; + uint16_t i; + + req_hdr = (struct mbox_hdr *)((uintptr_t)mdev->mbase + mbox->rx_start); + if (req_hdr->num_msgs == 0) + return 0; + + offset = mbox->rx_start + PLT_ALIGN(sizeof(*req_hdr), MBOX_MSG_ALIGN); + + for (i = 0; i < req_hdr->num_msgs; i++) { + msg = (struct mbox_msghdr *)((uintptr_t)mdev->mbase + offset); + + msgs_acked++; + /* RVU_PF_FUNC_S */ + msg->pcifunc = dev_pf_func(dev->pf, vf); + + switch (msg->id) { + case MBOX_MSG_CGX_LINK_EVENT: + plt_base_dbg("PF: Msg 0x%x (%s) fn:0x%x (pf:%d,vf:%d)", + msg->id, mbox_id2name(msg->id), + msg->pcifunc, dev_get_pf(msg->pcifunc), + dev_get_vf(msg->pcifunc)); + break; + case MBOX_MSG_CGX_PTP_RX_INFO: + plt_base_dbg("PF: Msg 0x%x (%s) fn:0x%x (pf:%d,vf:%d)", + msg->id, mbox_id2name(msg->id), + msg->pcifunc, dev_get_pf(msg->pcifunc), + dev_get_vf(msg->pcifunc)); + break; + default: + plt_err("Not handled UP msg 0x%x (%s) func:0x%x", + msg->id, mbox_id2name(msg->id), msg->pcifunc); + } + offset = mbox->rx_start + msg->next_msgoff; + } + mbox_reset(mbox, vf); + mdev->msgs_acked = msgs_acked; + plt_wmb(); + + return i; +} + +static void +roc_vf_pf_mbox_handle_msg(void *param) +{ + uint16_t vf, max_vf, max_bits; + struct dev *dev = param; + + max_bits = sizeof(dev->intr.bits[0]) * sizeof(uint64_t); + max_vf = max_bits * MAX_VFPF_DWORD_BITS; + + for (vf = 0; vf < max_vf; vf++) { + if (dev->intr.bits[vf / max_bits] & BIT_ULL(vf % max_bits)) { + plt_base_dbg("Process vf:%d request (pf:%d, vf:%d)", vf, + dev->pf, dev->vf); + vf_pf_process_msgs(dev, vf); + /* UP messages */ + vf_pf_process_up_msgs(dev, vf); + dev->intr.bits[vf / max_bits] &= + ~(BIT_ULL(vf % max_bits)); + } + } + dev->timer_set = 0; +} + +static void +roc_vf_pf_mbox_irq(void *param) +{ + struct dev *dev = param; + bool alarm_set = false; + uint64_t intr; + int vfpf; + + for (vfpf = 0; vfpf < MAX_VFPF_DWORD_BITS; ++vfpf) { + intr = plt_read64(dev->bar2 + RVU_PF_VFPF_MBOX_INTX(vfpf)); + if (!intr) + continue; + + plt_base_dbg("vfpf: %d intr: 0x%" PRIx64 " (pf:%d, vf:%d)", + vfpf, intr, dev->pf, dev->vf); + + /* Save and clear intr bits */ + dev->intr.bits[vfpf] |= intr; + plt_write64(intr, dev->bar2 + RVU_PF_VFPF_MBOX_INTX(vfpf)); + alarm_set = true; + } + + if (!dev->timer_set && alarm_set) { + dev->timer_set = 1; + /* Start timer to handle messages */ + plt_alarm_set(VF_PF_MBOX_TIMER_MS, roc_vf_pf_mbox_handle_msg, + dev); + } +} + static void process_msgs(struct dev *dev, struct mbox *mbox) { @@ -62,6 +393,112 @@ process_msgs(struct dev *dev, struct mbox *mbox) plt_wmb(); } +/* Copies the message received from AF and sends it to VF */ +static void +pf_vf_mbox_send_up_msg(struct dev *dev, void *rec_msg) +{ + uint16_t max_bits = sizeof(dev->active_vfs[0]) * sizeof(uint64_t); + struct mbox *vf_mbox = &dev->mbox_vfpf_up; + struct msg_req *msg = rec_msg; + struct mbox_msghdr *vf_msg; + uint16_t vf; + size_t size; + + size = PLT_ALIGN(mbox_id2size(msg->hdr.id), MBOX_MSG_ALIGN); + /* Send UP message to all VF's */ + for (vf = 0; vf < vf_mbox->ndevs; vf++) { + /* VF active */ + if (!(dev->active_vfs[vf / max_bits] & (BIT_ULL(vf)))) + continue; + + plt_base_dbg("(%s) size: %zx to VF: %d", + mbox_id2name(msg->hdr.id), size, vf); + + /* Reserve PF/VF mbox message */ + vf_msg = mbox_alloc_msg(vf_mbox, vf, size); + if (!vf_msg) { + plt_err("Failed to alloc VF%d UP message", vf); + continue; + } + mbox_req_init(msg->hdr.id, vf_msg); + + /* + * Copy message from AF<->PF UP mbox + * to PF<->VF UP mbox + */ + mbox_memcpy((uint8_t *)vf_msg + sizeof(struct mbox_msghdr), + (uint8_t *)msg + sizeof(struct mbox_msghdr), + size - sizeof(struct mbox_msghdr)); + + vf_msg->rc = msg->hdr.rc; + /* Set PF to be a sender */ + vf_msg->pcifunc = dev->pf_func; + + /* Send to VF */ + mbox_msg_send(vf_mbox, vf); + } +} + +static int +mbox_up_handler_cgx_link_event(struct dev *dev, struct cgx_link_info_msg *msg, + struct msg_rsp *rsp) +{ + struct cgx_link_user_info *linfo = &msg->link_info; + void *roc_nix = dev->roc_nix; + + plt_base_dbg("pf:%d/vf:%d NIC Link %s --> 0x%x (%s) from: pf:%d/vf:%d", + dev_get_pf(dev->pf_func), dev_get_vf(dev->pf_func), + linfo->link_up ? "UP" : "DOWN", msg->hdr.id, + mbox_id2name(msg->hdr.id), dev_get_pf(msg->hdr.pcifunc), + dev_get_vf(msg->hdr.pcifunc)); + + /* PF gets link notification from AF */ + if (dev_get_pf(msg->hdr.pcifunc) == 0) { + if (dev->ops && dev->ops->link_status_update) + dev->ops->link_status_update(roc_nix, linfo); + + /* Forward the same message as received from AF to VF */ + pf_vf_mbox_send_up_msg(dev, msg); + } else { + /* VF gets link up notification */ + if (dev->ops && dev->ops->link_status_update) + dev->ops->link_status_update(roc_nix, linfo); + } + + rsp->hdr.rc = 0; + return 0; +} + +static int +mbox_up_handler_cgx_ptp_rx_info(struct dev *dev, + struct cgx_ptp_rx_info_msg *msg, + struct msg_rsp *rsp) +{ + void *roc_nix = dev->roc_nix; + + plt_base_dbg("pf:%d/vf:%d PTP mode %s --> 0x%x (%s) from: pf:%d/vf:%d", + dev_get_pf(dev->pf_func), dev_get_vf(dev->pf_func), + msg->ptp_en ? "ENABLED" : "DISABLED", msg->hdr.id, + mbox_id2name(msg->hdr.id), dev_get_pf(msg->hdr.pcifunc), + dev_get_vf(msg->hdr.pcifunc)); + + /* PF gets PTP notification from AF */ + if (dev_get_pf(msg->hdr.pcifunc) == 0) { + if (dev->ops && dev->ops->ptp_info_update) + dev->ops->ptp_info_update(roc_nix, msg->ptp_en); + + /* Forward the same message as received from AF to VF */ + pf_vf_mbox_send_up_msg(dev, msg); + } else { + /* VF gets PTP notification */ + if (dev->ops && dev->ops->ptp_info_update) + dev->ops->ptp_info_update(roc_nix, msg->ptp_en); + } + + rsp->hdr.rc = 0; + return 0; +} + static int mbox_process_msgs_up(struct dev *dev, struct mbox_msghdr *req) { @@ -73,6 +510,24 @@ mbox_process_msgs_up(struct dev *dev, struct mbox_msghdr *req) default: reply_invalid_msg(&dev->mbox_up, 0, 0, req->id); break; +#define M(_name, _id, _fn_name, _req_type, _rsp_type) \ + case _id: { \ + struct _rsp_type *rsp; \ + int err; \ + rsp = (struct _rsp_type *)mbox_alloc_msg( \ + &dev->mbox_up, 0, sizeof(struct _rsp_type)); \ + if (!rsp) \ + return -ENOMEM; \ + rsp->hdr.id = _id; \ + rsp->hdr.sig = MBOX_RSP_SIG; \ + rsp->hdr.pcifunc = dev->pf_func; \ + rsp->hdr.rc = 0; \ + err = mbox_up_handler_##_fn_name(dev, (struct _req_type *)req, \ + rsp); \ + return err; \ + } + MBOX_UP_CGX_MESSAGES +#undef M } return -ENODEV; @@ -111,6 +566,26 @@ process_msgs_up(struct dev *dev, struct mbox *mbox) } static void +roc_pf_vf_mbox_irq(void *param) +{ + struct dev *dev = param; + uint64_t intr; + + intr = plt_read64(dev->bar2 + RVU_VF_INT); + if (intr == 0) + plt_base_dbg("Proceeding to check mbox UP messages if any"); + + plt_write64(intr, dev->bar2 + RVU_VF_INT); + plt_base_dbg("Irq 0x%" PRIx64 "(pf:%d,vf:%d)", intr, dev->pf, dev->vf); + + /* First process all configuration messages */ + process_msgs(dev, dev->mbox); + + /* Process Uplink messages */ + process_msgs_up(dev, &dev->mbox_up); +} + +static void roc_af_pf_mbox_irq(void *param) { struct dev *dev = param; @@ -121,7 +596,7 @@ roc_af_pf_mbox_irq(void *param) plt_base_dbg("Proceeding to check mbox UP messages if any"); plt_write64(intr, dev->bar2 + RVU_PF_INT); - plt_base_dbg("Irq 0x%" PRIx64 "(pf:%d)", intr, dev->pf); + plt_base_dbg("Irq 0x%" PRIx64 "(pf:%d,vf:%d)", intr, dev->pf, dev->vf); /* First process all configuration messages */ process_msgs(dev, dev->mbox); @@ -134,10 +609,33 @@ static int mbox_register_pf_irq(struct plt_pci_device *pci_dev, struct dev *dev) { struct plt_intr_handle *intr_handle = &pci_dev->intr_handle; - int rc; + int i, rc; + + /* HW clear irq */ + for (i = 0; i < MAX_VFPF_DWORD_BITS; ++i) + plt_write64(~0ull, + dev->bar2 + RVU_PF_VFPF_MBOX_INT_ENA_W1CX(i)); plt_write64(~0ull, dev->bar2 + RVU_PF_INT_ENA_W1C); + dev->timer_set = 0; + + /* MBOX interrupt for VF(0...63) <-> PF */ + rc = dev_irq_register(intr_handle, roc_vf_pf_mbox_irq, dev, + RVU_PF_INT_VEC_VFPF_MBOX0); + + if (rc) { + plt_err("Fail to register PF(VF0-63) mbox irq"); + return rc; + } + /* MBOX interrupt for VF(64...128) <-> PF */ + rc = dev_irq_register(intr_handle, roc_vf_pf_mbox_irq, dev, + RVU_PF_INT_VEC_VFPF_MBOX1); + + if (rc) { + plt_err("Fail to register PF(VF64-128) mbox irq"); + return rc; + } /* MBOX interrupt AF <-> PF */ rc = dev_irq_register(intr_handle, roc_af_pf_mbox_irq, dev, RVU_PF_INT_VEC_AFPF_MBOX); @@ -146,6 +644,11 @@ mbox_register_pf_irq(struct plt_pci_device *pci_dev, struct dev *dev) return rc; } + /* HW enable intr */ + for (i = 0; i < MAX_VFPF_DWORD_BITS; ++i) + plt_write64(~0ull, + dev->bar2 + RVU_PF_VFPF_MBOX_INT_ENA_W1SX(i)); + plt_write64(~0ull, dev->bar2 + RVU_PF_INT); plt_write64(~0ull, dev->bar2 + RVU_PF_INT_ENA_W1S); @@ -153,27 +656,263 @@ mbox_register_pf_irq(struct plt_pci_device *pci_dev, struct dev *dev) } static int +mbox_register_vf_irq(struct plt_pci_device *pci_dev, struct dev *dev) +{ + struct plt_intr_handle *intr_handle = &pci_dev->intr_handle; + int rc; + + /* Clear irq */ + plt_write64(~0ull, dev->bar2 + RVU_VF_INT_ENA_W1C); + + /* MBOX interrupt PF <-> VF */ + rc = dev_irq_register(intr_handle, roc_pf_vf_mbox_irq, dev, + RVU_VF_INT_VEC_MBOX); + if (rc) { + plt_err("Fail to register PF<->VF mbox irq"); + return rc; + } + + /* HW enable intr */ + plt_write64(~0ull, dev->bar2 + RVU_VF_INT); + plt_write64(~0ull, dev->bar2 + RVU_VF_INT_ENA_W1S); + + return rc; +} + +static int mbox_register_irq(struct plt_pci_device *pci_dev, struct dev *dev) { - return mbox_register_pf_irq(pci_dev, dev); + if (dev_is_vf(dev)) + return mbox_register_vf_irq(pci_dev, dev); + else + return mbox_register_pf_irq(pci_dev, dev); } static void mbox_unregister_pf_irq(struct plt_pci_device *pci_dev, struct dev *dev) { struct plt_intr_handle *intr_handle = &pci_dev->intr_handle; + int i; + + /* HW clear irq */ + for (i = 0; i < MAX_VFPF_DWORD_BITS; ++i) + plt_write64(~0ull, + dev->bar2 + RVU_PF_VFPF_MBOX_INT_ENA_W1CX(i)); plt_write64(~0ull, dev->bar2 + RVU_PF_INT_ENA_W1C); + dev->timer_set = 0; + + plt_alarm_cancel(roc_vf_pf_mbox_handle_msg, dev); + + /* Unregister the interrupt handler for each vectors */ + /* MBOX interrupt for VF(0...63) <-> PF */ + dev_irq_unregister(intr_handle, roc_vf_pf_mbox_irq, dev, + RVU_PF_INT_VEC_VFPF_MBOX0); + + /* MBOX interrupt for VF(64...128) <-> PF */ + dev_irq_unregister(intr_handle, roc_vf_pf_mbox_irq, dev, + RVU_PF_INT_VEC_VFPF_MBOX1); + /* MBOX interrupt AF <-> PF */ dev_irq_unregister(intr_handle, roc_af_pf_mbox_irq, dev, RVU_PF_INT_VEC_AFPF_MBOX); } static void +mbox_unregister_vf_irq(struct plt_pci_device *pci_dev, struct dev *dev) +{ + struct plt_intr_handle *intr_handle = &pci_dev->intr_handle; + + /* Clear irq */ + plt_write64(~0ull, dev->bar2 + RVU_VF_INT_ENA_W1C); + + /* Unregister the interrupt handler */ + dev_irq_unregister(intr_handle, roc_pf_vf_mbox_irq, dev, + RVU_VF_INT_VEC_MBOX); +} + +static void mbox_unregister_irq(struct plt_pci_device *pci_dev, struct dev *dev) { - mbox_unregister_pf_irq(pci_dev, dev); + if (dev_is_vf(dev)) + mbox_unregister_vf_irq(pci_dev, dev); + else + mbox_unregister_pf_irq(pci_dev, dev); +} + +static int +vf_flr_send_msg(struct dev *dev, uint16_t vf) +{ + struct mbox *mbox = dev->mbox; + struct msg_req *req; + int rc; + + req = mbox_alloc_msg_vf_flr(mbox); + if (req == NULL) + return -ENOSPC; + /* Overwrite pcifunc to indicate VF */ + req->hdr.pcifunc = dev_pf_func(dev->pf, vf); + + /* Sync message in interrupt context */ + rc = pf_af_sync_msg(dev, NULL); + if (rc) + plt_err("Failed to send VF FLR mbox msg, rc=%d", rc); + + return rc; +} + +static void +roc_pf_vf_flr_irq(void *param) +{ + struct dev *dev = (struct dev *)param; + uint16_t max_vf = 64, vf; + uintptr_t bar2; + uint64_t intr; + int i; + + max_vf = (dev->maxvf > 0) ? dev->maxvf : 64; + bar2 = dev->bar2; + + plt_base_dbg("FLR VF interrupt: max_vf: %d", max_vf); + + for (i = 0; i < MAX_VFPF_DWORD_BITS; ++i) { + intr = plt_read64(bar2 + RVU_PF_VFFLR_INTX(i)); + if (!intr) + continue; + + for (vf = 0; vf < max_vf; vf++) { + if (!(intr & (1ULL << vf))) + continue; + + plt_base_dbg("FLR: i :%d intr: 0x%" PRIx64 ", vf-%d", i, + intr, (64 * i + vf)); + /* Clear interrupt */ + plt_write64(BIT_ULL(vf), bar2 + RVU_PF_VFFLR_INTX(i)); + /* Disable the interrupt */ + plt_write64(BIT_ULL(vf), + bar2 + RVU_PF_VFFLR_INT_ENA_W1CX(i)); + /* Inform AF about VF reset */ + vf_flr_send_msg(dev, vf); + + /* Signal FLR finish */ + plt_write64(BIT_ULL(vf), bar2 + RVU_PF_VFTRPENDX(i)); + /* Enable interrupt */ + plt_write64(~0ull, bar2 + RVU_PF_VFFLR_INT_ENA_W1SX(i)); + } + } +} + +static int +vf_flr_unregister_irqs(struct plt_pci_device *pci_dev, struct dev *dev) +{ + struct plt_intr_handle *intr_handle = &pci_dev->intr_handle; + int i; + + plt_base_dbg("Unregister VF FLR interrupts for %s", pci_dev->name); + + /* HW clear irq */ + for (i = 0; i < MAX_VFPF_DWORD_BITS; i++) + plt_write64(~0ull, dev->bar2 + RVU_PF_VFFLR_INT_ENA_W1CX(i)); + + dev_irq_unregister(intr_handle, roc_pf_vf_flr_irq, dev, + RVU_PF_INT_VEC_VFFLR0); + + dev_irq_unregister(intr_handle, roc_pf_vf_flr_irq, dev, + RVU_PF_INT_VEC_VFFLR1); + + return 0; +} + +static int +vf_flr_register_irqs(struct plt_pci_device *pci_dev, struct dev *dev) +{ + struct plt_intr_handle *handle = &pci_dev->intr_handle; + int i, rc; + + plt_base_dbg("Register VF FLR interrupts for %s", pci_dev->name); + + rc = dev_irq_register(handle, roc_pf_vf_flr_irq, dev, + RVU_PF_INT_VEC_VFFLR0); + if (rc) + plt_err("Failed to init RVU_PF_INT_VEC_VFFLR0 rc=%d", rc); + + rc = dev_irq_register(handle, roc_pf_vf_flr_irq, dev, + RVU_PF_INT_VEC_VFFLR1); + if (rc) + plt_err("Failed to init RVU_PF_INT_VEC_VFFLR1 rc=%d", rc); + + /* Enable HW interrupt */ + for (i = 0; i < MAX_VFPF_DWORD_BITS; ++i) { + plt_write64(~0ull, dev->bar2 + RVU_PF_VFFLR_INTX(i)); + plt_write64(~0ull, dev->bar2 + RVU_PF_VFTRPENDX(i)); + plt_write64(~0ull, dev->bar2 + RVU_PF_VFFLR_INT_ENA_W1SX(i)); + } + return 0; +} + +int +dev_active_vfs(struct dev *dev) +{ + int i, count = 0; + + for (i = 0; i < MAX_VFPF_DWORD_BITS; i++) + count += __builtin_popcount(dev->active_vfs[i]); + + return count; +} + +static void +dev_vf_hwcap_update(struct plt_pci_device *pci_dev, struct dev *dev) +{ + switch (pci_dev->id.device_id) { + case PCI_DEVID_CNXK_RVU_PF: + break; + case PCI_DEVID_CNXK_RVU_SSO_TIM_VF: + case PCI_DEVID_CNXK_RVU_NPA_VF: + case PCI_DEVID_CNXK_RVU_AF_VF: + case PCI_DEVID_CNXK_RVU_VF: + case PCI_DEVID_CNXK_RVU_SDP_VF: + dev->hwcap |= DEV_HWCAP_F_VF; + break; + } +} + +static uintptr_t +dev_vf_mbase_get(struct plt_pci_device *pci_dev, struct dev *dev) +{ + void *vf_mbase = NULL; + uintptr_t pa; + + if (dev_is_vf(dev)) + return 0; + + /* For CN10K onwards, it is just after PF MBOX */ + if (!roc_model_is_cn9k()) + return dev->bar4 + MBOX_SIZE; + + pa = plt_read64(dev->bar2 + RVU_PF_VF_BAR4_ADDR); + if (!pa) { + plt_err("Invalid VF mbox base pa"); + return pa; + } + + vf_mbase = mbox_mem_map(pa, MBOX_SIZE * pci_dev->max_vfs); + if (vf_mbase == MAP_FAILED) { + plt_err("Failed to mmap vf mbase at pa 0x%lx, rc=%d", pa, + errno); + return 0; + } + return (uintptr_t)vf_mbase; +} + +static void +dev_vf_mbase_put(struct plt_pci_device *pci_dev, uintptr_t vf_mbase) +{ + if (!vf_mbase || !pci_dev->max_vfs || !roc_model_is_cn9k()) + return; + + mbox_mem_unmap((void *)vf_mbase, MBOX_SIZE * pci_dev->max_vfs); } static uint16_t @@ -213,7 +952,6 @@ dev_setup_shared_lmt_region(struct mbox *mbox) static int dev_lmt_setup(struct plt_pci_device *pci_dev, struct dev *dev) { - uint64_t bar4_mbox_sz = MBOX_SIZE; struct idev_cfg *idev; int rc; @@ -237,19 +975,34 @@ dev_lmt_setup(struct plt_pci_device *pci_dev, struct dev *dev) dev->pf_func, rc); } - /* PF BAR4 should always be sufficient enough to - * hold PF-AF MBOX + PF-VF MBOX + LMT lines. - */ - if (pci_dev->mem_resource[4].len < - (bar4_mbox_sz + (RVU_LMT_LINE_MAX * RVU_LMT_SZ))) { - plt_err("Not enough bar4 space for lmt lines and mbox"); - return -EFAULT; + if (dev_is_vf(dev)) { + /* VF BAR4 should always be sufficient enough to + * hold LMT lines. + */ + if (pci_dev->mem_resource[4].len < + (RVU_LMT_LINE_MAX * RVU_LMT_SZ)) { + plt_err("Not enough bar4 space for lmt lines"); + return -EFAULT; + } + + dev->lmt_base = dev->bar4; + } else { + uint64_t bar4_mbox_sz = MBOX_SIZE; + + /* PF BAR4 should always be sufficient enough to + * hold PF-AF MBOX + PF-VF MBOX + LMT lines. + */ + if (pci_dev->mem_resource[4].len < + (bar4_mbox_sz + (RVU_LMT_LINE_MAX * RVU_LMT_SZ))) { + plt_err("Not enough bar4 space for lmt lines and mbox"); + return -EFAULT; + } + + /* LMT base is just after total VF MBOX area */ + bar4_mbox_sz += (MBOX_SIZE * dev_pf_total_vfs(pci_dev)); + dev->lmt_base = dev->bar4 + bar4_mbox_sz; } - /* LMT base is just after total VF MBOX area */ - bar4_mbox_sz += (MBOX_SIZE * dev_pf_total_vfs(pci_dev)); - dev->lmt_base = dev->bar4 + bar4_mbox_sz; - /* Base LMT address should be chosen from only those pci funcs which * participate in LMT shared mode. */ @@ -270,6 +1023,7 @@ dev_init(struct dev *dev, struct plt_pci_device *pci_dev) { int direction, up_direction, rc; uintptr_t bar2, bar4, mbox; + uintptr_t vf_mbase = 0; uint64_t intr_offset; bar2 = (uintptr_t)pci_dev->mem_resource[2].addr; @@ -293,13 +1047,23 @@ dev_init(struct dev *dev, struct plt_pci_device *pci_dev) goto error; } + dev->maxvf = pci_dev->max_vfs; dev->bar2 = bar2; dev->bar4 = bar4; + dev_vf_hwcap_update(pci_dev, dev); - mbox = bar4; - direction = MBOX_DIR_PFAF; - up_direction = MBOX_DIR_PFAF_UP; - intr_offset = RVU_PF_INT; + if (dev_is_vf(dev)) { + mbox = (roc_model_is_cn9k() ? + bar4 : (bar2 + RVU_VF_MBOX_REGION)); + direction = MBOX_DIR_VFPF; + up_direction = MBOX_DIR_VFPF_UP; + intr_offset = RVU_VF_INT; + } else { + mbox = bar4; + direction = MBOX_DIR_PFAF; + up_direction = MBOX_DIR_PFAF_UP; + intr_offset = RVU_PF_INT; + } /* Initialize the local mbox */ rc = mbox_init(&dev->mbox_local, mbox, bar2, direction, 1, intr_offset); @@ -322,7 +1086,43 @@ dev_init(struct dev *dev, struct plt_pci_device *pci_dev) goto mbox_unregister; dev->pf = dev_get_pf(dev->pf_func); + dev->vf = dev_get_vf(dev->pf_func); + memset(&dev->active_vfs, 0, sizeof(dev->active_vfs)); + /* Allocate memory for device ops */ + dev->ops = plt_zmalloc(sizeof(struct dev_ops), 0); + if (dev->ops == NULL) { + rc = -ENOMEM; + goto mbox_unregister; + } + + /* Found VF devices in a PF device */ + if (pci_dev->max_vfs > 0) { + /* Remap mbox area for all vf's */ + vf_mbase = dev_vf_mbase_get(pci_dev, dev); + if (!vf_mbase) { + rc = -ENODEV; + goto mbox_unregister; + } + /* Init mbox object */ + rc = mbox_init(&dev->mbox_vfpf, vf_mbase, bar2, MBOX_DIR_PFVF, + pci_dev->max_vfs, intr_offset); + if (rc) + goto iounmap; + + /* PF -> VF UP messages */ + rc = mbox_init(&dev->mbox_vfpf_up, vf_mbase, bar2, + MBOX_DIR_PFVF_UP, pci_dev->max_vfs, intr_offset); + if (rc) + goto iounmap; + } + + /* Register VF-FLR irq handlers */ + if (!dev_is_vf(dev)) { + rc = vf_flr_register_irqs(pci_dev, dev); + if (rc) + goto iounmap; + } dev->mbox_active = 1; /* Setup LMT line base */ @@ -332,8 +1132,11 @@ dev_init(struct dev *dev, struct plt_pci_device *pci_dev) return rc; iounmap: + dev_vf_mbase_put(pci_dev, vf_mbase); mbox_unregister: mbox_unregister_irq(pci_dev, dev); + if (dev->ops) + plt_free(dev->ops); mbox_fini: mbox_fini(dev->mbox); mbox_fini(&dev->mbox_up); @@ -349,6 +1152,20 @@ dev_fini(struct dev *dev, struct plt_pci_device *pci_dev) mbox_unregister_irq(pci_dev, dev); + if (!dev_is_vf(dev)) + vf_flr_unregister_irqs(pci_dev, dev); + /* Release PF - VF */ + mbox = &dev->mbox_vfpf; + if (mbox->hwbase && mbox->dev) + dev_vf_mbase_put(pci_dev, mbox->hwbase); + + if (dev->ops) + plt_free(dev->ops); + + mbox_fini(mbox); + mbox = &dev->mbox_vfpf_up; + mbox_fini(mbox); + /* Release PF - AF */ mbox = dev->mbox; mbox_fini(mbox); diff --git a/drivers/common/cnxk/roc_dev_priv.h b/drivers/common/cnxk/roc_dev_priv.h index 655c58b..0996ec4 100644 --- a/drivers/common/cnxk/roc_dev_priv.h +++ b/drivers/common/cnxk/roc_dev_priv.h @@ -5,12 +5,38 @@ #ifndef _ROC_DEV_PRIV_H #define _ROC_DEV_PRIV_H +#define DEV_HWCAP_F_VF BIT_ULL(0) /* VF device */ + #define RVU_PFVF_PF_SHIFT 10 #define RVU_PFVF_PF_MASK 0x3F #define RVU_PFVF_FUNC_SHIFT 0 #define RVU_PFVF_FUNC_MASK 0x3FF +#define RVU_MAX_VF 64 /* RVU_PF_VFPF_MBOX_INT(0..1) */ #define RVU_MAX_INT_RETRY 3 +/* PF/VF message handling timer */ +#define VF_PF_MBOX_TIMER_MS (20 * 1000) + +typedef struct { +/* 128 devices translate to two 64 bits dwords */ +#define MAX_VFPF_DWORD_BITS 2 + uint64_t bits[MAX_VFPF_DWORD_BITS]; +} dev_intr_t; + +/* Link status update callback */ +typedef void (*link_info_t)(void *roc_nix, + struct cgx_link_user_info *link); + +/* PTP info callback */ +typedef int (*ptp_info_t)(void *roc_nix, bool enable); + +struct dev_ops { + link_info_t link_status_update; + ptp_info_t ptp_info_update; +}; + +#define dev_is_vf(dev) ((dev)->hwcap & DEV_HWCAP_F_VF) + static inline int dev_get_vf(uint16_t pf_func) { @@ -29,18 +55,33 @@ dev_pf_func(int pf, int vf) return (pf << RVU_PFVF_PF_SHIFT) | ((vf << RVU_PFVF_FUNC_SHIFT) + 1); } +static inline int +dev_is_afvf(uint16_t pf_func) +{ + return !(pf_func & ~RVU_PFVF_FUNC_MASK); +} + struct dev { uint16_t pf; + int16_t vf; uint16_t pf_func; uint8_t mbox_active; bool drv_inited; + uint64_t active_vfs[MAX_VFPF_DWORD_BITS]; uintptr_t bar2; uintptr_t bar4; uintptr_t lmt_base; struct mbox mbox_local; struct mbox mbox_up; + struct mbox mbox_vfpf; + struct mbox mbox_vfpf_up; + dev_intr_t intr; + int timer_set; /* ~0 : no alarm handling */ uint64_t hwcap; struct mbox *mbox; + uint16_t maxvf; + struct dev_ops *ops; + void *roc_nix; bool disable_shared_lmt; /* false(default): shared lmt mode enabled */ } __plt_cache_aligned; @@ -49,6 +90,7 @@ extern uint16_t dev_sclk_freq; int dev_init(struct dev *dev, struct plt_pci_device *pci_dev); int dev_fini(struct dev *dev, struct plt_pci_device *pci_dev); +int dev_active_vfs(struct dev *dev); int dev_irq_register(struct plt_intr_handle *intr_handle, plt_intr_callback_fn cb, void *data, unsigned int vec);