From patchwork Tue Jan 2 15:06:49 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Oded Gabbay X-Patchwork-Id: 184381 Return-Path: Delivered-To: ouuuleilei@gmail.com Received: by 2002:a05:7301:6f82:b0:100:9c79:88ff with SMTP id tb2csp4498851dyb; Tue, 2 Jan 2024 07:07:20 -0800 (PST) X-Google-Smtp-Source: AGHT+IEpzmyVEkFW7eDnhH/QS7OLwXZz6OWXSjNpLolNlPpu+YrSNlH0qIU8kKQB6K3gjLKfVn71 X-Received: by 2002:a05:6a21:7989:b0:195:2f17:9884 with SMTP id bh9-20020a056a21798900b001952f179884mr23540110pzc.20.1704208040198; Tue, 02 Jan 2024 07:07:20 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1704208040; cv=none; d=google.com; s=arc-20160816; b=wDibHUkPkjQ4HPTJeexuTRkzaroWSP5gcoNlxXcm6OtTuR/yDimf00E89ousAhexgR gV9aAZnOypFuKmjDV7UMLUGO8OLP+F5HCbVUoKhknqTJ1INdg3M9+eg28o5VM/S9mZ6G PK9xsH6KKEG1iU95J4TQFmoKDd5eLqo927swmPo3fAwFzqtB9RrLuoSfiCNwmexCOIke s0cFb/3nLBhTXZ4fvlycuCIz8sHtMv1TVIP6NNXDB/nJxdqSPG51uROmWzbKn+fZ9x/+ I4D0WJBsB+z4EHsv6kr4eJLY3DvLSRFFjSgl/PmKgU/qBJsffIN45Q+wcC3qMb61xJSe gArw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=content-transfer-encoding:mime-version:list-unsubscribe :list-subscribe:list-id:precedence:message-id:date:subject:cc:to :from:dkim-signature; bh=ZSfP2UlJLnQNFqju7YDr20QMtQ7dZMS55SDFaIw9zEU=; fh=FP+ffrafBoziKtAld7u50Ql/WsbhgFISxH3NOU3Jvlo=; b=hSxGanbamdhOpBes37P0Eo7TIlBXV73Ey8wD3moknmCJ4wDI9r4Sc6Lw4NDq6oSuZP AhVJeVdb0GrGWVQ4sQ9H9RYZGO+c2VjFCIPRUZsESEoLcVomPrkNPcf1y1pmX4iFuTOn AsSkGZ35fZZldn8Vx1y0WP8tnd5wABAWm/itI+436NJO9F665SSohdZcqatiO8JL+w/+ +j3IjsqzOx1W2f7Cor6uMFDxfPwCupXEFFhgJMvKtxG6utRCoMnRJ5fODengg3G3nxke 8BXomzb5QfXCCEDsD/TUvnks4gKIAL4/YLdqxgEV0VbO89rTeuJ2SvoJOfrfDkI9fCDr DkGg== ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=mUVzo72I; spf=pass (google.com: domain of linux-kernel+bounces-14526-ouuuleilei=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14526-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from sv.mirrors.kernel.org (sv.mirrors.kernel.org. [2604:1380:45e3:2400::1]) by mx.google.com with ESMTPS id z16-20020a63e550000000b00573f94e8b83si20018596pgj.265.2024.01.02.07.07.20 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 02 Jan 2024 07:07:20 -0800 (PST) Received-SPF: pass (google.com: domain of linux-kernel+bounces-14526-ouuuleilei=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1; Authentication-Results: mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=mUVzo72I; spf=pass (google.com: domain of linux-kernel+bounces-14526-ouuuleilei=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14526-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from smtp.subspace.kernel.org (wormhole.subspace.kernel.org [52.25.139.140]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by sv.mirrors.kernel.org (Postfix) with ESMTPS id EFCC6283B29 for ; Tue, 2 Jan 2024 15:07:19 +0000 (UTC) Received: from localhost.localdomain (localhost.localdomain [127.0.0.1]) by smtp.subspace.kernel.org (Postfix) with ESMTP id 8705514F8B; Tue, 2 Jan 2024 15:07:03 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="mUVzo72I" X-Original-To: linux-kernel@vger.kernel.org Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 0588314F63 for ; Tue, 2 Jan 2024 15:07:00 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 7A9A8C433C7; Tue, 2 Jan 2024 15:06:59 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1704208020; bh=G/yeMzSU+JFZOgE81eouKJsysPfDu7kxpb8J125k16Y=; h=From:To:Cc:Subject:Date:From; b=mUVzo72Imn03zBMvWwvM+Vip2RYb+XW6etshLoNmcZKrAbCYVnNEet6b3zU71MKvl bsOYiQGGswJw861qdjwb1fuleljWPXT16PXzanXcdZp1t/uJqWLB0CfvN3rVJ4FLNe 9/vH9zt3rWfSdRv2aEh8XoC1OdM0P9zikrD3bhH+UesI8EB3EwlacXzO9fx1a4RloA 51qqnp7NHIPZ23UxgcmJT03W3PT2oxh+tcFCQNbztta1esIp5y4zy8e2y0RAXc+jk4 7ll9+NzO41h4B3ZmLiGJ7WXJSviGTLAkoLS/8EcfKI+V0Ec/VvNtw8SrtKip1vYF4W MhSKWDh0StRJw== From: Oded Gabbay To: dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org Cc: Dafna Hirschfeld Subject: [PATCH 1/6] accel/habanalabs: check failure of eventfd_signal Date: Tue, 2 Jan 2024 17:06:49 +0200 Message-Id: <20240102150654.522555-1-ogabbay@kernel.org> X-Mailer: git-send-email 2.34.1 Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1786991650172741798 X-GMAIL-MSGID: 1786991650172741798 From: Dafna Hirschfeld print err log if eventfd_signal return val is not 1 as expected. Signed-off-by: Dafna Hirschfeld Reviewed-by: Oded Gabbay Signed-off-by: Oded Gabbay --- drivers/accel/habanalabs/common/device.c | 14 ++++++++++---- 1 file changed, 10 insertions(+), 4 deletions(-) diff --git a/drivers/accel/habanalabs/common/device.c b/drivers/accel/habanalabs/common/device.c index a365791a9f5c..e81ccafddfb3 100644 --- a/drivers/accel/habanalabs/common/device.c +++ b/drivers/accel/habanalabs/common/device.c @@ -2043,13 +2043,19 @@ int hl_device_cond_reset(struct hl_device *hdev, u32 flags, u64 event_mask) return hl_device_reset(hdev, flags | HL_DRV_RESET_HARD); } -static void hl_notifier_event_send(struct hl_notifier_event *notifier_event, u64 event_mask) +static void hl_notifier_event_send(struct hl_device *hdev, + struct hl_notifier_event *notifier_event, u64 event_mask) { + u64 n; + mutex_lock(¬ifier_event->lock); notifier_event->events_mask |= event_mask; - if (notifier_event->eventfd) - eventfd_signal(notifier_event->eventfd, 1); + if (notifier_event->eventfd) { + n = eventfd_signal(notifier_event->eventfd, 1); + if (n != 1) + dev_err(hdev->dev, "eventfd signal failed, incremented by %llu\n", n); + } mutex_unlock(¬ifier_event->lock); } @@ -2073,7 +2079,7 @@ void hl_notifier_event_send_all(struct hl_device *hdev, u64 event_mask) mutex_lock(&hdev->fpriv_list_lock); list_for_each_entry(hpriv, &hdev->fpriv_list, dev_node) - hl_notifier_event_send(&hpriv->notifier_event, event_mask); + hl_notifier_event_send(hdev, &hpriv->notifier_event, event_mask); mutex_unlock(&hdev->fpriv_list_lock); } From patchwork Tue Jan 2 15:06:50 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Oded Gabbay X-Patchwork-Id: 184382 Return-Path: Delivered-To: ouuuleilei@gmail.com Received: by 2002:a05:7301:6f82:b0:100:9c79:88ff with SMTP id tb2csp4499156dyb; Tue, 2 Jan 2024 07:07:41 -0800 (PST) X-Google-Smtp-Source: AGHT+IFVY26NoYxl2AemFflXJP/O0XJ1yItodvLALODN/7N3rWmINzAWY93rbAdqb+zxFjXXPkCd X-Received: by 2002:a05:6a21:3392:b0:196:ac18:380f with SMTP id yy18-20020a056a21339200b00196ac18380fmr7257474pzb.106.1704208060792; Tue, 02 Jan 2024 07:07:40 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1704208060; cv=none; d=google.com; s=arc-20160816; b=1KPrVKR/+7pPR+7jlR6aIo5wgOzhG94YKAC52gV6DBQlGBn8Om5YSBIOez5M1bys/p 4K2StBRUfcJQ39zyzXUcXu1TPCsoiqfokUmSvniLknTFUy5bo8IOVN6h64wUCDm28JAV xhN92/Jg97xH1ak7HZ7YKX+5QYA8q5Z/iuf6xO4HHeISuGvtzqdu0MqDxG8ivL+VvTQw 1kUjf7vnD6icg/xlHrermD1lNmpZtiXZDvrGzatoTWyxSCeUqivFVocwnFbO38RuTXna H2VT1Z1X9hYsITxK+T/JnKQwydyzvZjn0ao3uCDOP8gqvbPdqCK2WDEAsfhcrGrev+LM 9/Uw== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=content-transfer-encoding:mime-version:list-unsubscribe :list-subscribe:list-id:precedence:references:in-reply-to:message-id :date:subject:cc:to:from:dkim-signature; bh=4gQGXqJQJ5lW5I/Js1+ZfuJy6IjAsBe1/8sWzw2U5sM=; fh=+fEw3/zQrmpgsgtgl9f9CsaDRovpb+9b/IJZeo8A5zw=; b=kVagOQAZzvTPTE7ZtVxS3NASC5/1idr1/L/nicap19rn07r6SDHAc0RMsUqGYqub5k QLhaLLxzoVdg2doQSO4QhUbP3wAGBqTU/+LYGP5jjZUEx9GejJPPmLZinwtFyMnObKKe fayVAsRMrm1qi/ydgW8PtvhdNVJs3V70kBClvkE6wFOmGPjmv+6xAcLPv+lfPf6cRNAl B9D73Ch2LctMmqE2BjPMyme2IviQnu2fBtzHzZW9R3V7mTKZGdbBwe1+o+5cb/L4rFoe XJm5vjQLkWihJdX/RAeozpElUoPM9lMLMz7kP8Vv+mgy/3wWk/fl8czQQIPWUNuGIQW8 JF5Q== ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=SXbvuShq; spf=pass (google.com: domain of linux-kernel+bounces-14527-ouuuleilei=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14527-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from sv.mirrors.kernel.org (sv.mirrors.kernel.org. [2604:1380:45e3:2400::1]) by mx.google.com with ESMTPS id j16-20020a635510000000b005ce0f4a19f9si14933089pgb.340.2024.01.02.07.07.40 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 02 Jan 2024 07:07:40 -0800 (PST) Received-SPF: pass (google.com: domain of linux-kernel+bounces-14527-ouuuleilei=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1; Authentication-Results: mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=SXbvuShq; spf=pass (google.com: domain of linux-kernel+bounces-14527-ouuuleilei=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14527-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from smtp.subspace.kernel.org (wormhole.subspace.kernel.org [52.25.139.140]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by sv.mirrors.kernel.org (Postfix) with ESMTPS id 90961283AC5 for ; Tue, 2 Jan 2024 15:07:40 +0000 (UTC) Received: from localhost.localdomain (localhost.localdomain [127.0.0.1]) by smtp.subspace.kernel.org (Postfix) with ESMTP id 383E214A82; Tue, 2 Jan 2024 15:07:05 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="SXbvuShq" X-Original-To: linux-kernel@vger.kernel.org Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id A2EAD14F76 for ; Tue, 2 Jan 2024 15:07:02 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 2808AC433C7; Tue, 2 Jan 2024 15:07:00 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1704208022; bh=ngDURZBrEZ6nJ6PufU0TBhtE8mc4FbnsCbyVBq3BT6k=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=SXbvuShqZhDJsRcKp/SJ1c21pa875vVyuzNFjzDuV+HncxvShQ0x7h2TP3EBUbL8J XP09rSCbAXzE7JR2VTHdavFjL2zboylGd4qRnIUPSBjg38l/uokXnuIpRwhHG0A/Dk CWakqv+GgFXI+rpcZOsjyjKI5+NUgf/vr6kiVp7nREvLmWT691VTXiKHpdDM02VuiA AVGEYdYa2vSjrRQRKPpGarhMQ/MaQV5emRbyvS2eSKMr8zae6DtiGMVCdVbCwX2qIj WZ9BmmNLKYDQy2aadlDYW+LVoWrV5hQeqrYGoLBIXg///wHQOBuh8o+tTnEltb3AIP kKptNmqOAVXUw== From: Oded Gabbay To: dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org Cc: Dani Liberman Subject: [PATCH 2/6] accel/habanalabs/gaudi2: add interrupt affinity for user interrupts Date: Tue, 2 Jan 2024 17:06:50 +0200 Message-Id: <20240102150654.522555-2-ogabbay@kernel.org> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20240102150654.522555-1-ogabbay@kernel.org> References: <20240102150654.522555-1-ogabbay@kernel.org> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1786991671665417805 X-GMAIL-MSGID: 1786991671665417805 From: Dani Liberman User interrupts are MSIx interrupts coming from Gaudi2, that have specific range of IDs and are assigned to the sole use of the user process that opened the Gaudi2 device (reminder: there can be only a single user process running on Gaudi2 at any given time). The interrupts are allocated and managed by the driver and therefore, the user expects the driver to initialize them properly, which also includes setting the affinity to the related CPU cores of the device's NUMA node to get maximum performance. Signed-off-by: Dani Liberman Reviewed-by: Oded Gabbay Signed-off-by: Oded Gabbay --- drivers/accel/habanalabs/common/device.c | 30 ++++++++++++++++++++ drivers/accel/habanalabs/common/habanalabs.h | 5 ++++ drivers/accel/habanalabs/gaudi2/gaudi2.c | 5 ++++ 3 files changed, 40 insertions(+) diff --git a/drivers/accel/habanalabs/common/device.c b/drivers/accel/habanalabs/common/device.c index e81ccafddfb3..514459396818 100644 --- a/drivers/accel/habanalabs/common/device.c +++ b/drivers/accel/habanalabs/common/device.c @@ -2807,3 +2807,33 @@ void hl_enable_err_info_capture(struct hl_error_info *captured_err_info) atomic_set(&captured_err_info->cs_timeout.write_enable, 1); captured_err_info->undef_opcode.write_enable = true; } + +void hl_init_cpu_for_irq(struct hl_device *hdev) +{ + struct cpumask *available_mask = &hdev->irq_affinity_mask; + int numa_node = hdev->pdev->dev.numa_node, i; + static struct cpumask cpu_mask; + + if (numa_node < 0) + return; + + if (!cpumask_and(&cpu_mask, cpumask_of_node(numa_node), cpu_online_mask)) { + dev_err(hdev->dev, "No available affinities in current numa node\n"); + return; + } + + /* Remove HT siblings */ + for_each_cpu(i, &cpu_mask) + cpumask_set_cpu(cpumask_first(topology_sibling_cpumask(i)), available_mask); +} + +void hl_set_irq_affinity(struct hl_device *hdev, int irq) +{ + if (cpumask_empty(&hdev->irq_affinity_mask)) { + dev_dbg(hdev->dev, "affinity mask is empty\n"); + return; + } + + if (irq_set_affinity_hint(irq, &hdev->irq_affinity_mask)) + dev_err(hdev->dev, "Failed setting irq %d affinity\n", irq); +} diff --git a/drivers/accel/habanalabs/common/habanalabs.h b/drivers/accel/habanalabs/common/habanalabs.h index 2a900c9941fe..b1a7b229e161 100644 --- a/drivers/accel/habanalabs/common/habanalabs.h +++ b/drivers/accel/habanalabs/common/habanalabs.h @@ -3257,6 +3257,7 @@ struct hl_reset_info { * @clk_throttling: holds information about current/previous clock throttling events * @captured_err_info: holds information about errors. * @reset_info: holds current device reset information. + * @irq_affinity_mask: mask of available CPU cores for user and decoder interrupt handling. * @stream_master_qid_arr: pointer to array with QIDs of master streams. * @fw_inner_major_ver: the major of current loaded preboot inner version. * @fw_inner_minor_ver: the minor of current loaded preboot inner version. @@ -3446,6 +3447,8 @@ struct hl_device { struct hl_reset_info reset_info; + cpumask_t irq_affinity_mask; + u32 *stream_master_qid_arr; u32 fw_inner_major_ver; u32 fw_inner_minor_ver; @@ -4032,6 +4035,8 @@ void hl_handle_critical_hw_err(struct hl_device *hdev, u16 event_id, u64 *event_ void hl_handle_fw_err(struct hl_device *hdev, struct hl_info_fw_err_info *info); void hl_capture_engine_err(struct hl_device *hdev, u16 engine_id, u16 error_count); void hl_enable_err_info_capture(struct hl_error_info *captured_err_info); +void hl_init_cpu_for_irq(struct hl_device *hdev); +void hl_set_irq_affinity(struct hl_device *hdev, int irq); #ifdef CONFIG_DEBUG_FS diff --git a/drivers/accel/habanalabs/gaudi2/gaudi2.c b/drivers/accel/habanalabs/gaudi2/gaudi2.c index e0e5615ef9b0..fd01525b1ea2 100644 --- a/drivers/accel/habanalabs/gaudi2/gaudi2.c +++ b/drivers/accel/habanalabs/gaudi2/gaudi2.c @@ -4254,6 +4254,8 @@ static int gaudi2_enable_msix(struct hl_device *hdev) if (gaudi2->hw_cap_initialized & HW_CAP_MSIX) return 0; + hl_init_cpu_for_irq(hdev); + rc = pci_alloc_irq_vectors(hdev->pdev, GAUDI2_MSIX_ENTRIES, GAUDI2_MSIX_ENTRIES, PCI_IRQ_MSIX); if (rc < 0) { @@ -4307,6 +4309,7 @@ static int gaudi2_enable_msix(struct hl_device *hdev) i++, j++, user_irq_init_cnt++) { irq = pci_irq_vector(hdev->pdev, i); + hl_set_irq_affinity(hdev, irq); rc = request_irq(irq, hl_irq_user_interrupt_handler, 0, gaudi2_irq_name(i), &hdev->user_interrupt[j]); if (rc) { @@ -4333,6 +4336,7 @@ static int gaudi2_enable_msix(struct hl_device *hdev) i < GAUDI2_IRQ_NUM_USER_FIRST + user_irq_init_cnt ; i++, j++) { irq = pci_irq_vector(hdev->pdev, i); + irq_set_affinity_hint(irq, NULL); free_irq(irq, &hdev->user_interrupt[j]); } irq = pci_irq_vector(hdev->pdev, GAUDI2_IRQ_NUM_UNEXPECTED_ERROR); @@ -4413,6 +4417,7 @@ static void gaudi2_disable_msix(struct hl_device *hdev) k < hdev->asic_prop.user_interrupt_count ; i++, j++, k++) { irq = pci_irq_vector(hdev->pdev, i); + irq_set_affinity_hint(irq, NULL); free_irq(irq, &hdev->user_interrupt[j]); } From patchwork Tue Jan 2 15:06:51 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-Patchwork-Submitter: Oded Gabbay X-Patchwork-Id: 184383 Return-Path: Delivered-To: ouuuleilei@gmail.com Received: by 2002:a05:7301:6f82:b0:100:9c79:88ff with SMTP id tb2csp4499416dyb; Tue, 2 Jan 2024 07:08:03 -0800 (PST) X-Google-Smtp-Source: AGHT+IF3szPtzee9F54h+B6NPqdfu4KlzwEaDd5+c8FTpzgSwVkspwNGzDPwUqTZpDs21GiSkLy3 X-Received: by 2002:a05:6808:14d5:b0:3bc:12f6:e575 with SMTP id f21-20020a05680814d500b003bc12f6e575mr2012264oiw.55.1704208083612; Tue, 02 Jan 2024 07:08:03 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1704208083; cv=none; d=google.com; s=arc-20160816; b=amK21Lp0K5zrzYl4fTH9kV0eY8dTMrpHNCMne82SZN+WV34f5/2XtQJVOmKqxm8D+A +iy/OXHo9NDrh5kIDINyJHKEybdNX3NTupimwJEamzNGlgyLiMYwEqA2xifaC7DgE81P K3SebbTsEtC3L4HoZQR+QqR78qLqmSV5Bd98LGYw1swjC6PH78AQ9QP5czrCPcJAqEDr ul4Irh6+becvE12s3M7zVtce+BJVOpFik8JI5spkt0ysIh2No22DstJtpqBx1YUPVKe5 HYf4MNkjzTrjvGWNgz2k6KmQUE6RAoCML8OwT3VyPLT7BPGxUt4KSl5dtkjrqDfs5oME rQLA== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=content-transfer-encoding:mime-version:list-unsubscribe :list-subscribe:list-id:precedence:references:in-reply-to:message-id :date:subject:cc:to:from:dkim-signature; bh=zS5qedOfE4UmAX36/Z6FB6orDilloP8EnCHWsdpMRaU=; fh=WNgYbmxwWcfeNadeuX+WeG2RxddHnoAGemnB56x+2vw=; b=kh11i9JFdCgpPFGybYA72AINgb6VAeYKw8Y7Fu3SsAi57b8ZzqboQYwEw4C7OBcqDT egruV9bJqi3jGTBhzz6Nut7InROA2AfgJAK4fimVFEQl/UmK3bcQehpux4q5XEXc7J0K GIPE+wxjnn8fV6q9pxaJeRnDc3NH6JcTUm2SV30MAxVgzniGjvdY67VJwZargvbk61yn ddwGVfP53PeP+GFkRWlQKhNlgdb8yJpbiow7yV9nDnOFmqNwAAmTW4YonPT+g9Hkh9mL Dq91sJmcSoul+EnnmV72bJ1Mo99onf72ttoej2wCVxnc1s9COiaiLxu8/TvJVXfrPcc8 XVQg== ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=gaBJwM7F; spf=pass (google.com: domain of linux-kernel+bounces-14529-ouuuleilei=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14529-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from ny.mirrors.kernel.org (ny.mirrors.kernel.org. [147.75.199.223]) by mx.google.com with ESMTPS id v9-20020a67f1c9000000b004670adb82ddsi1798894vsm.269.2024.01.02.07.08.03 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 02 Jan 2024 07:08:03 -0800 (PST) Received-SPF: pass (google.com: domain of linux-kernel+bounces-14529-ouuuleilei=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223; Authentication-Results: mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=gaBJwM7F; spf=pass (google.com: domain of linux-kernel+bounces-14529-ouuuleilei=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14529-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from smtp.subspace.kernel.org (wormhole.subspace.kernel.org [52.25.139.140]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ny.mirrors.kernel.org (Postfix) with ESMTPS id 293D71C223F8 for ; Tue, 2 Jan 2024 15:08:03 +0000 (UTC) Received: from localhost.localdomain (localhost.localdomain [127.0.0.1]) by smtp.subspace.kernel.org (Postfix) with ESMTP id DCE2715491; Tue, 2 Jan 2024 15:07:06 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="gaBJwM7F" X-Original-To: linux-kernel@vger.kernel.org Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 580BD12E7E for ; Tue, 2 Jan 2024 15:07:04 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id C610BC433CB; Tue, 2 Jan 2024 15:07:02 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1704208023; bh=QDhoQSaSpo9hrMc6cyInKkOcqMGrIdfUTWR31DFTkoo=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=gaBJwM7FcrWhNZXpLFWahC00ere2bPHL3R4R5XVWiT4oK40lXtL9slx0hKJ2P5vtq m2R3MAdyLtWj8kxYhgeR/qdw51jsC8p50dfyeNECB5arRU7YEib+RTDhUjBHa368+x YAMRNt7vpo/RMxY1+USfx4I/C1CnH1GN9lfIkwWV79YSURb/j3Aq5UeDVD6gf0JB6L RB5tQ3+IdCFpc1vZzHs1WZxNdFxrgNG0oLS03CNeEUttpeSbWQGyJx12yCJNMbhRxF 7UHEFp8xewAUOYN16ZIqUejqXDVALsBaxv9G5YXwqKfNa0dTYWnSV1JCei1lWaWqcP NBB3kVk66nicg== From: Oded Gabbay To: dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org Cc: Koby Elbaz Subject: [PATCH 3/6] accel/habanalabs: increase HL_MAX_STR to 64 bytes to avoid warnings Date: Tue, 2 Jan 2024 17:06:51 +0200 Message-Id: <20240102150654.522555-3-ogabbay@kernel.org> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20240102150654.522555-1-ogabbay@kernel.org> References: <20240102150654.522555-1-ogabbay@kernel.org> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1786991695796277820 X-GMAIL-MSGID: 1786991695796277820 From: Koby Elbaz Fix a warning of a buffer overflow: ‘snprintf’ output between 38 and 47 bytes into a destination of size 32 Signed-off-by: Koby Elbaz Reviewed-by: Oded Gabbay Signed-off-by: Oded Gabbay --- drivers/accel/habanalabs/common/habanalabs.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/accel/habanalabs/common/habanalabs.h b/drivers/accel/habanalabs/common/habanalabs.h index b1a7b229e161..253873315888 100644 --- a/drivers/accel/habanalabs/common/habanalabs.h +++ b/drivers/accel/habanalabs/common/habanalabs.h @@ -2547,7 +2547,7 @@ struct hl_state_dump_specs { * DEVICES */ -#define HL_STR_MAX 32 +#define HL_STR_MAX 64 #define HL_DEV_STS_MAX (HL_DEVICE_STATUS_LAST + 1) From patchwork Tue Jan 2 15:06:52 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Oded Gabbay X-Patchwork-Id: 184384 Return-Path: Delivered-To: ouuuleilei@gmail.com Received: by 2002:a05:7301:6f82:b0:100:9c79:88ff with SMTP id tb2csp4499525dyb; Tue, 2 Jan 2024 07:08:14 -0800 (PST) X-Google-Smtp-Source: AGHT+IHerLsiR1DfOw3aL148laSR6KQBH9NTw8Db2IHoKGN1y5j2QI6O1nFwgjoUNSaQhCEB9cA3 X-Received: by 2002:a05:6a20:9189:b0:197:49ff:55be with SMTP id v9-20020a056a20918900b0019749ff55bemr927130pzd.58.1704208094623; Tue, 02 Jan 2024 07:08:14 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1704208094; cv=none; d=google.com; s=arc-20160816; b=sUbyF3BuoNvc8brQT9dxXOGExloVoGac5HO+ZMIYskE+X6O9se3hOrQ2UCPFNfhymT xP9AMMAQhMd1a4f6VQuPxOODLiKfoBbozlffGDq2vQd3K2MrJZfJg+N9kVkkGkrSwrjy xjmzxyzj+oJ+RjxKNBM35EHcIqqB2ALnhnNLdLnGIBiQ8NqAXI2m/hChWzDkUHVb/qYs l/xHyQKdQk4jXxvM3GOy8axHwHebxcdtJ5BeJDtV4sZdZmeH1F4tbPD+B8CEZtGeE2P0 edOsyj0+bfBLwcgW5m5tOLlFiCshP+fc74nnv6ZY18Zo4eOlC/u33+uhbq/cDjBrCl+D pvCg== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=content-transfer-encoding:mime-version:list-unsubscribe :list-subscribe:list-id:precedence:references:in-reply-to:message-id :date:subject:cc:to:from:dkim-signature; bh=UJ0RLncwiO2op7cWaWHluy+Mns/9sKZr2605hoR32wY=; fh=PceJZ/KGF6ujSpGwn68DXpj5qiXzqwvBM4Q702GUgLM=; b=Pa8KY9aTRenClzNn7XoSBxON+tKqRZwmy/qFlOTzluXNfu3yMnVJZ+65uxkmeYr4Hz bpzmBrg85y00faTaYLzoeooc769lUGEbufIzESLaS/VbNNVjjc7v0B/45odKHCEyiIW1 y60eeGQgOjcM6FrOVgiCGR3KnZaf1kWbT96U4YI2+XeZ5e5epE4xDm7/ONFReUhg/dul +RiBJrVPwVfnbfFzdKU/DWwRKFd1zmQ2i/M0yh+f+zTN4BHgdzgWvA3sS/2FAqQ9jflP +YlmtXi8MylArMMxCQ9tEZTFOBlh/IT7IsgkMAObb2tYcbc3/ked5gL0g7zo5ulvmMqu 0t1g== ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=mOSLFDWQ; spf=pass (google.com: domain of linux-kernel+bounces-14530-ouuuleilei=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14530-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from sv.mirrors.kernel.org (sv.mirrors.kernel.org. [139.178.88.99]) by mx.google.com with ESMTPS id b8-20020a63d808000000b005c660acad7asi20379551pgh.4.2024.01.02.07.08.14 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 02 Jan 2024 07:08:14 -0800 (PST) Received-SPF: pass (google.com: domain of linux-kernel+bounces-14530-ouuuleilei=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99; Authentication-Results: mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=mOSLFDWQ; spf=pass (google.com: domain of linux-kernel+bounces-14530-ouuuleilei=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14530-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from smtp.subspace.kernel.org (wormhole.subspace.kernel.org [52.25.139.140]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by sv.mirrors.kernel.org (Postfix) with ESMTPS id 6D82D283AE5 for ; Tue, 2 Jan 2024 15:08:14 +0000 (UTC) Received: from localhost.localdomain (localhost.localdomain [127.0.0.1]) by smtp.subspace.kernel.org (Postfix) with ESMTP id 88BA6154AE; Tue, 2 Jan 2024 15:07:08 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="mOSLFDWQ" X-Original-To: linux-kernel@vger.kernel.org Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 018F614F63 for ; Tue, 2 Jan 2024 15:07:05 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 70731C433CC; Tue, 2 Jan 2024 15:07:04 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1704208025; bh=UlaJEziqzdhCw7f0wAaTCtw2xwU8224gk817BjROfoQ=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=mOSLFDWQ2FEnN28+kaDYYNEa95I80yaYBaFJiI4uHrywdVmuO+em4uFdcUUHeZ9z0 TT7aU9FslIDvH/zEUoymI/PcpbA4p21OAYk6vRoclbzfxxThwylc3Gob6qPSl6K8iu ejoERQ8Qtp3Xe4XLvunc3XmYf30Vll4/Xn4K5xB4UMU40Emxk3dxgdoYtm3cpfU/jg Ks0QUwECbyR5+jUmdUAnWmHtcdFpL02tSFCzhWmD/NtqC5XT5SEFicA/fIVRuww0vr h2F0+XVi4LOnkQ94rE34LeemaGajAlMLbd0vnSjTScO8oVifTkH+L3I6s52xIBsvZl 9/Zr7pWy9nI+w== From: Oded Gabbay To: dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org Cc: Tomer Tayar Subject: [PATCH 4/6] accel/habanalabs: fix DRAM BAR base address calculation Date: Tue, 2 Jan 2024 17:06:52 +0200 Message-Id: <20240102150654.522555-4-ogabbay@kernel.org> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20240102150654.522555-1-ogabbay@kernel.org> References: <20240102150654.522555-1-ogabbay@kernel.org> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1786991706853032982 X-GMAIL-MSGID: 1786991706853032982 From: Tomer Tayar When the DRAM region size in the BAR is not a power of 2, calculating the corresponding BAR base address should be done using the offset from the DRAM start address, and not using directly the DRAM address. Signed-off-by: Tomer Tayar Reviewed-by: Oded Gabbay Signed-off-by: Oded Gabbay --- drivers/accel/habanalabs/common/device.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/accel/habanalabs/common/device.c b/drivers/accel/habanalabs/common/device.c index 514459396818..15891de6cf39 100644 --- a/drivers/accel/habanalabs/common/device.c +++ b/drivers/accel/habanalabs/common/device.c @@ -55,7 +55,8 @@ static u64 hl_set_dram_bar(struct hl_device *hdev, u64 addr, struct pci_mem_regi if (is_power_of_2(prop->dram_pci_bar_size)) bar_base_addr = addr & ~(prop->dram_pci_bar_size - 0x1ull); else - bar_base_addr = DIV_ROUND_DOWN_ULL(addr, prop->dram_pci_bar_size) * + bar_base_addr = region->region_base + + div64_u64((addr - region->region_base), prop->dram_pci_bar_size) * prop->dram_pci_bar_size; old_base = hdev->asic_funcs->set_dram_bar_base(hdev, bar_base_addr); From patchwork Tue Jan 2 15:06:53 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Oded Gabbay X-Patchwork-Id: 184385 Return-Path: Delivered-To: ouuuleilei@gmail.com Received: by 2002:a05:7301:6f82:b0:100:9c79:88ff with SMTP id tb2csp4499758dyb; Tue, 2 Jan 2024 07:08:38 -0800 (PST) X-Google-Smtp-Source: AGHT+IGZQG0mWodd/oDBeQeD4YSTU8lv5WhZ0XASnYTIKcjvvBYAS3Fmgr9bspxmMB9q9yXtL5RX X-Received: by 2002:a50:a6dd:0:b0:553:82de:b01 with SMTP id f29-20020a50a6dd000000b0055382de0b01mr9707655edc.22.1704208118411; Tue, 02 Jan 2024 07:08:38 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1704208118; cv=none; d=google.com; s=arc-20160816; b=RrTbbbAf+Mk+I4qhQxRDkIiNYW8tUSGg/IFVywfpqQO2kt93XIxumzVPUppCqLhxFU pbAH0YUzyiwZrtygEsQFrzH0wxznbjggJjPZ4CVLI1pcUBEl4mWziRzOI+BeHbF5aPlZ LUNxR8e1UVjT3KDeZDcd5CMzAQbDlgGK99dfZX88c5NU7LTPcok7lZyjE/IiaR963Dmq eO5gRllyYGik+AFW6okVBEuXA4hH53tb7AnX6kSMDzKZb8VHOaT1Gr4c71ImvplKM8O4 eFPYSOfYaDM2s3TtrCRUpy7f5j0J8XYg9Y1vWJv/jSIJqu0OJl2V67bb1iR41z5Ub4fJ Kj0g== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=content-transfer-encoding:mime-version:list-unsubscribe :list-subscribe:list-id:precedence:references:in-reply-to:message-id :date:subject:cc:to:from:dkim-signature; bh=f0qGT2OWl6ZuiNACBVs2dRrwg74FmJhk7+BdIApBsaY=; fh=V54BVPs49mTldOrR7dHX4twJjw0Ji4qGRFqX5qRehjA=; b=lVFbwPwvSNWaHCz7snfqT9qx75pxmhLi9cOyUa2PTDiskZ0xxId0mm050jLXo44pcl 30kM52qopWODV7+rEmKkennBeNAx7Fhr8vqpO5viNnwAzPhwKlCS8GgpLnE0MM9Vf6VB j2zma2zq3aJw/piSSzpRSB0JZCA/55K9ylUBM8Ue/aYM/tgQgldpzTwDR5IiqrpXN7Rd vgKl8LepS9VWcmBxZ1O+hT7eDDqlKz9elSgQcWq9EBJ50OAxE81eno/BMVW9c7qb49NQ iLVEumiSb/ARRASCU8xKXwQQFkf6JC+YFU91WcsRPTHATcN2fxSNrhbrZ/KypGMZCjgt GRTA== ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=ecTyiNML; spf=pass (google.com: domain of linux-kernel+bounces-14531-ouuuleilei=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14531-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from am.mirrors.kernel.org (am.mirrors.kernel.org. [147.75.80.249]) by mx.google.com with ESMTPS id p20-20020a50cd94000000b0055366d82a99si11374954edi.434.2024.01.02.07.08.38 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 02 Jan 2024 07:08:38 -0800 (PST) Received-SPF: pass (google.com: domain of linux-kernel+bounces-14531-ouuuleilei=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249; Authentication-Results: mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=ecTyiNML; spf=pass (google.com: domain of linux-kernel+bounces-14531-ouuuleilei=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14531-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from smtp.subspace.kernel.org (wormhole.subspace.kernel.org [52.25.139.140]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by am.mirrors.kernel.org (Postfix) with ESMTPS id 9C9601F22EE7 for ; Tue, 2 Jan 2024 15:08:37 +0000 (UTC) Received: from localhost.localdomain (localhost.localdomain [127.0.0.1]) by smtp.subspace.kernel.org (Postfix) with ESMTP id CF986156E4; Tue, 2 Jan 2024 15:07:10 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="ecTyiNML" X-Original-To: linux-kernel@vger.kernel.org Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 35088154A9 for ; Tue, 2 Jan 2024 15:07:07 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 1C25BC433CB; Tue, 2 Jan 2024 15:07:05 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1704208027; bh=QSXbMuxdnjCa66UQ+SaInPOrrZUBavWZBAYCShYwlnI=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=ecTyiNML5RDRXTiFkiBn2GCKVC54KLzyOp7BqvwKRmBaoPB+eQc/MoYmukfqVuhlh hIKdoAKRhNvr4cadEO8w1InKkD5/1Qju0G0gbCI18tY/TmFpzZxyGqLBQOogetG9Rh E3+0zR3fMnvfU3sF7shZEaKwCXgUeHz4y4rslyZUYUKcw0aAjFFjR0n7fVud1IQdyG zPDSBorlNN9e8TldjuSuHgkRBjm2bjw/UBklwHXmIksOVXbScOrRZn7imiJuDD7zHN iUoJYOnW1ixvI9rKIfvH0+weFOZdCmQbu/UVUKuObimNOAQrTeadGhNdKBSjVScWWB tHp6pKkrnQN3w== From: Oded Gabbay To: dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org Cc: Farah Kassabri Subject: [PATCH 5/6] accel/habanalabs/gaudi2: move HMMU page tables to device memory Date: Tue, 2 Jan 2024 17:06:53 +0200 Message-Id: <20240102150654.522555-5-ogabbay@kernel.org> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20240102150654.522555-1-ogabbay@kernel.org> References: <20240102150654.522555-1-ogabbay@kernel.org> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1786991731954654555 X-GMAIL-MSGID: 1786991731954654555 From: Farah Kassabri Currently the HMMU page tables reside in the host memory, which will cause host access from the device for every page walk. This can affect PCIe bandwidth in certain scenarios. To prevent that problem, HMMU page tables will be moved to the device memory so the miss transaction will read the hops from there instead of going to the host. Signed-off-by: Farah Kassabri Reviewed-by: Oded Gabbay Signed-off-by: Oded Gabbay --- drivers/accel/habanalabs/common/habanalabs.h | 26 ++ drivers/accel/habanalabs/common/hw_queue.c | 17 + drivers/accel/habanalabs/common/mmu/Makefile | 2 +- drivers/accel/habanalabs/common/mmu/mmu.c | 223 ++++++++++- drivers/accel/habanalabs/common/mmu/mmu_v1.c | 352 +++--------------- drivers/accel/habanalabs/common/mmu/mmu_v2.c | 338 +++++++++++++++++ drivers/accel/habanalabs/gaudi/gaudi.c | 1 + drivers/accel/habanalabs/gaudi2/gaudi2.c | 245 ++++++++---- drivers/accel/habanalabs/gaudi2/gaudi2P.h | 12 +- .../include/hw_ip/mmu/mmu_general.h | 2 + 10 files changed, 836 insertions(+), 382 deletions(-) create mode 100644 drivers/accel/habanalabs/common/mmu/mmu_v2.c diff --git a/drivers/accel/habanalabs/common/habanalabs.h b/drivers/accel/habanalabs/common/habanalabs.h index 253873315888..7397ce86b7f0 100644 --- a/drivers/accel/habanalabs/common/habanalabs.h +++ b/drivers/accel/habanalabs/common/habanalabs.h @@ -443,18 +443,22 @@ enum hl_collective_mode { * a CB handle can be provided for jobs on this queue. * Otherwise, a CB address must be provided. * @collective_mode: collective mode of current queue + * @q_dram_bd_address: PQ dram address, used when PQ need to reside in DRAM. * @driver_only: true if only the driver is allowed to send a job to this queue, * false otherwise. * @binned: True if the queue is binned out and should not be used * @supports_sync_stream: True if queue supports sync stream + * @dram_bd: True if the bd should be copied to dram, needed for PQ which has been allocated on dram */ struct hw_queue_properties { enum hl_queue_type type; enum queue_cb_alloc_flags cb_alloc_flags; enum hl_collective_mode collective_mode; + u64 q_dram_bd_address; u8 driver_only; u8 binned; u8 supports_sync_stream; + u8 dram_bd; }; /** @@ -1052,6 +1056,8 @@ struct hl_encaps_signals_mgr { * @collective_mode: collective mode of current queue * @kernel_address: holds the queue's kernel virtual address. * @bus_address: holds the queue's DMA address. + * @pq_dram_address: hold the dram address when the PQ is allocated, used when dram_bd is true in + * queue properites. * @pi: holds the queue's pi value. * @ci: holds the queue's ci value, AS CALCULATED BY THE DRIVER (not real ci). * @hw_queue_id: the id of the H/W queue. @@ -1061,6 +1067,7 @@ struct hl_encaps_signals_mgr { * @valid: is the queue valid (we have array of 32 queues, not all of them * exist). * @supports_sync_stream: True if queue supports sync stream + * @dram_bd: True if the bd should be copied to dram, needed for PQ which has been allocated on dram */ struct hl_hw_queue { struct hl_cs_job **shadow_queue; @@ -1069,6 +1076,7 @@ struct hl_hw_queue { enum hl_collective_mode collective_mode; void *kernel_address; dma_addr_t bus_address; + u64 pq_dram_address; u32 pi; atomic_t ci; u32 hw_queue_id; @@ -1077,6 +1085,7 @@ struct hl_hw_queue { u16 int_queue_len; u8 valid; u8 supports_sync_stream; + u8 dram_bd; }; /** @@ -3889,6 +3898,7 @@ int hl_mmu_hr_get_tlb_info(struct hl_ctx *ctx, u64 virt_addr, struct hl_mmu_hop_ struct hl_hr_mmu_funcs *hr_func); int hl_mmu_if_set_funcs(struct hl_device *hdev); void hl_mmu_v1_set_funcs(struct hl_device *hdev, struct hl_mmu_funcs *mmu); +void hl_mmu_v2_set_funcs(struct hl_device *hdev, struct hl_mmu_funcs *mmu); void hl_mmu_v2_hr_set_funcs(struct hl_device *hdev, struct hl_mmu_funcs *mmu); int hl_mmu_va_to_pa(struct hl_ctx *ctx, u64 virt_addr, u64 *phys_addr); int hl_mmu_get_tlb_info(struct hl_ctx *ctx, u64 virt_addr, @@ -3896,6 +3906,22 @@ int hl_mmu_get_tlb_info(struct hl_ctx *ctx, u64 virt_addr, u64 hl_mmu_scramble_addr(struct hl_device *hdev, u64 addr); u64 hl_mmu_descramble_addr(struct hl_device *hdev, u64 addr); bool hl_is_dram_va(struct hl_device *hdev, u64 virt_addr); +struct pgt_info *hl_mmu_dr_get_pgt_info(struct hl_ctx *ctx, u64 hop_addr); +void hl_mmu_dr_free_hop(struct hl_ctx *ctx, u64 hop_addr); +void hl_mmu_dr_free_pgt_node(struct hl_ctx *ctx, struct pgt_info *pgt_info); +u64 hl_mmu_dr_get_phys_hop0_addr(struct hl_ctx *ctx); +u64 hl_mmu_dr_get_hop0_addr(struct hl_ctx *ctx); +void hl_mmu_dr_write_pte(struct hl_ctx *ctx, u64 shadow_pte_addr, u64 val); +void hl_mmu_dr_write_final_pte(struct hl_ctx *ctx, u64 shadow_pte_addr, u64 val); +void hl_mmu_dr_clear_pte(struct hl_ctx *ctx, u64 pte_addr); +u64 hl_mmu_dr_get_phys_addr(struct hl_ctx *ctx, u64 shadow_addr); +void hl_mmu_dr_get_pte(struct hl_ctx *ctx, u64 hop_addr); +int hl_mmu_dr_put_pte(struct hl_ctx *ctx, u64 hop_addr); +u64 hl_mmu_dr_get_alloc_next_hop_addr(struct hl_ctx *ctx, u64 curr_pte, bool *is_new_hop); +u64 hl_mmu_dr_alloc_hop(struct hl_ctx *ctx); +void hl_mmu_dr_flush(struct hl_ctx *ctx); +int hl_mmu_dr_init(struct hl_device *hdev); +void hl_mmu_dr_fini(struct hl_device *hdev); int hl_fw_load_fw_to_device(struct hl_device *hdev, const char *fw_name, void __iomem *dst, u32 src_offset, u32 size); diff --git a/drivers/accel/habanalabs/common/hw_queue.c b/drivers/accel/habanalabs/common/hw_queue.c index d0087c0ec48c..3d04a7507cce 100644 --- a/drivers/accel/habanalabs/common/hw_queue.c +++ b/drivers/accel/habanalabs/common/hw_queue.c @@ -84,6 +84,8 @@ void hl_hw_queue_submit_bd(struct hl_device *hdev, struct hl_hw_queue *q, u32 ctl, u32 len, u64 ptr) { struct hl_bd *bd; + u64 addr; + int i; bd = q->kernel_address; bd += hl_pi_2_offset(q->pi); @@ -91,7 +93,16 @@ void hl_hw_queue_submit_bd(struct hl_device *hdev, struct hl_hw_queue *q, bd->len = cpu_to_le32(len); bd->ptr = cpu_to_le64(ptr); + if (q->dram_bd) + for (i = 0 ; i < 2 ; i++) { + addr = q->pq_dram_address + + ((hl_pi_2_offset(q->pi) * sizeof(struct hl_bd)) + (i * sizeof(u64))); + hdev->asic_funcs->access_dev_mem(hdev, PCI_REGION_DRAM, addr, + (u64 *)(bd) + i, DEBUGFS_WRITE64); + } + q->pi = hl_queue_inc_ptr(q->pi); + hdev->asic_funcs->ring_doorbell(hdev, q->hw_queue_id, q->pi); } @@ -1087,12 +1098,18 @@ int hl_hw_queues_create(struct hl_device *hdev) q->supports_sync_stream = asic->hw_queues_props[i].supports_sync_stream; q->collective_mode = asic->hw_queues_props[i].collective_mode; + q->dram_bd = asic->hw_queues_props[i].dram_bd; + rc = queue_init(hdev, q, i); if (rc) { dev_err(hdev->dev, "failed to initialize queue %d\n", i); goto release_queues; } + + /* Set DRAM PQ address for the queue if it should be at DRAM */ + if (q->dram_bd) + q->pq_dram_address = asic->hw_queues_props[i].q_dram_bd_address; } return 0; diff --git a/drivers/accel/habanalabs/common/mmu/Makefile b/drivers/accel/habanalabs/common/mmu/Makefile index 1806c524e04a..f4b815bf4f7d 100644 --- a/drivers/accel/habanalabs/common/mmu/Makefile +++ b/drivers/accel/habanalabs/common/mmu/Makefile @@ -1,3 +1,3 @@ # SPDX-License-Identifier: GPL-2.0-only HL_COMMON_MMU_FILES := common/mmu/mmu.o common/mmu/mmu_v1.o \ - common/mmu/mmu_v2_hr.o + common/mmu/mmu_v2.o common/mmu/mmu_v2_hr.o diff --git a/drivers/accel/habanalabs/common/mmu/mmu.c b/drivers/accel/habanalabs/common/mmu/mmu.c index b654302a68fc..fa7919dba783 100644 --- a/drivers/accel/habanalabs/common/mmu/mmu.c +++ b/drivers/accel/habanalabs/common/mmu/mmu.c @@ -585,6 +585,8 @@ int hl_mmu_get_tlb_info(struct hl_ctx *ctx, u64 virt_addr, int hl_mmu_if_set_funcs(struct hl_device *hdev) { + struct asic_fixed_properties *prop = &hdev->asic_prop; + if (hdev->mmu_disable) return 0; @@ -597,8 +599,9 @@ int hl_mmu_if_set_funcs(struct hl_device *hdev) case ASIC_GAUDI2: case ASIC_GAUDI2B: case ASIC_GAUDI2C: - /* MMUs in Gaudi2 are always host resident */ - hl_mmu_v2_hr_set_funcs(hdev, &hdev->mmu_func[MMU_HR_PGT]); + hl_mmu_v2_set_funcs(hdev, &hdev->mmu_func[MMU_DR_PGT]); + if (prop->pmmu.host_resident) + hl_mmu_v2_hr_set_funcs(hdev, &hdev->mmu_func[MMU_HR_PGT]); break; default: dev_err(hdev->dev, "Unrecognized ASIC type %d\n", @@ -1209,3 +1212,219 @@ int hl_mmu_hr_get_tlb_info(struct hl_ctx *ctx, u64 virt_addr, struct hl_mmu_hop_ return 0; } +struct pgt_info *hl_mmu_dr_get_pgt_info(struct hl_ctx *ctx, u64 hop_addr) +{ + struct pgt_info *pgt_info = NULL; + + hash_for_each_possible(ctx->mmu_shadow_hash, pgt_info, node, + (unsigned long) hop_addr) + if (hop_addr == pgt_info->shadow_addr) + break; + + return pgt_info; +} + +void hl_mmu_dr_free_hop(struct hl_ctx *ctx, u64 hop_addr) +{ + struct pgt_info *pgt_info = hl_mmu_dr_get_pgt_info(ctx, hop_addr); + + hl_mmu_dr_free_pgt_node(ctx, pgt_info); +} + +void hl_mmu_dr_free_pgt_node(struct hl_ctx *ctx, struct pgt_info *pgt_info) +{ + struct hl_device *hdev = ctx->hdev; + + gen_pool_free(hdev->mmu_priv.dr.mmu_pgt_pool, pgt_info->phys_addr, + hdev->asic_prop.mmu_hop_table_size); + hash_del(&pgt_info->node); + kfree((u64 *) (uintptr_t) pgt_info->shadow_addr); + kfree(pgt_info); +} + +u64 hl_mmu_dr_get_phys_hop0_addr(struct hl_ctx *ctx) +{ + return ctx->hdev->asic_prop.mmu_pgt_addr + + (ctx->asid * ctx->hdev->asic_prop.mmu_hop_table_size); +} + +u64 hl_mmu_dr_get_hop0_addr(struct hl_ctx *ctx) +{ + return (u64) (uintptr_t) ctx->hdev->mmu_priv.dr.mmu_shadow_hop0 + + (ctx->asid * ctx->hdev->asic_prop.mmu_hop_table_size); +} + +u64 hl_mmu_dr_get_phys_addr(struct hl_ctx *ctx, u64 shadow_addr) +{ + u64 page_mask = ctx->hdev->asic_prop.mmu_hop_table_size - 1; + u64 shadow_hop_addr = shadow_addr & (~page_mask); + u64 pte_offset = shadow_addr & page_mask; + u64 phys_hop_addr; + + if (shadow_hop_addr != hl_mmu_dr_get_hop0_addr(ctx)) + phys_hop_addr = hl_mmu_dr_get_pgt_info(ctx, shadow_hop_addr)->phys_addr; + else + phys_hop_addr = hl_mmu_dr_get_phys_hop0_addr(ctx); + + return phys_hop_addr + pte_offset; +} + +void hl_mmu_dr_write_pte(struct hl_ctx *ctx, u64 shadow_pte_addr, u64 val) +{ + u64 phys_val = hl_mmu_dr_get_phys_addr(ctx, val); + + ctx->hdev->asic_funcs->write_pte(ctx->hdev, hl_mmu_dr_get_phys_addr(ctx, shadow_pte_addr), + phys_val); + + *(u64 *) (uintptr_t) shadow_pte_addr = val; +} + +void hl_mmu_dr_write_final_pte(struct hl_ctx *ctx, u64 shadow_pte_addr, u64 val) +{ + ctx->hdev->asic_funcs->write_pte(ctx->hdev, + hl_mmu_dr_get_phys_addr(ctx, shadow_pte_addr), val); + *(u64 *) (uintptr_t) shadow_pte_addr = val; +} + +void hl_mmu_dr_clear_pte(struct hl_ctx *ctx, u64 pte_addr) +{ + hl_mmu_dr_write_final_pte(ctx, pte_addr, 0); +} + +void hl_mmu_dr_get_pte(struct hl_ctx *ctx, u64 hop_addr) +{ + hl_mmu_dr_get_pgt_info(ctx, hop_addr)->num_of_ptes++; +} + +int hl_mmu_dr_put_pte(struct hl_ctx *ctx, u64 hop_addr) +{ + struct pgt_info *pgt_info = hl_mmu_dr_get_pgt_info(ctx, hop_addr); + int num_of_ptes_left; + + pgt_info->num_of_ptes--; + + /* + * Need to save the number of ptes left because hl_mmu_free_hop might free + * the pgt_info + */ + num_of_ptes_left = pgt_info->num_of_ptes; + if (!num_of_ptes_left) + hl_mmu_dr_free_pgt_node(ctx, pgt_info); + + return num_of_ptes_left; +} + +u64 hl_mmu_dr_alloc_hop(struct hl_ctx *ctx) +{ + struct hl_device *hdev = ctx->hdev; + struct asic_fixed_properties *prop = &hdev->asic_prop; + struct pgt_info *pgt_info; + u64 phys_addr, shadow_addr; + + pgt_info = kmalloc(sizeof(*pgt_info), GFP_KERNEL); + if (!pgt_info) + return ULLONG_MAX; + + phys_addr = (u64) gen_pool_alloc(hdev->mmu_priv.dr.mmu_pgt_pool, + prop->mmu_hop_table_size); + if (!phys_addr) { + dev_err(hdev->dev, "failed to allocate page\n"); + goto pool_add_err; + } + + shadow_addr = (u64) (uintptr_t) kzalloc(prop->mmu_hop_table_size, + GFP_KERNEL); + if (!shadow_addr) + goto shadow_err; + + pgt_info->phys_addr = phys_addr; + pgt_info->shadow_addr = shadow_addr; + pgt_info->ctx = ctx; + pgt_info->num_of_ptes = 0; + hash_add(ctx->mmu_shadow_hash, &pgt_info->node, shadow_addr); + + return shadow_addr; + +shadow_err: + gen_pool_free(hdev->mmu_priv.dr.mmu_pgt_pool, + phys_addr, prop->mmu_hop_table_size); +pool_add_err: + kfree(pgt_info); + + return ULLONG_MAX; +} + +u64 hl_mmu_dr_get_alloc_next_hop_addr(struct hl_ctx *ctx, u64 curr_pte, bool *is_new_hop) +{ + u64 hop_addr = hl_mmu_get_next_hop_addr(ctx, curr_pte); + + if (hop_addr == ULLONG_MAX) { + hop_addr = hl_mmu_dr_alloc_hop(ctx); + *is_new_hop = (hop_addr != ULLONG_MAX); + } + + return hop_addr; +} + +void hl_mmu_dr_flush(struct hl_ctx *ctx) +{ + /* flush all writes from all cores to reach PCI */ + mb(); + ctx->hdev->asic_funcs->read_pte(ctx->hdev, hl_mmu_dr_get_phys_hop0_addr(ctx)); +} + +int hl_mmu_dr_init(struct hl_device *hdev) +{ + struct asic_fixed_properties *prop = &hdev->asic_prop; + int rc; + + hdev->mmu_priv.dr.mmu_pgt_pool = + gen_pool_create(__ffs(prop->mmu_hop_table_size), -1); + + if (!hdev->mmu_priv.dr.mmu_pgt_pool) { + dev_err(hdev->dev, "Failed to create page gen pool\n"); + return -ENOMEM; + } + + rc = gen_pool_add(hdev->mmu_priv.dr.mmu_pgt_pool, prop->mmu_pgt_addr + + prop->mmu_hop0_tables_total_size, + prop->dmmu.pgt_size - prop->mmu_hop0_tables_total_size, + -1); + if (rc) { + dev_err(hdev->dev, "Failed to add memory to page gen pool\n"); + goto err_pool_add; + } + + hdev->mmu_priv.dr.mmu_shadow_hop0 = kvcalloc(prop->max_asid, + prop->mmu_hop_table_size, GFP_KERNEL); + if (ZERO_OR_NULL_PTR(hdev->mmu_priv.dr.mmu_shadow_hop0)) { + rc = -ENOMEM; + goto err_pool_add; + } + + /* MMU H/W init will be done in device hw_init() */ + + return 0; + +err_pool_add: + gen_pool_destroy(hdev->mmu_priv.dr.mmu_pgt_pool); + + return rc; +} + +void hl_mmu_dr_fini(struct hl_device *hdev) +{ + /* MMU H/W fini was already done in device hw_fini() */ + + if (ZERO_OR_NULL_PTR(hdev->mmu_priv.dr.mmu_shadow_hop0)) + return; + + kvfree(hdev->mmu_priv.dr.mmu_shadow_hop0); + gen_pool_destroy(hdev->mmu_priv.dr.mmu_pgt_pool); + + /* Make sure that if we arrive here again without init was + * called we won't cause kernel panic. This can happen for + * example if we fail during hard reset code at certain points + */ + hdev->mmu_priv.dr.mmu_shadow_hop0 = NULL; +} diff --git a/drivers/accel/habanalabs/common/mmu/mmu_v1.c b/drivers/accel/habanalabs/common/mmu/mmu_v1.c index d925dc4dd097..64b5c8fbb166 100644 --- a/drivers/accel/habanalabs/common/mmu/mmu_v1.c +++ b/drivers/accel/habanalabs/common/mmu/mmu_v1.c @@ -12,166 +12,6 @@ #define MMU_V1_MAX_HOPS (MMU_HOP4 + 1) -static inline u64 get_phys_addr(struct hl_ctx *ctx, u64 shadow_addr); - -static struct pgt_info *get_pgt_info(struct hl_ctx *ctx, u64 hop_addr) -{ - struct pgt_info *pgt_info = NULL; - - hash_for_each_possible(ctx->mmu_shadow_hash, pgt_info, node, - (unsigned long) hop_addr) - if (hop_addr == pgt_info->shadow_addr) - break; - - return pgt_info; -} - -static void _free_hop(struct hl_ctx *ctx, struct pgt_info *pgt_info) -{ - struct hl_device *hdev = ctx->hdev; - - gen_pool_free(hdev->mmu_priv.dr.mmu_pgt_pool, pgt_info->phys_addr, - hdev->asic_prop.mmu_hop_table_size); - hash_del(&pgt_info->node); - kfree((u64 *) (uintptr_t) pgt_info->shadow_addr); - kfree(pgt_info); -} - -static void free_hop(struct hl_ctx *ctx, u64 hop_addr) -{ - struct pgt_info *pgt_info = get_pgt_info(ctx, hop_addr); - - _free_hop(ctx, pgt_info); -} - -static u64 alloc_hop(struct hl_ctx *ctx) -{ - struct hl_device *hdev = ctx->hdev; - struct asic_fixed_properties *prop = &hdev->asic_prop; - struct pgt_info *pgt_info; - u64 phys_addr, shadow_addr; - - pgt_info = kmalloc(sizeof(*pgt_info), GFP_KERNEL); - if (!pgt_info) - return ULLONG_MAX; - - phys_addr = (u64) gen_pool_alloc(hdev->mmu_priv.dr.mmu_pgt_pool, - prop->mmu_hop_table_size); - if (!phys_addr) { - dev_err(hdev->dev, "failed to allocate page\n"); - goto pool_add_err; - } - - shadow_addr = (u64) (uintptr_t) kzalloc(prop->mmu_hop_table_size, - GFP_KERNEL); - if (!shadow_addr) - goto shadow_err; - - pgt_info->phys_addr = phys_addr; - pgt_info->shadow_addr = shadow_addr; - pgt_info->ctx = ctx; - pgt_info->num_of_ptes = 0; - hash_add(ctx->mmu_shadow_hash, &pgt_info->node, shadow_addr); - - return shadow_addr; - -shadow_err: - gen_pool_free(hdev->mmu_priv.dr.mmu_pgt_pool, phys_addr, - prop->mmu_hop_table_size); -pool_add_err: - kfree(pgt_info); - - return ULLONG_MAX; -} - -static inline u64 get_phys_hop0_addr(struct hl_ctx *ctx) -{ - return ctx->hdev->asic_prop.mmu_pgt_addr + - (ctx->asid * ctx->hdev->asic_prop.mmu_hop_table_size); -} - -static inline u64 get_hop0_addr(struct hl_ctx *ctx) -{ - return (u64) (uintptr_t) ctx->hdev->mmu_priv.dr.mmu_shadow_hop0 + - (ctx->asid * ctx->hdev->asic_prop.mmu_hop_table_size); -} - -static void flush(struct hl_ctx *ctx) -{ - /* flush all writes from all cores to reach PCI */ - mb(); - ctx->hdev->asic_funcs->read_pte(ctx->hdev, get_phys_hop0_addr(ctx)); -} - -/* transform the value to physical address when writing to H/W */ -static inline void write_pte(struct hl_ctx *ctx, u64 shadow_pte_addr, u64 val) -{ - /* - * The value to write is actually the address of the next shadow hop + - * flags at the 12 LSBs. - * Hence in order to get the value to write to the physical PTE, we - * clear the 12 LSBs and translate the shadow hop to its associated - * physical hop, and add back the original 12 LSBs. - */ - u64 phys_val = get_phys_addr(ctx, val & HOP_PHYS_ADDR_MASK) | - (val & FLAGS_MASK); - - ctx->hdev->asic_funcs->write_pte(ctx->hdev, - get_phys_addr(ctx, shadow_pte_addr), - phys_val); - - *(u64 *) (uintptr_t) shadow_pte_addr = val; -} - -/* do not transform the value to physical address when writing to H/W */ -static inline void write_final_pte(struct hl_ctx *ctx, u64 shadow_pte_addr, - u64 val) -{ - ctx->hdev->asic_funcs->write_pte(ctx->hdev, - get_phys_addr(ctx, shadow_pte_addr), - val); - *(u64 *) (uintptr_t) shadow_pte_addr = val; -} - -/* clear the last and present bits */ -static inline void clear_pte(struct hl_ctx *ctx, u64 pte_addr) -{ - /* no need to transform the value to physical address */ - write_final_pte(ctx, pte_addr, 0); -} - -static inline void get_pte(struct hl_ctx *ctx, u64 hop_addr) -{ - get_pgt_info(ctx, hop_addr)->num_of_ptes++; -} - -/* - * put_pte - decrement the num of ptes and free the hop if possible - * - * @ctx: pointer to the context structure - * @hop_addr: addr of the hop - * - * This function returns the number of ptes left on this hop. If the number is - * 0, it means the pte was freed. - */ -static inline int put_pte(struct hl_ctx *ctx, u64 hop_addr) -{ - struct pgt_info *pgt_info = get_pgt_info(ctx, hop_addr); - int num_of_ptes_left; - - pgt_info->num_of_ptes--; - - /* - * Need to save the number of ptes left because free_hop might free - * the pgt_info - */ - num_of_ptes_left = pgt_info->num_of_ptes; - if (!num_of_ptes_left) - _free_hop(ctx, pgt_info); - - return num_of_ptes_left; -} - static inline u64 get_hop_pte_addr(struct hl_ctx *ctx, struct hl_mmu_properties *mmu_prop, u64 *hop_addr_arr, u64 virt_addr, enum mmu_hop_num hop_idx) { @@ -183,35 +23,6 @@ static inline u64 get_hop_pte_addr(struct hl_ctx *ctx, struct hl_mmu_properties ctx->hdev->asic_prop.mmu_pte_size * ((virt_addr & mask) >> shift); } -static inline u64 get_alloc_next_hop_addr(struct hl_ctx *ctx, u64 curr_pte, - bool *is_new_hop) -{ - u64 hop_addr = hl_mmu_get_next_hop_addr(ctx, curr_pte); - - if (hop_addr == ULLONG_MAX) { - hop_addr = alloc_hop(ctx); - *is_new_hop = (hop_addr != ULLONG_MAX); - } - - return hop_addr; -} - -/* translates shadow address inside hop to a physical address */ -static inline u64 get_phys_addr(struct hl_ctx *ctx, u64 shadow_addr) -{ - u64 page_mask = (ctx->hdev->asic_prop.mmu_hop_table_size - 1); - u64 shadow_hop_addr = shadow_addr & ~page_mask; - u64 pte_offset = shadow_addr & page_mask; - u64 phys_hop_addr; - - if (shadow_hop_addr != get_hop0_addr(ctx)) - phys_hop_addr = get_pgt_info(ctx, shadow_hop_addr)->phys_addr; - else - phys_hop_addr = get_phys_hop0_addr(ctx); - - return phys_hop_addr + pte_offset; -} - static int dram_default_mapping_init(struct hl_ctx *ctx) { struct hl_device *hdev = ctx->hdev; @@ -236,9 +47,9 @@ static int dram_default_mapping_init(struct hl_ctx *ctx) if (!ctx->dram_default_hops) return -ENOMEM; - hop0_addr = get_hop0_addr(ctx); + hop0_addr = hl_mmu_dr_get_hop0_addr(ctx); - hop1_addr = alloc_hop(ctx); + hop1_addr = hl_mmu_dr_alloc_hop(ctx); if (hop1_addr == ULLONG_MAX) { dev_err(hdev->dev, "failed to alloc hop 1\n"); rc = -ENOMEM; @@ -247,7 +58,7 @@ static int dram_default_mapping_init(struct hl_ctx *ctx) ctx->dram_default_hops[total_hops - 1] = hop1_addr; - hop2_addr = alloc_hop(ctx); + hop2_addr = hl_mmu_dr_alloc_hop(ctx); if (hop2_addr == ULLONG_MAX) { dev_err(hdev->dev, "failed to alloc hop 2\n"); rc = -ENOMEM; @@ -257,7 +68,7 @@ static int dram_default_mapping_init(struct hl_ctx *ctx) ctx->dram_default_hops[total_hops - 2] = hop2_addr; for (i = 0 ; i < num_of_hop3 ; i++) { - ctx->dram_default_hops[i] = alloc_hop(ctx); + ctx->dram_default_hops[i] = hl_mmu_dr_alloc_hop(ctx); if (ctx->dram_default_hops[i] == ULLONG_MAX) { dev_err(hdev->dev, "failed to alloc hop 3, i: %d\n", i); rc = -ENOMEM; @@ -268,18 +79,18 @@ static int dram_default_mapping_init(struct hl_ctx *ctx) /* need only pte 0 in hops 0 and 1 */ pte_val = (hop1_addr & HOP_PHYS_ADDR_MASK) | PAGE_PRESENT_MASK; - write_pte(ctx, hop0_addr, pte_val); + hl_mmu_dr_write_pte(ctx, hop0_addr, pte_val); pte_val = (hop2_addr & HOP_PHYS_ADDR_MASK) | PAGE_PRESENT_MASK; - write_pte(ctx, hop1_addr, pte_val); - get_pte(ctx, hop1_addr); + hl_mmu_dr_write_pte(ctx, hop1_addr, pte_val); + hl_mmu_dr_get_pte(ctx, hop1_addr); hop2_pte_addr = hop2_addr; for (i = 0 ; i < num_of_hop3 ; i++) { pte_val = (ctx->dram_default_hops[i] & HOP_PHYS_ADDR_MASK) | PAGE_PRESENT_MASK; - write_pte(ctx, hop2_pte_addr, pte_val); - get_pte(ctx, hop2_addr); + hl_mmu_dr_write_pte(ctx, hop2_pte_addr, pte_val); + hl_mmu_dr_get_pte(ctx, hop2_addr); hop2_pte_addr += HL_PTE_SIZE; } @@ -289,23 +100,23 @@ static int dram_default_mapping_init(struct hl_ctx *ctx) for (i = 0 ; i < num_of_hop3 ; i++) { hop3_pte_addr = ctx->dram_default_hops[i]; for (j = 0 ; j < HOP_PTE_ENTRIES_512 ; j++) { - write_final_pte(ctx, hop3_pte_addr, pte_val); - get_pte(ctx, ctx->dram_default_hops[i]); + hl_mmu_dr_write_final_pte(ctx, hop3_pte_addr, pte_val); + hl_mmu_dr_get_pte(ctx, ctx->dram_default_hops[i]); hop3_pte_addr += HL_PTE_SIZE; } } - flush(ctx); + hl_mmu_dr_flush(ctx); return 0; hop3_err: for (i = 0 ; i < hop3_allocated ; i++) - free_hop(ctx, ctx->dram_default_hops[i]); + hl_mmu_dr_free_hop(ctx, ctx->dram_default_hops[i]); - free_hop(ctx, hop2_addr); + hl_mmu_dr_free_hop(ctx, hop2_addr); hop2_err: - free_hop(ctx, hop1_addr); + hl_mmu_dr_free_hop(ctx, hop1_addr); hop1_err: kfree(ctx->dram_default_hops); @@ -329,7 +140,7 @@ static void dram_default_mapping_fini(struct hl_ctx *ctx) do_div(num_of_hop3, prop->dram_page_size); do_div(num_of_hop3, HOP_PTE_ENTRIES_512); - hop0_addr = get_hop0_addr(ctx); + hop0_addr = hl_mmu_dr_get_hop0_addr(ctx); /* add hop1 and hop2 */ total_hops = num_of_hop3 + 2; hop1_addr = ctx->dram_default_hops[total_hops - 1]; @@ -338,101 +149,26 @@ static void dram_default_mapping_fini(struct hl_ctx *ctx) for (i = 0 ; i < num_of_hop3 ; i++) { hop3_pte_addr = ctx->dram_default_hops[i]; for (j = 0 ; j < HOP_PTE_ENTRIES_512 ; j++) { - clear_pte(ctx, hop3_pte_addr); - put_pte(ctx, ctx->dram_default_hops[i]); + hl_mmu_dr_clear_pte(ctx, hop3_pte_addr); + hl_mmu_dr_put_pte(ctx, ctx->dram_default_hops[i]); hop3_pte_addr += HL_PTE_SIZE; } } hop2_pte_addr = hop2_addr; for (i = 0 ; i < num_of_hop3 ; i++) { - clear_pte(ctx, hop2_pte_addr); - put_pte(ctx, hop2_addr); + hl_mmu_dr_clear_pte(ctx, hop2_pte_addr); + hl_mmu_dr_put_pte(ctx, hop2_addr); hop2_pte_addr += HL_PTE_SIZE; } - clear_pte(ctx, hop1_addr); - put_pte(ctx, hop1_addr); - clear_pte(ctx, hop0_addr); + hl_mmu_dr_clear_pte(ctx, hop1_addr); + hl_mmu_dr_put_pte(ctx, hop1_addr); + hl_mmu_dr_clear_pte(ctx, hop0_addr); kfree(ctx->dram_default_hops); - flush(ctx); -} - -/** - * hl_mmu_v1_init() - initialize the MMU module. - * @hdev: habanalabs device structure. - * - * This function does the following: - * - Create a pool of pages for pgt_infos. - * - Create a shadow table for pgt - * - * Return: 0 for success, non-zero for failure. - */ -static int hl_mmu_v1_init(struct hl_device *hdev) -{ - struct asic_fixed_properties *prop = &hdev->asic_prop; - int rc; - - hdev->mmu_priv.dr.mmu_pgt_pool = - gen_pool_create(__ffs(prop->mmu_hop_table_size), -1); - - if (!hdev->mmu_priv.dr.mmu_pgt_pool) { - dev_err(hdev->dev, "Failed to create page gen pool\n"); - return -ENOMEM; - } - - rc = gen_pool_add(hdev->mmu_priv.dr.mmu_pgt_pool, prop->mmu_pgt_addr + - prop->mmu_hop0_tables_total_size, - prop->mmu_pgt_size - prop->mmu_hop0_tables_total_size, - -1); - if (rc) { - dev_err(hdev->dev, "Failed to add memory to page gen pool\n"); - goto err_pool_add; - } - - hdev->mmu_priv.dr.mmu_shadow_hop0 = kvcalloc(prop->max_asid, prop->mmu_hop_table_size, - GFP_KERNEL); - if (ZERO_OR_NULL_PTR(hdev->mmu_priv.dr.mmu_shadow_hop0)) { - rc = -ENOMEM; - goto err_pool_add; - } - - /* MMU H/W init will be done in device hw_init() */ - - return 0; - -err_pool_add: - gen_pool_destroy(hdev->mmu_priv.dr.mmu_pgt_pool); - - return rc; -} - -/** - * hl_mmu_v1_fini() - release the MMU module. - * @hdev: habanalabs device structure. - * - * This function does the following: - * - Disable MMU in H/W. - * - Free the pgt_infos pool. - * - * All contexts should be freed before calling this function. - */ -static void hl_mmu_v1_fini(struct hl_device *hdev) -{ - /* MMU H/W fini was already done in device hw_fini() */ - - if (!ZERO_OR_NULL_PTR(hdev->mmu_priv.dr.mmu_shadow_hop0)) { - kvfree(hdev->mmu_priv.dr.mmu_shadow_hop0); - gen_pool_destroy(hdev->mmu_priv.dr.mmu_pgt_pool); - - /* Make sure that if we arrive here again without init was - * called we won't cause kernel panic. This can happen for - * example if we fail during hard reset code at certain points - */ - hdev->mmu_priv.dr.mmu_shadow_hop0 = NULL; - } + hl_mmu_dr_flush(ctx); } /** @@ -476,7 +212,7 @@ static void hl_mmu_v1_ctx_fini(struct hl_ctx *ctx) dev_err_ratelimited(hdev->dev, "pgt_info of addr 0x%llx of asid %d was not destroyed, num_ptes: %d\n", pgt_info->phys_addr, ctx->asid, pgt_info->num_of_ptes); - _free_hop(ctx, pgt_info); + hl_mmu_dr_free_pgt_node(ctx, pgt_info); } } @@ -495,7 +231,7 @@ static int hl_mmu_v1_unmap(struct hl_ctx *ctx, for (hop_idx = MMU_HOP0; hop_idx < MMU_HOP4; hop_idx++) { if (hop_idx == MMU_HOP0) { - hop_addr[hop_idx] = get_hop0_addr(ctx); + hop_addr[hop_idx] = hl_mmu_dr_get_hop0_addr(ctx); } else { hop_addr[hop_idx] = hl_mmu_get_next_hop_addr(ctx, curr_pte); if (hop_addr[hop_idx] == ULLONG_MAX) @@ -546,30 +282,30 @@ static int hl_mmu_v1_unmap(struct hl_ctx *ctx, } hop_idx = MMU_HOP3; - write_final_pte(ctx, hop_pte_addr[hop_idx], default_pte); - put_pte(ctx, hop_addr[hop_idx]); + hl_mmu_dr_write_final_pte(ctx, hop_pte_addr[hop_idx], default_pte); + hl_mmu_dr_put_pte(ctx, hop_addr[hop_idx]); } else { if (!(curr_pte & PAGE_PRESENT_MASK)) goto not_mapped; if (hop_addr[MMU_HOP4]) - clear_pte(ctx, hop_pte_addr[MMU_HOP4]); + hl_mmu_dr_clear_pte(ctx, hop_pte_addr[MMU_HOP4]); else - clear_pte(ctx, hop_pte_addr[MMU_HOP3]); + hl_mmu_dr_clear_pte(ctx, hop_pte_addr[MMU_HOP3]); - if (hop_addr[MMU_HOP4] && !put_pte(ctx, hop_addr[MMU_HOP4])) + if (hop_addr[MMU_HOP4] && !hl_mmu_dr_put_pte(ctx, hop_addr[MMU_HOP4])) clear_hop3 = true; if (!clear_hop3) goto mapped; for (hop_idx = MMU_HOP3; hop_idx >= 0; hop_idx--) { - clear_pte(ctx, hop_pte_addr[hop_idx]); + hl_mmu_dr_clear_pte(ctx, hop_pte_addr[hop_idx]); if (hop_idx == MMU_HOP0) break; - if (put_pte(ctx, hop_addr[hop_idx])) + if (hl_mmu_dr_put_pte(ctx, hop_addr[hop_idx])) goto mapped; } } @@ -616,10 +352,10 @@ static int hl_mmu_v1_map(struct hl_ctx *ctx, u64 virt_addr, u64 phys_addr, for (hop_idx = MMU_HOP0; hop_idx < num_hops; hop_idx++) { if (hop_idx == MMU_HOP0) { - hop_addr[hop_idx] = get_hop0_addr(ctx); + hop_addr[hop_idx] = hl_mmu_dr_get_hop0_addr(ctx); } else { hop_addr[hop_idx] = - get_alloc_next_hop_addr(ctx, curr_pte, &hop_new[hop_idx]); + hl_mmu_dr_get_alloc_next_hop_addr(ctx, curr_pte, &hop_new[hop_idx]); if (hop_addr[hop_idx] == ULLONG_MAX) goto err; } @@ -666,27 +402,27 @@ static int hl_mmu_v1_map(struct hl_ctx *ctx, u64 virt_addr, u64 phys_addr, curr_pte = (phys_addr & HOP_PHYS_ADDR_MASK) | mmu_prop->last_mask | PAGE_PRESENT_MASK; - write_final_pte(ctx, hop_pte_addr[num_hops - 1], curr_pte); + hl_mmu_dr_write_final_pte(ctx, hop_pte_addr[num_hops - 1], curr_pte); for (hop_idx = MMU_HOP1; hop_idx < num_hops; hop_idx++) { prev_hop = hop_idx - 1; if (hop_new[hop_idx]) { curr_pte = (hop_addr[hop_idx] & HOP_PHYS_ADDR_MASK) | PAGE_PRESENT_MASK; - write_pte(ctx, hop_pte_addr[prev_hop], curr_pte); + hl_mmu_dr_write_pte(ctx, hop_pte_addr[prev_hop], curr_pte); if (hop_idx != MMU_HOP1) - get_pte(ctx, hop_addr[prev_hop]); + hl_mmu_dr_get_pte(ctx, hop_addr[prev_hop]); } } - get_pte(ctx, hop_addr[num_hops - 1]); + hl_mmu_dr_get_pte(ctx, hop_addr[num_hops - 1]); return 0; err: for (hop_idx = num_hops; hop_idx > MMU_HOP0; hop_idx--) { if (hop_new[hop_idx]) - free_hop(ctx, hop_addr[hop_idx]); + hl_mmu_dr_free_hop(ctx, hop_addr[hop_idx]); } return rc; @@ -752,7 +488,7 @@ static int hl_mmu_v1_get_tlb_info(struct hl_ctx *ctx, u64 virt_addr, if (is_huge) used_hops--; - hops->hop_info[0].hop_addr = get_phys_hop0_addr(ctx); + hops->hop_info[0].hop_addr = hl_mmu_dr_get_phys_hop0_addr(ctx); hops->hop_info[0].hop_pte_addr = hl_mmu_get_hop_pte_phys_addr(ctx, mmu_prop, 0, hops->hop_info[0].hop_addr, virt_addr); @@ -801,13 +537,13 @@ static int hl_mmu_v1_get_tlb_info(struct hl_ctx *ctx, u64 virt_addr, */ void hl_mmu_v1_set_funcs(struct hl_device *hdev, struct hl_mmu_funcs *mmu) { - mmu->init = hl_mmu_v1_init; - mmu->fini = hl_mmu_v1_fini; + mmu->init = hl_mmu_dr_init; + mmu->fini = hl_mmu_dr_fini; mmu->ctx_init = hl_mmu_v1_ctx_init; mmu->ctx_fini = hl_mmu_v1_ctx_fini; mmu->map = hl_mmu_v1_map; mmu->unmap = hl_mmu_v1_unmap; - mmu->flush = flush; + mmu->flush = hl_mmu_dr_flush; mmu->swap_out = hl_mmu_v1_swap_out; mmu->swap_in = hl_mmu_v1_swap_in; mmu->get_tlb_info = hl_mmu_v1_get_tlb_info; diff --git a/drivers/accel/habanalabs/common/mmu/mmu_v2.c b/drivers/accel/habanalabs/common/mmu/mmu_v2.c new file mode 100644 index 000000000000..4bc0268fff1c --- /dev/null +++ b/drivers/accel/habanalabs/common/mmu/mmu_v2.c @@ -0,0 +1,338 @@ +// SPDX-License-Identifier: GPL-2.0 + +/* + * Copyright 2016-2020 HabanaLabs, Ltd. + * All Rights Reserved. + */ + +#include "../habanalabs.h" +#include "../../include/hw_ip/mmu/mmu_general.h" +#include "../../include/hw_ip/mmu/mmu_v2_0.h" + +#include + +/** + * hl_mmu_v2_ctx_init() - initialize a context for using the MMU module. + * @ctx: pointer to the context structure to initialize. + * + * Initialize a mutex to protect the concurrent mapping flow, a hash to hold all + * page tables hops related to this context. + * Return: 0 on success, non-zero otherwise. + */ +static int hl_mmu_v2_ctx_init(struct hl_ctx *ctx) +{ + hash_init(ctx->mmu_shadow_hash); + + return 0; +} + +/* + * hl_mmu_v2_ctx_fini - disable a ctx from using the mmu module + * + * @ctx: pointer to the context structure + * + * This function does the following: + * - Free any pgts which were not freed yet + * - Free the mutex + * - Free DRAM default page mapping hops + */ +static void hl_mmu_v2_ctx_fini(struct hl_ctx *ctx) +{ + struct hl_device *hdev = ctx->hdev; + struct pgt_info *pgt_info; + struct hlist_node *tmp; + int i; + + if (!hash_empty(ctx->mmu_shadow_hash)) + dev_err(hdev->dev, "ctx %d is freed while it has pgts in use\n", + ctx->asid); + + hash_for_each_safe(ctx->mmu_shadow_hash, i, tmp, pgt_info, node) { + dev_err_ratelimited(hdev->dev, + "pgt_info of addr 0x%llx of asid %d was not destroyed, num_ptes: %d\n", + pgt_info->phys_addr, ctx->asid, pgt_info->num_of_ptes); + hl_mmu_dr_free_pgt_node(ctx, pgt_info); + } +} + +static int hl_mmu_v2_unmap(struct hl_ctx *ctx, u64 virt_addr, bool is_dram_addr) +{ + u64 hop_addr[MMU_ARCH_6_HOPS] = { 0 }, hop_pte_addr[MMU_ARCH_6_HOPS] = { 0 }, curr_pte, + scrambled_virt_addr; + struct asic_fixed_properties *prop = &ctx->hdev->asic_prop; + struct hl_device *hdev = ctx->hdev; + struct hl_mmu_properties *mmu_prop; + bool is_huge = false; + int i, hop_last; + + /* device resident in V2 are allowed only for HMMU */ + if (!is_dram_addr) + return -EINVAL; + + mmu_prop = &prop->dmmu; + + hop_last = mmu_prop->num_hops - 1; + + scrambled_virt_addr = hdev->asic_funcs->scramble_addr(hdev, virt_addr); + + hop_addr[0] = hl_mmu_dr_get_hop0_addr(ctx); + hop_pte_addr[0] = hl_mmu_get_hop_pte_phys_addr(ctx, mmu_prop, 0, + hop_addr[0], scrambled_virt_addr); + if (hop_pte_addr[0] == U64_MAX) + return -EFAULT; + + curr_pte = *(u64 *) (uintptr_t) hop_pte_addr[0]; + + for (i = 1 ; i < mmu_prop->num_hops ; i++) { + hop_addr[i] = hl_mmu_get_next_hop_addr(ctx, curr_pte); + if (hop_addr[i] == ULLONG_MAX) + goto not_mapped; + + hop_pte_addr[i] = hl_mmu_get_hop_pte_phys_addr(ctx, mmu_prop, i, + hop_addr[i], scrambled_virt_addr); + if (hop_pte_addr[i] == U64_MAX) + return -EFAULT; + + curr_pte = *(u64 *) (uintptr_t) hop_pte_addr[i]; + + if ((i <= hop_last) && (curr_pte & mmu_prop->last_mask)) { + hop_last = i; + is_huge = true; + break; + } + } + + if (is_dram_addr && !is_huge) { + dev_err(hdev->dev, "DRAM unmapping should use huge pages only\n"); + return -EFAULT; + } + + if (!(curr_pte & PAGE_PRESENT_MASK)) + goto not_mapped; + + for (i = hop_last ; i > 0 ; i--) { + hl_mmu_dr_clear_pte(ctx, hop_pte_addr[i]); + if (hl_mmu_dr_put_pte(ctx, hop_addr[i])) + goto mapped; + } + hl_mmu_dr_clear_pte(ctx, hop_pte_addr[0]); + +mapped: + return 0; + +not_mapped: + dev_err(hdev->dev, "virt addr 0x%llx is not mapped to phys addr\n", + virt_addr); + + return -EINVAL; +} + +static int hl_mmu_v2_map(struct hl_ctx *ctx, u64 virt_addr, u64 phys_addr, + u32 page_size, bool is_dram_addr) +{ + u64 hop_addr[MMU_ARCH_6_HOPS] = { 0 }, hop_pte_addr[MMU_ARCH_6_HOPS] = { 0 }, + curr_pte = 0, scrambled_virt_addr, scrambled_phys_addr; + struct asic_fixed_properties *prop = &ctx->hdev->asic_prop; + bool hop_new[MMU_ARCH_6_HOPS] = { false }; + struct hl_device *hdev = ctx->hdev; + struct hl_mmu_properties *mmu_prop; + int rc, i, hop_last; + + /* device resident in V2 are allowed only for HMMU */ + if (!is_dram_addr) + return -EINVAL; + + mmu_prop = &prop->dmmu; + + hop_last = mmu_prop->num_hops - 1; + + scrambled_virt_addr = hdev->asic_funcs->scramble_addr(hdev, virt_addr); + scrambled_phys_addr = hdev->asic_funcs->scramble_addr(hdev, phys_addr); + + /* First hop is preallocated therefore it is treated differently */ + hop_addr[0] = hl_mmu_dr_get_hop0_addr(ctx); + hop_pte_addr[0] = hl_mmu_get_hop_pte_phys_addr(ctx, mmu_prop, 0, + hop_addr[0], scrambled_virt_addr); + curr_pte = *(u64 *) (uintptr_t) hop_pte_addr[0]; + + /* Handle hop1 to hop_last */ + for (i = 1 ; i <= hop_last ; i++) { + hop_addr[i] = hl_mmu_dr_get_alloc_next_hop_addr(ctx, curr_pte, &hop_new[i]); + if (hop_addr[i] == ULLONG_MAX) { + rc = -ENOMEM; + goto err; + } + + hop_pte_addr[i] = hl_mmu_get_hop_pte_phys_addr(ctx, mmu_prop, i, + hop_addr[i], scrambled_virt_addr); + if (hop_pte_addr[i] == U64_MAX) { + rc = -EINVAL; + goto err; + } + + if (!hop_pte_addr[i]) { + rc = -EINVAL; + goto err; + } + + curr_pte = *(u64 *) (uintptr_t) hop_pte_addr[i]; + } + + if (curr_pte & PAGE_PRESENT_MASK) { + dev_err(hdev->dev, + "mapping already exists for virt_addr 0x%llx\n", + virt_addr); + + for (i = 0 ; i <= hop_last ; i++) + dev_dbg(hdev->dev, "hop%d pte: 0x%llx (0x%llx)\n", + i, *(u64 *) (uintptr_t) hop_pte_addr[i], + hop_pte_addr[i]); + + rc = -EINVAL; + goto err; + } + + curr_pte = (scrambled_phys_addr & HOP_PHYS_ADDR_MASK) + | mmu_prop->last_mask | PAGE_PRESENT_MASK; + + /* Write the PTEs */ + hl_mmu_dr_write_final_pte(ctx, hop_pte_addr[hop_last], curr_pte); + + /* for each new hop, add its address to the table of previous-hop */ + for (i = 1 ; i <= hop_last ; i++) { + if (hop_new[i]) { + curr_pte = (hop_addr[i] & HOP_PHYS_ADDR_MASK) | PAGE_PRESENT_MASK; + hl_mmu_dr_write_pte(ctx, hop_pte_addr[i - 1], curr_pte); + + if (i - 1) + hl_mmu_dr_get_pte(ctx, hop_addr[i - 1]); + } + } + hl_mmu_dr_get_pte(ctx, hop_addr[hop_last]); + + return 0; + +err: + for (i = 1 ; i <= hop_last ; i++) + if (hop_new[i] && (hop_addr[i] != U64_MAX)) + hl_mmu_dr_free_hop(ctx, hop_addr[i]); + + return rc; +} + +/* + * hl_mmu_v2_swap_out - marks all mapping of the given ctx as swapped out + * + * @ctx: pointer to the context structure + * + */ +static void hl_mmu_v2_swap_out(struct hl_ctx *ctx) +{ + +} + +/* + * hl_mmu_v2_swap_in - marks all mapping of the given ctx as swapped in + * + * @ctx: pointer to the context structure + * + */ +static void hl_mmu_v2_swap_in(struct hl_ctx *ctx) +{ + +} + +static int hl_mmu_v2_get_tlb_info(struct hl_ctx *ctx, u64 virt_addr, struct hl_mmu_hop_info *hops) +{ + struct asic_fixed_properties *prop = &ctx->hdev->asic_prop; + struct hl_device *hdev = ctx->hdev; + struct hl_mmu_properties *mmu_prop; + bool is_dram_addr; + int i; + + is_dram_addr = hl_mem_area_inside_range(virt_addr, prop->dmmu.page_size, + prop->dmmu.start_addr, + prop->dmmu.end_addr); + + /* device resident in V2 are allowed only for HMMU */ + if (!is_dram_addr) + return -EINVAL; + + mmu_prop = &prop->dmmu; + hops->range_type = HL_VA_RANGE_TYPE_DRAM; + + hops->scrambled_vaddr = hdev->asic_funcs->scramble_addr(hdev, virt_addr); + + hops->hop_info[0].hop_addr = hl_mmu_dr_get_phys_hop0_addr(ctx); + hops->hop_info[0].hop_pte_addr = hl_mmu_get_hop_pte_phys_addr(ctx, mmu_prop, 0, + hops->hop_info[0].hop_addr, + hops->scrambled_vaddr); + if (hops->hop_info[0].hop_pte_addr == U64_MAX) + return -EFAULT; + + hops->hop_info[0].hop_pte_val = hdev->asic_funcs->read_pte(hdev, + hops->hop_info[0].hop_pte_addr); + if (hops->hop_info[0].hop_pte_val == U64_MAX) + return -EFAULT; + + for (i = 1 ; i < mmu_prop->num_hops ; i++) { + hops->hop_info[i].hop_addr = + hl_mmu_get_next_hop_addr(ctx, hops->hop_info[i - 1].hop_pte_val); + if (hops->hop_info[i].hop_addr == ULLONG_MAX) + return -EFAULT; + + hops->hop_info[i].hop_pte_addr = + hl_mmu_get_hop_pte_phys_addr(ctx, mmu_prop, i, + hops->hop_info[i].hop_addr, + hops->scrambled_vaddr); + if (hops->hop_info[i].hop_pte_addr == U64_MAX) + return -EFAULT; + + hops->hop_info[i].hop_pte_val = + hdev->asic_funcs->read_pte(hdev, + hops->hop_info[i].hop_pte_addr); + + if (!(hops->hop_info[i].hop_pte_val & PAGE_PRESENT_MASK)) + return -EFAULT; + + if (hops->hop_info[i].hop_pte_val & mmu_prop->last_mask) + break; + } + + /* if passed over all hops then no last hop was found */ + if (i == mmu_prop->num_hops) + return -EFAULT; + + if (!(hops->hop_info[i].hop_pte_val & PAGE_PRESENT_MASK)) + return -EFAULT; + + if (hops->scrambled_vaddr != virt_addr) + hops->unscrambled_paddr = hdev->asic_funcs->descramble_addr + (hdev, hops->hop_info[i].hop_pte_val); + else + hops->unscrambled_paddr = hops->hop_info[i].hop_pte_val; + + hops->used_hops = i + 1; + + return 0; +} + +/* + * hl_mmu_v2_prepare - prepare mmu_if for working with mmu v2 + * + * @hdev: pointer to the device structure + * @mmu_if: pointer to the mmu interface structure + */ +void hl_mmu_v2_set_funcs(struct hl_device *hdev, struct hl_mmu_funcs *mmu) +{ + mmu->init = hl_mmu_dr_init; + mmu->fini = hl_mmu_dr_fini; + mmu->ctx_init = hl_mmu_v2_ctx_init; + mmu->ctx_fini = hl_mmu_v2_ctx_fini; + mmu->map = hl_mmu_v2_map; + mmu->unmap = hl_mmu_v2_unmap; + mmu->flush = hl_mmu_dr_flush; + mmu->swap_out = hl_mmu_v2_swap_out; + mmu->swap_in = hl_mmu_v2_swap_in; + mmu->get_tlb_info = hl_mmu_v2_get_tlb_info; +} diff --git a/drivers/accel/habanalabs/gaudi/gaudi.c b/drivers/accel/habanalabs/gaudi/gaudi.c index 53292d4c15c8..dde3839fe0e0 100644 --- a/drivers/accel/habanalabs/gaudi/gaudi.c +++ b/drivers/accel/habanalabs/gaudi/gaudi.c @@ -649,6 +649,7 @@ static int gaudi_set_fixed_properties(struct hl_device *hdev) prop->dmmu.start_addr = (VA_HOST_SPACE_START + VA_HOST_SPACE_SIZE / 2); prop->dmmu.end_addr = VA_HOST_SPACE_END; prop->dmmu.page_size = PAGE_SIZE_2MB; + prop->dmmu.pgt_size = prop->mmu_pgt_size; prop->cfg_size = CFG_SIZE; prop->max_asid = MAX_ASID; diff --git a/drivers/accel/habanalabs/gaudi2/gaudi2.c b/drivers/accel/habanalabs/gaudi2/gaudi2.c index fd01525b1ea2..c95809120a07 100644 --- a/drivers/accel/habanalabs/gaudi2/gaudi2.c +++ b/drivers/accel/habanalabs/gaudi2/gaudi2.c @@ -2308,11 +2308,26 @@ static int set_number_of_functional_hbms(struct hl_device *hdev) return 0; } +bool gaudi2_is_edma_queue_id(u32 queue_id) +{ + + switch (queue_id) { + case GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE0_EDMA_1_3: + case GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE1_EDMA_1_3: + case GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE2_EDMA_1_3: + case GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE3_EDMA_1_3: + return true; + default: + return false; + } +} + static int gaudi2_set_dram_properties(struct hl_device *hdev) { struct asic_fixed_properties *prop = &hdev->asic_prop; - u32 basic_hbm_page_size; - int rc; + u64 hbm_drv_base_offset = 0, edma_pq_base_addr; + u32 basic_hbm_page_size, edma_idx = 0; + int rc, i; rc = set_number_of_functional_hbms(hdev); if (rc) @@ -2356,9 +2371,35 @@ static int gaudi2_set_dram_properties(struct hl_device *hdev) prop->dmmu.start_addr = prop->dram_base_address + (prop->dram_page_size * DIV_ROUND_UP_SECTOR_T(prop->dram_size, prop->dram_page_size)); - prop->dmmu.end_addr = prop->dmmu.start_addr + prop->dram_page_size * div_u64((VA_HBM_SPACE_END - prop->dmmu.start_addr), prop->dmmu.page_size); + /* + * Driver can't share an (48MB) HBM page with the F/W in order to prevent FW to block + * the driver part by range register, so it must start at the next (48MB) page + */ + hbm_drv_base_offset = roundup(CPU_FW_IMAGE_SIZE, prop->num_functional_hbms * SZ_8M); + + /* + * The NIC driver section size and the HMMU page tables section in the HBM needs + * to be the remaining size in the first dram page after taking into + * account the F/W image size + */ + + /* Reserve region in HBM for HMMU page tables */ + prop->mmu_pgt_addr = DRAM_PHYS_BASE + hbm_drv_base_offset + + ((prop->dram_page_size - hbm_drv_base_offset) - + (HMMU_PAGE_TABLES_SIZE + EDMA_PQS_SIZE + EDMA_SCRATCHPAD_SIZE)); + + /* Set EDMA PQs HBM addresses */ + edma_pq_base_addr = prop->mmu_pgt_addr + HMMU_PAGE_TABLES_SIZE; + + for (i = 0 ; i < GAUDI2_QUEUE_ID_CPU_PQ ; i++) { + if (gaudi2_is_edma_queue_id(i)) { + prop->hw_queues_props[i].q_dram_bd_address = edma_pq_base_addr + + (edma_idx * HL_QUEUE_SIZE_IN_BYTES); + edma_idx++; + } + } return 0; } @@ -2368,7 +2409,7 @@ static int gaudi2_set_fixed_properties(struct hl_device *hdev) struct asic_fixed_properties *prop = &hdev->asic_prop; struct hw_queue_properties *q_props; u32 num_sync_stream_queues = 0; - int i; + int i, rc; prop->max_queues = GAUDI2_QUEUE_ID_SIZE; prop->hw_queues_props = kcalloc(prop->max_queues, sizeof(struct hw_queue_properties), @@ -2391,6 +2432,9 @@ static int gaudi2_set_fixed_properties(struct hl_device *hdev) } q_props[i].cb_alloc_flags = CB_ALLOC_USER; + + if (gaudi2_is_edma_queue_id(i)) + q_props[i].dram_bd = 1; } q_props[GAUDI2_QUEUE_ID_CPU_PQ].type = QUEUE_TYPE_CPU; @@ -2419,40 +2463,39 @@ static int gaudi2_set_fixed_properties(struct hl_device *hdev) prop->rotator_enabled_mask = BIT(NUM_OF_ROT) - 1; - if (hdev->pldm) - prop->mmu_pgt_size = 0x800000; /* 8MB */ - else - prop->mmu_pgt_size = MMU_PAGE_TABLES_INITIAL_SIZE; + prop->max_asid = 2; + prop->dmmu.pgt_size = HMMU_PAGE_TABLES_SIZE; prop->mmu_pte_size = HL_PTE_SIZE; prop->mmu_hop_table_size = HOP_TABLE_SIZE_512_PTE; - prop->mmu_hop0_tables_total_size = HOP0_512_PTE_TABLES_TOTAL_SIZE; + prop->mmu_hop0_tables_total_size = HOP_TABLE_SIZE_512_PTE * prop->max_asid; prop->dmmu.hop_shifts[MMU_HOP0] = DHOP0_SHIFT; prop->dmmu.hop_shifts[MMU_HOP1] = DHOP1_SHIFT; prop->dmmu.hop_shifts[MMU_HOP2] = DHOP2_SHIFT; prop->dmmu.hop_shifts[MMU_HOP3] = DHOP3_SHIFT; - prop->dmmu.hop_shifts[MMU_HOP4] = DHOP4_SHIFT; prop->dmmu.hop_masks[MMU_HOP0] = DHOP0_MASK; prop->dmmu.hop_masks[MMU_HOP1] = DHOP1_MASK; prop->dmmu.hop_masks[MMU_HOP2] = DHOP2_MASK; prop->dmmu.hop_masks[MMU_HOP3] = DHOP3_MASK; - prop->dmmu.hop_masks[MMU_HOP4] = DHOP4_MASK; prop->dmmu.page_size = PAGE_SIZE_1GB; - prop->dmmu.num_hops = MMU_ARCH_6_HOPS; + prop->dmmu.num_hops = MMU_ARCH_4_HOPS; prop->dmmu.last_mask = LAST_MASK; - prop->dmmu.host_resident = 1; + prop->dmmu.host_resident = 0; prop->dmmu.hop_table_size = prop->mmu_hop_table_size; prop->dmmu.hop0_tables_total_size = prop->mmu_hop0_tables_total_size; - /* - * this is done in order to be able to validate FW descriptor (i.e. validating that - * the addresses and allocated space for FW image does not cross memory bounds). - * for this reason we set the DRAM size to the minimum possible and later it will - * be modified according to what reported in the cpucp info packet + /* As we need to set the pgt address in dram for HMMU init so we cannot + * wait to the fw cpucp info to set the dram props as mmu init comes before + * hw init */ - prop->dram_size = (GAUDI2_HBM_NUM - 1) * SZ_16G; + rc = hdev->asic_funcs->set_dram_properties(hdev); + if (rc) + goto free_qprops; + + prop->mmu_pgt_size = PMMU_PAGE_TABLES_SIZE; + prop->pmmu.pgt_size = prop->mmu_pgt_size; hdev->pmmu_huge_range = true; prop->pmmu.host_resident = 1; prop->pmmu.num_hops = MMU_ARCH_6_HOPS; @@ -2516,7 +2559,6 @@ static int gaudi2_set_fixed_properties(struct hl_device *hdev) prop->max_num_of_engines = GAUDI2_ENGINE_ID_SIZE; prop->num_engine_cores = CPU_ID_MAX; prop->cfg_size = CFG_SIZE; - prop->max_asid = MAX_ASID; prop->num_of_events = GAUDI2_EVENT_SIZE; prop->supports_engine_modes = true; @@ -2560,6 +2602,10 @@ static int gaudi2_set_fixed_properties(struct hl_device *hdev) prop->hbw_flush_reg = mmPCIE_WRAP_SPECIAL_GLBL_SPARE_0; return 0; + +free_qprops: + kfree(prop->hw_queues_props); + return rc; } static int gaudi2_pci_bars_map(struct hl_device *hdev) @@ -3033,6 +3079,25 @@ static int gaudi2_fetch_psoc_frequency(struct hl_device *hdev) return 0; } +static int gaudi2_mmu_clear_pgt_range(struct hl_device *hdev) +{ + struct gaudi2_device *gaudi2 = hdev->asic_specific; + struct asic_fixed_properties *prop = &hdev->asic_prop; + int rc; + + if (!(gaudi2->hw_cap_initialized & HW_CAP_MMU_MASK)) + return 0; + + if (prop->dmmu.host_resident) + return 0; + + rc = gaudi2_memset_device_memory(hdev, prop->mmu_pgt_addr, prop->dmmu.pgt_size, 0); + if (rc) + dev_err(hdev->dev, "Failed to clear mmu pgt"); + + return rc; +} + static int gaudi2_early_init(struct hl_device *hdev) { struct asic_fixed_properties *prop = &hdev->asic_prop; @@ -3258,6 +3323,12 @@ static int gaudi2_late_init(struct hl_device *hdev) goto disable_pci_access; } + rc = gaudi2_mmu_clear_pgt_range(hdev); + if (rc) { + dev_err(hdev->dev, "Failed to clear MMU page tables range\n"); + goto disable_pci_access; + } + gaudi2_init_arcs(hdev); rc = gaudi2_scrub_arcs_dccm(hdev); @@ -3697,13 +3768,7 @@ static int gaudi2_sw_init(struct hl_device *hdev) spin_lock_init(&gaudi2->hw_queues_lock); - gaudi2->scratchpad_kernel_address = hl_asic_dma_alloc_coherent(hdev, PAGE_SIZE, - &gaudi2->scratchpad_bus_address, - GFP_KERNEL | __GFP_ZERO); - if (!gaudi2->scratchpad_kernel_address) { - rc = -ENOMEM; - goto free_virt_msix_db_mem; - } + gaudi2->scratchpad_bus_address = prop->mmu_pgt_addr + HMMU_PAGE_TABLES_SIZE + EDMA_PQS_SIZE; gaudi2_user_mapped_blocks_init(hdev); @@ -3727,7 +3792,7 @@ static int gaudi2_sw_init(struct hl_device *hdev) rc = gaudi2_special_blocks_iterator_config(hdev); if (rc) - goto free_scratchpad_mem; + goto free_virt_msix_db_mem; rc = gaudi2_test_queues_msgs_alloc(hdev); if (rc) @@ -3737,9 +3802,6 @@ static int gaudi2_sw_init(struct hl_device *hdev) special_blocks_free: gaudi2_special_blocks_iterator_free(hdev); -free_scratchpad_mem: - hl_asic_dma_free_coherent(hdev, PAGE_SIZE, gaudi2->scratchpad_kernel_address, - gaudi2->scratchpad_bus_address); free_virt_msix_db_mem: hl_cpu_accessible_dma_pool_free(hdev, prop->pmmu.page_size, gaudi2->virt_msix_db_cpu_addr); free_cpu_accessible_dma_pool: @@ -3770,9 +3832,6 @@ static int gaudi2_sw_fini(struct hl_device *hdev) hl_asic_dma_free_coherent(hdev, HL_CPU_ACCESSIBLE_MEM_SIZE, hdev->cpu_accessible_dma_mem, hdev->cpu_accessible_dma_address); - hl_asic_dma_free_coherent(hdev, PAGE_SIZE, gaudi2->scratchpad_kernel_address, - gaudi2->scratchpad_bus_address); - dma_pool_destroy(hdev->dma_pool); kfree(gaudi2); @@ -4962,10 +5021,17 @@ static void gaudi2_init_qman_pq(struct hl_device *hdev, u32 reg_base, q = &hdev->kernel_queues[queue_id_base + pq_id]; pq_offset = pq_id * 4; - WREG32(reg_base + QM_PQ_BASE_LO_0_OFFSET + pq_offset, - lower_32_bits(q->bus_address)); - WREG32(reg_base + QM_PQ_BASE_HI_0_OFFSET + pq_offset, - upper_32_bits(q->bus_address)); + if (q->dram_bd) { + WREG32(reg_base + QM_PQ_BASE_LO_0_OFFSET + pq_offset, + lower_32_bits(q->pq_dram_address)); + WREG32(reg_base + QM_PQ_BASE_HI_0_OFFSET + pq_offset, + upper_32_bits(q->pq_dram_address)); + } else { + WREG32(reg_base + QM_PQ_BASE_LO_0_OFFSET + pq_offset, + lower_32_bits(q->bus_address)); + WREG32(reg_base + QM_PQ_BASE_HI_0_OFFSET + pq_offset, + upper_32_bits(q->bus_address)); + } WREG32(reg_base + QM_PQ_SIZE_0_OFFSET + pq_offset, ilog2(HL_QUEUE_LENGTH)); WREG32(reg_base + QM_PQ_PI_0_OFFSET + pq_offset, 0); WREG32(reg_base + QM_PQ_CI_0_OFFSET + pq_offset, 0); @@ -5852,7 +5918,8 @@ static int gaudi2_mmu_invalidate_cache_range(struct hl_device *hdev, bool is_har return rc; } -static int gaudi2_mmu_update_hop0_addr(struct hl_device *hdev, u32 stlb_base) +static int gaudi2_mmu_update_hop0_addr(struct hl_device *hdev, u32 stlb_base, + bool host_resident_pgt) { struct asic_fixed_properties *prop = &hdev->asic_prop; u64 hop0_addr; @@ -5864,7 +5931,11 @@ static int gaudi2_mmu_update_hop0_addr(struct hl_device *hdev, u32 stlb_base) max_asid = min((u32) 8, max_asid); for (asid = 0 ; asid < max_asid ; asid++) { - hop0_addr = hdev->mmu_priv.hr.mmu_asid_hop0[asid].phys_addr; + if (host_resident_pgt) + hop0_addr = hdev->mmu_priv.hr.mmu_asid_hop0[asid].phys_addr; + else + hop0_addr = prop->mmu_pgt_addr + (asid * prop->mmu_hop_table_size); + rc = gaudi2_mmu_update_asid_hop0_addr(hdev, stlb_base, asid, hop0_addr); if (rc) { dev_err(hdev->dev, "failed to set hop0 addr for asid %d\n", asid); @@ -5875,7 +5946,8 @@ static int gaudi2_mmu_update_hop0_addr(struct hl_device *hdev, u32 stlb_base) return 0; } -static int gaudi2_mmu_init_common(struct hl_device *hdev, u32 mmu_base, u32 stlb_base) +static int gaudi2_mmu_init_common(struct hl_device *hdev, u32 mmu_base, u32 stlb_base, + bool host_resident_pgt) { u32 status, timeout_usec; int rc; @@ -5898,7 +5970,7 @@ static int gaudi2_mmu_init_common(struct hl_device *hdev, u32 mmu_base, u32 stlb if (rc) dev_notice_ratelimited(hdev->dev, "Timeout when waiting for MMU SRAM init\n"); - rc = gaudi2_mmu_update_hop0_addr(hdev, stlb_base); + rc = gaudi2_mmu_update_hop0_addr(hdev, stlb_base, host_resident_pgt); if (rc) return rc; @@ -5922,6 +5994,7 @@ static int gaudi2_mmu_init_common(struct hl_device *hdev, u32 mmu_base, u32 stlb static int gaudi2_pci_mmu_init(struct hl_device *hdev) { + struct asic_fixed_properties *prop = &hdev->asic_prop; struct gaudi2_device *gaudi2 = hdev->asic_specific; u32 mmu_base, stlb_base; int rc; @@ -5961,7 +6034,7 @@ static int gaudi2_pci_mmu_init(struct hl_device *hdev) WREG32(mmu_base + MMU_SPI_SEI_MASK_OFFSET, GAUDI2_PMMU_SPI_SEI_ENABLE_MASK); - rc = gaudi2_mmu_init_common(hdev, mmu_base, stlb_base); + rc = gaudi2_mmu_init_common(hdev, mmu_base, stlb_base, prop->pmmu.host_resident); if (rc) return rc; @@ -6013,7 +6086,7 @@ static int gaudi2_dcore_hmmu_init(struct hl_device *hdev, int dcore_id, WREG32(mmu_base + MMU_SPI_SEI_MASK_OFFSET, GAUDI2_HMMU_SPI_SEI_ENABLE_MASK); - rc = gaudi2_mmu_init_common(hdev, mmu_base, stlb_base); + rc = gaudi2_mmu_init_common(hdev, mmu_base, stlb_base, prop->dmmu.host_resident); if (rc) return rc; @@ -7051,7 +7124,7 @@ static int gaudi2_test_queues(struct hl_device *hdev) /* send test message on all enabled Qs */ for (i = GAUDI2_QUEUE_ID_PDMA_0_0 ; i < GAUDI2_QUEUE_ID_CPU_PQ; i++) { - if (!gaudi2_is_queue_enabled(hdev, i)) + if (!gaudi2_is_queue_enabled(hdev, i) || gaudi2_is_edma_queue_id(i)) continue; msg_info = &gaudi2->queues_test_info[i - GAUDI2_QUEUE_ID_PDMA_0_0]; @@ -7068,7 +7141,7 @@ static int gaudi2_test_queues(struct hl_device *hdev) /* verify that all messages were processed */ for (i = GAUDI2_QUEUE_ID_PDMA_0_0 ; i < GAUDI2_QUEUE_ID_CPU_PQ; i++) { - if (!gaudi2_is_queue_enabled(hdev, i)) + if (!gaudi2_is_queue_enabled(hdev, i) || gaudi2_is_edma_queue_id(i)) continue; rc = gaudi2_test_queue_wait_completion(hdev, i, sob_val); @@ -8988,7 +9061,6 @@ static void gaudi2_handle_page_error(struct hl_device *hdev, u64 mmu_base, bool if (is_pmmu) { dev_err_ratelimited(hdev->dev, "PMMU page fault on va 0x%llx\n", addr); } else { - addr = gaudi2_mmu_descramble_addr(hdev, addr); addr &= HW_UNSCRAMBLED_BITS_MASK; dev_err_ratelimited(hdev->dev, "HMMU page fault on va range 0x%llx - 0x%llx\n", @@ -10255,11 +10327,11 @@ static void gaudi2_handle_eqe(struct hl_device *hdev, struct hl_eq_entry *eq_ent } static int gaudi2_memset_memory_chunk_using_edma_qm(struct hl_device *hdev, - struct packet_lin_dma *lin_dma_pkt, dma_addr_t pkt_dma_addr, - u32 hw_queue_id, u32 size, u64 addr, u32 val) + struct packet_lin_dma *lin_dma_pkt, + u64 phys_addr, u32 hw_queue_id, u32 size, u64 addr, u32 val) { u32 ctl, pkt_size; - int rc = 0; + int rc = 0, i; ctl = FIELD_PREP(GAUDI2_PKT_CTL_OPCODE_MASK, PACKET_LIN_DMA); ctl |= FIELD_PREP(GAUDI2_PKT_LIN_DMA_CTL_MEMSET_MASK, 1); @@ -10273,7 +10345,12 @@ static int gaudi2_memset_memory_chunk_using_edma_qm(struct hl_device *hdev, pkt_size = sizeof(struct packet_lin_dma); - rc = hl_hw_queue_send_cb_no_cmpl(hdev, hw_queue_id, pkt_size, pkt_dma_addr); + for (i = 0; i < 3; i++) + rc = hdev->asic_funcs->access_dev_mem(hdev, PCI_REGION_DRAM, + phys_addr + (i * sizeof(u64)), + ((u64 *)(lin_dma_pkt)) + i, DEBUGFS_WRITE64); + + rc = hl_hw_queue_send_cb_no_cmpl(hdev, hw_queue_id, pkt_size, phys_addr); if (rc) dev_err(hdev->dev, "Failed to send lin dma packet to H/W queue %d\n", hw_queue_id); @@ -10288,12 +10365,11 @@ static int gaudi2_memset_device_memory(struct hl_device *hdev, u64 addr, u64 siz GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0, GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0}; u32 chunk_size, dcore, edma_idx, sob_offset, sob_addr, comp_val, - old_mmubp, mmubp, num_of_pkts, busy, pkt_size; + old_mmubp, mmubp, num_of_pkts, busy, pkt_size, cb_len; u64 comp_addr, cur_addr = addr, end_addr = addr + size; struct asic_fixed_properties *prop = &hdev->asic_prop; + int rc = 0, dma_num = 0, i; void *lin_dma_pkts_arr; - dma_addr_t pkt_dma_addr; - int rc = 0, dma_num = 0; if (prop->edma_enabled_mask == 0) { dev_info(hdev->dev, "non of the EDMA engines is enabled - skip dram scrubbing\n"); @@ -10311,9 +10387,19 @@ static int gaudi2_memset_device_memory(struct hl_device *hdev, u64 addr, u64 siz /* Calculate how many lin dma pkts we'll need */ num_of_pkts = div64_u64(round_up(size, SZ_2G), SZ_2G); pkt_size = sizeof(struct packet_lin_dma); + cb_len = pkt_size * num_of_pkts; + + /* + * if we're not scrubing HMMU or NIC reserved sections in hbm, + * then it the scrubing of the user section, as we use the start of the user section + * to store the CB of the EDMA QM, so shift the start address of the scrubbing accordingly + * and scrub the CB section before leaving this function. + */ + if ((addr >= prop->dram_user_base_address) && + (addr < prop->dram_user_base_address + cb_len)) + cur_addr += (prop->dram_user_base_address + cb_len) - addr; - lin_dma_pkts_arr = hl_asic_dma_alloc_coherent(hdev, pkt_size * num_of_pkts, - &pkt_dma_addr, GFP_KERNEL); + lin_dma_pkts_arr = kvcalloc(num_of_pkts, pkt_size, GFP_KERNEL); if (!lin_dma_pkts_arr) return -ENOMEM; @@ -10359,7 +10445,7 @@ static int gaudi2_memset_device_memory(struct hl_device *hdev, u64 addr, u64 siz rc = gaudi2_memset_memory_chunk_using_edma_qm(hdev, (struct packet_lin_dma *)lin_dma_pkts_arr + dma_num, - pkt_dma_addr + dma_num * pkt_size, + prop->dram_user_base_address + (dma_num * pkt_size), edma_queues_id[dcore] + edma_idx * 4, chunk_size, cur_addr, val); if (rc) @@ -10368,14 +10454,16 @@ static int gaudi2_memset_device_memory(struct hl_device *hdev, u64 addr, u64 siz dma_num++; cur_addr += chunk_size; if (cur_addr == end_addr) - break; + goto edma_wait; } } } +edma_wait: rc = hl_poll_timeout(hdev, sob_addr, busy, (busy == dma_num), 1000, 1000000); if (rc) { - dev_err(hdev->dev, "DMA Timeout during HBM scrubbing\n"); + dev_err(hdev->dev, "DMA Timeout during HBM scrubbing(sob: 0x%x, dma_num: 0x%x)\n", + busy, dma_num); goto end; } end: @@ -10396,8 +10484,16 @@ static int gaudi2_memset_device_memory(struct hl_device *hdev, u64 addr, u64 siz } } + memset(lin_dma_pkts_arr, 0, sizeof(u64)); + + /* Zero the HBM area where we copied the CB */ + for (i = 0; i < cb_len / sizeof(u64); i += sizeof(u64)) + rc = hdev->asic_funcs->access_dev_mem(hdev, PCI_REGION_DRAM, + prop->dram_user_base_address + i, + (u64 *)(lin_dma_pkts_arr), DEBUGFS_WRITE64); WREG32(sob_addr, 0); - hl_asic_dma_free_coherent(hdev, pkt_size * num_of_pkts, lin_dma_pkts_arr, pkt_dma_addr); + + kfree(lin_dma_pkts_arr); return rc; } @@ -11455,7 +11551,7 @@ static int gaudi2_mmu_get_real_page_size(struct hl_device *hdev, struct hl_mmu_p return 0; page_size_err: - dev_err(hdev->dev, "page size of %u is not %uKB aligned, can't map\n", + dev_err(hdev->dev, "page size of 0x%X is not 0x%X aligned, can't map\n", page_size, mmu_prop->page_size >> 10); return -EFAULT; } @@ -11475,6 +11571,29 @@ int gaudi2_send_device_activity(struct hl_device *hdev, bool open) return hl_fw_send_device_activity(hdev, open); } +static u64 gaudi2_read_pte(struct hl_device *hdev, u64 addr) +{ + struct gaudi2_device *gaudi2 = hdev->asic_specific; + u64 val; + + if (hdev->reset_info.hard_reset_pending) + return U64_MAX; + + val = readq(hdev->pcie_bar[DRAM_BAR_ID] + (addr - gaudi2->dram_bar_cur_addr)); + + return val; +} + +static void gaudi2_write_pte(struct hl_device *hdev, u64 addr, u64 val) +{ + struct gaudi2_device *gaudi2 = hdev->asic_specific; + + if (hdev->reset_info.hard_reset_pending) + return; + + writeq(val, hdev->pcie_bar[DRAM_BAR_ID] + (addr - gaudi2->dram_bar_cur_addr)); +} + static const struct hl_asic_funcs gaudi2_funcs = { .early_init = gaudi2_early_init, .early_fini = gaudi2_early_fini, @@ -11511,8 +11630,8 @@ static const struct hl_asic_funcs gaudi2_funcs = { .add_device_attr = gaudi2_add_device_attr, .handle_eqe = gaudi2_handle_eqe, .get_events_stat = gaudi2_get_events_stat, - .read_pte = NULL, - .write_pte = NULL, + .read_pte = gaudi2_read_pte, + .write_pte = gaudi2_write_pte, .mmu_invalidate_cache = gaudi2_mmu_invalidate_cache, .mmu_invalidate_cache_range = gaudi2_mmu_invalidate_cache_range, .mmu_prefetch_cache_range = NULL, diff --git a/drivers/accel/habanalabs/gaudi2/gaudi2P.h b/drivers/accel/habanalabs/gaudi2/gaudi2P.h index 9b9eef0d97d6..bc508c9cee5c 100644 --- a/drivers/accel/habanalabs/gaudi2/gaudi2P.h +++ b/drivers/accel/habanalabs/gaudi2/gaudi2P.h @@ -19,8 +19,6 @@ #define GAUDI2_LINUX_FW_FILE "habanalabs/gaudi2/gaudi2-fit.itb" #define GAUDI2_BOOT_FIT_FILE "habanalabs/gaudi2/gaudi2-boot-fit.itb" -#define MMU_PAGE_TABLES_INITIAL_SIZE 0x10000000 /* 256MB */ - #define GAUDI2_CPU_TIMEOUT_USEC 30000000 /* 30s */ #define NUMBER_OF_PDMA_QUEUES 2 @@ -109,13 +107,11 @@ /* DRAM Memory Map */ #define CPU_FW_IMAGE_SIZE 0x10000000 /* 256MB */ - -/* This define should be used only when working in a debug mode without dram. - * When working with dram, the driver size will be calculated dynamically. - */ -#define NIC_DEFAULT_DRV_SIZE 0x20000000 /* 512MB */ - #define CPU_FW_IMAGE_ADDR DRAM_PHYS_BASE +#define PMMU_PAGE_TABLES_SIZE 0x10000000 /* 256MB */ +#define EDMA_PQS_SIZE SZ_2M +#define EDMA_SCRATCHPAD_SIZE SZ_1M +#define HMMU_PAGE_TABLES_SIZE SZ_1M #define NIC_NUMBER_OF_PORTS NIC_NUMBER_OF_ENGINES diff --git a/drivers/accel/habanalabs/include/hw_ip/mmu/mmu_general.h b/drivers/accel/habanalabs/include/hw_ip/mmu/mmu_general.h index d408feecd483..b4a5e95be354 100644 --- a/drivers/accel/habanalabs/include/hw_ip/mmu/mmu_general.h +++ b/drivers/accel/habanalabs/include/hw_ip/mmu/mmu_general.h @@ -26,6 +26,8 @@ #define LAST_MASK 0x0000000000800ull #define FLAGS_MASK 0x0000000000FFFull +#define MMU_ARCH_3_HOPS 3 +#define MMU_ARCH_4_HOPS 4 #define MMU_ARCH_5_HOPS 5 #define MMU_ARCH_6_HOPS 6 From patchwork Tue Jan 2 15:06:54 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Oded Gabbay X-Patchwork-Id: 184386 Return-Path: Delivered-To: ouuuleilei@gmail.com Received: by 2002:a05:7301:6f82:b0:100:9c79:88ff with SMTP id tb2csp4499839dyb; Tue, 2 Jan 2024 07:08:46 -0800 (PST) X-Google-Smtp-Source: AGHT+IHJFjYcV0IdFBkR2lAmyzwSX8IaOWM5dLnKoej5dkXD2QTgEes47f+EfnncsuicqAY2s7Eh X-Received: by 2002:a05:6870:3929:b0:203:6297:b554 with SMTP id b41-20020a056870392900b002036297b554mr17170465oap.111.1704208125995; Tue, 02 Jan 2024 07:08:45 -0800 (PST) ARC-Seal: i=1; a=rsa-sha256; t=1704208125; cv=none; d=google.com; s=arc-20160816; b=vOziftnzRNDcFqHfB7eDjplb06B0lmhEY4fg8ZHnzRzHhcbzmPLlEqwDSJqrYGgTrm GbwfiIqHc/7AiFTuYC1GrSsAijHJB9Ddrrgj1SOTmY2nOZNwr+tRuTR6fDmu8bXGQag+ 1eH/yAsAePlF4k7n2revipJUV6HGwFyuYmzKZxnMRMNrfpqrAetxiaHrZg74PuXTMRNU Q775n360SUK7+3S++DDGChEAyQjOlnUUXyvVr3lXg/OqUA6zBZM7n3DwG2G4OQPK1N4P mfUxzWGHr4+RKg0xjTH4Qf3mu9760lN5qvAZfnJcen9NqFfvU/IF/WmewLNmIgcifxUa U8YA== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=content-transfer-encoding:mime-version:list-unsubscribe :list-subscribe:list-id:precedence:references:in-reply-to:message-id :date:subject:cc:to:from:dkim-signature; bh=KnjyZ8tLmXwFVZkLvKYIRzeuewg9lZUFf8BqUVY7Q7I=; fh=PceJZ/KGF6ujSpGwn68DXpj5qiXzqwvBM4Q702GUgLM=; b=sRkofD5TiwAJv03bzhuinZl4meqCCqBHbZwpAIloSlUk78BDqA7e2rFosxMYCNRX/C eEZREoWnKb52b1hxcEhZGW1Hqmb+ViGpxW1IQU797HBl/nYikyJgqGPMrbNQHGaCRBcq p/ZQpAQiFCH4mle6SJL3Y0RUxFhafoegEObOQaKpnjlnWf4lRwAibZ+8OgWTSKx3J7cH 3abdn4eex9cw7TZ641hKn/hlf2Ps3lBMn487IMEg7G+fWMRw12LcJqUtqcOUzY50sVqv RyKEr/7+nNBCQf4ir7OHXbVJygsA2iFdPmQxmNWDOY1OHqbRLEBiWM3PsTdcvOEz84t6 PwnQ== ARC-Authentication-Results: i=1; mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=KP1U0QKO; spf=pass (google.com: domain of linux-kernel+bounces-14532-ouuuleilei=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14532-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from ny.mirrors.kernel.org (ny.mirrors.kernel.org. [147.75.199.223]) by mx.google.com with ESMTPS id w2-20020a1f6202000000b004b71865ad31si3237791vkb.130.2024.01.02.07.08.45 for (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Tue, 02 Jan 2024 07:08:45 -0800 (PST) Received-SPF: pass (google.com: domain of linux-kernel+bounces-14532-ouuuleilei=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223; Authentication-Results: mx.google.com; dkim=pass header.i=@kernel.org header.s=k20201202 header.b=KP1U0QKO; spf=pass (google.com: domain of linux-kernel+bounces-14532-ouuuleilei=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) smtp.mailfrom="linux-kernel+bounces-14532-ouuuleilei=gmail.com@vger.kernel.org"; dmarc=pass (p=NONE sp=NONE dis=NONE) header.from=kernel.org Received: from smtp.subspace.kernel.org (wormhole.subspace.kernel.org [52.25.139.140]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ny.mirrors.kernel.org (Postfix) with ESMTPS id B17CF1C223FF for ; Tue, 2 Jan 2024 15:08:45 +0000 (UTC) Received: from localhost.localdomain (localhost.localdomain [127.0.0.1]) by smtp.subspace.kernel.org (Postfix) with ESMTP id 0ADC2156EE; Tue, 2 Jan 2024 15:07:12 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="KP1U0QKO" X-Original-To: linux-kernel@vger.kernel.org Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 7A080156C4 for ; Tue, 2 Jan 2024 15:07:09 +0000 (UTC) Received: by smtp.kernel.org (Postfix) with ESMTPSA id 38FA2C433C9; Tue, 2 Jan 2024 15:07:08 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1704208029; bh=8W1aookR1BcvR/0gXIR7y7VRCd9tBn4MkxLxXyNZgJs=; h=From:To:Cc:Subject:Date:In-Reply-To:References:From; b=KP1U0QKObEFiNMkX3FGzm2UxsHn/uU05/ePalYvzSZ7YucStUmiOsGh7gw8A03mp+ i4Y2X9Qto03hggabk2/UiA0RgME3yeAcVm8wnGRpcPz+kpBEubtP6tqdcaQunpbgZx V1KYlAZPDom3hrgQgvUPfa693RmUuKE5fYXtUcY1F4AdaW5axCtiIq/F5g3wsoHgaB H1GoMJglQ2SRVUSKU+7g5w5NvKJIPu0il6f6Qeg5t6LWLY8OFlK3ucyG2jUJDyArZ/ c2aKJR7qN5RdpwYyA/dIDawiXjPKgOBui7Pff4DxBMXS4fraILosRSV2jPzIitf/E5 SoOWo5BeBfa7w== From: Oded Gabbay To: dri-devel@lists.freedesktop.org, linux-kernel@vger.kernel.org Cc: Tomer Tayar Subject: [PATCH 6/6] accel/habanalabs: abort device reset for consecutive heartbeat failures Date: Tue, 2 Jan 2024 17:06:54 +0200 Message-Id: <20240102150654.522555-6-ogabbay@kernel.org> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20240102150654.522555-1-ogabbay@kernel.org> References: <20240102150654.522555-1-ogabbay@kernel.org> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 X-getmail-retrieved-from-mailbox: INBOX X-GMAIL-THRID: 1786991739877889866 X-GMAIL-MSGID: 1786991739877889866 From: Tomer Tayar The mechanism of aborting device reset for consecutive fatal errors is currently only for fatal errors that are reported by FW. A non-responsive FW and consecutive heartbeat failures is also considered fatal, so add them as well to this mechanism to avoid recurring device reset in such a case. Signed-off-by: Tomer Tayar Reviewed-by: Oded Gabbay Signed-off-by: Oded Gabbay --- drivers/accel/habanalabs/common/device.c | 14 ++++++++------ 1 file changed, 8 insertions(+), 6 deletions(-) diff --git a/drivers/accel/habanalabs/common/device.c b/drivers/accel/habanalabs/common/device.c index 15891de6cf39..581fc99ad89b 100644 --- a/drivers/accel/habanalabs/common/device.c +++ b/drivers/accel/habanalabs/common/device.c @@ -1769,14 +1769,16 @@ int hl_device_reset(struct hl_device *hdev, u32 flags) hdev->device_cpu_disabled = false; hdev->reset_info.hard_reset_pending = false; + /* + * Put the device in an unusable state if there are 2 back to back resets due to + * fatal errors. + */ if (hdev->reset_info.reset_trigger_repeated && - (hdev->reset_info.prev_reset_trigger == - HL_DRV_RESET_FW_FATAL_ERR)) { - /* if there 2 back to back resets from FW, - * ensure driver puts the driver in a unusable state - */ + (hdev->reset_info.prev_reset_trigger == HL_DRV_RESET_FW_FATAL_ERR || + hdev->reset_info.prev_reset_trigger == + HL_DRV_RESET_HEARTBEAT)) { dev_crit(hdev->dev, - "%s Consecutive FW fatal errors received, stopping hard reset\n", + "%s Consecutive fatal errors, stopping hard reset\n", dev_name(&(hdev)->pdev->dev)); rc = -EIO; goto out_err;