From patchwork Thu Apr 13 14:26:13 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Jonathan Cameron X-Patchwork-Id: 13210349 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id E429FC77B61 for ; Thu, 13 Apr 2023 14:26:51 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S229936AbjDMO0v (ORCPT ); Thu, 13 Apr 2023 10:26:51 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:56332 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S231217AbjDMO0u (ORCPT ); Thu, 13 Apr 2023 10:26:50 -0400 Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id B1B6B9EE3; Thu, 13 Apr 2023 07:26:49 -0700 (PDT) Received: from lhrpeml500005.china.huawei.com (unknown [172.18.147.207]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4Py21S54Z5z67NYF; Thu, 13 Apr 2023 22:25:48 +0800 (CST) Received: from SecurePC-101-06.china.huawei.com (10.122.247.231) by lhrpeml500005.china.huawei.com (7.191.163.240) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2507.23; Thu, 13 Apr 2023 15:26:46 +0100 From: Jonathan Cameron To: Liang Kan , , , , CC: , , , , , , Davidlohr Bueso , Dave Jiang Subject: [PATCH v6 1/5] perf: Allow a PMU to have a parent Date: Thu, 13 Apr 2023 15:26:13 +0100 Message-ID: <20230413142617.15995-2-Jonathan.Cameron@huawei.com> X-Mailer: git-send-email 2.37.2 In-Reply-To: <20230413142617.15995-1-Jonathan.Cameron@huawei.com> References: <20230413142617.15995-1-Jonathan.Cameron@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.122.247.231] X-ClientProxiedBy: lhrpeml500001.china.huawei.com (7.191.163.213) To lhrpeml500005.china.huawei.com (7.191.163.240) X-CFilter-Loop: Reflected Precedence: bulk List-ID: X-Mailing-List: linux-cxl@vger.kernel.org Some PMUs have well defined parents such as PCI devices. As the device_initialize() and device_add() are all within pmu_dev_alloc() which is called from perf_pmu_register() there is no opportunity to set the parent from within a driver. Add a struct device *parent field to struct pmu and use that to set the parent. Signed-off-by: Jonathan Cameron Reviewed-by: Dan Williams Reviewed-by: Greg Kroah-Hartman Acked-by: Peter Zijlstra (Intel) --- v6: No change v5: Move to head of series as may merge as part of: https://lore.kernel.org/all/20230404134225.13408-1-Jonathan.Cameron@huawei.com/ --- include/linux/perf_event.h | 1 + kernel/events/core.c | 1 + 2 files changed, 2 insertions(+) diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h index d5628a7b5eaa..b99db1eda72c 100644 --- a/include/linux/perf_event.h +++ b/include/linux/perf_event.h @@ -303,6 +303,7 @@ struct pmu { struct module *module; struct device *dev; + struct device *parent; const struct attribute_group **attr_groups; const struct attribute_group **attr_update; const char *name; diff --git a/kernel/events/core.c b/kernel/events/core.c index 435815d3be3f..32079a332480 100644 --- a/kernel/events/core.c +++ b/kernel/events/core.c @@ -11367,6 +11367,7 @@ static int pmu_dev_alloc(struct pmu *pmu) dev_set_drvdata(pmu->dev, pmu); pmu->dev->bus = &pmu_bus; + pmu->dev->parent = pmu->parent; pmu->dev->release = pmu_dev_release; ret = dev_set_name(pmu->dev, "%s", pmu->name); From patchwork Thu Apr 13 14:26:14 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Jonathan Cameron X-Patchwork-Id: 13210350 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 60A2CC77B61 for ; Thu, 13 Apr 2023 14:27:22 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S231466AbjDMO1V (ORCPT ); Thu, 13 Apr 2023 10:27:21 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:56928 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S230084AbjDMO1U (ORCPT ); Thu, 13 Apr 2023 10:27:20 -0400 Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id D460A9EE0; Thu, 13 Apr 2023 07:27:19 -0700 (PDT) Received: from lhrpeml500005.china.huawei.com (unknown [172.18.147.200]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4Py20N0l6Xz6J7JS; Thu, 13 Apr 2023 22:24:52 +0800 (CST) Received: from SecurePC-101-06.china.huawei.com (10.122.247.231) by lhrpeml500005.china.huawei.com (7.191.163.240) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2507.23; Thu, 13 Apr 2023 15:27:17 +0100 From: Jonathan Cameron To: Liang Kan , , , , CC: , , , , , , Davidlohr Bueso , Dave Jiang Subject: [PATCH v6 2/5] cxl: Add functions to get an instance of / count regblocks of a given type Date: Thu, 13 Apr 2023 15:26:14 +0100 Message-ID: <20230413142617.15995-3-Jonathan.Cameron@huawei.com> X-Mailer: git-send-email 2.37.2 In-Reply-To: <20230413142617.15995-1-Jonathan.Cameron@huawei.com> References: <20230413142617.15995-1-Jonathan.Cameron@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.122.247.231] X-ClientProxiedBy: lhrpeml500001.china.huawei.com (7.191.163.213) To lhrpeml500005.china.huawei.com (7.191.163.240) X-CFilter-Loop: Reflected Precedence: bulk List-ID: X-Mailing-List: linux-cxl@vger.kernel.org Until the recently release CXL 3.0 specification, there was only ever one instance of any given register block pointed to by the Register Block Locator DVSEC. Now, the specification allows for multiple CXL PMU instances, each with their own register block. To enable this add cxl_find_regblock_instance() that takes an index parameter and use that to implement cxl_count_regblock() and cxl_find_regblock(). Signed-off-by: Jonathan Cameron --- v6: No change Dropped RBs as significant refactor for v5. v5: Thanks to Dan Williams, - As there is only one case of repeated regblocks so far keep cxl_find_regblock() as finding the first instance but implement it using a new cxl_find_regblock_instance() function, also used to implement the counting function. --- drivers/cxl/core/regs.c | 59 ++++++++++++++++++++++++++++++++++++----- drivers/cxl/cxl.h | 3 +++ 2 files changed, 56 insertions(+), 6 deletions(-) diff --git a/drivers/cxl/core/regs.c b/drivers/cxl/core/regs.c index 1476a0299c9b..4b9672db867d 100644 --- a/drivers/cxl/core/regs.c +++ b/drivers/cxl/core/regs.c @@ -286,20 +286,23 @@ static bool cxl_decode_regblock(struct pci_dev *pdev, u32 reg_lo, u32 reg_hi, } /** - * cxl_find_regblock() - Locate register blocks by type + * cxl_find_regblock_instance() - Locate a register block by type / index * @pdev: The CXL PCI device to enumerate. * @type: Register Block Indicator id * @map: Enumeration output, clobbered on error + * @index: Index into which particular instance of a regblock wanted in the + * order found in register locator DVSEC. * * Return: 0 if register block enumerated, negative error code otherwise * * A CXL DVSEC may point to one or more register blocks, search for them - * by @type. + * by @type and @index. */ -int cxl_find_regblock(struct pci_dev *pdev, enum cxl_regloc_type type, - struct cxl_register_map *map) +int cxl_find_regblock_instance(struct pci_dev *pdev, enum cxl_regloc_type type, + struct cxl_register_map *map, int index) { u32 regloc_size, regblocks; + int instance = 0; int regloc, i; map->resource = CXL_RESOURCE_NONE; @@ -323,15 +326,59 @@ int cxl_find_regblock(struct pci_dev *pdev, enum cxl_regloc_type type, if (!cxl_decode_regblock(pdev, reg_lo, reg_hi, map)) continue; - if (map->reg_type == type) - return 0; + if (map->reg_type == type) { + if (index == instance) + return 0; + instance++; + } } map->resource = CXL_RESOURCE_NONE; return -ENODEV; } +EXPORT_SYMBOL_NS_GPL(cxl_find_regblock_instance, CXL); + +/** + * cxl_find_regblock() - Locate register blocks by type + * @pdev: The CXL PCI device to enumerate. + * @type: Register Block Indicator id + * @map: Enumeration output, clobbered on error + * + * Return: 0 if register block enumerated, negative error code otherwise + * + * A CXL DVSEC may point to one or more register blocks, search for them + * by @type. + */ +int cxl_find_regblock(struct pci_dev *pdev, enum cxl_regloc_type type, + struct cxl_register_map *map) +{ + return cxl_find_regblock_instance(pdev, type, map, 0); +} EXPORT_SYMBOL_NS_GPL(cxl_find_regblock, CXL); +/** + * cxl_count_regblock() - Count instances of a given regblock type. + * @pdev: The CXL PCI device to enumerate. + * @type: Register Block Indicator id + * + * Some regblocks may be repeated. Count how many instances. + * + * Return: count of matching regblocks. + */ +int cxl_count_regblock(struct pci_dev *pdev, enum cxl_regloc_type type) +{ + struct cxl_register_map map; + int rc, count = 0; + + while (1) { + rc = cxl_find_regblock_instance(pdev, type, &map, count); + if (rc) + return count; + count++; + } +} +EXPORT_SYMBOL_NS_GPL(cxl_count_regblock, CXL); + resource_size_t cxl_rcrb_to_component(struct device *dev, resource_size_t rcrb, enum cxl_rcrb which) diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h index 044a92d9813e..f6e2a9ea5f41 100644 --- a/drivers/cxl/cxl.h +++ b/drivers/cxl/cxl.h @@ -260,6 +260,9 @@ int cxl_map_device_regs(struct device *dev, struct cxl_device_regs *regs, struct cxl_register_map *map); enum cxl_regloc_type; +int cxl_count_regblock(struct pci_dev *pdev, enum cxl_regloc_type type); +int cxl_find_regblock_instance(struct pci_dev *pdev, enum cxl_regloc_type type, + struct cxl_register_map *map, int index); int cxl_find_regblock(struct pci_dev *pdev, enum cxl_regloc_type type, struct cxl_register_map *map); From patchwork Thu Apr 13 14:26:15 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Jonathan Cameron X-Patchwork-Id: 13210351 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 4240AC77B6E for ; Thu, 13 Apr 2023 14:27:54 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S229900AbjDMO1x (ORCPT ); Thu, 13 Apr 2023 10:27:53 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:57432 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S230137AbjDMO1w (ORCPT ); Thu, 13 Apr 2023 10:27:52 -0400 Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 76BEE9EEE; Thu, 13 Apr 2023 07:27:50 -0700 (PDT) Received: from lhrpeml500005.china.huawei.com (unknown [172.18.147.200]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4Py20y52SGz6J7JS; Thu, 13 Apr 2023 22:25:22 +0800 (CST) Received: from SecurePC-101-06.china.huawei.com (10.122.247.231) by lhrpeml500005.china.huawei.com (7.191.163.240) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2507.23; Thu, 13 Apr 2023 15:27:48 +0100 From: Jonathan Cameron To: Liang Kan , , , , CC: , , , , , , Davidlohr Bueso , Dave Jiang Subject: [PATCH v6 3/5] cxl/pci: Find and register CXL PMU devices Date: Thu, 13 Apr 2023 15:26:15 +0100 Message-ID: <20230413142617.15995-4-Jonathan.Cameron@huawei.com> X-Mailer: git-send-email 2.37.2 In-Reply-To: <20230413142617.15995-1-Jonathan.Cameron@huawei.com> References: <20230413142617.15995-1-Jonathan.Cameron@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.122.247.231] X-ClientProxiedBy: lhrpeml500001.china.huawei.com (7.191.163.213) To lhrpeml500005.china.huawei.com (7.191.163.240) X-CFilter-Loop: Reflected Precedence: bulk List-ID: X-Mailing-List: linux-cxl@vger.kernel.org CXL PMU devices can be found from entries in the Register Locator DVSEC. Reviewed-by: Dan Williams Reviewed-by: Dave Jiang Signed-off-by: Jonathan Cameron --- v6: No change v5: - Switch to cxl_find_regblock_instance() as replacement for indexed version of cxl_find_reblock() - Dropped stale patch description. - Rename all structure, functions etc from cpmu to pmu / cxl_pmu as appropriate. - Register definitions pushed down into drivers/perf/pmu.c as not needed from drivers/cxl/ - Allow richer device naming based on which CXL memdev the CPMU is associated with. /sys/bus/cxl/devices/pmu_memX.Y for instance Y on device /sys/bus/cxl/devices/memX - Tidy up definitions by adding include to linux/device.h for struct device and a forwards reference for struct cxl_pmu_regs. --- drivers/cxl/core/Makefile | 1 + drivers/cxl/core/core.h | 1 + drivers/cxl/core/pmu.c | 69 +++++++++++++++++++++++++++++++++++++++ drivers/cxl/core/port.c | 2 ++ drivers/cxl/core/regs.c | 16 +++++++++ drivers/cxl/cxl.h | 13 ++++++++ drivers/cxl/cxlpci.h | 1 + drivers/cxl/pci.c | 26 ++++++++++++++- drivers/cxl/pmu.h | 28 ++++++++++++++++ 9 files changed, 156 insertions(+), 1 deletion(-) diff --git a/drivers/cxl/core/Makefile b/drivers/cxl/core/Makefile index ca4ae31d8f57..1f66b5d4d935 100644 --- a/drivers/cxl/core/Makefile +++ b/drivers/cxl/core/Makefile @@ -12,5 +12,6 @@ cxl_core-y += memdev.o cxl_core-y += mbox.o cxl_core-y += pci.o cxl_core-y += hdm.o +cxl_core-y += pmu.o cxl_core-$(CONFIG_TRACING) += trace.o cxl_core-$(CONFIG_CXL_REGION) += region.o diff --git a/drivers/cxl/core/core.h b/drivers/cxl/core/core.h index cde475e13216..c9504dfddfed 100644 --- a/drivers/cxl/core/core.h +++ b/drivers/cxl/core/core.h @@ -6,6 +6,7 @@ extern const struct device_type cxl_nvdimm_bridge_type; extern const struct device_type cxl_nvdimm_type; +extern const struct device_type cxl_pmu_type; extern struct attribute_group cxl_base_attribute_group; diff --git a/drivers/cxl/core/pmu.c b/drivers/cxl/core/pmu.c new file mode 100644 index 000000000000..b55eb962235f --- /dev/null +++ b/drivers/cxl/core/pmu.c @@ -0,0 +1,69 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* Copyright(c) 2023 Huawei. All rights reserved. */ + +#include +#include +#include +#include +#include +#include +#include "core.h" + +static void cxl_pmu_release(struct device *dev) +{ + struct cxl_pmu *pmu = to_cxl_pmu(dev); + + kfree(pmu); +} + +const struct device_type cxl_pmu_type = { + .name = "cxl_pmu", + .release = cxl_pmu_release, +}; + +static void remove_dev(void *dev) +{ + device_del(dev); +} + +int devm_cxl_pmu_add(struct device *parent, struct cxl_pmu_regs *regs, + int assoc_id, int index, enum cxl_pmu_type type) +{ + struct cxl_pmu *pmu; + struct device *dev; + int rc; + + pmu = kzalloc(sizeof(*pmu), GFP_KERNEL); + if (!pmu) + return -ENOMEM; + + pmu->assoc_id = assoc_id; + pmu->index = index; + pmu->type = type; + pmu->base = regs->pmu; + dev = &pmu->dev; + device_initialize(dev); + device_set_pm_not_required(dev); + dev->parent = parent; + dev->bus = &cxl_bus_type; + dev->type = &cxl_pmu_type; + switch (pmu->type) { + case CXL_PMU_MEMDEV: + rc = dev_set_name(dev, "pmu_mem%d.%d", assoc_id, index); + break; + } + if (rc) + goto err; + + rc = device_add(dev); + if (rc) + goto err; + + return devm_add_action_or_reset(parent, remove_dev, dev); + +err: + put_device(&pmu->dev); + return rc; +} +EXPORT_SYMBOL_NS_GPL(devm_cxl_pmu_add, CXL); + diff --git a/drivers/cxl/core/port.c b/drivers/cxl/core/port.c index 4d1f9c5b5029..9a71e249f458 100644 --- a/drivers/cxl/core/port.c +++ b/drivers/cxl/core/port.c @@ -57,6 +57,8 @@ static int cxl_device_id(const struct device *dev) return CXL_DEVICE_MEMORY_EXPANDER; if (dev->type == CXL_REGION_TYPE()) return CXL_DEVICE_REGION; + if (dev->type == &cxl_pmu_type) + return CXL_DEVICE_PMU; return 0; } diff --git a/drivers/cxl/core/regs.c b/drivers/cxl/core/regs.c index 4b9672db867d..518bc2ad2c1e 100644 --- a/drivers/cxl/core/regs.c +++ b/drivers/cxl/core/regs.c @@ -6,6 +6,7 @@ #include #include #include +#include #include "core.h" @@ -379,6 +380,21 @@ int cxl_count_regblock(struct pci_dev *pdev, enum cxl_regloc_type type) } EXPORT_SYMBOL_NS_GPL(cxl_count_regblock, CXL); +int cxl_map_pmu_regs(struct pci_dev *pdev, struct cxl_pmu_regs *regs, + struct cxl_register_map *map) +{ + struct device *dev = &pdev->dev; + resource_size_t phys_addr; + + phys_addr = map->resource; + regs->pmu = devm_cxl_iomap_block(dev, phys_addr, CXL_PMU_REGMAP_SIZE); + if (!regs->pmu) + return -ENOMEM; + + return 0; +} +EXPORT_SYMBOL_NS_GPL(cxl_map_pmu_regs, CXL); + resource_size_t cxl_rcrb_to_component(struct device *dev, resource_size_t rcrb, enum cxl_rcrb which) diff --git a/drivers/cxl/cxl.h b/drivers/cxl/cxl.h index f6e2a9ea5f41..610aa430276b 100644 --- a/drivers/cxl/cxl.h +++ b/drivers/cxl/cxl.h @@ -209,6 +209,10 @@ struct cxl_regs { struct_group_tagged(cxl_device_regs, device_regs, void __iomem *status, *mbox, *memdev; ); + + struct_group_tagged(cxl_pmu_regs, pmu_regs, + void __iomem *pmu; + ); }; struct cxl_reg_map { @@ -229,6 +233,10 @@ struct cxl_device_reg_map { struct cxl_reg_map memdev; }; +struct cxl_pmu_reg_map { + struct cxl_reg_map pmu; +}; + /** * struct cxl_register_map - DVSEC harvested register block mapping parameters * @base: virtual base of the register-block-BAR + @block_offset @@ -237,6 +245,7 @@ struct cxl_device_reg_map { * @reg_type: see enum cxl_regloc_type * @component_map: cxl_reg_map for component registers * @device_map: cxl_reg_maps for device registers + * @pmu_map: cxl_reg_maps for CXL Performance Monitoring Units */ struct cxl_register_map { void __iomem *base; @@ -246,6 +255,7 @@ struct cxl_register_map { union { struct cxl_component_reg_map component_map; struct cxl_device_reg_map device_map; + struct cxl_pmu_reg_map pmu_map; }; }; @@ -258,6 +268,8 @@ int cxl_map_component_regs(struct device *dev, struct cxl_component_regs *regs, unsigned long map_mask); int cxl_map_device_regs(struct device *dev, struct cxl_device_regs *regs, struct cxl_register_map *map); +int cxl_map_pmu_regs(struct pci_dev *pdev, struct cxl_pmu_regs *regs, + struct cxl_register_map *map); enum cxl_regloc_type; int cxl_count_regblock(struct pci_dev *pdev, enum cxl_regloc_type type); @@ -752,6 +764,7 @@ void cxl_driver_unregister(struct cxl_driver *cxl_drv); #define CXL_DEVICE_REGION 6 #define CXL_DEVICE_PMEM_REGION 7 #define CXL_DEVICE_DAX_REGION 8 +#define CXL_DEVICE_PMU 9 #define MODULE_ALIAS_CXL(type) MODULE_ALIAS("cxl:t" __stringify(type) "*") #define CXL_MODALIAS_FMT "cxl:t%d" diff --git a/drivers/cxl/cxlpci.h b/drivers/cxl/cxlpci.h index 0465ef963cd6..b1bbf69b62c5 100644 --- a/drivers/cxl/cxlpci.h +++ b/drivers/cxl/cxlpci.h @@ -65,6 +65,7 @@ enum cxl_regloc_type { CXL_REGLOC_RBI_COMPONENT, CXL_REGLOC_RBI_VIRT, CXL_REGLOC_RBI_MEMDEV, + CXL_REGLOC_RBI_PMU, CXL_REGLOC_RBI_TYPES }; diff --git a/drivers/cxl/pci.c b/drivers/cxl/pci.c index 60b23624d167..8a39e9ef081f 100644 --- a/drivers/cxl/pci.c +++ b/drivers/cxl/pci.c @@ -14,6 +14,7 @@ #include "cxlmem.h" #include "cxlpci.h" #include "cxl.h" +#include "pmu.h" /** * DOC: cxl pci @@ -704,7 +705,7 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id) struct cxl_register_map map; struct cxl_memdev *cxlmd; struct cxl_dev_state *cxlds; - int rc; + int i, rc, pmu_count; /* * Double check the anonymous union trickery in struct cxl_regs @@ -785,6 +786,29 @@ static int cxl_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id) if (IS_ERR(cxlmd)) return PTR_ERR(cxlmd); + pmu_count = cxl_count_regblock(pdev, CXL_REGLOC_RBI_PMU); + for (i = 0; i < pmu_count; i++) { + struct cxl_pmu_regs pmu_regs; + + rc = cxl_find_regblock_instance(pdev, CXL_REGLOC_RBI_PMU, &map, i); + if (rc) { + dev_dbg(&pdev->dev, "Could not find PMU regblock\n"); + break; + } + + rc = cxl_map_pmu_regs(pdev, &pmu_regs, &map); + if (rc) { + dev_dbg(&pdev->dev, "Could not map PMU regs\n"); + break; + } + + rc = devm_cxl_pmu_add(cxlds->dev, &pmu_regs, cxlmd->id, i, CXL_PMU_MEMDEV); + if (rc) { + dev_dbg(&pdev->dev, "Could not add PMU instance\n"); + break; + } + } + rc = cxl_event_config(host_bridge, cxlds); if (rc) return rc; diff --git a/drivers/cxl/pmu.h b/drivers/cxl/pmu.h new file mode 100644 index 000000000000..b1e9bcd9f28c --- /dev/null +++ b/drivers/cxl/pmu.h @@ -0,0 +1,28 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Copyright(c) 2023 Huawei + * CXL Specification rev 3.0 Setion 8.2.7 (CPMU Register Interface) + */ +#ifndef CXL_PMU_H +#define CXL_PMU_H +#include + +enum cxl_pmu_type { + CXL_PMU_MEMDEV, +}; + +#define CXL_PMU_REGMAP_SIZE 0xe00 /* Table 8-32 CXL 3.0 specification */ +struct cxl_pmu { + struct device dev; + void __iomem *base; + int assoc_id; + int index; + enum cxl_pmu_type type; +}; + +#define to_cxl_pmu(dev) container_of(dev, struct cxl_pmu, dev) +struct cxl_pmu_regs; +int devm_cxl_pmu_add(struct device *parent, struct cxl_pmu_regs *regs, + int assoc_id, int idx, enum cxl_pmu_type type); + +#endif From patchwork Thu Apr 13 14:26:16 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Jonathan Cameron X-Patchwork-Id: 13210352 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id B2851C77B61 for ; Thu, 13 Apr 2023 14:28:26 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S229864AbjDMO20 (ORCPT ); Thu, 13 Apr 2023 10:28:26 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:58130 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S229712AbjDMO2Y (ORCPT ); Thu, 13 Apr 2023 10:28:24 -0400 Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 3ABA430E5; Thu, 13 Apr 2023 07:28:21 -0700 (PDT) Received: from lhrpeml500005.china.huawei.com (unknown [172.18.147.226]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4Py1zG2dQvz67Rl1; Thu, 13 Apr 2023 22:23:54 +0800 (CST) Received: from SecurePC-101-06.china.huawei.com (10.122.247.231) by lhrpeml500005.china.huawei.com (7.191.163.240) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2507.23; Thu, 13 Apr 2023 15:28:18 +0100 From: Jonathan Cameron To: Liang Kan , , , , CC: , , , , , , Davidlohr Bueso , Dave Jiang Subject: [PATCH v6 4/5] perf: CXL Performance Monitoring Unit driver Date: Thu, 13 Apr 2023 15:26:16 +0100 Message-ID: <20230413142617.15995-5-Jonathan.Cameron@huawei.com> X-Mailer: git-send-email 2.37.2 In-Reply-To: <20230413142617.15995-1-Jonathan.Cameron@huawei.com> References: <20230413142617.15995-1-Jonathan.Cameron@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.122.247.231] X-ClientProxiedBy: lhrpeml500001.china.huawei.com (7.191.163.213) To lhrpeml500005.china.huawei.com (7.191.163.240) X-CFilter-Loop: Reflected Precedence: bulk List-ID: X-Mailing-List: linux-cxl@vger.kernel.org CXL rev 3.0 introduces a standard performance monitoring hardware block to CXL. Instances are discovered using CXL Register Locator DVSEC entries. Each CXL component may have multiple PMUs. This initial driver supports a subset of types of counter. It supports counters that are either fixed or configurable, but requires that they support the ability to freeze and write value whilst frozen. Development done with QEMU model which will be posted shortly. Example: $ perf stat -e cxl_pmu_mem0.0/h2d_req_snpcur/ -e cpmu0/h2d_req_snpdata/ -e cpmu0/clock_ticks/ sleep 1 Performance counter stats for 'system wide': 96,757,023,244,321 cxl_pmu_mem0.0/h2d_req_snpcur/ 96,757,023,244,365 cxl_pmu_mem0.0/h2d_req_snpdata/ 193,514,046,488,653 cxl_pmu_mem0.0/clock_ticks/ 1.090539600 seconds time elapsed Reviewed-by: Dave Jiang Reviewed-by: Kan Liang Signed-off-by: Jonathan Cameron --- v6: Thanks to Kan again! - Update PMU naming in example above. - Tags applied. v5: Thanks to Kan Liang and Dan Williams for review. - Move driver to drivers/perf and rename patch to reflect that. - Add a MAINTAINERS entry now driver doesn't fall under drivers/cxl. - Fix mask - Drop duplication of counter reset and config writing (which also gets rid of a write to an entirely read only register) - Lots of cpmu->pmu and similar renames. - Rename the cxl_pmu_event and related to cxl_pmu_ev_cap so as to avoid impression that there is one of these for every event. They reflect the event capability registers hence the new naming. Rename the two lists to reflect this as well. - Comments on WARN_ON cases. - Fixed a bug in cpuhp offline callback where we could move to the cpu that is going away. - Use devm_add_action_or_reset () for remaining few bits of remove() - Richer PMU naming as cxl_pmu_memX.Y to associate with /sys/bus/cxl/device/memX instance of PMU Y. Scheme extents for USP cxl_pmu_uspX.Y and DSP cxl_pmu_dspX.Y.Z for DSP Y of USP X, PMU instance Z. --- MAINTAINERS | 6 + drivers/cxl/Kconfig | 13 + drivers/perf/Kconfig | 13 + drivers/perf/Makefile | 1 + drivers/perf/cxl_pmu.c | 984 +++++++++++++++++++++++++++++++++++++++++ 5 files changed, 1017 insertions(+) diff --git a/MAINTAINERS b/MAINTAINERS index 90abe83c02f3..51234eb7858e 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -5174,6 +5174,12 @@ S: Maintained F: drivers/cxl/ F: include/uapi/linux/cxl_mem.h +COMPUTE EXPRESS LINK PMU (CPMU) +M: Jonathan Cameron +L: linux-cxl@vger.kernel.org +S: Maintained +F: drivers/perf/cxl_pmu.c + CONEXANT ACCESSRUNNER USB DRIVER L: accessrunner-general@lists.sourceforge.net S: Orphan diff --git a/drivers/cxl/Kconfig b/drivers/cxl/Kconfig index ff4e78117b31..d0a4b4cfa50c 100644 --- a/drivers/cxl/Kconfig +++ b/drivers/cxl/Kconfig @@ -139,4 +139,17 @@ config CXL_REGION_INVALIDATION_TEST If unsure, or if this kernel is meant for production environments, say N. +config CXL_PMU + tristate "CXL Performance Monitoring Unit" + default CXL_BUS + depends on PERF_EVENTS + help + Support performance monitoring as defined in CXL rev 3.0 + section 13.2: Performance Monitoring. CXL components may have + one or more CXL Performance Monitoring Units (CPMUs). + + Say 'y/m' to enable a driver that will attach to performance + monitoring units and provide standard perf based interfaces. + + If unsure say 'm'. endif diff --git a/drivers/perf/Kconfig b/drivers/perf/Kconfig index 66c259000a44..ffeabb14df9e 100644 --- a/drivers/perf/Kconfig +++ b/drivers/perf/Kconfig @@ -203,4 +203,17 @@ source "drivers/perf/arm_cspmu/Kconfig" source "drivers/perf/amlogic/Kconfig" +config CXL_PMU + tristate "CXL Performance Monitoring Unit" + depends on CXL_BUS + help + Support performance monitoring as defined in CXL rev 3.0 + section 13.2: Performance Monitoring. CXL components may have + one or more CXL Performance Monitoring Units (CPMUs). + + Say 'y/m' to enable a driver that will attach to performance + monitoring units and provide standard perf based interfaces. + + If unsure say 'm'. + endmenu diff --git a/drivers/perf/Makefile b/drivers/perf/Makefile index 13e45da61100..900af2c6c4f0 100644 --- a/drivers/perf/Makefile +++ b/drivers/perf/Makefile @@ -23,3 +23,4 @@ obj-$(CONFIG_APPLE_M1_CPU_PMU) += apple_m1_cpu_pmu.o obj-$(CONFIG_ALIBABA_UNCORE_DRW_PMU) += alibaba_uncore_drw_pmu.o obj-$(CONFIG_ARM_CORESIGHT_PMU_ARCH_SYSTEM_PMU) += arm_cspmu/ obj-$(CONFIG_MESON_DDR_PMU) += amlogic/ +obj-$(CONFIG_CXL_PMU) += cxl_pmu.o diff --git a/drivers/perf/cxl_pmu.c b/drivers/perf/cxl_pmu.c new file mode 100644 index 000000000000..29d8b64a11b6 --- /dev/null +++ b/drivers/perf/cxl_pmu.c @@ -0,0 +1,984 @@ +// SPDX-License-Identifier: GPL-2.0-only + +/* + * Copyright(c) 2023 Huawei + * + * The CXL 3.0 specification includes a standard Performance Monitoring Unit, + * called the CXL PMU, or CPMU. In order to allow a high degree of + * implementation flexibility the specification provides a wide range of + * options all of which are self describing. + * + * Details in CXL rev 3.0 section 8.2.7 CPMU Register Interface + */ + +#include +#include +#include +#include +#include +#include +#include +#include + +#include "../cxl/cxlpci.h" +#include "../cxl/cxl.h" +#include "../cxl/pmu.h" + +#define CXL_PMU_CAP_REG 0x0 +#define CXL_PMU_CAP_NUM_COUNTERS_MSK GENMASK_ULL(4, 0) +#define CXL_PMU_CAP_COUNTER_WIDTH_MSK GENMASK_ULL(15, 8) +#define CXL_PMU_CAP_NUM_EVN_CAP_REG_SUP_MSK GENMASK_ULL(24, 20) +#define CXL_PMU_CAP_FILTERS_SUP_MSK GENMASK_ULL(39, 32) +#define CXL_PMU_FILTER_HDM BIT(0) +#define CXL_PMU_FILTER_CHAN_RANK_BANK BIT(1) +#define CXL_PMU_CAP_MSI_N_MSK GENMASK_ULL(47, 44) +#define CXL_PMU_CAP_WRITEABLE_WHEN_FROZEN BIT_ULL(48) +#define CXL_PMU_CAP_FREEZE BIT_ULL(49) +#define CXL_PMU_CAP_INT BIT_ULL(50) +#define CXL_PMU_CAP_VERSION_MSK GENMASK_ULL(63, 60) + +#define CXL_PMU_OVERFLOW_REG 0x10 +#define CXL_PMU_FREEZE_REG 0x18 +#define CXL_PMU_EVENT_CAP_REG(n) (0x100 + 8 * (n)) +#define CXL_PMU_EVENT_CAP_SUPPORTED_EVENTS_MSK GENMASK_ULL(31, 0) +#define CXL_PMU_EVENT_CAP_GROUP_ID_MSK GENMASK_ULL(47, 32) +#define CXL_PMU_EVENT_CAP_VENDOR_ID_MSK GENMASK_ULL(63, 48) + +#define CXL_PMU_COUNTER_CFG_REG(n) (0x200 + 8 * (n)) +#define CXL_PMU_COUNTER_CFG_TYPE_MSK GENMASK_ULL(1, 0) +#define CXL_PMU_COUNTER_CFG_TYPE_FREE_RUN 0 +#define CXL_PMU_COUNTER_CFG_TYPE_FIXED_FUN 1 +#define CXL_PMU_COUNTER_CFG_TYPE_CONFIGURABLE 2 +#define CXL_PMU_COUNTER_CFG_ENABLE BIT_ULL(8) +#define CXL_PMU_COUNTER_CFG_INT_ON_OVRFLW BIT_ULL(9) +#define CXL_PMU_COUNTER_CFG_FREEZE_ON_OVRFLW BIT_ULL(10) +#define CXL_PMU_COUNTER_CFG_EDGE BIT_ULL(11) +#define CXL_PMU_COUNTER_CFG_INVERT BIT_ULL(12) +#define CXL_PMU_COUNTER_CFG_THRESHOLD_MSK GENMASK_ULL(23, 16) +#define CXL_PMU_COUNTER_CFG_EVENTS_MSK GENMASK_ULL(55, 24) +#define CXL_PMU_COUNTER_CFG_EVENT_GRP_ID_IDX_MSK GENMASK_ULL(63, 59) + +#define CXL_PMU_FILTER_CFG_REG(n, f) (0x400 + 4 * ((f) + (n) * 8)) +#define CXL_PMU_FILTER_CFG_VALUE_MSK GENMASK(15, 0) + +#define CXL_PMU_COUNTER_REG(n) (0xc00 + 8 * (n)) + +/* CXL rev 3.0 Table 13-5 Events under CXL Vendor ID */ +#define CXL_PMU_GID_CLOCK_TICKS 0x00 +#define CXL_PMU_GID_D2H_REQ 0x0010 +#define CXL_PMU_GID_D2H_RSP 0x0011 +#define CXL_PMU_GID_H2D_REQ 0x0012 +#define CXL_PMU_GID_H2D_RSP 0x0013 +#define CXL_PMU_GID_CACHE_DATA 0x0014 +#define CXL_PMU_GID_M2S_REQ 0x0020 +#define CXL_PMU_GID_M2S_RWD 0x0021 +#define CXL_PMU_GID_M2S_BIRSP 0x0022 +#define CXL_PMU_GID_S2M_BISNP 0x0023 +#define CXL_PMU_GID_S2M_NDR 0x0024 +#define CXL_PMU_GID_S2M_DRS 0x0025 +#define CXL_PMU_GID_DDR 0x8000 + +static int cxl_pmu_cpuhp_state_num; + +struct cxl_pmu_ev_cap { + u16 vid; + u16 gid; + u32 msk; + union { + int counter_idx; /* fixed counters */ + int event_idx; /* configurable counters */ + }; + struct list_head node; +}; + +#define CXL_PMU_MAX_COUNTERS 64 +struct cxl_pmu_info { + struct pmu pmu; + void __iomem *base; + struct perf_event **hw_events; + struct list_head event_caps_configurable; + struct list_head event_caps_fixed; + DECLARE_BITMAP(used_counter_bm, CXL_PMU_MAX_COUNTERS); + DECLARE_BITMAP(conf_counter_bm, CXL_PMU_MAX_COUNTERS); + u16 counter_width; + u8 num_counters; + u8 num_event_capabilities; + int on_cpu; + struct hlist_node node; + bool filter_hdm; + int irq; +}; + +#define pmu_to_cxl_pmu_info(_pmu) container_of(_pmu, struct cxl_pmu_info, pmu) + +/* + * All CPMU counters are discoverable via the Event Capabilities Registers. + * Each Event Capability register contains a a VID / GroupID. + * A counter may then count any combination (by summing) of events in + * that group which are in the Supported Events Bitmask. + * However, there are some complexities to the scheme. + * - Fixed function counters refer to an Event Capabilities register. + * That event capability register is not then used for Configurable + * counters. + */ +static int cxl_pmu_parse_caps(struct device *dev, struct cxl_pmu_info *info) +{ + unsigned long fixed_counter_event_cap_bm = 0; + void __iomem *base = info->base; + bool freeze_for_enable; + u64 val, eval; + int i; + + val = readq(base + CXL_PMU_CAP_REG); + freeze_for_enable = FIELD_GET(CXL_PMU_CAP_WRITEABLE_WHEN_FROZEN, val) && + FIELD_GET(CXL_PMU_CAP_FREEZE, val); + if (!freeze_for_enable) { + dev_err(dev, "Counters not writable while frozen\n"); + return -ENODEV; + } + + info->num_counters = FIELD_GET(CXL_PMU_CAP_NUM_COUNTERS_MSK, val) + 1; + info->counter_width = FIELD_GET(CXL_PMU_CAP_COUNTER_WIDTH_MSK, val); + info->num_event_capabilities = FIELD_GET(CXL_PMU_CAP_NUM_EVN_CAP_REG_SUP_MSK, val) + 1; + + info->filter_hdm = FIELD_GET(CXL_PMU_CAP_FILTERS_SUP_MSK, val) & CXL_PMU_FILTER_HDM; + if (FIELD_GET(CXL_PMU_CAP_INT, val)) + info->irq = FIELD_GET(CXL_PMU_CAP_MSI_N_MSK, val); + else + info->irq = -1; + + /* First handle fixed function counters; note if configurable counters found */ + for (i = 0; i < info->num_counters; i++) { + struct cxl_pmu_ev_cap *pmu_ev; + u32 events_msk; + u8 group_idx; + + val = readq(base + CXL_PMU_COUNTER_CFG_REG(i)); + + if (FIELD_GET(CXL_PMU_COUNTER_CFG_TYPE_MSK, val) == + CXL_PMU_COUNTER_CFG_TYPE_CONFIGURABLE) { + set_bit(i, info->conf_counter_bm); + } + + if (FIELD_GET(CXL_PMU_COUNTER_CFG_TYPE_MSK, val) != + CXL_PMU_COUNTER_CFG_TYPE_FIXED_FUN) + continue; + + /* In this case we know which fields are const */ + group_idx = FIELD_GET(CXL_PMU_COUNTER_CFG_EVENT_GRP_ID_IDX_MSK, val); + events_msk = FIELD_GET(CXL_PMU_COUNTER_CFG_EVENTS_MSK, val); + eval = readq(base + CXL_PMU_EVENT_CAP_REG(group_idx)); + pmu_ev = devm_kzalloc(dev, sizeof(*pmu_ev), GFP_KERNEL); + if (!pmu_ev) + return -ENOMEM; + + pmu_ev->vid = FIELD_GET(CXL_PMU_EVENT_CAP_VENDOR_ID_MSK, eval); + pmu_ev->gid = FIELD_GET(CXL_PMU_EVENT_CAP_GROUP_ID_MSK, eval); + /* For a fixed purpose counter use the events mask from the counter CFG */ + pmu_ev->msk = events_msk; + pmu_ev->counter_idx = i; + /* This list add is never unwound as all entries deleted on remove */ + list_add(&pmu_ev->node, &info->event_caps_fixed); + /* + * Configurable counters must not use an Event Capability registers that + * is in use for a Fixed counter + */ + set_bit(group_idx, &fixed_counter_event_cap_bm); + } + + if (!bitmap_empty(info->conf_counter_bm, CXL_PMU_MAX_COUNTERS)) { + struct cxl_pmu_ev_cap *pmu_ev; + int j; + /* Walk event capabilities unused by fixed counters */ + for_each_clear_bit(j, &fixed_counter_event_cap_bm, + info->num_event_capabilities) { + pmu_ev = devm_kzalloc(dev, sizeof(*pmu_ev), GFP_KERNEL); + if (!pmu_ev) + return -ENOMEM; + + eval = readq(base + CXL_PMU_EVENT_CAP_REG(j)); + pmu_ev->vid = FIELD_GET(CXL_PMU_EVENT_CAP_VENDOR_ID_MSK, eval); + pmu_ev->gid = FIELD_GET(CXL_PMU_EVENT_CAP_GROUP_ID_MSK, eval); + pmu_ev->msk = FIELD_GET(CXL_PMU_EVENT_CAP_SUPPORTED_EVENTS_MSK, eval); + pmu_ev->event_idx = j; + list_add(&pmu_ev->node, &info->event_caps_configurable); + } + } + + return 0; +} + +static ssize_t cxl_pmu_format_sysfs_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct dev_ext_attribute *eattr; + + eattr = container_of(attr, struct dev_ext_attribute, attr); + + return sysfs_emit(buf, "%s\n", (char *)eattr->var); +} + +#define CXL_PMU_FORMAT_ATTR(_name, _format)\ + (&((struct dev_ext_attribute[]) { \ + { \ + .attr = __ATTR(_name, 0444, \ + cxl_pmu_format_sysfs_show, NULL), \ + .var = (void *)_format \ + } \ + })[0].attr.attr) + +enum { + cxl_pmu_mask_attr, + cxl_pmu_gid_attr, + cxl_pmu_vid_attr, + cxl_pmu_threshold_attr, + cxl_pmu_invert_attr, + cxl_pmu_edge_attr, + cxl_pmu_hdm_filter_en_attr, + cxl_pmu_hdm_attr, +}; + +static struct attribute *cxl_pmu_format_attr[] = { + [cxl_pmu_mask_attr] = CXL_PMU_FORMAT_ATTR(mask, "config:0-31"), + [cxl_pmu_gid_attr] = CXL_PMU_FORMAT_ATTR(gid, "config:32-47"), + [cxl_pmu_vid_attr] = CXL_PMU_FORMAT_ATTR(vid, "config:48-63"), + [cxl_pmu_threshold_attr] = CXL_PMU_FORMAT_ATTR(threshold, "config1:0-15"), + [cxl_pmu_invert_attr] = CXL_PMU_FORMAT_ATTR(invert, "config1:16"), + [cxl_pmu_edge_attr] = CXL_PMU_FORMAT_ATTR(edge, "config1:17"), + [cxl_pmu_hdm_filter_en_attr] = CXL_PMU_FORMAT_ATTR(hdm_filter_en, "config1:18"), + [cxl_pmu_hdm_attr] = CXL_PMU_FORMAT_ATTR(hdm, "config2:0-15"), + NULL +}; + +#define CXL_PMU_ATTR_CONFIG_MASK_MSK GENMASK_ULL(31, 0) +#define CXL_PMU_ATTR_CONFIG_GID_MSK GENMASK_ULL(47, 32) +#define CXL_PMU_ATTR_CONFIG_VID_MSK GENMASK_ULL(63, 48) +#define CXL_PMU_ATTR_CONFIG1_THRESHOLD_MSK GENMASK_ULL(15, 0) +#define CXL_PMU_ATTR_CONFIG1_INVERT_MSK BIT(16) +#define CXL_PMU_ATTR_CONFIG1_EDGE_MSK BIT(17) +#define CXL_PMU_ATTR_CONFIG1_FILTER_EN_MSK BIT(18) +#define CXL_PMU_ATTR_CONFIG2_HDM_MSK GENMASK(15, 0) + +static umode_t cxl_pmu_format_is_visible(struct kobject *kobj, + struct attribute *attr, int a) +{ + struct device *dev = kobj_to_dev(kobj); + struct cxl_pmu_info *info = dev_get_drvdata(dev); + + /* + * Filter capability at the CPMU level, so hide the attributes if the particular + * filter is not supported. + */ + if (!info->filter_hdm && + (attr == cxl_pmu_format_attr[cxl_pmu_hdm_filter_en_attr] || + attr == cxl_pmu_format_attr[cxl_pmu_hdm_attr])) + return 0; + + return attr->mode; +} + +static const struct attribute_group cxl_pmu_format_group = { + .name = "format", + .attrs = cxl_pmu_format_attr, + .is_visible = cxl_pmu_format_is_visible, +}; + +static u32 cxl_pmu_config_get_mask(struct perf_event *event) +{ + return FIELD_GET(CXL_PMU_ATTR_CONFIG_MASK_MSK, event->attr.config); +} + +static u16 cxl_pmu_config_get_gid(struct perf_event *event) +{ + return FIELD_GET(CXL_PMU_ATTR_CONFIG_GID_MSK, event->attr.config); +} + +static u16 cxl_pmu_config_get_vid(struct perf_event *event) +{ + return FIELD_GET(CXL_PMU_ATTR_CONFIG_VID_MSK, event->attr.config); +} + +static u8 cxl_pmu_config1_get_threshold(struct perf_event *event) +{ + return FIELD_GET(CXL_PMU_ATTR_CONFIG1_THRESHOLD_MSK, event->attr.config1); +} + +static bool cxl_pmu_config1_get_invert(struct perf_event *event) +{ + return FIELD_GET(CXL_PMU_ATTR_CONFIG1_INVERT_MSK, event->attr.config1); +} + +static bool cxl_pmu_config1_get_edge(struct perf_event *event) +{ + return FIELD_GET(CXL_PMU_ATTR_CONFIG1_EDGE_MSK, event->attr.config1); +} + +/* + * CPMU specification allows for 8 filters, each with a 16 bit value... + * So we need to find 8x16bits to store it in. + * As the value used for disable is 0xffff, a separate enable switch + * is needed. + */ + +static bool cxl_pmu_config1_hdm_filter_en(struct perf_event *event) +{ + return FIELD_GET(CXL_PMU_ATTR_CONFIG1_FILTER_EN_MSK, event->attr.config1); +} + +static u16 cxl_pmu_config2_get_hdm_decoder(struct perf_event *event) +{ + return FIELD_GET(CXL_PMU_ATTR_CONFIG2_HDM_MSK, event->attr.config2); +} + +static ssize_t cxl_pmu_event_sysfs_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct perf_pmu_events_attr *pmu_attr = + container_of(attr, struct perf_pmu_events_attr, attr); + + return sysfs_emit(buf, "config=%#llx\n", pmu_attr->id); +} + +#define CXL_PMU_EVENT_ATTR(_name, _vid, _gid, _msk) \ + PMU_EVENT_ATTR_ID(_name, cxl_pmu_event_sysfs_show, \ + ((u64)(_vid) << 48) | ((u64)(_gid) << 32) | (u64)(_msk)) + +/* For CXL spec defined events */ +#define CXL_PMU_EVENT_CXL_ATTR(_name, _gid, _msk) \ + CXL_PMU_EVENT_ATTR(_name, PCI_DVSEC_VENDOR_ID_CXL, _gid, _msk) + +static struct attribute *cxl_pmu_event_attrs[] = { + CXL_PMU_EVENT_CXL_ATTR(clock_ticks, CXL_PMU_GID_CLOCK_TICKS, BIT(0)), + /* CXL rev 3.0 Table 3-17 - Device to Host Requests */ + CXL_PMU_EVENT_CXL_ATTR(d2h_req_rdcurr, CXL_PMU_GID_D2H_REQ, BIT(1)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_rdown, CXL_PMU_GID_D2H_REQ, BIT(2)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_rdshared, CXL_PMU_GID_D2H_REQ, BIT(3)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_rdany, CXL_PMU_GID_D2H_REQ, BIT(4)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_rdownnodata, CXL_PMU_GID_D2H_REQ, BIT(5)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_itomwr, CXL_PMU_GID_D2H_REQ, BIT(6)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_wrcurr, CXL_PMU_GID_D2H_REQ, BIT(7)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_clflush, CXL_PMU_GID_D2H_REQ, BIT(8)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_cleanevict, CXL_PMU_GID_D2H_REQ, BIT(9)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_dirtyevict, CXL_PMU_GID_D2H_REQ, BIT(10)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_cleanevictnodata, CXL_PMU_GID_D2H_REQ, BIT(11)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_wowrinv, CXL_PMU_GID_D2H_REQ, BIT(12)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_wowrinvf, CXL_PMU_GID_D2H_REQ, BIT(13)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_wrinv, CXL_PMU_GID_D2H_REQ, BIT(14)), + CXL_PMU_EVENT_CXL_ATTR(d2h_req_cacheflushed, CXL_PMU_GID_D2H_REQ, BIT(16)), + /* CXL rev 3.0 Table 3-20 - D2H Repsonse Encodings */ + CXL_PMU_EVENT_CXL_ATTR(d2h_rsp_rspihiti, CXL_PMU_GID_D2H_RSP, BIT(4)), + CXL_PMU_EVENT_CXL_ATTR(d2h_rsp_rspvhitv, CXL_PMU_GID_D2H_RSP, BIT(6)), + CXL_PMU_EVENT_CXL_ATTR(d2h_rsp_rspihitse, CXL_PMU_GID_D2H_RSP, BIT(5)), + CXL_PMU_EVENT_CXL_ATTR(d2h_rsp_rspshitse, CXL_PMU_GID_D2H_RSP, BIT(1)), + CXL_PMU_EVENT_CXL_ATTR(d2h_rsp_rspsfwdm, CXL_PMU_GID_D2H_RSP, BIT(7)), + CXL_PMU_EVENT_CXL_ATTR(d2h_rsp_rspifwdm, CXL_PMU_GID_D2H_RSP, BIT(15)), + CXL_PMU_EVENT_CXL_ATTR(d2h_rsp_rspvfwdv, CXL_PMU_GID_D2H_RSP, BIT(22)), + /* CXL rev 3.0 Table 3-21 - CXL.cache - Mapping of H2D Requests to D2H Responses */ + CXL_PMU_EVENT_CXL_ATTR(h2d_req_snpdata, CXL_PMU_GID_H2D_REQ, BIT(1)), + CXL_PMU_EVENT_CXL_ATTR(h2d_req_snpinv, CXL_PMU_GID_H2D_REQ, BIT(2)), + CXL_PMU_EVENT_CXL_ATTR(h2d_req_snpcur, CXL_PMU_GID_H2D_REQ, BIT(3)), + /* CXL rev 3.0 Table 3-22 - H2D Response Opcode Encodings */ + CXL_PMU_EVENT_CXL_ATTR(h2d_rsp_writepull, CXL_PMU_GID_H2D_RSP, BIT(1)), + CXL_PMU_EVENT_CXL_ATTR(h2d_rsp_go, CXL_PMU_GID_H2D_RSP, BIT(4)), + CXL_PMU_EVENT_CXL_ATTR(h2d_rsp_gowritepull, CXL_PMU_GID_H2D_RSP, BIT(5)), + CXL_PMU_EVENT_CXL_ATTR(h2d_rsp_extcmp, CXL_PMU_GID_H2D_RSP, BIT(6)), + CXL_PMU_EVENT_CXL_ATTR(h2d_rsp_gowritepulldrop, CXL_PMU_GID_H2D_RSP, BIT(8)), + CXL_PMU_EVENT_CXL_ATTR(h2d_rsp_fastgowritepull, CXL_PMU_GID_H2D_RSP, BIT(13)), + CXL_PMU_EVENT_CXL_ATTR(h2d_rsp_goerrwritepull, CXL_PMU_GID_H2D_RSP, BIT(15)), + /* CXL rev 3.0 Table 13-5 directly lists these */ + CXL_PMU_EVENT_CXL_ATTR(cachedata_d2h_data, CXL_PMU_GID_CACHE_DATA, BIT(0)), + CXL_PMU_EVENT_CXL_ATTR(cachedata_h2d_data, CXL_PMU_GID_CACHE_DATA, BIT(1)), + /* CXL rev 3.0 Table 3-29 M2S Req Memory Opcodes */ + CXL_PMU_EVENT_CXL_ATTR(m2s_req_meminv, CXL_PMU_GID_M2S_REQ, BIT(0)), + CXL_PMU_EVENT_CXL_ATTR(m2s_req_memrd, CXL_PMU_GID_M2S_REQ, BIT(1)), + CXL_PMU_EVENT_CXL_ATTR(m2s_req_memrddata, CXL_PMU_GID_M2S_REQ, BIT(2)), + CXL_PMU_EVENT_CXL_ATTR(m2s_req_memrdfwd, CXL_PMU_GID_M2S_REQ, BIT(3)), + CXL_PMU_EVENT_CXL_ATTR(m2s_req_memwrfwd, CXL_PMU_GID_M2S_REQ, BIT(4)), + CXL_PMU_EVENT_CXL_ATTR(m2s_req_memspecrd, CXL_PMU_GID_M2S_REQ, BIT(8)), + CXL_PMU_EVENT_CXL_ATTR(m2s_req_meminvnt, CXL_PMU_GID_M2S_REQ, BIT(9)), + CXL_PMU_EVENT_CXL_ATTR(m2s_req_memcleanevict, CXL_PMU_GID_M2S_REQ, BIT(10)), + /* CXL rev 3.0 Table 3-35 M2S RwD Memory Opcodes */ + CXL_PMU_EVENT_CXL_ATTR(m2s_rwd_memwr, CXL_PMU_GID_M2S_RWD, BIT(1)), + CXL_PMU_EVENT_CXL_ATTR(m2s_rwd_memwrptl, CXL_PMU_GID_M2S_RWD, BIT(2)), + CXL_PMU_EVENT_CXL_ATTR(m2s_rwd_biconflict, CXL_PMU_GID_M2S_RWD, BIT(4)), + /* CXL rev 3.0 Table 3-38 M2S BIRsp Memory Opcodes */ + CXL_PMU_EVENT_CXL_ATTR(m2s_birsp_i, CXL_PMU_GID_M2S_BIRSP, BIT(0)), + CXL_PMU_EVENT_CXL_ATTR(m2s_birsp_s, CXL_PMU_GID_M2S_BIRSP, BIT(1)), + CXL_PMU_EVENT_CXL_ATTR(m2s_birsp_e, CXL_PMU_GID_M2S_BIRSP, BIT(2)), + CXL_PMU_EVENT_CXL_ATTR(m2s_birsp_iblk, CXL_PMU_GID_M2S_BIRSP, BIT(4)), + CXL_PMU_EVENT_CXL_ATTR(m2s_birsp_sblk, CXL_PMU_GID_M2S_BIRSP, BIT(5)), + CXL_PMU_EVENT_CXL_ATTR(m2s_birsp_eblk, CXL_PMU_GID_M2S_BIRSP, BIT(6)), + /* CXL rev 3.0 Table 3-40 S2M BISnp Opcodes */ + CXL_PMU_EVENT_CXL_ATTR(s2m_bisnp_cur, CXL_PMU_GID_S2M_BISNP, BIT(0)), + CXL_PMU_EVENT_CXL_ATTR(s2m_bisnp_data, CXL_PMU_GID_S2M_BISNP, BIT(1)), + CXL_PMU_EVENT_CXL_ATTR(s2m_bisnp_inv, CXL_PMU_GID_S2M_BISNP, BIT(2)), + CXL_PMU_EVENT_CXL_ATTR(s2m_bisnp_curblk, CXL_PMU_GID_S2M_BISNP, BIT(4)), + CXL_PMU_EVENT_CXL_ATTR(s2m_bisnp_datblk, CXL_PMU_GID_S2M_BISNP, BIT(5)), + CXL_PMU_EVENT_CXL_ATTR(s2m_bisnp_invblk, CXL_PMU_GID_S2M_BISNP, BIT(6)), + /* CXL rev 3.0 Table 3-43 S2M NDR Opcopdes */ + CXL_PMU_EVENT_CXL_ATTR(s2m_ndr_cmp, CXL_PMU_GID_S2M_NDR, BIT(0)), + CXL_PMU_EVENT_CXL_ATTR(s2m_ndr_cmps, CXL_PMU_GID_S2M_NDR, BIT(1)), + CXL_PMU_EVENT_CXL_ATTR(s2m_ndr_cmpe, CXL_PMU_GID_S2M_NDR, BIT(2)), + CXL_PMU_EVENT_CXL_ATTR(s2m_ndr_biconflictack, CXL_PMU_GID_S2M_NDR, BIT(3)), + /* CXL rev 3.0 Table 3-46 S2M DRS opcodes */ + CXL_PMU_EVENT_CXL_ATTR(s2m_drs_memdata, CXL_PMU_GID_S2M_DRS, BIT(0)), + CXL_PMU_EVENT_CXL_ATTR(s2m_drs_memdatanxm, CXL_PMU_GID_S2M_DRS, BIT(1)), + /* CXL rev 3.0 Table 13-5 directly lists these */ + CXL_PMU_EVENT_CXL_ATTR(ddr_act, CXL_PMU_GID_DDR, BIT(0)), + CXL_PMU_EVENT_CXL_ATTR(ddr_pre, CXL_PMU_GID_DDR, BIT(1)), + CXL_PMU_EVENT_CXL_ATTR(ddr_casrd, CXL_PMU_GID_DDR, BIT(2)), + CXL_PMU_EVENT_CXL_ATTR(ddr_caswr, CXL_PMU_GID_DDR, BIT(3)), + CXL_PMU_EVENT_CXL_ATTR(ddr_refresh, CXL_PMU_GID_DDR, BIT(4)), + CXL_PMU_EVENT_CXL_ATTR(ddr_selfrefreshent, CXL_PMU_GID_DDR, BIT(5)), + CXL_PMU_EVENT_CXL_ATTR(ddr_rfm, CXL_PMU_GID_DDR, BIT(6)), + NULL +}; + +static struct cxl_pmu_ev_cap *cxl_pmu_find_fixed_counter_ev_cap(struct cxl_pmu_info *info, + int vid, int gid, int msk) +{ + struct cxl_pmu_ev_cap *pmu_ev; + + list_for_each_entry(pmu_ev, &info->event_caps_fixed, node) { + if (vid != pmu_ev->vid || gid != pmu_ev->gid) + continue; + + /* Precise match for fixed counter */ + if (msk == pmu_ev->msk) + return pmu_ev; + } + + return ERR_PTR(-EINVAL); +} + +static struct cxl_pmu_ev_cap *cxl_pmu_find_config_counter_ev_cap(struct cxl_pmu_info *info, + int vid, int gid, int msk) +{ + struct cxl_pmu_ev_cap *pmu_ev; + + list_for_each_entry(pmu_ev, &info->event_caps_configurable, node) { + if (vid != pmu_ev->vid || gid != pmu_ev->gid) + continue; + + /* Request mask must be subset of supported */ + if (msk & ~pmu_ev->msk) + continue; + + return pmu_ev; + } + + return ERR_PTR(-EINVAL); +} + +static umode_t cxl_pmu_event_is_visible(struct kobject *kobj, struct attribute *attr, int a) +{ + struct device_attribute *dev_attr = container_of(attr, struct device_attribute, attr); + struct perf_pmu_events_attr *pmu_attr = + container_of(dev_attr, struct perf_pmu_events_attr, attr); + struct device *dev = kobj_to_dev(kobj); + struct cxl_pmu_info *info = dev_get_drvdata(dev); + int vid = FIELD_GET(CXL_PMU_ATTR_CONFIG_VID_MSK, pmu_attr->id); + int gid = FIELD_GET(CXL_PMU_ATTR_CONFIG_GID_MSK, pmu_attr->id); + int msk = FIELD_GET(CXL_PMU_ATTR_CONFIG_MASK_MSK, pmu_attr->id); + + if (!IS_ERR(cxl_pmu_find_fixed_counter_ev_cap(info, vid, gid, msk))) + return attr->mode; + + if (!IS_ERR(cxl_pmu_find_config_counter_ev_cap(info, vid, gid, msk))) + return attr->mode; + + return 0; +} + +static const struct attribute_group cxl_pmu_events = { + .name = "events", + .attrs = cxl_pmu_event_attrs, + .is_visible = cxl_pmu_event_is_visible, +}; + +static ssize_t cpumask_show(struct device *dev, struct device_attribute *attr, + char *buf) +{ + struct cxl_pmu_info *info = dev_get_drvdata(dev); + + return cpumap_print_to_pagebuf(true, buf, cpumask_of(info->on_cpu)); +} +static DEVICE_ATTR_RO(cpumask); + +static struct attribute *cxl_pmu_cpumask_attrs[] = { + &dev_attr_cpumask.attr, + NULL +}; + +static const struct attribute_group cxl_pmu_cpumask_group = { + .attrs = cxl_pmu_cpumask_attrs, +}; + +static const struct attribute_group *cxl_pmu_attr_groups[] = { + &cxl_pmu_events, + &cxl_pmu_format_group, + &cxl_pmu_cpumask_group, + NULL +}; + +/* If counter_idx == NULL, don't try to allocate a counter. */ +static int cxl_pmu_get_event_idx(struct perf_event *event, int *counter_idx, + int *event_idx) +{ + struct cxl_pmu_info *info = pmu_to_cxl_pmu_info(event->pmu); + DECLARE_BITMAP(configurable_and_free, CXL_PMU_MAX_COUNTERS); + struct cxl_pmu_ev_cap *pmu_ev; + u32 mask; + u16 gid, vid; + int i; + + vid = cxl_pmu_config_get_vid(event); + gid = cxl_pmu_config_get_gid(event); + mask = cxl_pmu_config_get_mask(event); + + pmu_ev = cxl_pmu_find_fixed_counter_ev_cap(info, vid, gid, mask); + if (!IS_ERR(pmu_ev)) { + if (!counter_idx) + return 0; + if (!test_bit(pmu_ev->counter_idx, info->used_counter_bm)) { + *counter_idx = pmu_ev->counter_idx; + return 0; + } + /* Fixed counter is in use, but maybe a configurable one? */ + } + + pmu_ev = cxl_pmu_find_config_counter_ev_cap(info, vid, gid, mask); + if (!IS_ERR(pmu_ev)) { + if (!counter_idx) + return 0; + + bitmap_andnot(configurable_and_free, info->conf_counter_bm, + info->used_counter_bm, CXL_PMU_MAX_COUNTERS); + + i = find_first_bit(configurable_and_free, CXL_PMU_MAX_COUNTERS); + if (i == CXL_PMU_MAX_COUNTERS) + return -EINVAL; + + *counter_idx = i; + return 0; + } + + return -EINVAL; +} + +static int cxl_pmu_event_init(struct perf_event *event) +{ + struct cxl_pmu_info *info = pmu_to_cxl_pmu_info(event->pmu); + + event->cpu = info->on_cpu; + /* Top level type sanity check - is this a Hardware Event being requested */ + if (event->attr.type != event->pmu->type) + return -ENOENT; + + if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK) + return -EOPNOTSUPP; + /* TODO: Validation of any filter */ + + /* + * Verify that it is possible to count what was requested. Either must + * be a fixed counter that is a precise match or a configurable counter + * where this is a subset. + */ + return cxl_pmu_get_event_idx(event, NULL, NULL); +} + +static void cxl_pmu_enable(struct pmu *pmu) +{ + struct cxl_pmu_info *info = pmu_to_cxl_pmu_info(pmu); + void __iomem *base = info->base; + + /* Can assume frozen at this stage */ + writeq(0, base + CXL_PMU_FREEZE_REG); +} + +static void cxl_pmu_disable(struct pmu *pmu) +{ + struct cxl_pmu_info *info = pmu_to_cxl_pmu_info(pmu); + void __iomem *base = info->base; + + /* + * Whilst bits above number of counters are RsvdZ + * they are unlikely to be repurposed given + * number of counters is allowed to be 64 leaving + * no reserved bits. Hence this is only slightly + * naughty. + */ + writeq(GENMASK_ULL(63, 0), base + CXL_PMU_FREEZE_REG); +} + +static void cxl_pmu_event_start(struct perf_event *event, int flags) +{ + struct cxl_pmu_info *info = pmu_to_cxl_pmu_info(event->pmu); + struct hw_perf_event *hwc = &event->hw; + void __iomem *base = info->base; + u64 cfg; + + /* + * All paths to here should either set these flags directly or + * call cxl_pmu_event_stop() which will ensure the correct state. + */ + if (WARN_ON_ONCE(!(hwc->state & PERF_HES_STOPPED))) + return; + + WARN_ON_ONCE(!(hwc->state & PERF_HES_UPTODATE)); + hwc->state = 0; + + /* + * Currently only hdm filter control is implemnted, this code will + * want generalizing when more filters are added. + */ + if (info->filter_hdm) { + if (cxl_pmu_config1_hdm_filter_en(event)) + cfg = cxl_pmu_config2_get_hdm_decoder(event); + else + cfg = GENMASK(15, 0); /* No filtering if 0xFFFF_FFFF */ + writeq(cfg, base + CXL_PMU_FILTER_CFG_REG(hwc->idx, 0)); + } + + cfg = readq(base + CXL_PMU_COUNTER_CFG_REG(hwc->idx)); + cfg |= FIELD_PREP(CXL_PMU_COUNTER_CFG_INT_ON_OVRFLW, 1); + cfg |= FIELD_PREP(CXL_PMU_COUNTER_CFG_FREEZE_ON_OVRFLW, 1); + cfg |= FIELD_PREP(CXL_PMU_COUNTER_CFG_ENABLE, 1); + cfg |= FIELD_PREP(CXL_PMU_COUNTER_CFG_EDGE, + cxl_pmu_config1_get_edge(event) ? 1 : 0); + cfg |= FIELD_PREP(CXL_PMU_COUNTER_CFG_INVERT, + cxl_pmu_config1_get_invert(event) ? 1 : 0); + + /* Fixed purpose counters have next two fields RO */ + if (test_bit(hwc->idx, info->conf_counter_bm)) { + cfg |= FIELD_PREP(CXL_PMU_COUNTER_CFG_EVENT_GRP_ID_IDX_MSK, + hwc->event_base); + cfg |= FIELD_PREP(CXL_PMU_COUNTER_CFG_EVENTS_MSK, + cxl_pmu_config_get_mask(event)); + } + cfg &= ~CXL_PMU_COUNTER_CFG_THRESHOLD_MSK; + /* + * For events that generate only 1 count per clock the CXL 3.0 spec + * states the threshold shall be set to 1 but if set to 0 it will + * count the raw value anwyay? + * There is no definition of what events will count multiple per cycle + * and hence to which non 1 values of threshold can apply. + * (CXL 3.0 8.2.7.2.1 Counter Configuration - threshold field definition) + */ + cfg |= FIELD_PREP(CXL_PMU_COUNTER_CFG_THRESHOLD_MSK, + cxl_pmu_config1_get_threshold(event)); + writeq(cfg, base + CXL_PMU_COUNTER_CFG_REG(hwc->idx)); + + local64_set(&hwc->prev_count, 0); + writeq(0, base + CXL_PMU_COUNTER_REG(hwc->idx)); + + perf_event_update_userpage(event); +} + +static u64 cxl_pmu_read_counter(struct perf_event *event) +{ + struct cxl_pmu_info *info = pmu_to_cxl_pmu_info(event->pmu); + void __iomem *base = info->base; + + return readq(base + CXL_PMU_COUNTER_REG(event->hw.idx)); +} + +static void __cxl_pmu_read(struct perf_event *event, bool overflow) +{ + struct cxl_pmu_info *info = pmu_to_cxl_pmu_info(event->pmu); + struct hw_perf_event *hwc = &event->hw; + u64 new_cnt, prev_cnt, delta; + + do { + prev_cnt = local64_read(&hwc->prev_count); + new_cnt = cxl_pmu_read_counter(event); + } while (local64_cmpxchg(&hwc->prev_count, prev_cnt, new_cnt) != prev_cnt); + + /* + * If we know an overflow occur then take that into account. + * Note counter is not reset as that would lose events + */ + delta = (new_cnt - prev_cnt) & GENMASK_ULL(info->counter_width - 1, 0); + if (overflow && delta < GENMASK_ULL(info->counter_width - 1, 0)) + delta += (1UL << info->counter_width); + + local64_add(delta, &event->count); +} + +static void cxl_pmu_read(struct perf_event *event) +{ + __cxl_pmu_read(event, false); +} + +static void cxl_pmu_event_stop(struct perf_event *event, int flags) +{ + struct cxl_pmu_info *info = pmu_to_cxl_pmu_info(event->pmu); + void __iomem *base = info->base; + struct hw_perf_event *hwc = &event->hw; + u64 cfg; + + cxl_pmu_read(event); + WARN_ON_ONCE(hwc->state & PERF_HES_STOPPED); + hwc->state |= PERF_HES_STOPPED; + + cfg = readq(base + CXL_PMU_COUNTER_CFG_REG(hwc->idx)); + cfg &= ~(FIELD_PREP(CXL_PMU_COUNTER_CFG_INT_ON_OVRFLW, 1) | + FIELD_PREP(CXL_PMU_COUNTER_CFG_ENABLE, 1)); + writeq(cfg, base + CXL_PMU_COUNTER_CFG_REG(hwc->idx)); + + hwc->state |= PERF_HES_UPTODATE; +} + +static int cxl_pmu_event_add(struct perf_event *event, int flags) +{ + struct cxl_pmu_info *info = pmu_to_cxl_pmu_info(event->pmu); + struct hw_perf_event *hwc = &event->hw; + int idx, rc; + int event_idx = 0; + + hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE; + + rc = cxl_pmu_get_event_idx(event, &idx, &event_idx); + if (rc < 0) + return rc; + + hwc->idx = idx; + + /* Only set for configurable counters */ + hwc->event_base = event_idx; + info->hw_events[idx] = event; + set_bit(idx, info->used_counter_bm); + + if (flags & PERF_EF_START) + cxl_pmu_event_start(event, PERF_EF_RELOAD); + + return 0; +} + +static void cxl_pmu_event_del(struct perf_event *event, int flags) +{ + struct cxl_pmu_info *info = pmu_to_cxl_pmu_info(event->pmu); + struct hw_perf_event *hwc = &event->hw; + + cxl_pmu_event_stop(event, PERF_EF_UPDATE); + clear_bit(hwc->idx, info->used_counter_bm); + info->hw_events[hwc->idx] = NULL; + perf_event_update_userpage(event); +} + +static irqreturn_t cxl_pmu_irq(int irq, void *data) +{ + struct cxl_pmu_info *info = data; + void __iomem *base = info->base; + u64 overflowed; + DECLARE_BITMAP(overflowedbm, 64); + int i; + + overflowed = readq(base + CXL_PMU_OVERFLOW_REG); + + /* Interrupt may be shared, so maybe it isn't ours */ + if (!overflowed) + return IRQ_NONE; + + bitmap_from_arr64(overflowedbm, &overflowed, 64); + for_each_set_bit(i, overflowedbm, info->num_counters) { + struct perf_event *event = info->hw_events[i]; + + if (!event) { + dev_dbg(info->pmu.dev, + "overflow but on non enabled counter %d\n", i); + continue; + } + + __cxl_pmu_read(event, true); + } + + writeq(overflowed, base + CXL_PMU_OVERFLOW_REG); + + return IRQ_HANDLED; +} + +static void cxl_pmu_perf_unregister(void *_info) +{ + struct cxl_pmu_info *info = _info; + + perf_pmu_unregister(&info->pmu); +} + +static void cxl_pmu_cpuhp_remove(void *_info) +{ + struct cxl_pmu_info *info = _info; + + cpuhp_state_remove_instance_nocalls(cxl_pmu_cpuhp_state_num, &info->node); +} + +static int cxl_pmu_probe(struct device *dev) +{ + struct cxl_pmu *pmu = to_cxl_pmu(dev); + struct pci_dev *pdev = to_pci_dev(dev->parent); + struct cxl_pmu_info *info; + char *irq_name; + char *dev_name; + int rc, irq; + + info = devm_kzalloc(dev, sizeof(*info), GFP_KERNEL); + if (!info) + return -ENOMEM; + + dev_set_drvdata(dev, info); + INIT_LIST_HEAD(&info->event_caps_fixed); + INIT_LIST_HEAD(&info->event_caps_configurable); + + info->base = pmu->base; + + info->on_cpu = -1; + rc = cxl_pmu_parse_caps(dev, info); + if (rc) + return rc; + + info->hw_events = devm_kcalloc(dev, sizeof(*info->hw_events), + info->num_counters, GFP_KERNEL); + if (!info->hw_events) + return -ENOMEM; + + switch (pmu->type) { + case CXL_PMU_MEMDEV: + dev_name = devm_kasprintf(dev, GFP_KERNEL, "cxl_pmu_mem%d.%d", + pmu->assoc_id, pmu->index); + break; + } + if (!dev_name) + return -ENOMEM; + + info->pmu = (struct pmu) { + .name = dev_name, + .parent = dev, + .module = THIS_MODULE, + .event_init = cxl_pmu_event_init, + .pmu_enable = cxl_pmu_enable, + .pmu_disable = cxl_pmu_disable, + .add = cxl_pmu_event_add, + .del = cxl_pmu_event_del, + .start = cxl_pmu_event_start, + .stop = cxl_pmu_event_stop, + .read = cxl_pmu_read, + .task_ctx_nr = perf_invalid_context, + .attr_groups = cxl_pmu_attr_groups, + .capabilities = PERF_PMU_CAP_NO_EXCLUDE, + }; + + if (info->irq <= 0) + return -EINVAL; + + rc = pci_irq_vector(pdev, info->irq); + if (rc < 0) + return rc; + irq = rc; + + irq_name = devm_kasprintf(dev, GFP_KERNEL, "%s_overflow\n", dev_name); + if (!irq_name) + return -ENOMEM; + + rc = devm_request_irq(dev, irq, cxl_pmu_irq, IRQF_SHARED | IRQF_ONESHOT, + irq_name, info); + if (rc) + return rc; + info->irq = irq; + + rc = cpuhp_state_add_instance(cxl_pmu_cpuhp_state_num, &info->node); + if (rc) + return rc; + + rc = devm_add_action_or_reset(dev, cxl_pmu_cpuhp_remove, info); + if (rc) + return rc; + + rc = perf_pmu_register(&info->pmu, info->pmu.name, -1); + if (rc) + return rc; + + rc = devm_add_action_or_reset(dev, cxl_pmu_perf_unregister, info); + if (rc) + return rc; + + return 0; +} + +static struct cxl_driver cxl_pmu_driver = { + .name = "cxl_pmu", + .probe = cxl_pmu_probe, + .id = CXL_DEVICE_PMU, +}; + +static int cxl_pmu_online_cpu(unsigned int cpu, struct hlist_node *node) +{ + struct cxl_pmu_info *info = hlist_entry_safe(node, struct cxl_pmu_info, node); + + if (info->on_cpu != -1) + return 0; + + info->on_cpu = cpu; + /* + * CPU HP lock is held so we should be guaranteed that the CPU hasn't yet + * gone away again. + */ + WARN_ON(irq_set_affinity(info->irq, cpumask_of(cpu))); + + return 0; +} + +static int cxl_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node) +{ + struct cxl_pmu_info *info = hlist_entry_safe(node, struct cxl_pmu_info, node); + unsigned int target; + + if (info->on_cpu != cpu) + return 0; + + info->on_cpu = -1; + target = cpumask_any_but(cpu_online_mask, cpu); + if (target >= nr_cpu_ids) { + dev_err(info->pmu.dev, "Unable to find a suitable CPU\n"); + return 0; + } + + perf_pmu_migrate_context(&info->pmu, cpu, target); + info->on_cpu = target; + /* + * CPU HP lock is held so we should be guaranteed that this CPU hasn't yet + * gone away. + */ + WARN_ON(irq_set_affinity(info->irq, cpumask_of(target))); + + return 0; +} + +static __init int cxl_pmu_init(void) +{ + int rc; + + rc = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, + "AP_PERF_CXL_PMU_ONLINE", + cxl_pmu_online_cpu, cxl_pmu_offline_cpu); + if (rc < 0) + return rc; + cxl_pmu_cpuhp_state_num = rc; + + rc = cxl_driver_register(&cxl_pmu_driver); + if (rc) + cpuhp_remove_multi_state(cxl_pmu_cpuhp_state_num); + + return rc; +} + +static __exit void cxl_pmu_exit(void) +{ + cxl_driver_unregister(&cxl_pmu_driver); + cpuhp_remove_multi_state(cxl_pmu_cpuhp_state_num); +} + +MODULE_LICENSE("GPL"); +MODULE_IMPORT_NS(CXL); +module_init(cxl_pmu_init); +module_exit(cxl_pmu_exit); +MODULE_ALIAS_CXL(CXL_DEVICE_PMU); From patchwork Thu Apr 13 14:26:17 2023 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Jonathan Cameron X-Patchwork-Id: 13210353 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by smtp.lore.kernel.org (Postfix) with ESMTP id 8BCE1C77B61 for ; Thu, 13 Apr 2023 14:28:58 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S229999AbjDMO25 (ORCPT ); Thu, 13 Apr 2023 10:28:57 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:58514 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S229638AbjDMO24 (ORCPT ); Thu, 13 Apr 2023 10:28:56 -0400 Received: from frasgout.his.huawei.com (frasgout.his.huawei.com [185.176.79.56]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id 9C21444B6; Thu, 13 Apr 2023 07:28:51 -0700 (PDT) Received: from lhrpeml500005.china.huawei.com (unknown [172.18.147.207]) by frasgout.his.huawei.com (SkyGuard) with ESMTP id 4Py23p5MXpz67ngg; Thu, 13 Apr 2023 22:27:50 +0800 (CST) Received: from SecurePC-101-06.china.huawei.com (10.122.247.231) by lhrpeml500005.china.huawei.com (7.191.163.240) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256) id 15.1.2507.23; Thu, 13 Apr 2023 15:28:49 +0100 From: Jonathan Cameron To: Liang Kan , , , , CC: , , , , , , Davidlohr Bueso , Dave Jiang Subject: [PATCH v6 5/5] docs: perf: Minimal introduction the the CXL PMU device and driver Date: Thu, 13 Apr 2023 15:26:17 +0100 Message-ID: <20230413142617.15995-6-Jonathan.Cameron@huawei.com> X-Mailer: git-send-email 2.37.2 In-Reply-To: <20230413142617.15995-1-Jonathan.Cameron@huawei.com> References: <20230413142617.15995-1-Jonathan.Cameron@huawei.com> MIME-Version: 1.0 X-Originating-IP: [10.122.247.231] X-ClientProxiedBy: lhrpeml500001.china.huawei.com (7.191.163.213) To lhrpeml500005.china.huawei.com (7.191.163.240) X-CFilter-Loop: Reflected Precedence: bulk List-ID: X-Mailing-List: linux-cxl@vger.kernel.org Very basic introduction to the device and the current driver support provided. I expect to expand on this in future versions of this patch set. Reviewed-by: Dave Jiang Reviewed-by: Kan Liang Signed-off-by: Jonathan Cameron --- v6: - Tags applied. v5: - Correct final statemnt about sampling support. (Kan) - Update CXL PMU MAINTAINERS entry to include this documentation - Update device naming to cxl_pmu_memX.Y etc --- Documentation/admin-guide/perf/cxl.rst | 68 ++++++++++++++++++++++++ Documentation/admin-guide/perf/index.rst | 1 + MAINTAINERS | 1 + 3 files changed, 70 insertions(+) diff --git a/Documentation/admin-guide/perf/cxl.rst b/Documentation/admin-guide/perf/cxl.rst new file mode 100644 index 000000000000..2a1c684603bb --- /dev/null +++ b/Documentation/admin-guide/perf/cxl.rst @@ -0,0 +1,68 @@ +.. SPDX-License-Identifier: GPL-2.0 + +====================================== +CXL Performance Monitoring Unit (CPMU) +====================================== + +The CXL rev 3.0 specification provides a definition of CXL Performance +Monitoring Unit in section 13.2: Performance Monitoring. + +CXL components (e.g. Root Port, Switch Upstream Port, End Point) may have +any number of CPMU instances. CPMU capabilities are fully discoverable from +the devices. The specification provides event definitions for all CXL protocol +message types and a set of additional events for things commonly counted on +CXL devices (e.g. DRAM events). + +CPMU driver +=========== + +The CPMU driver registers a perf PMU with the name pmu_mem. on the CXL bus +representing the Yth CPMU for memX. + + /sys/bus/cxl/device/pmu_mem. + +The associated PMU is registered as + + /sys/bus/event_sources/devices/cxl_pmu_mem. + +In common with other CXL bus devices, the id has no specific meaning and the +relationship to specific CXL device should be established via the device parent +of the device on the CXL bus. + +PMU driver provides description of available events and filter options in sysfs. + +The "format" directory describes all formats of the config (event vendor id, +group id and mask) config1 (threshold, filter enables) and config2 (filter +parameters) fields of the perf_event_attr structure. The "events" directory +describes all documented events show in perf list. + +The events shown in perf list are the most fine grained events with a single +bit of the event mask set. More general events may be enable by setting +multiple mask bits in config. For example, all Device to Host Read Requests +may be captured on a single counter by setting the bits for all of + +* d2h_req_rdcurr +* d2h_req_rdown +* d2h_req_rdshared +* d2h_req_rdany +* d2h_req_rdownnodata + +Example of usage:: + + $#perf list + cxl_pmu_mem0.0/clock_ticks/ [Kernel PMU event] + cxl_pmu_mem0.0/d2h_req_rdshared/ [Kernel PMU event] + cxl_pmu_mem0.0/h2d_req_snpcur/ [Kernel PMU event] + cxl_pmu_mem0.0/h2d_req_snpdata/ [Kernel PMU event] + cxl_pmu_mem0.0/h2d_req_snpinv/ [Kernel PMU event] + ----------------------------------------------------------- + + $# perf stat -e cxl_pmu_mem0.0/clock_ticks/ -e cxl_pmu_mem0.0/d2h_req_rdshared/ + +Vendor specific events may also be available and if so can be used via + + $# perf stat -e cxl_pmu_mem0.0/vid=VID,gid=GID,mask=MASK/ + +The driver does not support sampling so "perf record" is unsupported. +It only supports system-wide counting so attaching to a task is +unsupported. diff --git a/Documentation/admin-guide/perf/index.rst b/Documentation/admin-guide/perf/index.rst index 9de64a40adab..f60be04e4e33 100644 --- a/Documentation/admin-guide/perf/index.rst +++ b/Documentation/admin-guide/perf/index.rst @@ -21,3 +21,4 @@ Performance monitor support alibaba_pmu nvidia-pmu meson-ddr-pmu + cxl diff --git a/MAINTAINERS b/MAINTAINERS index 51234eb7858e..29a108afa065 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -5178,6 +5178,7 @@ COMPUTE EXPRESS LINK PMU (CPMU) M: Jonathan Cameron L: linux-cxl@vger.kernel.org S: Maintained +F: Documentation/admin-guide/perf/cxl.rst F: drivers/perf/cxl_pmu.c CONEXANT ACCESSRUNNER USB DRIVER