]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
EDAC: Add scrub control feature
authorShiju Jose <shiju.jose@huawei.com>
Wed, 12 Feb 2025 14:36:40 +0000 (14:36 +0000)
committerBorislav Petkov (AMD) <bp@alien8.de>
Tue, 25 Feb 2025 14:39:09 +0000 (15:39 +0100)
Add a scrub control to manage memory scrubbers in the system.

Devices with a scrub feature register with the EDAC device driver which
retrieves the scrub descriptor from the scrub driver and exposes the
control attributes for a instance to userspace at

  /sys/bus/edac/devices/<dev-name>/scrubX/.

The common sysfs scrub control interface abstracts the control of
arbitrary scrubbing functionality into a common set of functions. The
attribute nodes are only present if the client driver has implemented
the corresponding attribute callback function and passed the operations
to the device driver during registration.

  [ bp: Massage commit message, docs and code, simplify text a bit.
    Integrate fixup for: https://lore.kernel.org/r/202502251009.0sGkolEJ-lkp@intel.com
Reported-by: kernel test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@linaro.org> ]
Co-developed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Shiju Jose <shiju.jose@huawei.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Tested-by: Daniel Ferguson <danielf@os.amperecomputing.com>
Tested-by: Fan Ni <fan.ni@samsung.com>
Link: https://lore.kernel.org/r/20250212143654.1893-3-shiju.jose@huawei.com
Documentation/ABI/testing/sysfs-edac-scrub [new file with mode: 0644]
Documentation/edac/features.rst
Documentation/edac/index.rst
Documentation/edac/scrub.rst [new file with mode: 0644]
drivers/edac/Kconfig
drivers/edac/Makefile
drivers/edac/edac_device.c
drivers/edac/scrub.c [new file with mode: 0755]
include/linux/edac.h

diff --git a/Documentation/ABI/testing/sysfs-edac-scrub b/Documentation/ABI/testing/sysfs-edac-scrub
new file mode 100644 (file)
index 0000000..c43be90
--- /dev/null
@@ -0,0 +1,69 @@
+What:          /sys/bus/edac/devices/<dev-name>/scrubX
+Date:          March 2025
+KernelVersion: 6.15
+Contact:       linux-edac@vger.kernel.org
+Description:
+               The sysfs EDAC bus devices /<dev-name>/scrubX subdirectory
+               belongs to an instance of memory scrub control feature,
+               where <dev-name> directory corresponds to a device/memory
+               region registered with the EDAC device driver for the
+               scrub control feature.
+
+               The sysfs scrub attr nodes are only present if the parent
+               driver has implemented the corresponding attr callback
+               function and provided the necessary operations to the EDAC
+               device driver during registration.
+
+What:          /sys/bus/edac/devices/<dev-name>/scrubX/addr
+Date:          March 2025
+KernelVersion: 6.15
+Contact:       linux-edac@vger.kernel.org
+Description:
+               (RW) The base address of the memory region to be scrubbed
+               for on-demand scrubbing. Setting address starts scrubbing.
+               The size must be set before that.
+
+               The readback addr value is non-zero if the requested
+               on-demand scrubbing is in progress, zero otherwise.
+
+What:          /sys/bus/edac/devices/<dev-name>/scrubX/size
+Date:          March 2025
+KernelVersion: 6.15
+Contact:       linux-edac@vger.kernel.org
+Description:
+               (RW) The size of the memory region to be scrubbed
+               (on-demand scrubbing).
+
+What:          /sys/bus/edac/devices/<dev-name>/scrubX/enable_background
+Date:          March 2025
+KernelVersion: 6.15
+Contact:       linux-edac@vger.kernel.org
+Description:
+               (RW) Start/Stop background (patrol) scrubbing if supported.
+
+What:          /sys/bus/edac/devices/<dev-name>/scrubX/min_cycle_duration
+Date:          March 2025
+KernelVersion: 6.15
+Contact:       linux-edac@vger.kernel.org
+Description:
+               (RO) Supported minimum scrub cycle duration in seconds
+               by the memory scrubber.
+
+What:          /sys/bus/edac/devices/<dev-name>/scrubX/max_cycle_duration
+Date:          March 2025
+KernelVersion: 6.15
+Contact:       linux-edac@vger.kernel.org
+Description:
+               (RO) Supported maximum scrub cycle duration in seconds
+               by the memory scrubber.
+
+What:          /sys/bus/edac/devices/<dev-name>/scrubX/current_cycle_duration
+Date:          March 2025
+KernelVersion: 6.15
+Contact:       linux-edac@vger.kernel.org
+Description:
+               (RW) The current scrub cycle duration in seconds and must be
+               within the supported range by the memory scrubber.
+
+               Scrub has an overhead when running and that may want to be
+               reduced by taking longer to do it.
index 3c279d026bbd99d863edc348027cce78293f8dea..fdcecb3df10ee3fe79ca73e0e19027fafc76d604 100644 (file)
@@ -91,3 +91,9 @@ High level design is illustrated in the following diagram::
 3. RAS dynamic feature controller - Userspace sample modules in rasdaemon for
    dynamic scrub/repair control to issue scrubbing/repair when excess number
    of corrected memory errors are reported in a short span of time.
+
+RAS features
+------------
+1. Memory Scrub
+
+Memory scrub features are documented in `Documentation/edac/scrub.rst`.
index de4a3aa452cb172e9c108ca123dcae39a58f94c0..0a00c23838b60ed385336687f24878fe8ff1ec93 100644 (file)
@@ -8,3 +8,4 @@ EDAC Subsystem
    :maxdepth: 1
 
    features
+   scrub
diff --git a/Documentation/edac/scrub.rst b/Documentation/edac/scrub.rst
new file mode 100644 (file)
index 0000000..8b9611e
--- /dev/null
@@ -0,0 +1,264 @@
+.. SPDX-License-Identifier: GPL-2.0 OR GFDL-1.2-no-invariants-or-later
+
+=============
+Scrub Control
+=============
+
+Copyright (c) 2024-2025 HiSilicon Limited.
+
+:Author:   Shiju Jose <shiju.jose@huawei.com>
+:License:  The GNU Free Documentation License, Version 1.2 without
+           Invariant Sections, Front-Cover Texts nor Back-Cover Texts.
+           (dual licensed under the GPL v2)
+
+- Written for: 6.15
+
+Introduction
+------------
+
+Increasing DRAM size and cost have made memory subsystem reliability an
+important concern. These modules are used where potentially corrupted data
+could cause expensive or fatal issues. Memory errors are among the top
+hardware failures that cause server and workload crashes.
+
+Memory scrubbing is a feature where an ECC (Error-Correcting Code) engine
+reads data from each memory media location, corrects if necessary and writes
+the corrected data back to the same memory media location.
+
+DIMMs can be scrubbed at a configurable rate to detect uncorrected memory
+errors and attempt recovery from detected errors, providing the following
+benefits:
+
+1. Proactively scrubbing DIMMs reduces the chance of a correctable error
+   becoming uncorrectable.
+
+2. When detected, uncorrected errors caught in unallocated memory pages are
+   isolated and prevented from being allocated to an application or the OS.
+
+3. This reduces the likelihood of software or hardware products encountering
+   memory errors.
+
+4. The additional data on failures in memory may be used to build up
+   statistics that are later used to decide whether to use memory repair
+   technologies such as Post Package Repair or Sparing.
+
+There are 2 types of memory scrubbing:
+
+1. Background (patrol) scrubbing while the DRAM is otherwise idle.
+
+2. On-demand scrubbing for a specific address range or region of memory.
+
+Several types of interfaces to hardware memory scrubbers have been
+identified, such as CXL memory device patrol scrub, CXL DDR5 ECS, ACPI
+RAS2 memory scrubbing, and ACPI NVDIMM ARS (Address Range Scrub).
+
+The control mechanisms vary across different memory scrubbers. To enable
+standardized userspace tooling, there is a need to present these controls
+through a standardized ABI.
+
+A generic memory EDAC scrub control allows users to manage underlying
+scrubbers in the system through a standardized sysfs control interface.  It
+abstracts the management of various scrubbing functionalities into a unified
+set of functions.
+
+Use cases of common scrub control feature
+-----------------------------------------
+
+1. Several types of interfaces for hardware memory scrubbers have been
+   identified, including the CXL memory device patrol scrub, CXL DDR5 ECS,
+   ACPI RAS2 memory scrubbing features, ACPI NVDIMM ARS (Address Range Scrub),
+   and software-based memory scrubbers.
+
+   Of the identified interfaces to hardware memory scrubbers some support
+   control over patrol (background) scrubbing (e.g., ACPI RAS2, CXL) and/or
+   on-demand scrubbing (e.g., ACPI RAS2, ACPI ARS). However, the scrub control
+   interfaces vary between memory scrubbers, highlighting the need for
+   a standardized, generic sysfs scrub control interface that is accessible to
+   userspace for administration and use by scripts/tools.
+
+2. User-space scrub controls allow users to disable scrubbing if necessary,
+   for example, to disable background patrol scrubbing or adjust the scrub
+   rate for performance-aware operations where background activities need to
+   be minimized or disabled.
+
+3. User-space tools enable on-demand scrubbing for specific address ranges,
+   provided that the scrubber supports this functionality.
+
+4. User-space tools can also control memory DIMM scrubbing at a configurable
+   scrub rate via sysfs scrub controls. This approach offers several benefits:
+
+   4.1. Detects uncorrectable memory errors early, before user access to affected
+        memory, helping facilitate recovery.
+
+   4.2. Reduces the likelihood of correctable errors developing into uncorrectable
+        errors.
+
+5. Policy control for hotplugged memory is necessary because there may not
+   be a system-wide BIOS or similar control to manage scrub settings for a CXL
+   device added after boot. Determining these settings is a policy decision,
+   balancing reliability against performance, so userspace should control it.
+   Therefore, a unified interface is recommended for handling this function in
+   a way that aligns with other similar interfaces, rather than creating a
+   separate one.
+
+Scrubbing features
+------------------
+
+CXL Memory Scrubbing features
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+CXL spec r3.1 [1]_ section 8.2.9.9.11.1 describes the memory device patrol
+scrub control feature. The device patrol scrub proactively locates and makes
+corrections to errors in regular cycle. The patrol scrub control allows the
+userspace request to change CXL patrol scrubber's configurations.
+
+The patrol scrub control allows the requester to specify the number of
+hours in which the patrol scrub cycles must be completed, provided that
+the requested scrub rate must be within the supported range of the
+scrub rate that the device is capable of. In the CXL driver, the
+number of seconds per scrub cycles, which user requests via sysfs, is
+rescaled to hours per scrub cycles.
+
+In addition, they allow the host to disable the feature in case it interferes
+with performance-aware operations which require the background operations to
+be turned off.
+
+Error Check Scrub (ECS)
+~~~~~~~~~~~~~~~~~~~~~~~
+
+CXL spec r3.1 [1]_ section 8.2.9.9.11.2 describes Error Check Scrub (ECS)
+- a feature defined in the JEDEC DDR5 SDRAM Specification (JESD79-5) and
+allowing DRAM to internally read, correct single-bit errors, and write back
+corrected data bits to the DRAM array while providing transparency to error
+counts.
+
+The DDR5 device contains number of memory media Field Replaceable Units (FRU)
+per device. The DDR5 ECS feature and thus the ECS control driver supports
+configuring the ECS parameters per FRU.
+
+ACPI RAS2 Hardware-based Memory Scrubbing
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+ACPI spec 6.5 [2]_ section 5.2.21 ACPI RAS2 describes an ACPI RAS2 table
+which provides interfaces for platform RAS features and supports independent
+RAS controls and capabilities for a given RAS feature for multiple instances
+of the same component in a given system.
+
+Memory RAS features apply to RAS capabilities, controls and operations that
+are specific to memory. RAS2 PCC sub-spaces for memory-specific RAS features
+have a Feature Type of 0x00 (Memory).
+
+The platform can use the hardware-based memory scrubbing feature to expose
+controls and capabilities associated with hardware-based memory scrub
+engines. The RAS2 memory scrubbing feature supports as per spec,
+
+1. Independent memory scrubbing controls for each NUMA domain, identified
+   using its proximity domain.
+
+2. Provision for background (patrol) scrubbing of the entire memory system,
+   as well as on-demand scrubbing for a specific region of memory.
+
+ACPI Address Range Scrubbing (ARS)
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+ACPI spec 6.5 [2]_ section 9.19.7.2 describes Address Range Scrubbing (ARS).
+ARS allows the platform to communicate memory errors to system software.
+This capability allows system software to prevent accesses to addresses with
+uncorrectable errors in memory. ARS functions manage all NVDIMMs present in
+the system. Only one scrub can be in progress system wide at any given time.
+
+The following functions are supported as per the specification:
+
+1. Query ARS Capabilities for a given address range, indicates platform
+   supports the ACPI NVDIMM Root Device Unconsumed Error Notification.
+
+2. Start ARS triggers an Address Range Scrub for the given memory range.
+   Address scrubbing can be done for volatile or persistent memory, or both.
+
+3. Query ARS Status command allows software to get the status of ARS,
+   including the progress of ARS and ARS error record.
+
+4. Clear Uncorrectable Error.
+
+5. Translate SPA
+
+6. ARS Error Inject etc.
+
+The kernel supports an existing control for ARS and ARS is currently not
+supported in EDAC.
+
+.. [1] https://computeexpresslink.org/cxl-specification/
+
+.. [2] https://uefi.org/specs/ACPI/6.5/
+
+Comparison of various scrubbing features
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+
+ +--------------+-----------+-----------+-----------+-----------+
+ |              |   ACPI    | CXL patrol|  CXL ECS  |  ARS      |
+ |  Name        |   RAS2    | scrub     |           |           |
+ +--------------+-----------+-----------+-----------+-----------+
+ |              |           |           |           |           |
+ | On-demand    | Supported | No        | No        | Supported |
+ | Scrubbing    |           |           |           |           |
+ |              |           |           |           |           |
+ +--------------+-----------+-----------+-----------+-----------+
+ |              |           |           |           |           |
+ | Background   | Supported | Supported | Supported | No        |
+ | scrubbing    |           |           |           |           |
+ |              |           |           |           |           |
+ +--------------+-----------+-----------+-----------+-----------+
+ |              |           |           |           |           |
+ | Mode of      | Scrub ctrl| per device| per memory|  Unknown  |
+ | scrubbing    | per NUMA  |           | media     |           |
+ |              | domain.   |           |           |           |
+ +--------------+-----------+-----------+-----------+-----------+
+ |              |           |           |           |           |
+ | Query scrub  | Supported | Supported | Supported | Supported |
+ | capabilities |           |           |           |           |
+ |              |           |           |           |           |
+ +--------------+-----------+-----------+-----------+-----------+
+ |              |           |           |           |           |
+ | Setting      | Supported | No        | No        | Supported |
+ | address range|           |           |           |           |
+ |              |           |           |           |           |
+ +--------------+-----------+-----------+-----------+-----------+
+ |              |           |           |           |           |
+ | Setting      | Supported | Supported | No        | No        |
+ | scrub rate   |           |           |           |           |
+ |              |           |           |           |           |
+ +--------------+-----------+-----------+-----------+-----------+
+ |              |           |           |           |           |
+ | Unit for     | Not       | in hours  | No        | No        |
+ | scrub rate   | Defined   |           |           |           |
+ |              |           |           |           |           |
+ +--------------+-----------+-----------+-----------+-----------+
+ |              | Supported |           |           |           |
+ | Scrub        | on-demand | No        | No        | Supported |
+ | status/      | scrubbing |           |           |           |
+ | Completion   | only      |           |           |           |
+ +--------------+-----------+-----------+-----------+-----------+
+ | UC error     |           |CXL general|CXL general| ACPI UCE  |
+ | reporting    | Exception |media/DRAM |media/DRAM | notify and|
+ |              |           |event/media|event/media| query     |
+ |              |           |scan?      |scan?      | ARS status|
+ +--------------+-----------+-----------+-----------+-----------+
+ |              |           |           |           |           |
+ | Support for  | Supported | Supported | Supported | No        |
+ | EDAC control |           |           |           |           |
+ |              |           |           |           |           |
+ +--------------+-----------+-----------+-----------+-----------+
+
+The File System
+---------------
+
+The control attributes of a registered scrubber instance could be
+accessed in:
+
+/sys/bus/edac/devices/<dev-name>/scrubX/
+
+sysfs
+-----
+
+Sysfs files are documented in
+`Documentation/ABI/testing/sysfs-edac-scrub`
index 2051a7c944a58c1db71e3eae48b5016829be9056..175d706168abda54c8da492693f5481653d5d3de 100644 (file)
@@ -75,6 +75,15 @@ config EDAC_GHES
 
          In doubt, say 'Y'.
 
+config EDAC_SCRUB
+       bool "EDAC scrub feature"
+       help
+         The EDAC scrub feature is optional and is designed to control the
+         memory scrubbers in the system. The common sysfs scrub interface
+         abstracts the control of various arbitrary scrubbing functionalities
+         into a unified set of functions.
+         Say 'y/n' to enable/disable EDAC scrub feature.
+
 config EDAC_AMD64
        tristate "AMD64 (Opteron, Athlon64)"
        depends on AMD_NB && EDAC_DECODE_MCE
index 89789ba8275fe1347b4ea01698407174fe9474bb..cdbd0a07b9e6b4599969b8d2518815daa37e9f51 100644 (file)
@@ -12,6 +12,7 @@ edac_core-y   := edac_mc.o edac_device.o edac_mc_sysfs.o
 edac_core-y    += edac_module.o edac_device_sysfs.o wq.o
 
 edac_core-$(CONFIG_EDAC_DEBUG)         += debugfs.o
+edac_core-$(CONFIG_EDAC_SCRUB)         += scrub.o
 
 ifdef CONFIG_PCI
 edac_core-y    += edac_pci.o edac_pci_sysfs.o
index 6af0893cadc94282652bf55822a74ebf26c66880..54c1e2d024ac202a94781b5fae85984b22d56230 100644 (file)
@@ -575,6 +575,7 @@ static void edac_dev_release(struct device *dev)
 {
        struct edac_dev_feat_ctx *ctx = container_of(dev, struct edac_dev_feat_ctx, dev);
 
+       kfree(ctx->scrub);
        kfree(ctx->dev.groups);
        kfree(ctx);
 }
@@ -610,9 +611,11 @@ int edac_dev_register(struct device *parent, char *name,
                      const struct edac_dev_feature *ras_features)
 {
        const struct attribute_group **ras_attr_groups;
+       struct edac_dev_data *dev_data;
        struct edac_dev_feat_ctx *ctx;
        int attr_gcnt = 0;
        int ret = -ENOMEM;
+       int scrub_cnt = 0;
        int feat;
 
        if (!parent || !name || !num_features || !ras_features)
@@ -621,7 +624,10 @@ int edac_dev_register(struct device *parent, char *name,
        /* Double parse to make space for attributes */
        for (feat = 0; feat < num_features; feat++) {
                switch (ras_features[feat].ft_type) {
-               /* Add feature specific code */
+               case RAS_FEAT_SCRUB:
+                       attr_gcnt++;
+                       scrub_cnt++;
+                       break;
                default:
                        return -EINVAL;
                }
@@ -635,13 +641,37 @@ int edac_dev_register(struct device *parent, char *name,
        if (!ras_attr_groups)
                goto ctx_free;
 
+       if (scrub_cnt) {
+               ctx->scrub = kcalloc(scrub_cnt, sizeof(*ctx->scrub), GFP_KERNEL);
+               if (!ctx->scrub)
+                       goto groups_free;
+       }
+
        attr_gcnt = 0;
+       scrub_cnt = 0;
        for (feat = 0; feat < num_features; feat++, ras_features++) {
                switch (ras_features->ft_type) {
-               /* Add feature specific code */
+               case RAS_FEAT_SCRUB:
+                       if (!ras_features->scrub_ops || scrub_cnt != ras_features->instance) {
+                               ret = -EINVAL;
+                               goto data_mem_free;
+                       }
+
+                       dev_data = &ctx->scrub[scrub_cnt];
+                       dev_data->instance = scrub_cnt;
+                       dev_data->scrub_ops = ras_features->scrub_ops;
+                       dev_data->private = ras_features->ctx;
+                       ret = edac_scrub_get_desc(parent, &ras_attr_groups[attr_gcnt],
+                                                 ras_features->instance);
+                       if (ret)
+                               goto data_mem_free;
+
+                       scrub_cnt++;
+                       attr_gcnt++;
+                       break;
                default:
                        ret = -EINVAL;
-                       goto groups_free;
+                       goto data_mem_free;
                }
        }
 
@@ -654,7 +684,7 @@ int edac_dev_register(struct device *parent, char *name,
 
        ret = dev_set_name(&ctx->dev, name);
        if (ret)
-               goto groups_free;
+               goto data_mem_free;
 
        ret = device_register(&ctx->dev);
        if (ret) {
@@ -664,6 +694,8 @@ int edac_dev_register(struct device *parent, char *name,
 
        return devm_add_action_or_reset(parent, edac_dev_unreg, &ctx->dev);
 
+data_mem_free:
+       kfree(ctx->scrub);
 groups_free:
        kfree(ras_attr_groups);
 ctx_free:
diff --git a/drivers/edac/scrub.c b/drivers/edac/scrub.c
new file mode 100755 (executable)
index 0000000..e421d3e
--- /dev/null
@@ -0,0 +1,209 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * The generic EDAC scrub driver controls the memory scrubbers in the
+ * system. The common sysfs scrub interface abstracts the control of
+ * various arbitrary scrubbing functionalities into a unified set of
+ * functions.
+ *
+ * Copyright (c) 2024-2025 HiSilicon Limited.
+ */
+
+#include <linux/edac.h>
+
+enum edac_scrub_attributes {
+       SCRUB_ADDRESS,
+       SCRUB_SIZE,
+       SCRUB_ENABLE_BACKGROUND,
+       SCRUB_MIN_CYCLE_DURATION,
+       SCRUB_MAX_CYCLE_DURATION,
+       SCRUB_CUR_CYCLE_DURATION,
+       SCRUB_MAX_ATTRS
+};
+
+struct edac_scrub_dev_attr {
+       struct device_attribute dev_attr;
+       u8 instance;
+};
+
+struct edac_scrub_context {
+       char name[EDAC_FEAT_NAME_LEN];
+       struct edac_scrub_dev_attr scrub_dev_attr[SCRUB_MAX_ATTRS];
+       struct attribute *scrub_attrs[SCRUB_MAX_ATTRS + 1];
+       struct attribute_group group;
+};
+
+#define TO_SCRUB_DEV_ATTR(_dev_attr)      \
+               container_of(_dev_attr, struct edac_scrub_dev_attr, dev_attr)
+
+#define EDAC_SCRUB_ATTR_SHOW(attrib, cb, type, format)                         \
+static ssize_t attrib##_show(struct device *ras_feat_dev,                      \
+                            struct device_attribute *attr, char *buf)          \
+{                                                                              \
+       u8 inst = TO_SCRUB_DEV_ATTR(attr)->instance;                            \
+       struct edac_dev_feat_ctx *ctx = dev_get_drvdata(ras_feat_dev);          \
+       const struct edac_scrub_ops *ops = ctx->scrub[inst].scrub_ops;          \
+       type data;                                                              \
+       int ret;                                                                \
+                                                                               \
+       ret = ops->cb(ras_feat_dev->parent, ctx->scrub[inst].private, &data);   \
+       if (ret)                                                                \
+               return ret;                                                     \
+                                                                               \
+       return sysfs_emit(buf, format, data);                                   \
+}
+
+EDAC_SCRUB_ATTR_SHOW(addr, read_addr, u64, "0x%llx\n")
+EDAC_SCRUB_ATTR_SHOW(size, read_size, u64, "0x%llx\n")
+EDAC_SCRUB_ATTR_SHOW(enable_background, get_enabled_bg, bool, "%u\n")
+EDAC_SCRUB_ATTR_SHOW(min_cycle_duration, get_min_cycle, u32, "%u\n")
+EDAC_SCRUB_ATTR_SHOW(max_cycle_duration, get_max_cycle, u32, "%u\n")
+EDAC_SCRUB_ATTR_SHOW(current_cycle_duration, get_cycle_duration, u32, "%u\n")
+
+#define EDAC_SCRUB_ATTR_STORE(attrib, cb, type, conv_func)                     \
+static ssize_t attrib##_store(struct device *ras_feat_dev,                     \
+                             struct device_attribute *attr,                    \
+                             const char *buf, size_t len)                      \
+{                                                                              \
+       u8 inst = TO_SCRUB_DEV_ATTR(attr)->instance;                            \
+       struct edac_dev_feat_ctx *ctx = dev_get_drvdata(ras_feat_dev);          \
+       const struct edac_scrub_ops *ops = ctx->scrub[inst].scrub_ops;          \
+       type data;                                                              \
+       int ret;                                                                \
+                                                                               \
+       ret = conv_func(buf, 0, &data);                                         \
+       if (ret < 0)                                                            \
+               return ret;                                                     \
+                                                                               \
+       ret = ops->cb(ras_feat_dev->parent, ctx->scrub[inst].private, data);    \
+       if (ret)                                                                \
+               return ret;                                                     \
+                                                                               \
+       return len;                                                             \
+}
+
+EDAC_SCRUB_ATTR_STORE(addr, write_addr, u64, kstrtou64)
+EDAC_SCRUB_ATTR_STORE(size, write_size, u64, kstrtou64)
+EDAC_SCRUB_ATTR_STORE(enable_background, set_enabled_bg, unsigned long, kstrtoul)
+EDAC_SCRUB_ATTR_STORE(current_cycle_duration, set_cycle_duration, unsigned long, kstrtoul)
+
+static umode_t scrub_attr_visible(struct kobject *kobj, struct attribute *a, int attr_id)
+{
+       struct device *ras_feat_dev = kobj_to_dev(kobj);
+       struct device_attribute *dev_attr = container_of(a, struct device_attribute, attr);
+       u8 inst = TO_SCRUB_DEV_ATTR(dev_attr)->instance;
+       struct edac_dev_feat_ctx *ctx = dev_get_drvdata(ras_feat_dev);
+       const struct edac_scrub_ops *ops = ctx->scrub[inst].scrub_ops;
+
+       switch (attr_id) {
+       case SCRUB_ADDRESS:
+               if (ops->read_addr) {
+                       if (ops->write_addr)
+                               return a->mode;
+                       else
+                               return 0444;
+               }
+               break;
+       case SCRUB_SIZE:
+               if (ops->read_size) {
+                       if (ops->write_size)
+                               return a->mode;
+                       else
+                               return 0444;
+               }
+               break;
+       case SCRUB_ENABLE_BACKGROUND:
+               if (ops->get_enabled_bg) {
+                       if (ops->set_enabled_bg)
+                               return a->mode;
+                       else
+                               return 0444;
+               }
+               break;
+       case SCRUB_MIN_CYCLE_DURATION:
+               if (ops->get_min_cycle)
+                       return a->mode;
+               break;
+       case SCRUB_MAX_CYCLE_DURATION:
+               if (ops->get_max_cycle)
+                       return a->mode;
+               break;
+       case SCRUB_CUR_CYCLE_DURATION:
+               if (ops->get_cycle_duration) {
+                       if (ops->set_cycle_duration)
+                               return a->mode;
+                       else
+                               return 0444;
+               }
+               break;
+       default:
+               break;
+       }
+
+       return 0;
+}
+
+#define EDAC_SCRUB_ATTR_RO(_name, _instance)       \
+       ((struct edac_scrub_dev_attr) { .dev_attr = __ATTR_RO(_name), \
+                                       .instance = _instance })
+
+#define EDAC_SCRUB_ATTR_WO(_name, _instance)       \
+       ((struct edac_scrub_dev_attr) { .dev_attr = __ATTR_WO(_name), \
+                                       .instance = _instance })
+
+#define EDAC_SCRUB_ATTR_RW(_name, _instance)       \
+       ((struct edac_scrub_dev_attr) { .dev_attr = __ATTR_RW(_name), \
+                                       .instance = _instance })
+
+static int scrub_create_desc(struct device *scrub_dev,
+                            const struct attribute_group **attr_groups, u8 instance)
+{
+       struct edac_scrub_context *scrub_ctx;
+       struct attribute_group *group;
+       int i;
+       struct edac_scrub_dev_attr dev_attr[] = {
+               [SCRUB_ADDRESS] = EDAC_SCRUB_ATTR_RW(addr, instance),
+               [SCRUB_SIZE] = EDAC_SCRUB_ATTR_RW(size, instance),
+               [SCRUB_ENABLE_BACKGROUND] = EDAC_SCRUB_ATTR_RW(enable_background, instance),
+               [SCRUB_MIN_CYCLE_DURATION] = EDAC_SCRUB_ATTR_RO(min_cycle_duration, instance),
+               [SCRUB_MAX_CYCLE_DURATION] = EDAC_SCRUB_ATTR_RO(max_cycle_duration, instance),
+               [SCRUB_CUR_CYCLE_DURATION] = EDAC_SCRUB_ATTR_RW(current_cycle_duration, instance)
+       };
+
+       scrub_ctx = devm_kzalloc(scrub_dev, sizeof(*scrub_ctx), GFP_KERNEL);
+       if (!scrub_ctx)
+               return -ENOMEM;
+
+       group = &scrub_ctx->group;
+       for (i = 0; i < SCRUB_MAX_ATTRS; i++) {
+               memcpy(&scrub_ctx->scrub_dev_attr[i], &dev_attr[i], sizeof(dev_attr[i]));
+               scrub_ctx->scrub_attrs[i] = &scrub_ctx->scrub_dev_attr[i].dev_attr.attr;
+       }
+       sprintf(scrub_ctx->name, "%s%d", "scrub", instance);
+       group->name = scrub_ctx->name;
+       group->attrs = scrub_ctx->scrub_attrs;
+       group->is_visible  = scrub_attr_visible;
+
+       attr_groups[0] = group;
+
+       return 0;
+}
+
+/**
+ * edac_scrub_get_desc - get EDAC scrub descriptors
+ * @scrub_dev: client device, with scrub support
+ * @attr_groups: pointer to attribute group container
+ * @instance: device's scrub instance number.
+ *
+ * Return:
+ *  * %0       - Success.
+ *  * %-EINVAL - Invalid parameters passed.
+ *  * %-ENOMEM - Dynamic memory allocation failed.
+ */
+int edac_scrub_get_desc(struct device *scrub_dev,
+                       const struct attribute_group **attr_groups, u8 instance)
+{
+       if (!scrub_dev || !attr_groups)
+               return -EINVAL;
+
+       return scrub_create_desc(scrub_dev, attr_groups, instance);
+}
index 8c4b6ca2a994dca6f942c8b63a7bf1cefb12d323..1cbab08720df195dd6c6c5ef52b58de9a6fbaae5 100644 (file)
@@ -662,13 +662,54 @@ static inline struct dimm_info *edac_get_dimm(struct mem_ctl_info *mci,
        return mci->dimms[index];
 }
 
+#define EDAC_FEAT_NAME_LEN     128
+
 /* RAS feature type */
 enum edac_dev_feat {
+       RAS_FEAT_SCRUB,
        RAS_FEAT_MAX
 };
 
+/**
+ * struct edac_scrub_ops - scrub device operations (all elements optional)
+ * @read_addr: read base address of scrubbing range.
+ * @read_size: read offset of scrubbing range.
+ * @write_addr: set base address of the scrubbing range.
+ * @write_size: set offset of the scrubbing range.
+ * @get_enabled_bg: check if currently performing background scrub.
+ * @set_enabled_bg: start or stop a bg-scrub.
+ * @get_min_cycle: get minimum supported scrub cycle duration in seconds.
+ * @get_max_cycle: get maximum supported scrub cycle duration in seconds.
+ * @get_cycle_duration: get current scrub cycle duration in seconds.
+ * @set_cycle_duration: set current scrub cycle duration in seconds.
+ */
+struct edac_scrub_ops {
+       int (*read_addr)(struct device *dev, void *drv_data, u64 *base);
+       int (*read_size)(struct device *dev, void *drv_data, u64 *size);
+       int (*write_addr)(struct device *dev, void *drv_data, u64 base);
+       int (*write_size)(struct device *dev, void *drv_data, u64 size);
+       int (*get_enabled_bg)(struct device *dev, void *drv_data, bool *enable);
+       int (*set_enabled_bg)(struct device *dev, void *drv_data, bool enable);
+       int (*get_min_cycle)(struct device *dev, void *drv_data,  u32 *min);
+       int (*get_max_cycle)(struct device *dev, void *drv_data,  u32 *max);
+       int (*get_cycle_duration)(struct device *dev, void *drv_data, u32 *cycle);
+       int (*set_cycle_duration)(struct device *dev, void *drv_data, u32 cycle);
+};
+
+#if IS_ENABLED(CONFIG_EDAC_SCRUB)
+int edac_scrub_get_desc(struct device *scrub_dev,
+                       const struct attribute_group **attr_groups,
+                       u8 instance);
+#else
+static inline int edac_scrub_get_desc(struct device *scrub_dev,
+                                     const struct attribute_group **attr_groups,
+                                     u8 instance)
+{ return -EOPNOTSUPP; }
+#endif /* CONFIG_EDAC_SCRUB */
+
 /* EDAC device feature information structure */
 struct edac_dev_data {
+       const struct edac_scrub_ops *scrub_ops;
        u8 instance;
        void *private;
 };
@@ -676,11 +717,13 @@ struct edac_dev_data {
 struct edac_dev_feat_ctx {
        struct device dev;
        void *private;
+       struct edac_dev_data *scrub;
 };
 
 struct edac_dev_feature {
        enum edac_dev_feat ft_type;
        u8 instance;
+       const struct edac_scrub_ops *scrub_ops;
        void *ctx;
 };