Re: [PATCH V5 OLK-5.10] soc: hisilicon: hisi_hbmdev: Add hbm acls repair and query methods

v5 - 增加了两行注释 /* AREP will return a positive error code to represent error status */ /* AQRY will return a positive error code to represent error status */

v4 - remove "Offering HULK" to pass CI patch format check

v3

- Add #include <linux/mm.h> to avoid compile failure in arch x86_64

v2: - Change "u64 pfn" ot "unsigned long pfn" to avoid compile error - Add two stub to pretiffy the code

在 2023/11/6 11:10, Zhang Zekun 写道:

hulk inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8CCP5

---

Hbm memory device add support for acls hot repair. The patch add two methods for userpace:

• query a paddr if it support acls repair
• repair a paddr in hbm memory device

The feature of ACLS hot repair can help to fix a memory error from userspace by passing through the error physical address to HBM hardware.

Signed-off-by: Zhang Zekun zhangzekun11@huawei.com

drivers/soc/hisilicon/Kconfig | 10 +++ drivers/soc/hisilicon/hisi_hbmdev.c | 134 ++++++++++++++++++++++++++++ 2 files changed, 144 insertions(+)

diff --git a/drivers/soc/hisilicon/Kconfig b/drivers/soc/hisilicon/Kconfig index 383375f20cac..45754528873e 100644 --- a/drivers/soc/hisilicon/Kconfig +++ b/drivers/soc/hisilicon/Kconfig @@ -44,4 +44,14 @@ config HISI_HBMCACHE To compile the driver as a module, choose M here: the module will be called hisi_hbmcache.

+config HISI_HBMDEV_ACLS

• bool "Add support for HISI ACLS repair"
• depends on HISI_HBMDEV
• help

•  Add ACLS support for hbm device, which can be used to query and
  

•  repair hardware error in HBM devices. This feature need to work with
  

•  hardware firmwares.
  

•  If not sure say no.
  

• endmenu

diff --git a/drivers/soc/hisilicon/hisi_hbmdev.c b/drivers/soc/hisilicon/hisi_hbmdev.c index 5b6b1618148c..a9cc78bde81b 100644 --- a/drivers/soc/hisilicon/hisi_hbmdev.c +++ b/drivers/soc/hisilicon/hisi_hbmdev.c @@ -11,6 +11,7 @@ #include <linux/node.h> #include <linux/arch_topology.h> #include <linux/memory_hotplug.h> +#include <linux/mm.h>

#include "hisi_internal.h"

@@ -25,6 +26,9 @@ struct cdev_node { struct memory_dev { struct kobject *memdev_kobj; struct kobject *topo_kobj; +#ifdef CONFIG_HISI_HBMDEV_ACLS

• struct kobject *acls_kobj;

+#endif struct cdev_node cdev_list; nodemask_t cluster_cpumask[MAX_NUMNODES]; }; @@ -85,6 +89,134 @@ static void memory_topo_init(void) kobject_put(mdev->topo_kobj); }

+#ifdef CONFIG_HISI_HBMDEV_ACLS +static struct acpi_device *paddr_to_acpi_device(u64 paddr) +{

• unsigned long pfn;
• int nid;
• pfn = __phys_to_pfn(paddr);
• if (!pfn_valid(pfn))

• return NULL;
  

• nid = pfn_to_nid(pfn);
• if (nid < 0 && nid >= MAX_NUMNODES)

• return NULL;
  

• return hotplug_mdev[nid];

+}

+static ssize_t acls_query_store(struct kobject *kobj,

• 		struct kobj_attribute *attr,
  

• 		const char *buf, size_t count)
  

+{

• struct acpi_object_list arg_list;
• struct acpi_device *adev;
• union acpi_object obj;
• acpi_status status;
• u64 paddr, res;
• if (kstrtoull(buf, 16, &paddr))

• return -EINVAL;
  

• adev = paddr_to_acpi_device(paddr);
• if (!adev)

• return -EINVAL;
  

• obj.type = ACPI_TYPE_INTEGER;
• obj.integer.value = paddr;
• arg_list.count = 1;
• arg_list.pointer = &obj;
• status = acpi_evaluate_integer(adev->handle, "AQRY", &arg_list, &res);
• if (ACPI_FAILURE(status))

• return -ENODEV;
  

• /* AQRY will return a positive error code to represent error status */
• if (IS_ERR_VALUE(-res))

• return -res;
  

• else if (res)

• return -ENODEV;
  

• return count;

+}

+static struct kobj_attribute acls_query_store_attribute =

• __ATTR(acls_query, 0200, NULL, acls_query_store);

+static ssize_t acls_repair_store(struct kobject *kobj,

• 		struct kobj_attribute *attr,
  

• 		const char *buf, size_t count)
  

+{

• struct acpi_object_list arg_list;
• struct acpi_device *adev;
• union acpi_object obj;
• acpi_status status;
• u64 paddr, res;
• if (kstrtoull(buf, 16, &paddr))

• return -EINVAL;
  

• adev = paddr_to_acpi_device(paddr);
• if (!adev)

• return -EINVAL;
  

• obj.type = ACPI_TYPE_INTEGER;
• obj.integer.value = paddr;
• arg_list.count = 1;
• arg_list.pointer = &obj;
• status = acpi_evaluate_integer(adev->handle, "AREP", &arg_list, &res);
• if (ACPI_FAILURE(status))

• return -ENODEV;
  

• /* AREP will return a positive error code to represent error status */
• if (IS_ERR_VALUE(-res))

• return -res;
  

• else if (res)

• return -ENODEV;
  

• return count;

+} +static struct kobj_attribute acls_repair_store_attribute =

• __ATTR(acls_repair, 0200, NULL, acls_repair_store);

+static struct attribute *acls_attrs[] = {

• &acls_query_store_attribute.attr,
• &acls_repair_store_attribute.attr,
• NULL,

+};

+static struct attribute_group acls_attr_group = {

• .attrs = acls_attrs,

+};

+static void acls_init(void) +{

• int ret = -ENOMEM;
• mdev->acls_kobj = kobject_create_and_add("acls", mdev->memdev_kobj);
• if (!mdev->acls_kobj)

• goto out;
  

• ret = sysfs_create_group(mdev->acls_kobj, &acls_attr_group);
• if (ret)

• kobject_put(mdev->acls_kobj);
  

+out:

• if (ret)

• pr_err("ACLS hot repair is not enabled\n");
  

+}

+static void acls_remove(void) +{

• kobject_put(mdev->acls_kobj);

+} +#else +static void acls_init(void) {} +static void acls_remove(void) {} +#endif

• static int get_pxm(struct acpi_device *acpi_device, void *arg) { acpi_handle handle = acpi_device->handle;

@@ -284,6 +416,7 @@ static int __init mdev_init(void) }

memory_topo_init();

• acls_init(); return ret; } module_init(mdev_init);

@@ -293,6 +426,7 @@ static void __exit mdev_exit(void) container_remove(); kobject_put(mdev->memdev_kobj); kobject_put(mdev->topo_kobj);

• acls_remove(); kfree(mdev); } module_exit(mdev_exit);