x86/sgx: Rename fallback labels in sgx_init()

PATCH Superseded Jarkko Sakkinen 2 participants 7 messages Nov 3, 2021 – Nov 5, 2021

Subsystems (3)

x86 Driver core Intel SGX

lore

Share & tools

lore

Jarkko Sakkinen

author maintainer

2021-11-03 01:28 UTC #0 lore raw

PATCH v11 1/2 x86/sgx: Rename fallback labels in sgx_init()

It's hard to add new content to this function because it is time consuming to match fallback and its cause. Rename labels in a way that the name of error label refers to the site where failure happened. This way it is easier to keep on track what is going on.

Signed-off-by Jarkko Sakkinen <[email protected]>

---
v5:
* A new patch.

---
 arch/x86/kernel/cpu/sgx/main.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

arch/x86/kernel/cpu/sgx/main.c +6 -6

index 63d3de02bbcc..a6e313f1a82d 100644
@@ -803,12 +803,12 @@ static int __init sgx_init(void)
 
 	if (!sgx_page_reclaimer_init()) {
 		ret = -ENOMEM;
-		goto err_page_cache;
+		goto err_reclaimer;
 	}
 
 	ret = misc_register(&sgx_dev_provision);
 	if (ret)
-		goto err_kthread;
+		goto err_provision;
 
 	/*
 	 * Always try to initialize the native *and* KVM drivers.

@@ -821,17 +821,17 @@ static int __init sgx_init(void)
 	ret = sgx_drv_init();
 
 	if (sgx_vepc_init() && ret)
-		goto err_provision;
+		goto err_driver;
 
 	return 0;
 
-err_provision:
+err_driver:
 	misc_deregister(&sgx_dev_provision);
 
-err_kthread:
+err_provision:
 	kthread_stop(ksgxd_tsk);
 
-err_page_cache:
+err_reclaimer:
 	for (i = 0; i < sgx_nr_epc_sections; i++) {
 		vfree(sgx_epc_sections[i].pages);
 		memunmap(sgx_epc_sections[i].virt_addr);

Signature · 1 line

2.32.0

Jarkko Sakkinen

author maintainer

2021-11-03 01:28 UTC #1 lore raw

PATCH v11 2/2 x86/sgx: Add an attribute for the amount of SGX memory in a NUMA node

The amount of SGX memory on the system is determined by the BIOS and it varies wildly between systems. It can be from dozens of MB's on desktops or VM's, up to many GB's on servers. Just like for regular memory, it is sometimes useful to know the amount of usable SGX memory in the system.

Introduce CONFIG_HAVE_ARCH_NODE_DEV_GROUP opt-in flag to expose an arch specific attribute group, and add an attribute for the amount of SGX memory in bytes to each NUMA node:

/sys/devices/system/node/nodeX/x86/sgx_total_bytes

Cc Greg Kroah-Hartman <[email protected]>

Signed-off-by Jarkko Sakkinen <[email protected]>

---
v11:
* Fix documentation and the commit message.

v10:
* Change DEVICE_ATTR_RO() to static (Greg K-H)
* Change the attribute name as sgx_total_bytes, and attribute group name as "x86" (Dave).
* Add a new config flag HAVE_ARCH_NODE_DEV_GROUP to identify, whether an arch exports arch specific attribute group for NUMA nodes.

v9:
* Fix racy initialization of sysfs attributes:
https://lore.kernel.org/linux-sgx/[email protected]/

v8:
* Fix a bug in sgx_numa_init(): node->dev should be only set after sysfe_create_group(). Otherwise, sysfs_remove_group() will issue a warning in sgx_numa_exit(), when sgx_create_group() is unsuccessful, because the group does not exist.

v7:
* Shorten memory_size to size. The prefix makes the name only longer but does not clarify things more than "size" would.
* Use device_attribute instead of kobj_attribute.
* Use named attribute group instead of creating raw kobject just for the "sgx" subdirectory.

v6:
* Initialize node->size to zero in sgx_setup_epc_section(), when the node is first accessed.

v5
* A new patch based on the discussion on
https://lore.kernel.org/linux-sgx/[email protected]/T/#t

---
 Documentation/ABI/stable/sysfs-devices-node |  6 ++++
 arch/Kconfig                                |  4 +++
 arch/x86/Kconfig                            |  1 +
 arch/x86/kernel/cpu/sgx/main.c              | 31 +++++++++++++++++++++
 arch/x86/kernel/cpu/sgx/sgx.h               |  2 ++
 drivers/base/node.c                         | 13 ++++++++-
 include/linux/numa.h                        |  4 +++
 7 files changed, 60 insertions(+), 1 deletion(-)

Documentation/ABI/stable/sysfs-devices-node +6 -0

index 484fc04bcc25..8db67aa472f1 100644
@@ -176,3 +176,9 @@ Contact:	Keith Busch <[email protected]>
 Description:
 		The cache write policy: 0 for write-back, 1 for write-through,
 		other or unknown.
+
+What:		/sys/devices/system/node/nodeX/x86/sgx_total_bytes
+Date:		November 2021
+Contact:	Jarkko Sakkinen <[email protected]>
+Description:
+		The total amount of SGX physical memory in bytes.

arch/Kconfig +4 -0

index 98db63496bab..ca5d75b5a427 100644
@@ -1285,6 +1285,10 @@ config ARCH_HAS_ELFCORE_COMPAT
 config ARCH_HAS_PARANOID_L1D_FLUSH
 	bool
 
+# Select, if arch has a named attribute group bound to NUMA device nodes.
+config HAVE_ARCH_NODE_DEV_GROUP
+	bool
+
 source "kernel/gcov/Kconfig"
 
 source "scripts/gcc-plugins/Kconfig"

arch/x86/Kconfig +1 -0

index 421fa9e38c60..8503c3bdf63f 100644
@@ -266,6 +266,7 @@ config X86
 	select HAVE_ARCH_KCSAN			if X86_64
 	select X86_FEATURE_NAMES		if PROC_FS
 	select PROC_PID_ARCH_STATUS		if PROC_FS
+	select HAVE_ARCH_NODE_DEV_GROUP		if X86_SGX
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 
 config INSTRUCTION_DECODER

arch/x86/kernel/cpu/sgx/main.c +31 -0

index a6e313f1a82d..02ebb233c511 100644
@@ -714,9 +714,12 @@ static bool __init sgx_page_cache_init(void)
 			spin_lock_init(&sgx_numa_nodes[nid].lock);
 			INIT_LIST_HEAD(&sgx_numa_nodes[nid].free_page_list);
 			node_set(nid, sgx_numa_mask);
+			sgx_numa_nodes[nid].size = 0;
 		}
 
 		sgx_epc_sections[i].node =  &sgx_numa_nodes[nid];
+		sgx_numa_nodes[nid].dev = &node_devices[nid]->dev;
+		sgx_numa_nodes[nid].size += size;
 
 		sgx_nr_epc_sections++;
 	}

@@ -790,6 +793,34 @@ int sgx_set_attribute(unsigned long *allowed_attributes,
 }
 EXPORT_SYMBOL_GPL(sgx_set_attribute);
 
+#ifdef CONFIG_NUMA
+static ssize_t sgx_total_bytes_show(struct device *dev, struct device_attribute *attr, char *buf)
+{
+	unsigned long size = 0;
+	int nid;
+
+	for (nid = 0; nid < num_possible_nodes(); nid++) {
+		if (dev == sgx_numa_nodes[nid].dev) {
+			size = sgx_numa_nodes[nid].size;
+			break;
+		}
+	}
+
+	return sysfs_emit(buf, "%lu\n", size);
+}
+static DEVICE_ATTR_RO(sgx_total_bytes);
+
+static struct attribute *arch_node_dev_attrs[] = {
+	&dev_attr_sgx_total_bytes.attr,
+	NULL,
+};
+
+const struct attribute_group arch_node_dev_group = {
+	.name = "x86",
+	.attrs = arch_node_dev_attrs,
+};
+#endif /* CONFIG_NUMA */
+
 static int __init sgx_init(void)
 {
 	int ret;

arch/x86/kernel/cpu/sgx/sgx.h +2 -0

index 4628acec0009..1de8c627a286 100644
@@ -39,6 +39,8 @@ struct sgx_epc_page {
  */
 struct sgx_numa_node {
 	struct list_head free_page_list;
+	struct device *dev;
+	unsigned long size;
 	spinlock_t lock;
 };

drivers/base/node.c +12 -1

index 4a4ae868ad9f..8da977895b9a 100644
@@ -565,7 +565,18 @@ static struct attribute *node_dev_attrs[] = {
 	&dev_attr_vmstat.attr,
 	NULL
 };
-ATTRIBUTE_GROUPS(node_dev);
+
+static const struct attribute_group node_dev_group = {
+	.attrs = node_dev_attrs,
+};
+
+static const struct attribute_group *node_dev_groups[] = {
+	&node_dev_group,
+#ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
+	&arch_node_dev_group,
+#endif
+	NULL,
+};
 
 #ifdef CONFIG_HUGETLBFS
 /*

include/linux/numa.h +4 -0

index cb44cfe2b725..59df211d051f 100644
@@ -58,4 +58,8 @@ static inline int phys_to_target_node(u64 start)
 }
 #endif
 
+#ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
+extern const struct attribute_group arch_node_dev_group;
+#endif
+
 #endif /* _LINUX_NUMA_H */

Signature · 1 line

2.32.0

Greg Kroah-Hartman

maintainer

2021-11-03 08:22 UTC #2 lore raw

Re: [PATCH v11 2/2] x86/sgx: Add an attribute for the amount of SGX memory in a NUMA node

Quoted 104 lines · Jarkko Sakkinen

The amount of SGX memory on the system is determined by the BIOS and it varies wildly between systems. It can be from dozens of MB's on desktops or VM's, up to many GB's on servers. Just like for regular memory, it is sometimes useful to know the amount of usable SGX memory in the system.
Introduce CONFIG_HAVE_ARCH_NODE_DEV_GROUP opt-in flag to expose an arch specific attribute group, and add an attribute for the amount of SGX memory in bytes to each NUMA node:
/sys/devices/system/node/nodeX/x86/sgx_total_bytes
Cc Greg Kroah-Hartman <[email protected]>
Signed-off-by Jarkko Sakkinen <[email protected]>
---
v11:
* Fix documentation and the commit message.
v10:
* Change DEVICE_ATTR_RO() to static (Greg K-H)
* Change the attribute name as sgx_total_bytes, and attribute group name as "x86" (Dave).
* Add a new config flag HAVE_ARCH_NODE_DEV_GROUP to identify, whether an arch exports arch specific attribute group for NUMA nodes.
v9:
* Fix racy initialization of sysfs attributes:
https://lore.kernel.org/linux-sgx/[email protected]/
v8:
* Fix a bug in sgx_numa_init(): node->dev should be only set after sysfe_create_group(). Otherwise, sysfs_remove_group() will issue a warning in sgx_numa_exit(), when sgx_create_group() is unsuccessful, because the group does not exist.
v7:
* Shorten memory_size to size. The prefix makes the name only longer but does not clarify things more than "size" would.
* Use device_attribute instead of kobj_attribute.
* Use named attribute group instead of creating raw kobject just for the "sgx" subdirectory.
v6:
* Initialize node->size to zero in sgx_setup_epc_section(), when the node is first accessed.
v5
* A new patch based on the discussion on
https://lore.kernel.org/linux-sgx/[email protected]/T/#t
---
 Documentation/ABI/stable/sysfs-devices-node |  6 ++++
 arch/Kconfig                                |  4 +++
 arch/x86/Kconfig                            |  1 +
 arch/x86/kernel/cpu/sgx/main.c              | 31 +++++++++++++++++++++
 arch/x86/kernel/cpu/sgx/sgx.h               |  2 ++
 drivers/base/node.c                         | 13 ++++++++-
 include/linux/numa.h                        |  4 +++
 7 files changed, 60 insertions(+), 1 deletion(-)
Documentation/ABI/stable/sysfs-devices-node +6 -0
index 484fc04bcc25..8db67aa472f1 100644
@@ -176,3 +176,9 @@ Contact:	Keith Busch <[email protected]>
 Description:
 		The cache write policy: 0 for write-back, 1 for write-through,
 		other or unknown.
+
+What:		/sys/devices/system/node/nodeX/x86/sgx_total_bytes
+Date:		November 2021
+Contact:	Jarkko Sakkinen <[email protected]>
+Description:
+		The total amount of SGX physical memory in bytes.
arch/Kconfig +4 -0
index 98db63496bab..ca5d75b5a427 100644
@@ -1285,6 +1285,10 @@ config ARCH_HAS_ELFCORE_COMPAT
 config ARCH_HAS_PARANOID_L1D_FLUSH
 	bool
 
+# Select, if arch has a named attribute group bound to NUMA device nodes.
+config HAVE_ARCH_NODE_DEV_GROUP
+	bool
+
 source "kernel/gcov/Kconfig"
 
 source "scripts/gcc-plugins/Kconfig"
arch/x86/Kconfig +1 -0
index 421fa9e38c60..8503c3bdf63f 100644
@@ -266,6 +266,7 @@ config X86
 	select HAVE_ARCH_KCSAN			if X86_64
 	select X86_FEATURE_NAMES		if PROC_FS
 	select PROC_PID_ARCH_STATUS		if PROC_FS
+	select HAVE_ARCH_NODE_DEV_GROUP		if X86_SGX
 	imply IMA_SECURE_AND_OR_TRUSTED_BOOT    if EFI
 
 config INSTRUCTION_DECODER
arch/x86/kernel/cpu/sgx/main.c +0 -0
index a6e313f1a82d..02ebb233c511 100644

arch/x86/kernel/cpu/sgx/main.c static bool __init sgx_page_cache_init(void) · line 721

Quoted 8 lines

arch/x86/kernel/cpu/sgx/main.c +2 -0

@@ -714,9 +714,12 @@ static bool __init sgx_page_cache_init(void)
 			spin_lock_init(&sgx_numa_nodes[nid].lock);
 			INIT_LIST_HEAD(&sgx_numa_nodes[nid].free_page_list);
 			node_set(nid, sgx_numa_mask);
+			sgx_numa_nodes[nid].size = 0;
 		}
 
 		sgx_epc_sections[i].node =  &sgx_numa_nodes[nid];
+		sgx_numa_nodes[nid].dev = &node_devices[nid]->dev;

You are saving off a pointer to an object without incrementing the reference count of it? What prevents it from going away in the future?

Quoted 6 lines

arch/x86/kernel/cpu/sgx/main.c +1 -0

+		sgx_numa_nodes[nid].size += size;
 
 		sgx_nr_epc_sections++;
 	}

@@ -790,6 +793,34 @@ int sgx_set_attribute(unsigned long *allowed_attributes,
 }

arch/x86/kernel/cpu/sgx/main.c int sgx_set_attribute(unsigned long *allowed_attributes, · line 803

Quoted 10 lines

arch/x86/kernel/cpu/sgx/main.c +8 -0

 EXPORT_SYMBOL_GPL(sgx_set_attribute);
 
+#ifdef CONFIG_NUMA
+static ssize_t sgx_total_bytes_show(struct device *dev, struct device_attribute *attr, char *buf)
+{
+	unsigned long size = 0;
+	int nid;
+
+	for (nid = 0; nid < num_possible_nodes(); nid++) {
+		if (dev == sgx_numa_nodes[nid].dev) {

Why aren't these values all just part of the device that is being used here? You are walking some odd array, with no locking, and no reference counting on the objects you are looking at, just to find the same object that you started out with???

That's not the proper thing to do here at all, these values should be part of the device structure that this attribute lives on, in order to properly handle all of these reference counting and locking issues automatically.

Please fix the design of this code first, _before_ adding new attributes.

thanks,

greg k-h

Jarkko Sakkinen

author maintainer

2021-11-03 21:49 UTC #3 lore raw

Re: [PATCH v11 2/2] x86/sgx: Add an attribute for the amount of SGX memory in a NUMA node

Quoted 108 lines · Greg Kroah-Hartman

Quoted 107 lines · Jarkko Sakkinen
The amount of SGX memory on the system is determined by the BIOS and it varies wildly between systems. It can be from dozens of MB's on desktops or VM's, up to many GB's on servers. Just like for regular memory, it is sometimes useful to know the amount of usable SGX memory in the system.
Introduce CONFIG_HAVE_ARCH_NODE_DEV_GROUP opt-in flag to expose an arch specific attribute group, and add an attribute for the amount of SGX memory in bytes to each NUMA node:
/sys/devices/system/node/nodeX/x86/sgx_total_bytes
Cc Greg Kroah-Hartman <[email protected]>
Signed-off-by Jarkko Sakkinen <[email protected]>
---
v11:
* Fix documentation and the commit message.
v10:
* Change DEVICE_ATTR_RO() to static (Greg K-H)
* Change the attribute name as sgx_total_bytes, and attribute group name as "x86" (Dave).
* Add a new config flag HAVE_ARCH_NODE_DEV_GROUP to identify, whether an arch exports arch specific attribute group for NUMA nodes.
v9:
* Fix racy initialization of sysfs attributes:
https://lore.kernel.org/linux-sgx/[email protected]/
v8:
* Fix a bug in sgx_numa_init(): node->dev should be only set after sysfe_create_group(). Otherwise, sysfs_remove_group() will issue a warning in sgx_numa_exit(), when sgx_create_group() is unsuccessful, because the group does not exist.
v7:
* Shorten memory_size to size. The prefix makes the name only longer but does not clarify things more than "size" would.
* Use device_attribute instead of kobj_attribute.
* Use named attribute group instead of creating raw kobject just for the "sgx" subdirectory.
v6:
* Initialize node->size to zero in sgx_setup_epc_section(), when the node is first accessed.
v5
* A new patch based on the discussion on
https://lore.kernel.org/linux-sgx/[email protected]/T/#t
---
Documentation/ABI/stable/sysfs-devices-node | 6 ++++ arch/Kconfig | 4 +++ arch/x86/Kconfig | 1 + arch/x86/kernel/cpu/sgx/main.c | 31 +++++++++++++++++++++ arch/x86/kernel/cpu/sgx/sgx.h | 2 ++ drivers/base/node.c | 13 ++++++++- include/linux/numa.h | 4 +++
 7 files changed, 60 insertions(+), 1 deletion(-)
Documentation/ABI/stable/sysfs-devices-node +0 -0
index 484fc04bcc25..8db67aa472f1 100644
@@ -176,3 +176,9 @@ Contact:    Keith Busch <[email protected]>
Description:
The cache write policy: 0 for write-back, 1 for write-through, other or unknown.
Documentation/ABI/stable/sysfs-devices-node +6 -0
+
+What:          /sys/devices/system/node/nodeX/x86/sgx_total_bytes
+Date:          November 2021
+Contact:       Jarkko Sakkinen <[email protected]>
+Description:
+               The total amount of SGX physical memory in bytes.
arch/Kconfig +0 -0
index 98db63496bab..ca5d75b5a427 100644
@@ -1285,6 +1285,10 @@ config ARCH_HAS_ELFCORE_COMPAT
config ARCH_HAS_PARANOID_L1D_FLUSH
bool
arch/Kconfig +4 -0
+# Select, if arch has a named attribute group bound to NUMA device nodes.
+config HAVE_ARCH_NODE_DEV_GROUP
+       bool
+
source "kernel/gcov/Kconfig"

source "scripts/gcc-plugins/Kconfig"
arch/x86/Kconfig +0 -0
index 421fa9e38c60..8503c3bdf63f 100644
@@ -266,6 +266,7 @@ config X86
select HAVE_ARCH_KCSAN if X86_64 select X86_FEATURE_NAMES if PROC_FS select PROC_PID_ARCH_STATUS if PROC_FS + select HAVE_ARCH_NODE_DEV_GROUP if X86_SGX imply IMA_SECURE_AND_OR_TRUSTED_BOOT if EFI
config INSTRUCTION_DECODER
arch/x86/kernel/cpu/sgx/main.c +0 -0
index a6e313f1a82d..02ebb233c511 100644
@@ -714,9 +714,12 @@ static bool __init sgx_page_cache_init(void)
                         spin_lock_init(&sgx_numa_nodes[nid].lock);
                         INIT_LIST_HEAD(&sgx_numa_nodes[nid].free_page_list);
                         node_set(nid, sgx_numa_mask);

Quoted 8 lines

Quoted 5 lines
arch/x86/kernel/cpu/sgx/main.c +1 -0
+                       sgx_numa_nodes[nid].size = 0;
                 }
sgx_epc_sections[i].node = &sgx_numa_nodes[nid]; + sgx_numa_nodes[nid].dev = &node_devices[nid]->dev;
You are saving off a pointer to an object without incrementing the reference count of it? What prevents it from going away in the future?

Since the arch specific attribute group is part of the "groups" of that device, I'd presume that when sgx_total_bytes_show() is called the device has not gone away:

static const struct attribute_group *node_dev_groups[] = {
&node_dev_group,
#ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
&arch_node_dev_group,
#endif
NULL,
};

I mean the "dev" parameter in sgx_total_bytes_show() is probably valid, when sysfs framework calls it, right?

Quoted 13 lines

Quoted 13 lines
+               sgx_numa_nodes[nid].size += size;

                sgx_nr_epc_sections++;
        }
arch/x86/kernel/cpu/sgx/main.c +0 -0
@@ -790,6 +793,34 @@ int sgx_set_attribute(unsigned long *allowed_attributes,
  }
  EXPORT_SYMBOL_GPL(sgx_set_attribute);
arch/x86/kernel/cpu/sgx/main.c +5 -0
+#ifdef CONFIG_NUMA
+static ssize_t sgx_total_bytes_show(struct device *dev, struct device_attribute *attr, char *buf)
+{
+       unsigned long size = 0;
+       int nid;

Quoted 8 lines

Quoted 3 lines
+
+ for (nid = 0; nid < num_possible_nodes(); nid++) { + if (dev == sgx_numa_nodes[nid].dev) {
Why aren't these values all just part of the device that is being used here? You are walking some odd array, with no locking, and no reference counting on the objects you are looking at, just to find the same object that you started out with???

This code looks up the struct sgx_numa_node instance, and the array is indexed with NUMA node ID. "dev" is only used as lookup key.

Quoted 4 lines

That's not the proper thing to do here at all, these values should be part of the device structure that this attribute lives on, in order to properly handle all of these reference counting and locking issues automatically.

The attribute group is part of the device structure, so in that sense it is already like this. And I think that all NUMA node specific data (most of which is unrelated to device sysfs) is best kept in the private struct.

Quoted 2 lines

Please fix the design of this code first, _before_ adding new attributes.

I sincerely think it is correct now.

Quoted 4 lines

thanks,
greg k-h

/Jarkko

Jarkko Sakkinen

author maintainer

2021-11-03 21:54 UTC #4 lore raw

Re: [PATCH v11 2/2] x86/sgx: Add an attribute for the amount of SGX memory in a NUMA node

Quoted 10 lines · Jarkko Sakkinen

Quoted 4 lines · Greg Kroah-Hartman
That's not the proper thing to do here at all, these values should be part of the device structure that this attribute lives on, in order to properly handle all of these reference counting and locking issues automatically.
The attribute group is part of the device structure, so in that sense it is already like this. And I think that all NUMA node specific data (most of which is unrelated to device sysfs) is best kept in the private struct.

If you are concerned that the SGX driver might go away when NUMA code calls it, I can address that concern too.

The memory management code if arch/x86/kernel/cpu/sgx is not associated with the user space facing driver. E.g. you can use memory manager with KVM without having the driver even enabled in the kernel.

This include the numa code in the SGX subsystem. It's life-cycle is the same as the power-cycle. It is guaranteed that any possible time during the power cycle, when that sysfs callback is called, all the data is valid. That's why it is safe to be part of the attribute groups of the NUMA device.

/Jarkko

Greg Kroah-Hartman

maintainer

2021-11-04 07:25 UTC #5 lore raw

Re: [PATCH v11 2/2] x86/sgx: Add an attribute for the amount of SGX memory in a NUMA node

In reply to Jarkko Sakkinen

Quoted 122 lines · Jarkko Sakkinen

Quoted 108 lines · Greg Kroah-Hartman
Quoted 107 lines · Jarkko Sakkinen
The amount of SGX memory on the system is determined by the BIOS and it varies wildly between systems. It can be from dozens of MB's on desktops or VM's, up to many GB's on servers. Just like for regular memory, it is sometimes useful to know the amount of usable SGX memory in the system.
Introduce CONFIG_HAVE_ARCH_NODE_DEV_GROUP opt-in flag to expose an arch specific attribute group, and add an attribute for the amount of SGX memory in bytes to each NUMA node:
/sys/devices/system/node/nodeX/x86/sgx_total_bytes
Cc Greg Kroah-Hartman <[email protected]>
Signed-off-by Jarkko Sakkinen <[email protected]>
---
v11:
* Fix documentation and the commit message.
v10:
* Change DEVICE_ATTR_RO() to static (Greg K-H)
* Change the attribute name as sgx_total_bytes, and attribute group name as "x86" (Dave).
* Add a new config flag HAVE_ARCH_NODE_DEV_GROUP to identify, whether an arch exports arch specific attribute group for NUMA nodes.
v9:
* Fix racy initialization of sysfs attributes:
https://lore.kernel.org/linux-sgx/[email protected]/
v8:
* Fix a bug in sgx_numa_init(): node->dev should be only set after sysfe_create_group(). Otherwise, sysfs_remove_group() will issue a warning in sgx_numa_exit(), when sgx_create_group() is unsuccessful, because the group does not exist.
v7:
* Shorten memory_size to size. The prefix makes the name only longer but does not clarify things more than "size" would.
* Use device_attribute instead of kobj_attribute.
* Use named attribute group instead of creating raw kobject just for the "sgx" subdirectory.
v6:
* Initialize node->size to zero in sgx_setup_epc_section(), when the node is first accessed.
v5
* A new patch based on the discussion on
https://lore.kernel.org/linux-sgx/[email protected]/T/#t
---
Documentation/ABI/stable/sysfs-devices-node | 6 ++++ arch/Kconfig | 4 +++ arch/x86/Kconfig | 1 + arch/x86/kernel/cpu/sgx/main.c | 31 +++++++++++++++++++++ arch/x86/kernel/cpu/sgx/sgx.h | 2 ++ drivers/base/node.c | 13 ++++++++- include/linux/numa.h | 4 +++
 7 files changed, 60 insertions(+), 1 deletion(-)
Documentation/ABI/stable/sysfs-devices-node +0 -0
index 484fc04bcc25..8db67aa472f1 100644
@@ -176,3 +176,9 @@ Contact:    Keith Busch <[email protected]>
Description:
The cache write policy: 0 for write-back, 1 for write-through, other or unknown.
Documentation/ABI/stable/sysfs-devices-node +6 -0
+
+What:          /sys/devices/system/node/nodeX/x86/sgx_total_bytes
+Date:          November 2021
+Contact:       Jarkko Sakkinen <[email protected]>
+Description:
+               The total amount of SGX physical memory in bytes.
arch/Kconfig +0 -0
index 98db63496bab..ca5d75b5a427 100644
@@ -1285,6 +1285,10 @@ config ARCH_HAS_ELFCORE_COMPAT
config ARCH_HAS_PARANOID_L1D_FLUSH
bool
arch/Kconfig +4 -0
+# Select, if arch has a named attribute group bound to NUMA device nodes.
+config HAVE_ARCH_NODE_DEV_GROUP
+       bool
+
source "kernel/gcov/Kconfig"

source "scripts/gcc-plugins/Kconfig"
arch/x86/Kconfig +0 -0
index 421fa9e38c60..8503c3bdf63f 100644
@@ -266,6 +266,7 @@ config X86
select HAVE_ARCH_KCSAN if X86_64 select X86_FEATURE_NAMES if PROC_FS select PROC_PID_ARCH_STATUS if PROC_FS + select HAVE_ARCH_NODE_DEV_GROUP if X86_SGX imply IMA_SECURE_AND_OR_TRUSTED_BOOT if EFI
config INSTRUCTION_DECODER
arch/x86/kernel/cpu/sgx/main.c +0 -0
index a6e313f1a82d..02ebb233c511 100644
@@ -714,9 +714,12 @@ static bool __init sgx_page_cache_init(void)
                         spin_lock_init(&sgx_numa_nodes[nid].lock);
                         INIT_LIST_HEAD(&sgx_numa_nodes[nid].free_page_list);
                         node_set(nid, sgx_numa_mask);
Quoted 8 lines
Quoted 5 lines
arch/x86/kernel/cpu/sgx/main.c +1 -0
+                       sgx_numa_nodes[nid].size = 0;
                 }
sgx_epc_sections[i].node = &sgx_numa_nodes[nid]; + sgx_numa_nodes[nid].dev = &node_devices[nid]->dev;
You are saving off a pointer to an object without incrementing the reference count of it? What prevents it from going away in the future?
Since the arch specific attribute group is part of the "groups" of that device, I'd presume that when sgx_total_bytes_show() is called the device has not gone away:

Quoted 7 lines

static const struct attribute_group *node_dev_groups[] = {
&node_dev_group,
#ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP
&arch_node_dev_group,
#endif
NULL,
};

Yes, that is true for the dev pointer passed to your callback, but what about the dev pointers in this random array you are looping over?

Quoted 2 lines

I mean the "dev" parameter in sgx_total_bytes_show() is probably valid, when sysfs framework calls it, right?

Yes.

Quoted 22 lines

Quoted 13 lines
Quoted 13 lines
+               sgx_numa_nodes[nid].size += size;

                sgx_nr_epc_sections++;
        }
arch/x86/kernel/cpu/sgx/main.c +0 -0
@@ -790,6 +793,34 @@ int sgx_set_attribute(unsigned long *allowed_attributes,
  }
  EXPORT_SYMBOL_GPL(sgx_set_attribute);
arch/x86/kernel/cpu/sgx/main.c +5 -0
+#ifdef CONFIG_NUMA
+static ssize_t sgx_total_bytes_show(struct device *dev, struct device_attribute *attr, char *buf)
+{
+       unsigned long size = 0;
+       int nid;
Quoted 8 lines
Quoted 3 lines
+
+ for (nid = 0; nid < num_possible_nodes(); nid++) { + if (dev == sgx_numa_nodes[nid].dev) {
Why aren't these values all just part of the device that is being used here? You are walking some odd array, with no locking, and no reference counting on the objects you are looking at, just to find the same object that you started out with???

Quoted 2 lines

This code looks up the struct sgx_numa_node instance, and the array is indexed with NUMA node ID. "dev" is only used as lookup key.

And why isn't that pointer/key properly reference counted?

What happens when the memory numa node id goes away? Who removes that item from the list?

Quoted 9 lines

Quoted 4 lines
That's not the proper thing to do here at all, these values should be part of the device structure that this attribute lives on, in order to properly handle all of these reference counting and locking issues automatically.
The attribute group is part of the device structure, so in that sense it is already like this. And I think that all NUMA node specific data (most of which is unrelated to device sysfs) is best kept in the private struct.

No, the data associated with a specific numa node needs to be IN the structure that it belongs to, not in a random other data structure that is not locked or reference counted at all. That is not ok.

Quoted 4 lines

Quoted 2 lines
Please fix the design of this code first, _before_ adding new attributes.
I sincerely think it is correct now.

I disagree. Where is the locking and reference count logic that keeps these different structures in sync properly? Why are there different structures at all?

Yes, you are just working with what you have now, but you have exposed that what we have now is incorrect, and trying to use it in the manner you are wanting to here is also incorrect.

thanks,

greg k-h

Jarkko Sakkinen

author maintainer

2021-11-05 22:14 UTC #6 lore raw

Re: [PATCH v11 2/2] x86/sgx: Add an attribute for the amount of SGX memory in a NUMA node

Quoted 10 lines · Greg Kroah-Hartman

Quoted 7 lines
static const struct attribute_group *node_dev_groups[] = { &node_dev_group,
#ifdef CONFIG_HAVE_ARCH_NODE_DEV_GROUP &arch_node_dev_group,
#endif
NULL,
};
Yes, that is true for the dev pointer passed to your callback, but what about the dev pointers in this random array you are looping over?

Right. I got what you are saying.

I think the most legit place to mark an entry in this array would be just *before* device_register() in register_node(). It's different from hugetlb_register_node() because hugetlb code adds its attribute group with sysfs_create_group().

Similarly, the legit place to unmark an entry would be in unregister_node(), right after device_unregister().

After writing this I realized something: the device ID is the same as NUMA node ID. This means that I can rewrite my callback as

static ssize_t sgx_total_bytes_show(struct device *dev, struct device_attribute *attr, char *buf)

{
	unsigned long size = 0;
	int nid = dev->id;

	return sysfs_emit(buf, "%lu\n", sgx_numa_nodes[dev->id].size);
}

I.e no need to maintain a device pointer.

/Jarkko