Skip to contentSkip to conversation

Expose PCIe SSD Status LED Management DSM in sysfs

PATCH In review Stuart Hayes 6 participants 9 messages Nov 10, 2020 – Nov 17, 2020 PCI

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 Expose PCIe SSD Status LED Management DSM in sysfs

This patch will expose the PCIe SSD Status LED Management interface in sysfs for devices that have the relevant _DSM method, per the "_DSM additions for PCIe SSD Status LED Management" ECN to the PCI Firmware Specification revision 3.2.

The interface is exposed in two sysfs files, ssdleds_supported_states (RO) and ssdleds_current_state (RW).

---

v2: made PCI_SSDLEDS dependent on PCI and ACPI
remove unused variable
warn if call to sysfs_create_group fails
include header file with function prototypes
change logical OR to bitwise

---
 Documentation/ABI/testing/sysfs-bus-pci |  14 ++
 drivers/pci/Kconfig                     |   7 +
 drivers/pci/Makefile                    |   1 +
 drivers/pci/pci-ssdleds.c               | 194 ++++++++++++++++++++++++
 drivers/pci/pci-sysfs.c                 |   1 +
 drivers/pci/pci.h                       |  11 ++
 6 files changed, 228 insertions(+)
 create mode 100644 drivers/pci/pci-ssdleds.c
Documentation/ABI/testing/sysfs-bus-pci +14 -0
index 77ad9ec3c801..18ca73b764ac 100644
@@ -366,3 +366,17 @@ Contact:	Heiner Kallweit <[email protected]>
 Description:	If ASPM is supported for an endpoint, these files can be
 		used to disable or enable the individual power management
 		states. Write y/1/on to enable, n/0/off to disable.
+
+What:		/sys/bus/pci/devices/.../ssdleds_supported_states
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_supported_states (read only)
+		will show the LED states that are supported by the _DSM.
+
+What:		/sys/bus/pci/devices/.../ssdleds_current_state
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_current_state will show or set
+		the current LED states that are active.
drivers/pci/Kconfig +7 -0
index 0c473d75e625..48049866145f 100644
@@ -182,6 +182,13 @@ config PCI_LABEL
 	def_bool y if (DMI || ACPI)
 	select NLS
 
+config PCI_SSDLEDS
+	def_bool y if (ACPI && PCI)
+	depends on PCI && ACPI
+	help
+	  Enables userspace access to PCIe SSD LED management interface via
+	  sysfs.
+
 config PCI_HYPERV
 	tristate "Hyper-V PCI Frontend"
 	depends on X86_64 && HYPERV && PCI_MSI && PCI_MSI_IRQ_DOMAIN && SYSFS
drivers/pci/Makefile +1 -0
index 522d2b974e91..75d3d5a3b1ed 100644
@@ -23,6 +23,7 @@ obj-$(CONFIG_PCI_ATS)		+= ats.o
 obj-$(CONFIG_PCI_IOV)		+= iov.o
 obj-$(CONFIG_PCI_BRIDGE_EMUL)	+= pci-bridge-emul.o
 obj-$(CONFIG_PCI_LABEL)		+= pci-label.o
+obj-$(CONFIG_PCI_SSDLEDS)	+= pci-ssdleds.o
 obj-$(CONFIG_X86_INTEL_MID)	+= pci-mid.o
 obj-$(CONFIG_PCI_SYSCALL)	+= syscall.o
 obj-$(CONFIG_PCI_STUB)		+= pci-stub.o
drivers/pci/pci-ssdleds.c new file +194 -0

194 lines not on this page · raw message

drivers/pci/pci-sysfs.c +1 -0
index d15c881e2e7e..820f32956971 100644
@@ -1377,6 +1377,7 @@ int __must_check pci_create_sysfs_dev_files(struct pci_dev *pdev)
 		goto err_rom_file;
 
 	pci_create_firmware_label_files(pdev);
+	pci_create_ssdleds_files(pdev);
 
 	return 0;
 
drivers/pci/pci.h +11 -0
index f86cae9aa1f4..8e6883a1b701 100644
@@ -30,6 +30,17 @@ static inline void pci_remove_firmware_label_files(struct pci_dev *pdev)
 void pci_create_firmware_label_files(struct pci_dev *pdev);
 void pci_remove_firmware_label_files(struct pci_dev *pdev);
 #endif
+
+#if !defined(CONFIG_PCI_SSDLEDS)
+static inline void pci_create_ssdleds_files(struct pci_dev *pdev)
+{ return; }
+static inline void pci_remove_ssdleds_files(struct pci_dev *pdev)
+{ return; }
+#else
+void pci_create_ssdleds_files(struct pci_dev *pdev);
+void pci_remove_ssdleds_files(struct pci_dev *pdev);
+#endif
+
 void pci_cleanup_rom(struct pci_dev *dev);
 
 enum pci_mmap_api {
Signature · 1 line
2.18.1
#1 lore raw

Documentation/ABI/testing/sysfs-bus-pci Contact: Heiner Kallweit <[email protected]> · line 382

Quoted 12 lines · Stuart Hayes
Documentation/ABI/testing/sysfs-bus-pci +12 -0
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_supported_states (read only)
+		will show the LED states that are supported by the _DSM.
+
+What:		/sys/bus/pci/devices/.../ssdleds_current_state
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_current_state will show or set
+		the current LED states that are active.

Is the supported file really required? Doesn't the current_state one also show which LEDs exist?

drivers/pci/Kconfig config PCI_LABEL · line 187

Quoted 3 lines
drivers/pci/Kconfig +3 -0
+config PCI_SSDLEDS
+	def_bool y if (ACPI && PCI)
+	depends on PCI && ACPI

We really should not default new code to y.

drivers/pci/pci-ssdleds.c line 75

Quoted 2 lines
drivers/pci/pci-ssdleds.c new file +2 -0
+	if (dsm_output->status != 0 &&
+	    !(dsm_output->status == 4 && dsm_output->function_specific_err == 1)) {

overly longline. But to make this a little more obvious, maybe you want to

a) switch on dsm_output->status
b) add symbolic names for the magic numbers

#2 lore raw
Quoted 4 lines · Christoph Hellwig

Documentation/ABI/testing/sysfs-bus-pci Contact: Heiner Kallweit <[email protected]>

Quoted 3 lines · Stuart Hayes
Documentation/ABI/testing/sysfs-bus-pci +3 -0
+Date:              October 2020
+Contact:   Stuart Hayes <[email protected]>
+Description:       If the device supports the ACPI _DSM method to

control the

Quoted 7 lines

Documentation/ABI/testing/sysfs-bus-pci Contact: Heiner Kallweit <[email protected]>

Quoted 7 lines
Documentation/ABI/testing/sysfs-bus-pci +7 -0
+           PCIe SSD LED states, ssdleds_supported_states (read only)
+           will show the LED states that are supported by the _DSM.
+
+What:              /sys/bus/pci/devices/.../ssdleds_current_state
+Date:              October 2020
+Contact:   Stuart Hayes <[email protected]>
+Description:       If the device supports the ACPI _DSM method to

control the

Quoted 6 lines

Documentation/ABI/testing/sysfs-bus-pci Contact: Heiner Kallweit <[email protected]> · line 382

Quoted 2 lines
Documentation/ABI/testing/sysfs-bus-pci +2 -0
+           PCIe SSD LED states, ssdleds_current_state will show or set
+           the current LED states that are active.

Is the supported file really required? Doesn't the current_state one also show which LEDs exist?

The current_state just shows which LED states are currently active, not which
are supported by the system. I guess you could try to set all the states in current_state and read it back to see which ones went on to see which states are supported, but I'm not 100% if that would work, plus it might result in the drive LEDs flashing in a strange way briefly.

Quoted 6 lines

drivers/pci/Kconfig config PCI_LABEL · line 187

Quoted 3 lines
drivers/pci/Kconfig +3 -0
+config PCI_SSDLEDS
+   def_bool y if (ACPI && PCI)
+   depends on PCI && ACPI

We really should not default new code to y.

Good point, I agree.

Quoted 2 lines

Review

Quoted 2 lines
(quoted diff) +2 -0
+   if (dsm_output->status != 0 &&
+       !(dsm_output->status == 4 && dsm_output->function_specific_err

== 1)) {

Quoted 7 lines

overly longline. But to make this a little more obvious, maybe you want to

a) switch on dsm_output->status
b) add symbolic names for the magic numbers

Good feedback, thanks.

#3 lore raw
In reply to Stuart Hayes

Hi Stuart,

url:    https://github.com/0day-ci/linux/commits/Stuart-Hayes/Expose-PCIe-SSD-Status-LED-Management-DSM-in-sysfs/20201111-004557
base:   https://git.kernel.org/pub/scm/linux/kernel/git/helgaas/pci.git next
config: i386-randconfig-m021-20201111 (attached as .config)
compiler: gcc-9 (Debian 9.3.0-15) 9.3.0

If you fix the issue, kindly add following tag as appropriate

smatch warnings:
drivers/pci/pci-ssdleds.c:47 ssdleds_dsm_set() warn: impossible condition '(val > (~0)) => (0-u32max > u32max)'

vim +47 drivers/pci/pci-ssdleds.c

e6eac5bf04a7aca Stuart Hayes 2020-11-10  34  static int ssdleds_dsm_set(struct device *dev, const char *buf, u64 dsm_func)
e6eac5bf04a7aca Stuart Hayes 2020-11-10  35  {
e6eac5bf04a7aca Stuart Hayes 2020-11-10  36  	acpi_handle handle;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  37  	union acpi_object *out_obj, arg3[2];
e6eac5bf04a7aca Stuart Hayes 2020-11-10  38  	struct pci_ssdleds_dsm_output *dsm_output;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  39  	u32 val;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  40  	int err;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  41  
e6eac5bf04a7aca Stuart Hayes 2020-11-10  42  	handle = ACPI_HANDLE(dev);
e6eac5bf04a7aca Stuart Hayes 2020-11-10  43  	if (!handle)
e6eac5bf04a7aca Stuart Hayes 2020-11-10  44  		return -ENODEV;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  45  
e6eac5bf04a7aca Stuart Hayes 2020-11-10  46  	err = kstrtou32(buf, 0, &val);
e6eac5bf04a7aca Stuart Hayes 2020-11-10 @47  	if (err || val > U32_MAX)
                                                           ^^^^^^^^^^^^^

This is not required. Just "if (err) return err;"'

e6eac5bf04a7aca Stuart Hayes 2020-11-10  48  		return -EINVAL;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  49  
e6eac5bf04a7aca Stuart Hayes 2020-11-10  50  	arg3[0].type = ACPI_TYPE_PACKAGE;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  51  	arg3[0].package.count = 1;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  52  	arg3[0].package.elements = &arg3[1];
e6eac5bf04a7aca Stuart Hayes 2020-11-10  53  
e6eac5bf04a7aca Stuart Hayes 2020-11-10  54  	arg3[1].type = ACPI_TYPE_BUFFER;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  55  	arg3[1].buffer.length = 4;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  56  	arg3[1].buffer.pointer = (u8 *)&val;
e6eac5bf04a7aca Stuart Hayes 2020-11-10  57  
e6eac5bf04a7aca Stuart Hayes 2020-11-10  58  	out_obj = acpi_evaluate_dsm_typed(handle, &pci_ssdleds_dsm_guid, 0x1,

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/kbuild-all(a)lists.01.org

#4 lore raw
In reply to Stuart Hayes

[+cc Christoph, Dan, Greg (for "one value per file" question below)]

Quoted 85 lines · Stuart Hayes

This patch will expose the PCIe SSD Status LED Management interface in sysfs for devices that have the relevant _DSM method, per the "_DSM additions for PCIe SSD Status LED Management" ECN to the PCI Firmware Specification revision 3.2.

The interface is exposed in two sysfs files, ssdleds_supported_states (RO) and ssdleds_current_state (RW).

---

v2: made PCI_SSDLEDS dependent on PCI and ACPI
remove unused variable
warn if call to sysfs_create_group fails
include header file with function prototypes
change logical OR to bitwise

---
 Documentation/ABI/testing/sysfs-bus-pci |  14 ++
 drivers/pci/Kconfig                     |   7 +
 drivers/pci/Makefile                    |   1 +
 drivers/pci/pci-ssdleds.c               | 194 ++++++++++++++++++++++++
 drivers/pci/pci-sysfs.c                 |   1 +
 drivers/pci/pci.h                       |  11 ++
 6 files changed, 228 insertions(+)
 create mode 100644 drivers/pci/pci-ssdleds.c
Documentation/ABI/testing/sysfs-bus-pci +14 -0
index 77ad9ec3c801..18ca73b764ac 100644
@@ -366,3 +366,17 @@ Contact:	Heiner Kallweit <[email protected]>
 Description:	If ASPM is supported for an endpoint, these files can be
 		used to disable or enable the individual power management
 		states. Write y/1/on to enable, n/0/off to disable.
+
+What:		/sys/bus/pci/devices/.../ssdleds_supported_states
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_supported_states (read only)
+		will show the LED states that are supported by the _DSM.
+
+What:		/sys/bus/pci/devices/.../ssdleds_current_state
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_current_state will show or set
+		the current LED states that are active.
drivers/pci/Kconfig +7 -0
index 0c473d75e625..48049866145f 100644
@@ -182,6 +182,13 @@ config PCI_LABEL
 	def_bool y if (DMI || ACPI)
 	select NLS
 
+config PCI_SSDLEDS
+	def_bool y if (ACPI && PCI)
+	depends on PCI && ACPI
+	help
+	  Enables userspace access to PCIe SSD LED management interface via
+	  sysfs.
+
 config PCI_HYPERV
 	tristate "Hyper-V PCI Frontend"
 	depends on X86_64 && HYPERV && PCI_MSI && PCI_MSI_IRQ_DOMAIN && SYSFS
drivers/pci/Makefile +1 -0
index 522d2b974e91..75d3d5a3b1ed 100644
@@ -23,6 +23,7 @@ obj-$(CONFIG_PCI_ATS)		+= ats.o
 obj-$(CONFIG_PCI_IOV)		+= iov.o
 obj-$(CONFIG_PCI_BRIDGE_EMUL)	+= pci-bridge-emul.o
 obj-$(CONFIG_PCI_LABEL)		+= pci-label.o
+obj-$(CONFIG_PCI_SSDLEDS)	+= pci-ssdleds.o
 obj-$(CONFIG_X86_INTEL_MID)	+= pci-mid.o
 obj-$(CONFIG_PCI_SYSCALL)	+= syscall.o
 obj-$(CONFIG_PCI_STUB)		+= pci-stub.o
drivers/pci/pci-ssdleds.c new file +0 -0
new file mode 100644
index 000000000000..d35e4f3caa71

drivers/pci/pci-ssdleds.c line 5

Quoted 5 lines
drivers/pci/pci-ssdleds.c new file +5 -0
@@ -0,0 +1,194 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Provide userspace access to PCIe SSD Status LED Management
+ *
+ * The PCIe spec defines an ACPI _DSM method to allow PCIe SSD status LED

s/PCIe spec/PCI Firmware spec/

drivers/pci/pci-ssdleds.c line 10

Quoted 5 lines
drivers/pci/pci-ssdleds.c new file +5 -0
+ * management (see "_DSM additions for PCIe SSD Status LED Management" ECN,
+ * February 12, 2020). This code provides a sysfs interface to this _DSM.
+ *
+ * Copyright (C) 2020 Dell Inc.
+ *

Unnecessary blank line.

drivers/pci/pci-ssdleds.c line 21

Quoted 11 lines
drivers/pci/pci-ssdleds.c new file +11 -0
+ */
+
+#include <linux/sysfs.h>
+#include <linux/pci.h>
+#include <linux/device.h>
+#include <linux/acpi.h>
+#include "pci.h"
+
+const guid_t pci_ssdleds_dsm_guid =
+	GUID_INIT(0x5d524d9d, 0xfff9, 0x4d4b,
+		  0x8c, 0xb7, 0x74, 0x7e, 0xd5, 0x1e, 0x19, 0x4d);

Static?

Quoted 16 lines
drivers/pci/pci-ssdleds.c new file +16 -0
+#define SSDLEDS_GET_SUPPORTED_STATES_DSM	0x01
+#define SSDLEDS_GET_STATE_DSM			0x02
+#define SSDLEDS_SET_STATE_DSM			0x03
+
+struct pci_ssdleds_dsm_output {
+	u16 status;
+	u8 function_specific_err;
+	u8 vendor_specific_err;
+	u32 state;
+};
+
+static int ssdleds_dsm_set(struct device *dev, const char *buf, u64 dsm_func)
+{
+	acpi_handle handle;
+	union acpi_object *out_obj, arg3[2];
+	struct pci_ssdleds_dsm_output *dsm_output;

drivers/pci/pci-ssdleds.c line 48

Quoted 10 lines
drivers/pci/pci-ssdleds.c new file +10 -0
+	u32 val;
+	int err;
+
+	handle = ACPI_HANDLE(dev);
+	if (!handle)
+		return -ENODEV;
+
+	err = kstrtou32(buf, 0, &val);
+	if (err || val > U32_MAX)
+		return -EINVAL;

"buf" is really a sysfs concept, not a _DSM concept, so I would consider moving the kstrtou32() to the caller.

Quoted 16 lines
drivers/pci/pci-ssdleds.c new file +16 -0
+	arg3[0].type = ACPI_TYPE_PACKAGE;
+	arg3[0].package.count = 1;
+	arg3[0].package.elements = &arg3[1];
+
+	arg3[1].type = ACPI_TYPE_BUFFER;
+	arg3[1].buffer.length = 4;
+	arg3[1].buffer.pointer = (u8 *)&val;
+
+	out_obj = acpi_evaluate_dsm_typed(handle, &pci_ssdleds_dsm_guid, 0x1,
+				      dsm_func, &arg3[0], ACPI_TYPE_BUFFER);
+	if (!out_obj)
+		return -EIO;
+
+	if (out_obj->buffer.length < 8) {
+		ACPI_FREE(out_obj);
+		return -EIO;

drivers/pci/pci-ssdleds.c line 75

Quoted 10 lines
drivers/pci/pci-ssdleds.c new file +10 -0
+	}
+
+	dsm_output = (struct pci_ssdleds_dsm_output *)out_obj->buffer.pointer;
+	/*
+	 * Ignore function specific error bit 1 (some LED state bits weren't
+	 * set), since write was done. User can read current state to see which
+	 * bits were set.
+	 */
+	if (dsm_output->status != 0 &&
+	    !(dsm_output->status == 4 && dsm_output->function_specific_err == 1)) {

It looks like this error check is specific to SSDLEDS_SET_STATE_DSM.

I know this function is currently only called for SSDLEDS_SET_STATE_DSM, but I think you should either get rid of the dsm_func argument and specify SSDLEDS_SET_STATE_DSM explicitly in *this* function, or do this error check in the caller where you *know* you're passing SSDLEDS_SET_STATE_DSM.

I guess the dsm_output you're checking only exists in this function, so maybe the former?

Quoted 26 lines
drivers/pci/pci-ssdleds.c new file +26 -0
+		ACPI_FREE(out_obj);
+		return -EIO;
+	}
+
+	ACPI_FREE(out_obj);
+
+	return 0;
+}
+
+static int ssdleds_dsm_get(struct device *dev, char *buf, u64 dsm_func)
+{
+	acpi_handle handle;
+	union acpi_object *out_obj;
+	struct pci_ssdleds_dsm_output *dsm_output;
+
+	handle = ACPI_HANDLE(dev);
+	if (!handle)
+		return -ENODEV;
+
+	out_obj = acpi_evaluate_dsm_typed(handle, &pci_ssdleds_dsm_guid, 0x1,
+				      dsm_func, NULL, ACPI_TYPE_BUFFER);
+	if (!out_obj)
+		return -EIO;
+
+	if (out_obj->buffer.length < 8) {
+		ACPI_FREE(out_obj);

drivers/pci/pci-ssdleds.c line 111

Quoted 10 lines
drivers/pci/pci-ssdleds.c new file +10 -0
+		return -EIO;
+	}
+
+	dsm_output = (struct pci_ssdleds_dsm_output *)out_obj->buffer.pointer;
+	if (dsm_output->status != 0) {
+		ACPI_FREE(out_obj);
+		return -EIO;
+	}
+
+	scnprintf(buf, PAGE_SIZE, "%#x\n", dsm_output->state);

Similarly, I would consider returning dsm_output->state and moving the scnprintf() so the sysfs-specific stuff is in the caller.

It looks like you just expose the hex value that encodes things like:

OK
Locate
Fail
Rebuild
...

The hex value is sort of a pain to interpret, especially since the PCI specs are not public. Maybe consider decoding it? If you did decode it, that might be a way to reduce this to a single file instead of having both "supported_states" and "current_state" -- the single file could list all the supported states, with the current states highlighted somehow.

Not sure if that would run afoul of the sysfs "one value per file" rule. CC'd Greg in case he wants to chime in.

Quoted 45 lines
drivers/pci/pci-ssdleds.c new file +45 -0
+	ACPI_FREE(out_obj);
+
+	return strlen(buf) > 0 ? strlen(buf) : -EIO;
+}
+
+static bool device_has_dsm(struct device *dev)
+{
+	acpi_handle handle;
+
+	handle = ACPI_HANDLE(dev);
+	if (!handle)
+		return false;
+
+	return !!acpi_check_dsm(handle, &pci_ssdleds_dsm_guid, 0x1,
+		       1 << SSDLEDS_GET_SUPPORTED_STATES_DSM |
+		       1 << SSDLEDS_GET_STATE_DSM |
+		       1 << SSDLEDS_SET_STATE_DSM);
+}
+
+static ssize_t ssdleds_current_store(struct device *dev,
+				struct device_attribute *attr,
+				const char *buf, size_t count)
+{
+	int ret;
+
+	ret = ssdleds_dsm_set(dev, buf, SSDLEDS_SET_STATE_DSM);
+	return (ret < 0) ? ret : count;
+}
+
+static ssize_t ssdleds_current_show(struct device *dev,
+			      struct device_attribute *attr, char *buf)
+{
+	return ssdleds_dsm_get(dev, buf, SSDLEDS_GET_STATE_DSM);
+}
+
+static ssize_t ssdleds_supported_show(struct device *dev,
+			      struct device_attribute *attr, char *buf)
+{
+	return ssdleds_dsm_get(dev, buf, SSDLEDS_GET_SUPPORTED_STATES_DSM);
+}
+
+static umode_t ssdleds_attr_visible(struct kobject *kobj,
+				    struct attribute *attr, int n)
+{
+	struct device *dev = kobj_to_dev(kobj);

drivers/pci/pci-ssdleds.c line 167

Quoted 10 lines
drivers/pci/pci-ssdleds.c new file +10 -0
+	umode_t mode = attr->mode;
+
+	return device_has_dsm(dev) ? mode : 0;
+}
+
+static struct device_attribute ssdleds_attr_current = {
+	.attr = {.name = "ssdleds_current_state", .mode = 0644},
+	.show = ssdleds_current_show,
+	.store = ssdleds_current_store,
+};

Can these use DEVICE_ATTR_RW() and friends?

Quoted 16 lines
drivers/pci/pci-ssdleds.c new file +16 -0
+static struct device_attribute ssdleds_attr_supported = {
+	.attr = {.name = "ssdleds_supported_states", .mode = 0444},
+	.show = ssdleds_supported_show,
+};
+
+static struct attribute *ssdleds_attributes[] = {
+	&ssdleds_attr_current.attr,
+	&ssdleds_attr_supported.attr,
+	NULL,
+};
+
+static const struct attribute_group ssdleds_attr_group = {
+	.attrs = ssdleds_attributes,
+	.is_visible = ssdleds_attr_visible,
+};
+

drivers/pci/pci-ssdleds.c line 194

Quoted 10 lines
drivers/pci/pci-ssdleds.c new file +10 -0
+void pci_create_ssdleds_files(struct pci_dev *pdev)
+{
+	if (sysfs_create_group(&pdev->dev.kobj, &ssdleds_attr_group))
+		pci_warn(pdev, "Unable to create ssdleds attributes\n");
+}
+
+void pci_remove_ssdleds_files(struct pci_dev *pdev)
+{
+	sysfs_remove_group(&pdev->dev.kobj, &ssdleds_attr_group);
+}

I'm not a real sysfs expert, but I *think* that if you follow the pattern in pci_dev_attr_groups[] in pci-sysfs.c, you won't need these create/remove functions.

Quoted 37 lines
drivers/pci/pci-sysfs.c +1 -0
index d15c881e2e7e..820f32956971 100644
@@ -1377,6 +1377,7 @@ int __must_check pci_create_sysfs_dev_files(struct pci_dev *pdev)
 		goto err_rom_file;
 
 	pci_create_firmware_label_files(pdev);
+	pci_create_ssdleds_files(pdev);
 
 	return 0;
 
drivers/pci/pci.h +11 -0
index f86cae9aa1f4..8e6883a1b701 100644
@@ -30,6 +30,17 @@ static inline void pci_remove_firmware_label_files(struct pci_dev *pdev)
 void pci_create_firmware_label_files(struct pci_dev *pdev);
 void pci_remove_firmware_label_files(struct pci_dev *pdev);
 #endif
+
+#if !defined(CONFIG_PCI_SSDLEDS)
+static inline void pci_create_ssdleds_files(struct pci_dev *pdev)
+{ return; }
+static inline void pci_remove_ssdleds_files(struct pci_dev *pdev)
+{ return; }
+#else
+void pci_create_ssdleds_files(struct pci_dev *pdev);
+void pci_remove_ssdleds_files(struct pci_dev *pdev);
+#endif
+
 void pci_cleanup_rom(struct pci_dev *dev);
 
 enum pci_mmap_api {
Signature · 1 line
2.18.1
#5 lore raw
Quoted 60 lines · Bjorn Helgaas

[+cc Christoph, Dan, Greg (for "one value per file" question below)]

Quoted 57 lines · Stuart Hayes

This patch will expose the PCIe SSD Status LED Management interface in sysfs for devices that have the relevant _DSM method, per the "_DSM additions for PCIe SSD Status LED Management" ECN to the PCI Firmware Specification revision 3.2.

The interface is exposed in two sysfs files, ssdleds_supported_states (RO) and ssdleds_current_state (RW).

---

v2: made PCI_SSDLEDS dependent on PCI and ACPI
remove unused variable
warn if call to sysfs_create_group fails
include header file with function prototypes
change logical OR to bitwise

---
 Documentation/ABI/testing/sysfs-bus-pci |  14 ++
 drivers/pci/Kconfig                     |   7 +
 drivers/pci/Makefile                    |   1 +
 drivers/pci/pci-ssdleds.c               | 194 ++++++++++++++++++++++++
 drivers/pci/pci-sysfs.c                 |   1 +
 drivers/pci/pci.h                       |  11 ++
 6 files changed, 228 insertions(+)
 create mode 100644 drivers/pci/pci-ssdleds.c
Documentation/ABI/testing/sysfs-bus-pci +14 -0
index 77ad9ec3c801..18ca73b764ac 100644
@@ -366,3 +366,17 @@ Contact:	Heiner Kallweit <[email protected]>
 Description:	If ASPM is supported for an endpoint, these files can be
 		used to disable or enable the individual power management
 		states. Write y/1/on to enable, n/0/off to disable.
+
+What:		/sys/bus/pci/devices/.../ssdleds_supported_states
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_supported_states (read only)
+		will show the LED states that are supported by the _DSM.
+
+What:		/sys/bus/pci/devices/.../ssdleds_current_state
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_current_state will show or set
+		the current LED states that are active.
drivers/pci/Kconfig +0 -0
index 0c473d75e625..48049866145f 100644
@@ -182,6 +182,13 @@ config PCI_LABEL
 	def_bool y if (DMI || ACPI)
 	select NLS
 
Quoted 2 lines

drivers/pci/Kconfig config PCI_LABEL · line 186

Quoted 2 lines
drivers/pci/Kconfig +2 -0
+config PCI_SSDLEDS
+	def_bool y if (ACPI && PCI)

That only should be set if the machine can not boot without it.

For a blinky light, that's not the case.

And why isn't this code using the existing LED subsystem? Don't create random new driver-specific sysfs files that do things the existing class drivers do please.

Quoted 46 lines
Quoted 17 lines
drivers/pci/Kconfig +17 -0
+static int ssdleds_dsm_get(struct device *dev, char *buf, u64 dsm_func)
+{
+	acpi_handle handle;
+	union acpi_object *out_obj;
+	struct pci_ssdleds_dsm_output *dsm_output;
+
+	handle = ACPI_HANDLE(dev);
+	if (!handle)
+		return -ENODEV;
+
+	out_obj = acpi_evaluate_dsm_typed(handle, &pci_ssdleds_dsm_guid, 0x1,
+				      dsm_func, NULL, ACPI_TYPE_BUFFER);
+	if (!out_obj)
+		return -EIO;
+
+	if (out_obj->buffer.length < 8) {
+		ACPI_FREE(out_obj);

drivers/pci/Kconfig config PCI_LABEL

Quoted 10 lines
drivers/pci/Kconfig +10 -0
+		return -EIO;
+	}
+
+	dsm_output = (struct pci_ssdleds_dsm_output *)out_obj->buffer.pointer;
+	if (dsm_output->status != 0) {
+		ACPI_FREE(out_obj);
+		return -EIO;
+	}
+
+	scnprintf(buf, PAGE_SIZE, "%#x\n", dsm_output->state);

Similarly, I would consider returning dsm_output->state and moving the scnprintf() so the sysfs-specific stuff is in the caller.

It looks like you just expose the hex value that encodes things like:

OK
Locate
Fail
Rebuild
...

The hex value is sort of a pain to interpret, especially since the PCI specs are not public. Maybe consider decoding it? If you did decode it, that might be a way to reduce this to a single file instead of having both "supported_states" and "current_state" -- the single file could list all the supported states, with the current states highlighted somehow.

Quoted 2 lines

Not sure if that would run afoul of the sysfs "one value per file" rule. CC'd Greg in case he wants to chime in.

That's a valid way of displaying options for a sysfs file that can be specific unique values.

Quoted 7 lines

drivers/pci/Kconfig

Quoted 5 lines
drivers/pci/Kconfig +5 -0
+static struct device_attribute ssdleds_attr_current = {
+	.attr = {.name = "ssdleds_current_state", .mode = 0644},
+	.show = ssdleds_current_show,
+	.store = ssdleds_current_store,
+};

Can these use DEVICE_ATTR_RW() and friends?

They should, never open-code something like this.

Quoted 5 lines

drivers/pci/Kconfig

Quoted 5 lines
drivers/pci/Kconfig +4 -0
+static struct device_attribute ssdleds_attr_supported = {
+	.attr = {.name = "ssdleds_supported_states", .mode = 0444},
+	.show = ssdleds_supported_show,
+};

DEVICE_ATTR_RO();

Quoted 11 lines

drivers/pci/Kconfig

Quoted 11 lines
drivers/pci/Kconfig +11 -0
+
+static struct attribute *ssdleds_attributes[] = {
+	&ssdleds_attr_current.attr,
+	&ssdleds_attr_supported.attr,
+	NULL,
+};
+
+static const struct attribute_group ssdleds_attr_group = {
+	.attrs = ssdleds_attributes,
+	.is_visible = ssdleds_attr_visible,
+};

ATTRIBUTE_GROUPS()?

Quoted 4 lines

drivers/pci/Kconfig

Quoted 4 lines
drivers/pci/Kconfig +4 -0
+void pci_create_ssdleds_files(struct pci_dev *pdev)
+{
+	if (sysfs_create_group(&pdev->dev.kobj, &ssdleds_attr_group))
+		pci_warn(pdev, "Unable to create ssdleds attributes\n");

You just raced userspace and lost. Just add this to the default groups for the device and the core will handle it all for you automatically.

Hint, when you find yourself calling a sysfs_*() call from driver code, that's a huge flag that you are doing something wrong.

Quoted 8 lines

drivers/pci/Kconfig

Quoted 4 lines
drivers/pci/Kconfig +4 -0
+void pci_remove_ssdleds_files(struct pci_dev *pdev)
+{
+	sysfs_remove_group(&pdev->dev.kobj, &ssdleds_attr_group);
+}

I'm not a real sysfs expert, but I *think* that if you follow the pattern in pci_dev_attr_groups[] in pci-sysfs.c, you won't need these create/remove functions.

Yes, that is true.

thanks,

greg k-h

#6 lore raw
Quoted 66 lines · Greg Kroah-Hartman
Quoted 60 lines · Bjorn Helgaas

[+cc Christoph, Dan, Greg (for "one value per file" question below)]

Quoted 57 lines · Stuart Hayes

This patch will expose the PCIe SSD Status LED Management interface in sysfs for devices that have the relevant _DSM method, per the "_DSM additions for PCIe SSD Status LED Management" ECN to the PCI Firmware Specification revision 3.2.

The interface is exposed in two sysfs files, ssdleds_supported_states (RO) and ssdleds_current_state (RW).

---

v2: made PCI_SSDLEDS dependent on PCI and ACPI
remove unused variable
warn if call to sysfs_create_group fails
include header file with function prototypes
change logical OR to bitwise

---
 Documentation/ABI/testing/sysfs-bus-pci |  14 ++
 drivers/pci/Kconfig                     |   7 +
 drivers/pci/Makefile                    |   1 +
 drivers/pci/pci-ssdleds.c               | 194 ++++++++++++++++++++++++
 drivers/pci/pci-sysfs.c                 |   1 +
 drivers/pci/pci.h                       |  11 ++
 6 files changed, 228 insertions(+)
 create mode 100644 drivers/pci/pci-ssdleds.c
Documentation/ABI/testing/sysfs-bus-pci +14 -0
index 77ad9ec3c801..18ca73b764ac 100644
@@ -366,3 +366,17 @@ Contact:	Heiner Kallweit <[email protected]>
 Description:	If ASPM is supported for an endpoint, these files can be
 		used to disable or enable the individual power management
 		states. Write y/1/on to enable, n/0/off to disable.
+
+What:		/sys/bus/pci/devices/.../ssdleds_supported_states
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_supported_states (read only)
+		will show the LED states that are supported by the _DSM.
+
+What:		/sys/bus/pci/devices/.../ssdleds_current_state
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_current_state will show or set
+		the current LED states that are active.
drivers/pci/Kconfig +0 -0
index 0c473d75e625..48049866145f 100644
@@ -182,6 +182,13 @@ config PCI_LABEL
 	def_bool y if (DMI || ACPI)
 	select NLS
 
Quoted 2 lines

drivers/pci/Kconfig config PCI_LABEL · line 186

Quoted 2 lines
drivers/pci/Kconfig +2 -0
+config PCI_SSDLEDS
+	def_bool y if (ACPI && PCI)

That only should be set if the machine can not boot without it.

Quoted 6 lines

For a blinky light, that's not the case.

And why isn't this code using the existing LED subsystem? Don't create random new driver-specific sysfs files that do things the existing class drivers do please.

I did consider the LED subsystem, but I don't think it is a great match.

In spite of the name, this _DSM doesn't directly control individual LEDs--it sets the state(s) of the PCI port to which an SSD may be connected, so that it may be conveyed to the user... a processor (or at least some logic) will decide how to show which states are active, probably by setting combinations of LEDs to certain colors or blink patterns, or possibly on some other type of display. On the system I have, changing the active state(s) sends a message via IPMI to an embedded processor, which will decide the colors and/or flashing pattern of the LEDs. So brightness/color/blinking are not controlled, or even known, by the OS.

Also, not all of the states will necessarily always be available. For example, you may only be able to set the "rebuild" state when a drive is actually connected to the pci port that has this _DSM, while the "locate" state might be available all the time. Since there's no notification if/when the supported states change, I believe that would mean, to implement this using the LED subsystem, the driver would have to register an "LED" with the LED subsystem for each possible state, and either make reads or writes to that state fail if it isn't supported.

But I will re-write this using the LED subsystem if you think it's a better fit (though I don't think so).

Quoted 47 lines
Quoted 47 lines
Quoted 18 lines
drivers/pci/Kconfig +17 -0
+static int ssdleds_dsm_get(struct device *dev, char *buf, u64 dsm_func)
+{
+	acpi_handle handle;
+	union acpi_object *out_obj;
+	struct pci_ssdleds_dsm_output *dsm_output;
+
+	handle = ACPI_HANDLE(dev);
+	if (!handle)
+		return -ENODEV;
+
+	out_obj = acpi_evaluate_dsm_typed(handle, &pci_ssdleds_dsm_guid, 0x1,
+				      dsm_func, NULL, ACPI_TYPE_BUFFER);
+	if (!out_obj)
+		return -EIO;
+
+	if (out_obj->buffer.length < 8) {
+		ACPI_FREE(out_obj);

drivers/pci/Kconfig config PCI_LABEL

Quoted 10 lines
drivers/pci/Kconfig +10 -0
+		return -EIO;
+	}
+
+	dsm_output = (struct pci_ssdleds_dsm_output *)out_obj->buffer.pointer;
+	if (dsm_output->status != 0) {
+		ACPI_FREE(out_obj);
+		return -EIO;
+	}
+
+	scnprintf(buf, PAGE_SIZE, "%#x\n", dsm_output->state);

Similarly, I would consider returning dsm_output->state and moving the scnprintf() so the sysfs-specific stuff is in the caller.

It looks like you just expose the hex value that encodes things like:

OK
Locate
Fail
Rebuild
...

The hex value is sort of a pain to interpret, especially since the PCI specs are not public. Maybe consider decoding it? If you did decode it, that might be a way to reduce this to a single file instead of having both "supported_states" and "current_state" -- the single file could list all the supported states, with the current states highlighted somehow.

Quoted 6 lines
Quoted 2 lines

Not sure if that would run afoul of the sysfs "one value per file" rule. CC'd Greg in case he wants to chime in.

That's a valid way of displaying options for a sysfs file that can be specific unique values.

Thanks--I used hex to keep it a single value.

I'll try to find a way to highlight the current states as suggested so I can keep this to a single file.

Quoted 53 lines
Quoted 8 lines

drivers/pci/Kconfig

Quoted 6 lines
drivers/pci/Kconfig +5 -0
+static struct device_attribute ssdleds_attr_current = {
+	.attr = {.name = "ssdleds_current_state", .mode = 0644},
+	.show = ssdleds_current_show,
+	.store = ssdleds_current_store,
+};

Can these use DEVICE_ATTR_RW() and friends?

They should, never open-code something like this.

Quoted 5 lines

drivers/pci/Kconfig

Quoted 5 lines
drivers/pci/Kconfig +4 -0
+static struct device_attribute ssdleds_attr_supported = {
+	.attr = {.name = "ssdleds_supported_states", .mode = 0444},
+	.show = ssdleds_supported_show,
+};

DEVICE_ATTR_RO();

Quoted 11 lines

drivers/pci/Kconfig

Quoted 11 lines
drivers/pci/Kconfig +11 -0
+
+static struct attribute *ssdleds_attributes[] = {
+	&ssdleds_attr_current.attr,
+	&ssdleds_attr_supported.attr,
+	NULL,
+};
+
+static const struct attribute_group ssdleds_attr_group = {
+	.attrs = ssdleds_attributes,
+	.is_visible = ssdleds_attr_visible,
+};

ATTRIBUTE_GROUPS()?

Quoted 4 lines

drivers/pci/Kconfig

Quoted 4 lines
drivers/pci/Kconfig +4 -0
+void pci_create_ssdleds_files(struct pci_dev *pdev)
+{
+	if (sysfs_create_group(&pdev->dev.kobj, &ssdleds_attr_group))
+		pci_warn(pdev, "Unable to create ssdleds attributes\n");

You just raced userspace and lost. Just add this to the default groups for the device and the core will handle it all for you automatically.

Hint, when you find yourself calling a sysfs_*() call from driver code, that's a huge flag that you are doing something wrong.

Quoted 8 lines

drivers/pci/Kconfig

Quoted 4 lines
drivers/pci/Kconfig +4 -0
+void pci_remove_ssdleds_files(struct pci_dev *pdev)
+{
+	sysfs_remove_group(&pdev->dev.kobj, &ssdleds_attr_group);
+}

I'm not a real sysfs expert, but I *think* that if you follow the pattern in pci_dev_attr_groups[] in pci-sysfs.c, you won't need these create/remove functions.

Quoted 6 lines

Yes, that is true.

thanks,

greg k-h

I will try to incorporate the remainder of the feedback, thanks for taking time to look a this.

#7 lore raw
Quoted 78 lines · Stuart Hayes
Quoted 66 lines · Greg Kroah-Hartman
Quoted 60 lines · Bjorn Helgaas

[+cc Christoph, Dan, Greg (for "one value per file" question below)]

Quoted 57 lines · Stuart Hayes

This patch will expose the PCIe SSD Status LED Management interface in sysfs for devices that have the relevant _DSM method, per the "_DSM additions for PCIe SSD Status LED Management" ECN to the PCI Firmware Specification revision 3.2.

The interface is exposed in two sysfs files, ssdleds_supported_states (RO) and ssdleds_current_state (RW).

---

v2: made PCI_SSDLEDS dependent on PCI and ACPI
remove unused variable
warn if call to sysfs_create_group fails
include header file with function prototypes
change logical OR to bitwise

---
 Documentation/ABI/testing/sysfs-bus-pci |  14 ++
 drivers/pci/Kconfig                     |   7 +
 drivers/pci/Makefile                    |   1 +
 drivers/pci/pci-ssdleds.c               | 194 ++++++++++++++++++++++++
 drivers/pci/pci-sysfs.c                 |   1 +
 drivers/pci/pci.h                       |  11 ++
 6 files changed, 228 insertions(+)
 create mode 100644 drivers/pci/pci-ssdleds.c
Documentation/ABI/testing/sysfs-bus-pci +14 -0
index 77ad9ec3c801..18ca73b764ac 100644
@@ -366,3 +366,17 @@ Contact:	Heiner Kallweit <[email protected]>
 Description:	If ASPM is supported for an endpoint, these files can be
 		used to disable or enable the individual power management
 		states. Write y/1/on to enable, n/0/off to disable.
+
+What:		/sys/bus/pci/devices/.../ssdleds_supported_states
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_supported_states (read only)
+		will show the LED states that are supported by the _DSM.
+
+What:		/sys/bus/pci/devices/.../ssdleds_current_state
+Date:		October 2020
+Contact:	Stuart Hayes <[email protected]>
+Description:	If the device supports the ACPI _DSM method to control the
+		PCIe SSD LED states, ssdleds_current_state will show or set
+		the current LED states that are active.
drivers/pci/Kconfig +0 -0
index 0c473d75e625..48049866145f 100644
@@ -182,6 +182,13 @@ config PCI_LABEL
 	def_bool y if (DMI || ACPI)
 	select NLS
 
Quoted 2 lines

drivers/pci/Kconfig config PCI_LABEL · line 186

Quoted 2 lines
drivers/pci/Kconfig +2 -0
+config PCI_SSDLEDS
+	def_bool y if (ACPI && PCI)

That only should be set if the machine can not boot without it.

Quoted 6 lines

For a blinky light, that's not the case.

And why isn't this code using the existing LED subsystem? Don't create random new driver-specific sysfs files that do things the existing class drivers do please.

I did consider the LED subsystem, but I don't think it is a great match.

Quoted 9 lines

In spite of the name, this _DSM doesn't directly control individual LEDs--it sets the state(s) of the PCI port to which an SSD may be connected, so that it may be conveyed to the user... a processor (or at least some logic) will decide how to show which states are active, probably by setting combinations of LEDs to certain colors or blink patterns, or possibly on some other type of display. On the system I have, changing the active state(s) sends a message via IPMI to an embedded processor, which will decide the colors and/or flashing pattern of the LEDs. So brightness/color/blinking are not controlled, or even known, by the OS.

Then I STRONGLY suggest you change the name of all of this code then, as it is totally confusing.

Quoted 11 lines

Also, not all of the states will necessarily always be available. For example, you may only be able to set the "rebuild" state when a drive is actually connected to the pci port that has this _DSM, while the "locate" state might be available all the time. Since there's no notification if/when the supported states change, I believe that would mean, to implement this using the LED subsystem, the driver would have to register an "LED" with the LED subsystem for each possible state, and either make reads or writes to that state fail if it isn't supported.

But I will re-write this using the LED subsystem if you think it's a better fit (though I don't think so).

If you are allowing LEDs to be controlled by the user, then yes, you have to use the LED subsystem as you should never try to create a brand new driver-specific user/kernel API just for your tiny driver right? Please work with the subsystems we have, they are unified for a reason.

thanks,

greg k-h

#8 lore raw
In reply to Stuart Hayes

Hi Stuart,

On 20-11-10 09:37:35, Stuart Hayes wrote:

[...]

drivers/pci/pci-ssdleds.c line 115

Quoted 6 lines
drivers/pci/pci-ssdleds.c new file +6 -0
+
+	scnprintf(buf, PAGE_SIZE, "%#x\n", dsm_output->state);
+
+	ACPI_FREE(out_obj);
+
+	return strlen(buf) > 0 ? strlen(buf) : -EIO;

Question to you - since scnprintf() returns the number of characters written into a buffer, maybe it be possible to use this return value instead of using strlen(), what do you think?

Krzysztof