Skip to contentSkip to conversation

ACPI _CRS CSI-2 and MIPI DisCo for Imaging support

PATCH In review Sakari Ailus 4 participants 36 messages Mar 29, 2023 – May 23, 2023

Subsystems (5)
lore
Share & tools
lore

Sakari Ailus

author reviewer
#0 lore raw

v8 cover ACPI _CRS CSI-2 and MIPI DisCo for Imaging support

Hello all,

Here's an implementation of ACPI 6.4 _CRS CSI-2 resource descriptor and MIPI DisCo for Imaging 1.0 [1]. What the two basically provide is an officially sanctioned way to describe CSI-2 connected cameras to operating system software, something DT based systems have enjoyed for quite some time already.

The implementation digs the information from ACPI tables (_CRS descriptors and data + property extensions) and constructs software nodes that are compatible with Documentation/firmware-guide/acpi/dsd/graph.rst and Documentation/devicetree/bindings/media/video-interface-devices.yaml . No specific driver changes are needed.

The first patch of the set also removes the second namespace walk from DSDT root.

[1] https://www.mipi.org/specifications/mipi-disco-imaging

since v7:

- Release _CRS CSI2 resource list directly from acpi_scan_bus(), calling acpi_bus_scan_crs_csi2_release(), instead of acpi_bus_scan_crs_csi2() doing it.

- Change first "for" to "because" in comment explaining node allocation.

- Move list heads to the beginning of the struct where possible.

- Use Andy's rework of the NEXT_PROPERTY() macro.

- Unwrap acpi_bus_device_postpone() definition.

- Define MIPI_IMG_PORT_PREFIX when the string is needed instead of changing the just added definition.

since v6:

- Avoid including an extra patch in the set. Otherwise the set is unchanged.

since v5:

- Add a high level description of the mipi.c file in the heading.

- Add a missing memory allocation check (thanks to Dan Carpenter).

- Prepend a patch to the set that removes the second ACPI namespace walk due _DEP. Instead, the devices that can't be initialised yet are collected to a linked list that is revisited after initialising the rest of the devices.

- Collect information on _CRS CSI2 records during the same, single namespace walk. Effectively since v5, two namespace walks from the namespace root are removed.

- The former requires passing state information between functions during the _DSD property initialisation, among other things, introducing additional function arguments in _DSD parsing functions.

- The order of the patches has changed somewhat, as the parsing process has changed. Also a patch has been added to collect ACPI devices for which software nodes need to be initialised for DisCo for Imaging.

- Elsewhere, acpi_bus_scan() becomes easier to grasp. Comments are also added in acpi_bus_scan() to explain what is being done.

- Change the wrong node type warning from acpi_parse_string_ref() to use debug level.

since v4:

- Add leading dots to comment sentences.

- Use UINT_MAX - 1 to denote an unallocated port instead of ~1U.

- Unwrap a line.

- Get ACPI handle into a local variable in acpi_init_swnodes() for easier use, use acpi_device_handle() to obtain it.

- Rework "rotation" property checking and _PLD object access.

- Also obtain return value of acpi_get_name() into a local variable before testing it.

- Use a local variable for the first element string pointer in acpi_properties_prepare_mipi().

since v3:

- Add comments to data structures and functions, code inside functions.

- Use ACPI_FAILURE() for testing ACPI framework function return values.

- Unwrap a few lines.

- Rename list heads as "head", some were called just "list".

- Count ACPI handles related to _CRS CSI2 resources during tree walk.

- Reshape testing for CSI-2 port allocation in next_csi2_port_index().

- Move allocation of software nodes into a new function, acpi_crs_csi2_alloc_fill_swnodes().

- Comments: acpi_bus_scan_crs_csi2() is to be called on the namespace root.

- Use size_t for this_count variable in acpi_bus_scan_crs_csi2().

- Revert the NEXT_PROPERTY() macro condition to original (the suggested one was different).

- Remove u union in init_port_csi2_common().

- Fix val array size in init_port_csi2_common(). This issue was masked by the presence of a u32 field in the union.

- Use "-" for copyright year range.

since v2:

- Unwrap a few lines.

- Copy CSI-2 resource source string using strscpy() instead of memcpy() in scan_check_crs_csi2_instance.

- Fix GRAPH_PORT_NAME() sanity check bug introduced in v2.

- Fix snprintf() return value check for port node name in get_mipi_port_handle().

- Fix mipi-img-lane-polarities reading.

- Cast bit value to bool instead of using ... ? 1U : 0U.

- Get primary fwnode using acpi_fwnode_handle().

- Don't use MIPI_IMG_PREFIX in the array of renamed properties.

- Use tabs for indenting drivers/acpi/property.c authors.

- Add a comment on assigning ACPI device's secondary fwnode and assign ACPI device's secondary fwnode straight to NULL when unassigning it.

since v1:

- Update copyright notices.

- Include linux/types.h instead of linux/kernel.h in drivers/acpi/mipi.c.

- Use SWNODE_GRAPH_PORT_NAME_FMT instead of plain "port@%u" in GRAPH_PORT_NAME macro.

- Make the condition in NEXT_PROPERTY() macro easier to read.

- Unwrap lines to make them moderately longer than 80 characters.

- Use * BITS_PER_TYPE(u8) instead of << 3 to convert bytes to bits in init_port_csi2_common().

- Test ACPI framework call success using ACPI_SUCCESS() instead of comparing with AE_OK. Likewise for ACPI_FAILURE and != AE_OK.

- Use newly added SOFTWARE_NODE() macro to construct the root software node.

- Use str_has_prefix() to test for a string prefix instead of memcmp().

- Add pr_fmt() macro to drivers/acpi/property.c.

- Move logical or operators to the end of the line in acpi_properties_prepare().

- Improve bad node type error in acpi_parse_string_ref().

Sakari Ailus (10):
ACPI: scan: Remove the second DSDT traversal
ACPI: property: Parse data node string references in properties
ACPI: property: Parse _CRS CSI-2 descriptor
device property: Add SOFTWARE_NODE() macro for defining software nodes
ACPI: property: Prepare generating swnodes for ACPI and DisCo for
Imaging
ACPI: scan: Generate software nodes based on MIPI DisCo for Imaging
ACPI: property: Dig "rotation" property for devices with CSI2 _CRS
ACPI: property: Rename parsed MIPI DisCo for Imaging properties
ACPI: property: Skip MIPI property table without "mipi-img" prefix
ACPI: property: Document _CRS CSI-2 and DisCo for Imaging support

 drivers/acpi/Makefile    |   2 +-
 drivers/acpi/internal.h  |  41 +-
 drivers/acpi/mipi.c      | 796 +++++++++++++++++++++++++++++++++++++++
 drivers/acpi/power.c     |   2 +-
 drivers/acpi/property.c  | 180 +++++++--
 drivers/acpi/scan.c      | 227 +++++++++--
 include/acpi/acpi_bus.h  |  82 ++++
 include/linux/property.h |   7 +
 8 files changed, 1252 insertions(+), 85 deletions(-)
 create mode 100644 drivers/acpi/mipi.c
Signature · 32 lines
2.30.2



*** BLURB HERE ***

Sakari Ailus (10):
  ACPI: scan: Remove the second DSDT traversal
  ACPI: property: Parse data node string references in properties
  ACPI: property: Parse _CRS CSI-2 descriptor
  device property: Add SOFTWARE_NODE() macro for defining software nodes
  ACPI: property: Prepare generating swnodes for ACPI and DisCo for
    Imaging
  ACPI: scan: Generate software nodes based on MIPI DisCo for Imaging
  ACPI: property: Dig "rotation" property for devices with CSI2 _CRS
  ACPI: property: Rename parsed MIPI DisCo for Imaging properties
  ACPI: property: Skip MIPI property table without "mipi-img" prefix
  ACPI: property: Document _CRS CSI-2 and DisCo for Imaging support

 drivers/acpi/Makefile    |   2 +-
 drivers/acpi/internal.h  |  42 ++-
 drivers/acpi/mipi.c      | 794 +++++++++++++++++++++++++++++++++++++++
 drivers/acpi/power.c     |   2 +-
 drivers/acpi/property.c  | 180 +++++++--
 drivers/acpi/scan.c      | 230 ++++++++++--
 include/acpi/acpi_bus.h  |  82 ++++
 include/linux/property.h |   7 +
 8 files changed, 1254 insertions(+), 85 deletions(-)
 create mode 100644 drivers/acpi/mipi.c

-- 
2.30.2
author reviewer
#1 lore raw

PATCH v8 1/10 ACPI: scan: Remove the second DSDT traversal

Collect the devices with _DEP into a list and continue processing them after a full traversal, instead of doing a full second traversal of the tree.

This makes the second DSDT traversal pass unnecessary as we already have the nodes we're interested in in a linked list.

---
 drivers/acpi/scan.c | 125 ++++++++++++++++++++++++++++++++------------
 1 file changed, 92 insertions(+), 33 deletions(-)
drivers/acpi/scan.c +92 -33
index 0c6f06abe3f4..280d12e0aa2f 100644
@@ -2029,10 +2029,52 @@ static u32 acpi_scan_check_dep(acpi_handle handle, bool check_dep)
 	return count;
 }
 
-static bool acpi_bus_scan_second_pass;
+/**
+ * struct acpi_postponed_handle - A postponed ACPI handle
+ * @list: Entry in a postponed list
+ * @handle: The postponed handle
+ *
+ * One such entry represents an ACPI handle the scanning of which has been
+ * postponed.
+ */
+struct acpi_postponed_handle {
+	struct list_head list;
+	acpi_handle handle;
+};
+
+/**
+ * struct acpi_scan_context - Context for scanning ACPI devices
+ * @postponed_head: The list head of the postponed ACPI handles
+ * @device: The first encountered device, typically the root of the scanned tree
+ */
+struct acpi_scan_context {
+	struct list_head postponed_head;
+	struct acpi_device *device;
+};
+
+/**
+ * acpi_bus_handle_postpone - Add an ACPI handle to a given postponed list
+ * @handle: The ACPI handle
+ * @head: Postponed list head
+ *
+ * Add a given ACPI handle to a list of ACPI objects for which the creation
+ * of the device objects is to be postponed.
+ */
+static void acpi_bus_handle_postpone(acpi_handle handle,
+				     struct list_head *head)
+{
+	struct acpi_postponed_handle *ph;
+
+	ph = kzalloc(sizeof(*ph), GFP_KERNEL);
+	if (!ph)
+		return;
+
+	ph->handle = handle;
+	list_add(&ph->list, head);
+}
 
 static acpi_status acpi_bus_check_add(acpi_handle handle, bool check_dep,
-				      struct acpi_device **adev_p)
+				      struct acpi_scan_context *ctx)
 {
 	struct acpi_device *device = acpi_fetch_acpi_dev(handle);
 	acpi_object_type acpi_type;
@@ -2051,7 +2093,7 @@ static acpi_status acpi_bus_check_add(acpi_handle handle, bool check_dep,
 
 		/* Bail out if there are dependencies. */
 		if (acpi_scan_check_dep(handle, check_dep) > 0) {
-			acpi_bus_scan_second_pass = true;
+			acpi_bus_handle_postpone(handle, &ctx->postponed_head);
 			return AE_CTRL_DEPTH;
 		}
 
@@ -2086,22 +2128,22 @@ static acpi_status acpi_bus_check_add(acpi_handle handle, bool check_dep,
 	acpi_scan_init_hotplug(device);
 
 out:
-	if (!*adev_p)
-		*adev_p = device;
+	if (!ctx->device)
+		ctx->device = device;
 
 	return AE_OK;
 }
 
 static acpi_status acpi_bus_check_add_1(acpi_handle handle, u32 lvl_not_used,
-					void *not_used, void **ret_p)
+					void *ctx, void **unused)
 {
-	return acpi_bus_check_add(handle, true, (struct acpi_device **)ret_p);
+	return acpi_bus_check_add(handle, true, (struct acpi_scan_context *)ctx);
 }
 
 static acpi_status acpi_bus_check_add_2(acpi_handle handle, u32 lvl_not_used,
-					void *not_used, void **ret_p)
+					void *ctx, void **device)
 {
-	return acpi_bus_check_add(handle, false, (struct acpi_device **)ret_p);
+	return acpi_bus_check_add(handle, false, (struct acpi_scan_context *)ctx);
 }
 
 static void acpi_default_enumeration(struct acpi_device *device)
@@ -2422,37 +2464,54 @@ EXPORT_SYMBOL_GPL(acpi_dev_get_next_consumer_dev);
  */
 int acpi_bus_scan(acpi_handle handle)
 {
-	struct acpi_device *device = NULL;
-
-	acpi_bus_scan_second_pass = false;
-
-	/* Pass 1: Avoid enumerating devices with missing dependencies. */
+	struct acpi_scan_context ctx = {
+		.postponed_head = LIST_HEAD_INIT(ctx.postponed_head),
+	};
+	struct acpi_postponed_handle *ph, *tmp_ph;
+	int ret = 0;
 
-	if (ACPI_SUCCESS(acpi_bus_check_add(handle, true, &device)))
+	if (ACPI_SUCCESS(acpi_bus_check_add(handle, true, &ctx)))
 		acpi_walk_namespace(ACPI_TYPE_ANY, handle, ACPI_UINT32_MAX,
-				    acpi_bus_check_add_1, NULL, NULL,
-				    (void **)&device);
-
-	if (!device)
-		return -ENODEV;
-
-	acpi_bus_attach(device, (void *)true);
+				    acpi_bus_check_add_1, NULL, (void *)&ctx,
+				    NULL);
 
-	if (!acpi_bus_scan_second_pass)
-		return 0;
-
-	/* Pass 2: Enumerate all of the remaining devices. */
+	if (!ctx.device) {
+		ret = -ENODEV;
+		goto out_release;
+	}
 
-	device = NULL;
+	acpi_bus_attach(ctx.device, (void *)true);
 
-	if (ACPI_SUCCESS(acpi_bus_check_add(handle, false, &device)))
-		acpi_walk_namespace(ACPI_TYPE_ANY, handle, ACPI_UINT32_MAX,
-				    acpi_bus_check_add_2, NULL, NULL,
-				    (void **)&device);
+	/*
+	 * Proceed to register ACPI devices that were postponed due to _DEP
+	 * objects during the namespace walk.
+	 */
+	list_for_each_entry_safe(ph, tmp_ph, &ctx.postponed_head, list) {
+		list_del(&ph->list);
+		/* Set device NULL here to obtain the root for this sub-tree. */
+		ctx.device = NULL;
+		/*
+		 * Do this manually, as the namespace walk will only include
+		 * sub-nodes, not the node itself. ctx.device is set to the
+		 * ACPI device corresponding ph->handle.
+		 */
+		acpi_bus_check_add_2(ph->handle, 0, &ctx, NULL);
+		/* Walk the rest of the sub-namespace. */
+		acpi_walk_namespace(ACPI_TYPE_ANY, ph->handle, ACPI_UINT32_MAX,
+				    acpi_bus_check_add_2, NULL, (void *)&ctx,
+				    NULL);
+		if (ctx.device)
+			acpi_bus_attach(ctx.device, NULL);
+		kfree(ph);
+	}
 
-	acpi_bus_attach(device, NULL);
+out_release:
+	list_for_each_entry_safe(ph, tmp_ph, &ctx.postponed_head, list) {
+		list_del(&ph->list);
+		kfree(ph);
+	}
 
-	return 0;
+	return ret;
 }
 EXPORT_SYMBOL(acpi_bus_scan);
 
Signature · 1 line
2.30.2
#2 lore raw

Re: [PATCH v8 01/10] ACPI: scan: Remove the second DSDT traversal

Quoted 7 lines · Sakari Ailus

Collect the devices with _DEP into a list and continue processing them after a full traversal, instead of doing a full second traversal of the tree.

This makes the second DSDT traversal pass unnecessary as we already have the nodes we're interested in in a linked list.

The second traversal of the ACPI namespace (it may not be just the DSDT at that point to be precise) is not really about _DEP handling. In fact, the latter has been added on top of it.

It is about the PCI enumeration. Namely, when acpi_pci_root_add() runs for the PCI host bridge object in the ACPI namespace, the entire device hierarchy below it is walked and all of the ACPI device objects corresponding to the PCI devices on the bus are assumed to be present. This means that all of the ACPI device objects need to be created in the first walk, without binding any ACPI drivers or scan handlers to them, and the second walk is to find out what is actually represented by those objects.

It cannot be eliminated in any simple way.

Quoted 184 lines
---
 drivers/acpi/scan.c | 125 ++++++++++++++++++++++++++++++++------------
 1 file changed, 92 insertions(+), 33 deletions(-)
drivers/acpi/scan.c +92 -34
index 0c6f06abe3f4..280d12e0aa2f 100644
@@ -2029,10 +2029,52 @@ static u32 acpi_scan_check_dep(acpi_handle handle, bool check_dep)
        return count;
 }
 
-static bool acpi_bus_scan_second_pass;
+/**
+ * struct acpi_postponed_handle - A postponed ACPI handle
+ * @list: Entry in a postponed list
+ * @handle: The postponed handle
+ *
+ * One such entry represents an ACPI handle the scanning of which has been
+ * postponed.
+ */
+struct acpi_postponed_handle {
+       struct list_head list;
+       acpi_handle handle;
+};
+
+/**
+ * struct acpi_scan_context - Context for scanning ACPI devices
+ * @postponed_head: The list head of the postponed ACPI handles
+ * @device: The first encountered device, typically the root of the scanned tree
+ */
+struct acpi_scan_context {
+       struct list_head postponed_head;
+       struct acpi_device *device;
+};
+
+/**
+ * acpi_bus_handle_postpone - Add an ACPI handle to a given postponed list
+ * @handle: The ACPI handle
+ * @head: Postponed list head
+ *
+ * Add a given ACPI handle to a list of ACPI objects for which the creation
+ * of the device objects is to be postponed.
+ */
+static void acpi_bus_handle_postpone(acpi_handle handle,
+                                    struct list_head *head)
+{
+       struct acpi_postponed_handle *ph;
+
+       ph = kzalloc(sizeof(*ph), GFP_KERNEL);
+       if (!ph)
+               return;
+
+       ph->handle = handle;
+       list_add(&ph->list, head);
+}
 
 static acpi_status acpi_bus_check_add(acpi_handle handle, bool check_dep,
-                                     struct acpi_device **adev_p)
+                                     struct acpi_scan_context *ctx)
 {
        struct acpi_device *device = acpi_fetch_acpi_dev(handle);
        acpi_object_type acpi_type;
@@ -2051,7 +2093,7 @@ static acpi_status acpi_bus_check_add(acpi_handle handle, bool check_dep,
 
                /* Bail out if there are dependencies. */
                if (acpi_scan_check_dep(handle, check_dep) > 0) {
-                       acpi_bus_scan_second_pass = true;
+                       acpi_bus_handle_postpone(handle, &ctx->postponed_head);
                        return AE_CTRL_DEPTH;
                }
 
@@ -2086,22 +2128,22 @@ static acpi_status acpi_bus_check_add(acpi_handle handle, bool check_dep,
        acpi_scan_init_hotplug(device);
 
 out:
-       if (!*adev_p)
-               *adev_p = device;
+       if (!ctx->device)
+               ctx->device = device;
 
        return AE_OK;
 }
 
 static acpi_status acpi_bus_check_add_1(acpi_handle handle, u32 lvl_not_used,
-                                       void *not_used, void **ret_p)
+                                       void *ctx, void **unused)
 {
-       return acpi_bus_check_add(handle, true, (struct acpi_device **)ret_p);
+       return acpi_bus_check_add(handle, true, (struct acpi_scan_context *)ctx);
 }
 
 static acpi_status acpi_bus_check_add_2(acpi_handle handle, u32 lvl_not_used,
-                                       void *not_used, void **ret_p)
+                                       void *ctx, void **device)
 {
-       return acpi_bus_check_add(handle, false, (struct acpi_device **)ret_p);
+       return acpi_bus_check_add(handle, false, (struct acpi_scan_context *)ctx);
 }
 
 static void acpi_default_enumeration(struct acpi_device *device)
@@ -2422,37 +2464,54 @@ EXPORT_SYMBOL_GPL(acpi_dev_get_next_consumer_dev);
  */
 int acpi_bus_scan(acpi_handle handle)
 {
-       struct acpi_device *device = NULL;
-
-       acpi_bus_scan_second_pass = false;
-
-       /* Pass 1: Avoid enumerating devices with missing dependencies. */
+       struct acpi_scan_context ctx = {
+               .postponed_head = LIST_HEAD_INIT(ctx.postponed_head),
+       };
+       struct acpi_postponed_handle *ph, *tmp_ph;
+       int ret = 0;
 
-       if (ACPI_SUCCESS(acpi_bus_check_add(handle, true, &device)))
+       if (ACPI_SUCCESS(acpi_bus_check_add(handle, true, &ctx)))
                acpi_walk_namespace(ACPI_TYPE_ANY, handle, ACPI_UINT32_MAX,
-                                   acpi_bus_check_add_1, NULL, NULL,
-                                   (void **)&device);
-
-       if (!device)
-               return -ENODEV;
-
-       acpi_bus_attach(device, (void *)true);
+                                   acpi_bus_check_add_1, NULL, (void *)&ctx,
+                                   NULL);
 
-       if (!acpi_bus_scan_second_pass)
-               return 0;
-
-       /* Pass 2: Enumerate all of the remaining devices. */
+       if (!ctx.device) {
+               ret = -ENODEV;
+               goto out_release;
+       }
 
-       device = NULL;
+       acpi_bus_attach(ctx.device, (void *)true);
 
-       if (ACPI_SUCCESS(acpi_bus_check_add(handle, false, &device)))
-               acpi_walk_namespace(ACPI_TYPE_ANY, handle, ACPI_UINT32_MAX,
-                                   acpi_bus_check_add_2, NULL, NULL,
-                                   (void **)&device);
+       /*
+        * Proceed to register ACPI devices that were postponed due to _DEP
+        * objects during the namespace walk.
+        */
+       list_for_each_entry_safe(ph, tmp_ph, &ctx.postponed_head, list) {
+               list_del(&ph->list);
+               /* Set device NULL here to obtain the root for this sub-tree. */
+               ctx.device = NULL;
+               /*
+                * Do this manually, as the namespace walk will only include
+                * sub-nodes, not the node itself. ctx.device is set to the
+                * ACPI device corresponding ph->handle.
+                */
+               acpi_bus_check_add_2(ph->handle, 0, &ctx, NULL);
+               /* Walk the rest of the sub-namespace. */
+               acpi_walk_namespace(ACPI_TYPE_ANY, ph->handle, ACPI_UINT32_MAX,
+                                   acpi_bus_check_add_2, NULL, (void *)&ctx,
+                                   NULL);
+               if (ctx.device)
+                       acpi_bus_attach(ctx.device, NULL);
+               kfree(ph);
+       }
 
-       acpi_bus_attach(device, NULL);
+out_release:
+       list_for_each_entry_safe(ph, tmp_ph, &ctx.postponed_head, list) {
+               list_del(&ph->list);
+               kfree(ph);
+       }
 
-       return 0;
+       return ret;
 }
 EXPORT_SYMBOL(acpi_bus_scan);
 
--

2.30.2

author reviewer
#3 lore raw

Re: [PATCH v8 01/10] ACPI: scan: Remove the second DSDT traversal

Hi Rafael,

Thank you for the review.

Quoted 23 lines · Rafael J. Wysocki
Quoted 7 lines · Sakari Ailus

Collect the devices with _DEP into a list and continue processing them after a full traversal, instead of doing a full second traversal of the tree.

This makes the second DSDT traversal pass unnecessary as we already have the nodes we're interested in in a linked list.

The second traversal of the ACPI namespace (it may not be just the DSDT at that point to be precise) is not really about _DEP handling. In fact, the latter has been added on top of it.

It is about the PCI enumeration. Namely, when acpi_pci_root_add() runs for the PCI host bridge object in the ACPI namespace, the entire device hierarchy below it is walked and all of the ACPI device objects corresponding to the PCI devices on the bus are assumed to be present. This means that all of the ACPI device objects need to be created in the first walk, without binding any ACPI drivers or scan handlers to them, and the second walk is to find out what is actually represented by those objects.

Quoted 1 line

It cannot be eliminated in any simple way.

My understanding still remains that this patch does not (or other patches in this set) change the above. It is just how those nodes are reached:
instead of traversing the entire tree and ignoring the devices that have already an acpi_device created for them, a linked list of devices of interest is traversed.

Of course it is possible that I have missed something. The codebase isn't entirely trivial.

Signature · 3 lines
Kind regards,

Sakari Ailus
#4 lore raw

Re: [PATCH v8 01/10] ACPI: scan: Remove the second DSDT traversal

Quoted 30 lines · CEST Sakari Ailus

Hi Rafael,

Thank you for the review.

Quoted 23 lines · Rafael J. Wysocki
Quoted 7 lines · Sakari Ailus

Collect the devices with _DEP into a list and continue processing them after a full traversal, instead of doing a full second traversal of the tree.

This makes the second DSDT traversal pass unnecessary as we already have the nodes we're interested in in a linked list.

The second traversal of the ACPI namespace (it may not be just the DSDT at that point to be precise) is not really about _DEP handling. In fact, the latter has been added on top of it.

It is about the PCI enumeration. Namely, when acpi_pci_root_add() runs for the PCI host bridge object in the ACPI namespace, the entire device hierarchy below it is walked and all of the ACPI device objects corresponding to the PCI devices on the bus are assumed to be present. This means that all of the ACPI device objects need to be created in the first walk, without binding any ACPI drivers or scan handlers to them, and the second walk is to find out what is actually represented by those objects.

Quoted 1 line

It cannot be eliminated in any simple way.

Quoted 8 lines

My understanding still remains that this patch does not (or other patches in this set) change the above. It is just how those nodes are reached:
instead of traversing the entire tree and ignoring the devices that have already an acpi_device created for them, a linked list of devices of interest is traversed.

Of course it is possible that I have missed something. The codebase isn't entirely trivial.

You are right and I see what it is about now.

However, the implementation is rather fragile and the list added by the $subject patch is redundant AFAICS, because all of the objects in it are also present in acpi_dep_list as consumers (adding an object to acpi_dep_list as a consumer is necessary for that object to be added to the new list too). Of course, there may be multiple acpi_dep_list for the same consumer object, but it is not a fundamental problem.

So overall I'd prefer to do something like the appended (untested) patch instead.

---
From: Rafael J. Wysocki <[email protected]>
Subject: [PATCH] ACPI: scan: Reduce overhead related to devices with dependencies

Notice that all of the objects for which the acpi_scan_check_dep() return value is greater than 0 are present in acpi_dep_list as consumers (there may be multiple entries for one object, but that is not a problem), so after carrying out the initial ACPI namespace walk in which devices with dependencies are skipped, acpi_bus_scan() can simply walk acpi_dep_list and enumerate all of the unique consumer objects from there and their descendants instead of walking the entire target branch of the ACPI namespace and looking for device objects that have not been enumerated yet in it.

Because walking acpi_dep_list is generally less overhead than walking the entire ACPI namespace, use the observation above to reduce the system initialization overhead related to ACPI, which is particularly important on large systems.

---
 drivers/acpi/scan.c     |   71 ++++++++++++++++++++++++++++++++++--------------
 include/acpi/acpi_bus.h |    2 +
 2 files changed, 53 insertions(+), 20 deletions(-)
include/acpi/acpi_bus.h +2 -0
Index: linux-pm/include/acpi/acpi_bus.h
===================================================================
@@ -289,6 +289,8 @@ struct acpi_dep_data {
 	acpi_handle supplier;
 	acpi_handle consumer;
 	bool honor_dep;
+	bool met;
+	bool free_when_met;
 };
 
 /* Performance Management */
drivers/acpi/scan.c +51 -20

110 lines not on this page · raw message

author reviewer
#5 lore raw

PATCH v8 2/10 ACPI: property: Parse data node string references in properties

In reply to Sakari Ailus

Add support for parsing property references using strings, besides reference objects that were previously supported. This allows also referencing data nodes which was not possible with reference objects.

Also add pr_fmt() macro to prefix printouts.

While at it, update copyright.

---
 drivers/acpi/property.c | 110 ++++++++++++++++++++++++++++++++++------
 1 file changed, 94 insertions(+), 16 deletions(-)
drivers/acpi/property.c +94 -16
index b8d9eb9a433e..08831ffba26c 100644
@@ -2,14 +2,17 @@
 /*
  * ACPI device specific properties support.
  *
- * Copyright (C) 2014, Intel Corporation
+ * Copyright (C) 2014-2023, Intel Corporation
  * All rights reserved.
  *
  * Authors: Mika Westerberg <[email protected]>
- *          Darren Hart <[email protected]>
- *          Rafael J. Wysocki <[email protected]>
+ *	    Darren Hart <[email protected]>
+ *	    Rafael J. Wysocki <[email protected]>
+ *	    Sakari Ailus <[email protected]>
  */
 
+#define pr_fmt(fmt) "ACPI: " fmt
+
 #include <linux/acpi.h>
 #include <linux/device.h>
 #include <linux/export.h>
@@ -795,7 +798,8 @@ acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode,
 static int acpi_get_ref_args(struct fwnode_reference_args *args,
 			     struct fwnode_handle *ref_fwnode,
 			     const union acpi_object **element,
-			     const union acpi_object *end, size_t num_args)
+			     const union acpi_object *end, size_t num_args,
+			     bool subnode_string)
 {
 	u32 nargs = 0, i;
 
@@ -803,13 +807,16 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
 	 * Find the referred data extension node under the
 	 * referred device node.
 	 */
-	for (; *element < end && (*element)->type == ACPI_TYPE_STRING;
-	     (*element)++) {
-		const char *child_name = (*element)->string.pointer;
-
-		ref_fwnode = acpi_fwnode_get_named_child_node(ref_fwnode, child_name);
-		if (!ref_fwnode)
-			return -EINVAL;
+	if (subnode_string) {
+		for (; *element < end && (*element)->type == ACPI_TYPE_STRING;
+		     (*element)++) {
+			const char *child_name = (*element)->string.pointer;
+
+			ref_fwnode = acpi_fwnode_get_named_child_node(ref_fwnode,
+								      child_name);
+			if (!ref_fwnode)
+				return -EINVAL;
+		}
 	}
 
 	/*
@@ -820,7 +827,8 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
 	for (i = 0; (*element) + i < end && i < num_args; i++) {
 		acpi_object_type type = (*element)[i].type;
 
-		if (type == ACPI_TYPE_LOCAL_REFERENCE)
+		if (type == ACPI_TYPE_LOCAL_REFERENCE ||
+		    (!subnode_string && type == ACPI_TYPE_STRING))
 			break;
 
 		if (type == ACPI_TYPE_INTEGER)
@@ -844,6 +852,43 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
 	return 0;
 }
 
+static struct fwnode_handle *
+acpi_parse_string_ref(const struct fwnode_handle *fwnode, const char *refstring)
+{
+	acpi_handle scope, handle;
+	struct acpi_data_node *dn;
+	struct acpi_device *device;
+	acpi_status status;
+
+	if (is_acpi_device_node(fwnode)) {
+		scope = to_acpi_device_node(fwnode)->handle;
+	} else if (is_acpi_data_node(fwnode)) {
+		scope = to_acpi_data_node(fwnode)->handle;
+	} else {
+		pr_debug("bad node type for node %pfw\n", fwnode);
+		return ERR_PTR(-EINVAL);
+	}
+
+	status = acpi_get_handle(scope, refstring, &handle);
+	if (ACPI_FAILURE(status)) {
+		acpi_handle_debug(scope, "can't get handle for %s", refstring);
+		return ERR_PTR(-EINVAL);
+	}
+
+	device = acpi_fetch_acpi_dev(handle);
+	if (device)
+		return acpi_fwnode_handle(device);
+
+	status = acpi_get_data_full(handle, acpi_nondev_subnode_tag,
+				    (void **)&dn, NULL);
+	if (ACPI_FAILURE(status) || !dn) {
+		acpi_handle_debug(handle, "can't find subnode");
+		return ERR_PTR(-EINVAL);
+	}
+
+	return &dn->fwnode;
+}
+
 /**
  * __acpi_node_get_property_reference - returns handle to the referenced object
  * @fwnode: Firmware node to get the property from
@@ -886,6 +931,7 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
 	const union acpi_object *element, *end;
 	const union acpi_object *obj;
 	const struct acpi_device_data *data;
+	struct fwnode_handle *ref_fwnode;
 	struct acpi_device *device;
 	int ret, idx = 0;
 
@@ -909,16 +955,29 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
 
 		args->fwnode = acpi_fwnode_handle(device);
 		args->nargs = 0;
+		return 0;
+	case ACPI_TYPE_STRING:
+		if (index)
+			return -ENOENT;
+
+		ref_fwnode = acpi_parse_string_ref(fwnode, obj->string.pointer);
+		if (IS_ERR(ref_fwnode))
+			return PTR_ERR(ref_fwnode);
+
+		args->fwnode = ref_fwnode;
+		args->nargs = 0;
+
 		return 0;
 	case ACPI_TYPE_PACKAGE:
 		/*
 		 * If it is not a single reference, then it is a package of
-		 * references followed by number of ints as follows:
+		 * references, followed by number of ints as follows:
 		 *
 		 *  Package () { REF, INT, REF, INT, INT }
 		 *
-		 * The index argument is then used to determine which reference
-		 * the caller wants (along with the arguments).
+		 * Here, REF may be either a local reference or a string. The
+		 * index argument is then used to determine which reference the
+		 * caller wants (along with the arguments).
 		 */
 		break;
 	default:
@@ -942,7 +1001,26 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
 
 			ret = acpi_get_ref_args(idx == index ? args : NULL,
 						acpi_fwnode_handle(device),
-						&element, end, num_args);
+						&element, end, num_args, true);
+			if (ret < 0)
+				return ret;
+
+			if (idx == index)
+				return 0;
+
+			break;
+		case ACPI_TYPE_STRING:
+			ref_fwnode =
+				acpi_parse_string_ref(fwnode,
+						      element->string.pointer);
+			if (IS_ERR(ref_fwnode))
+				return PTR_ERR(ref_fwnode);
+
+			element++;
+
+			ret = acpi_get_ref_args(idx == index ? args : NULL,
+						ref_fwnode, &element, end,
+						num_args, false);
 			if (ret < 0)
 				return ret;
 
Signature · 1 line
2.30.2
maintainer Acked-by on v8 2/10
#6 lore raw

Re: [PATCH v8 02/10] ACPI: property: Parse data node string references in properties

Quoted 10 lines · Sakari Ailus

Add support for parsing property references using strings, besides reference objects that were previously supported. This allows also referencing data nodes which was not possible with reference objects.

Also add pr_fmt() macro to prefix printouts.

While at it, update copyright.

This one looks good to me, so formally

and please add this tag if you resubmit.

Alternatively, I can just queue it up, because IMV this is an improvement regardless of the rest of the series.

Quoted 189 lines
---
 drivers/acpi/property.c | 110 ++++++++++++++++++++++++++++++++++------
 1 file changed, 94 insertions(+), 16 deletions(-)
drivers/acpi/property.c +94 -17
index b8d9eb9a433e..08831ffba26c 100644
@@ -2,14 +2,17 @@
 /*
  * ACPI device specific properties support.
  *
- * Copyright (C) 2014, Intel Corporation
+ * Copyright (C) 2014-2023, Intel Corporation
  * All rights reserved.
  *
  * Authors: Mika Westerberg <[email protected]>
- *          Darren Hart <[email protected]>
- *          Rafael J. Wysocki <[email protected]>
+ *         Darren Hart <[email protected]>
+ *         Rafael J. Wysocki <[email protected]>
+ *         Sakari Ailus <[email protected]>
  */
 
+#define pr_fmt(fmt) "ACPI: " fmt
+
 #include <linux/acpi.h>
 #include <linux/device.h>
 #include <linux/export.h>
@@ -795,7 +798,8 @@ acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode,
 static int acpi_get_ref_args(struct fwnode_reference_args *args,
                             struct fwnode_handle *ref_fwnode,
                             const union acpi_object **element,
-                            const union acpi_object *end, size_t num_args)
+                            const union acpi_object *end, size_t num_args,
+                            bool subnode_string)
 {
        u32 nargs = 0, i;
 
@@ -803,13 +807,16 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
         * Find the referred data extension node under the
         * referred device node.
         */
-       for (; *element < end && (*element)->type == ACPI_TYPE_STRING;
-            (*element)++) {
-               const char *child_name = (*element)->string.pointer;
-
-               ref_fwnode = acpi_fwnode_get_named_child_node(ref_fwnode, child_name);
-               if (!ref_fwnode)
-                       return -EINVAL;
+       if (subnode_string) {
+               for (; *element < end && (*element)->type == ACPI_TYPE_STRING;
+                    (*element)++) {
+                       const char *child_name = (*element)->string.pointer;
+
+                       ref_fwnode = acpi_fwnode_get_named_child_node(ref_fwnode,
+                                                                     child_name);
+                       if (!ref_fwnode)
+                               return -EINVAL;
+               }
        }
 
        /*
@@ -820,7 +827,8 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
        for (i = 0; (*element) + i < end && i < num_args; i++) {
                acpi_object_type type = (*element)[i].type;
 
-               if (type == ACPI_TYPE_LOCAL_REFERENCE)
+               if (type == ACPI_TYPE_LOCAL_REFERENCE ||
+                   (!subnode_string && type == ACPI_TYPE_STRING))
                        break;
 
                if (type == ACPI_TYPE_INTEGER)
@@ -844,6 +852,43 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
        return 0;
 }
 
+static struct fwnode_handle *
+acpi_parse_string_ref(const struct fwnode_handle *fwnode, const char *refstring)
+{
+       acpi_handle scope, handle;
+       struct acpi_data_node *dn;
+       struct acpi_device *device;
+       acpi_status status;
+
+       if (is_acpi_device_node(fwnode)) {
+               scope = to_acpi_device_node(fwnode)->handle;
+       } else if (is_acpi_data_node(fwnode)) {
+               scope = to_acpi_data_node(fwnode)->handle;
+       } else {
+               pr_debug("bad node type for node %pfw\n", fwnode);
+               return ERR_PTR(-EINVAL);
+       }
+
+       status = acpi_get_handle(scope, refstring, &handle);
+       if (ACPI_FAILURE(status)) {
+               acpi_handle_debug(scope, "can't get handle for %s", refstring);
+               return ERR_PTR(-EINVAL);
+       }
+
+       device = acpi_fetch_acpi_dev(handle);
+       if (device)
+               return acpi_fwnode_handle(device);
+
+       status = acpi_get_data_full(handle, acpi_nondev_subnode_tag,
+                                   (void **)&dn, NULL);
+       if (ACPI_FAILURE(status) || !dn) {
+               acpi_handle_debug(handle, "can't find subnode");
+               return ERR_PTR(-EINVAL);
+       }
+
+       return &dn->fwnode;
+}
+
 /**
  * __acpi_node_get_property_reference - returns handle to the referenced object
  * @fwnode: Firmware node to get the property from
@@ -886,6 +931,7 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
        const union acpi_object *element, *end;
        const union acpi_object *obj;
        const struct acpi_device_data *data;
+       struct fwnode_handle *ref_fwnode;
        struct acpi_device *device;
        int ret, idx = 0;
 
@@ -909,16 +955,29 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
 
                args->fwnode = acpi_fwnode_handle(device);
                args->nargs = 0;
+               return 0;
+       case ACPI_TYPE_STRING:
+               if (index)
+                       return -ENOENT;
+
+               ref_fwnode = acpi_parse_string_ref(fwnode, obj->string.pointer);
+               if (IS_ERR(ref_fwnode))
+                       return PTR_ERR(ref_fwnode);
+
+               args->fwnode = ref_fwnode;
+               args->nargs = 0;
+
                return 0;
        case ACPI_TYPE_PACKAGE:
                /*
                 * If it is not a single reference, then it is a package of
-                * references followed by number of ints as follows:
+                * references, followed by number of ints as follows:
                 *
                 *  Package () { REF, INT, REF, INT, INT }
                 *
-                * The index argument is then used to determine which reference
-                * the caller wants (along with the arguments).
+                * Here, REF may be either a local reference or a string. The
+                * index argument is then used to determine which reference the
+                * caller wants (along with the arguments).
                 */
                break;
        default:
@@ -942,7 +1001,26 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
 
                        ret = acpi_get_ref_args(idx == index ? args : NULL,
                                                acpi_fwnode_handle(device),
-                                               &element, end, num_args);
+                                               &element, end, num_args, true);
+                       if (ret < 0)
+                               return ret;
+
+                       if (idx == index)
+                               return 0;
+
+                       break;
+               case ACPI_TYPE_STRING:
+                       ref_fwnode =
+                               acpi_parse_string_ref(fwnode,
+                                                     element->string.pointer);
+                       if (IS_ERR(ref_fwnode))
+                               return PTR_ERR(ref_fwnode);
+
+                       element++;
+
+                       ret = acpi_get_ref_args(idx == index ? args : NULL,
+                                               ref_fwnode, &element, end,
+                                               num_args, false);
                        if (ret < 0)
                                return ret;
 
--

2.30.2

#7 lore raw

Re: [PATCH v8 02/10] ACPI: property: Parse data node string references in properties

In reply to Sakari Ailus
Quoted 8 lines · Sakari Ailus

Add support for parsing property references using strings, besides reference objects that were previously supported. This allows also referencing data nodes which was not possible with reference objects.

Also add pr_fmt() macro to prefix printouts.

While at it, update copyright.

Although I said that it looked good to me, some minor improvements can still be made.

First off, the above changelog is a bit terse.

I think that it would help to provide an example of device properties that would not be parsed properly before the change and can be parsed now.

Quoted 12 lines
---
 drivers/acpi/property.c | 110 ++++++++++++++++++++++++++++++++++------
 1 file changed, 94 insertions(+), 16 deletions(-)
drivers/acpi/property.c +0 -0
index b8d9eb9a433e..08831ffba26c 100644
@@ -2,14 +2,17 @@
 /*
  * ACPI device specific properties support.

drivers/acpi/property.c line 10

Quoted 10 lines
drivers/acpi/property.c +3 -3
  *
- * Copyright (C) 2014, Intel Corporation
+ * Copyright (C) 2014-2023, Intel Corporation
  * All rights reserved.
  *
  * Authors: Mika Westerberg <[email protected]>
- *          Darren Hart <[email protected]>
- *          Rafael J. Wysocki <[email protected]>
+ *         Darren Hart <[email protected]>
+ *         Rafael J. Wysocki <[email protected]>

I'm not sure if the whitespace change here is really useful.

Quoted 9 lines
drivers/acpi/property.c +3 -0
+ *         Sakari Ailus <[email protected]>
  */
 
+#define pr_fmt(fmt) "ACPI: " fmt
+
 #include <linux/acpi.h>
 #include <linux/device.h>
 #include <linux/export.h>

drivers/acpi/property.c acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode, · line 802

Quoted 6 lines
drivers/acpi/property.c +2 -1
@@ -795,7 +798,8 @@ acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode,
 static int acpi_get_ref_args(struct fwnode_reference_args *args,
                             struct fwnode_handle *ref_fwnode,
                             const union acpi_object **element,
-                            const union acpi_object *end, size_t num_args)
+                            const union acpi_object *end, size_t num_args,
+                            bool subnode_string)

The meaning of the new argument isn't really clear. it would be good to somehow help a casual reader of the code to find this out more easily.

Quoted 66 lines
drivers/acpi/property.c +37 -8
 {
        u32 nargs = 0, i;
 
@@ -803,13 +807,16 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
         * Find the referred data extension node under the
         * referred device node.
         */
-       for (; *element < end && (*element)->type == ACPI_TYPE_STRING;
-            (*element)++) {
-               const char *child_name = (*element)->string.pointer;
-
-               ref_fwnode = acpi_fwnode_get_named_child_node(ref_fwnode, child_name);
-               if (!ref_fwnode)
-                       return -EINVAL;
+       if (subnode_string) {
+               for (; *element < end && (*element)->type == ACPI_TYPE_STRING;
+                    (*element)++) {
+                       const char *child_name = (*element)->string.pointer;
+
+                       ref_fwnode = acpi_fwnode_get_named_child_node(ref_fwnode,
+                                                                     child_name);
+                       if (!ref_fwnode)
+                               return -EINVAL;
+               }
        }
 
        /*
@@ -820,7 +827,8 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
        for (i = 0; (*element) + i < end && i < num_args; i++) {
                acpi_object_type type = (*element)[i].type;
 
-               if (type == ACPI_TYPE_LOCAL_REFERENCE)
+               if (type == ACPI_TYPE_LOCAL_REFERENCE ||
+                   (!subnode_string && type == ACPI_TYPE_STRING))
                        break;
 
                if (type == ACPI_TYPE_INTEGER)
@@ -844,6 +852,43 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
        return 0;
 }
 
+static struct fwnode_handle *
+acpi_parse_string_ref(const struct fwnode_handle *fwnode, const char *refstring)
+{
+       acpi_handle scope, handle;
+       struct acpi_data_node *dn;
+       struct acpi_device *device;
+       acpi_status status;
+
+       if (is_acpi_device_node(fwnode)) {
+               scope = to_acpi_device_node(fwnode)->handle;
+       } else if (is_acpi_data_node(fwnode)) {
+               scope = to_acpi_data_node(fwnode)->handle;
+       } else {
+               pr_debug("bad node type for node %pfw\n", fwnode);
+               return ERR_PTR(-EINVAL);
+       }
+
+       status = acpi_get_handle(scope, refstring, &handle);
+       if (ACPI_FAILURE(status)) {
+               acpi_handle_debug(scope, "can't get handle for %s", refstring);
+               return ERR_PTR(-EINVAL);
+       }
+
+       device = acpi_fetch_acpi_dev(handle);
+       if (device)

drivers/acpi/property.c static int acpi_get_ref_args(struct fwnode_reference_args *args, · line 889

Quoted 10 lines
drivers/acpi/property.c +10 -0
+               return acpi_fwnode_handle(device);
+
+       status = acpi_get_data_full(handle, acpi_nondev_subnode_tag,
+                                   (void **)&dn, NULL);
+       if (ACPI_FAILURE(status) || !dn) {
+               acpi_handle_debug(handle, "can't find subnode");
+               return ERR_PTR(-EINVAL);
+       }
+
+       return &dn->fwnode;

So on failure this function always returns the same error code. Can it return NULL instead which can be translated into an error code by the caller?

Quoted 75 lines
drivers/acpi/property.c +39 -5
+}
+
 /**
  * __acpi_node_get_property_reference - returns handle to the referenced object
  * @fwnode: Firmware node to get the property from
@@ -886,6 +931,7 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
        const union acpi_object *element, *end;
        const union acpi_object *obj;
        const struct acpi_device_data *data;
+       struct fwnode_handle *ref_fwnode;
        struct acpi_device *device;
        int ret, idx = 0;
 
@@ -909,16 +955,29 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
 
                args->fwnode = acpi_fwnode_handle(device);
                args->nargs = 0;
+               return 0;
+       case ACPI_TYPE_STRING:
+               if (index)
+                       return -ENOENT;
+
+               ref_fwnode = acpi_parse_string_ref(fwnode, obj->string.pointer);
+               if (IS_ERR(ref_fwnode))
+                       return PTR_ERR(ref_fwnode);
+
+               args->fwnode = ref_fwnode;
+               args->nargs = 0;
+
                return 0;
        case ACPI_TYPE_PACKAGE:
                /*
                 * If it is not a single reference, then it is a package of
-                * references followed by number of ints as follows:
+                * references, followed by number of ints as follows:
                 *
                 *  Package () { REF, INT, REF, INT, INT }
                 *
-                * The index argument is then used to determine which reference
-                * the caller wants (along with the arguments).
+                * Here, REF may be either a local reference or a string. The
+                * index argument is then used to determine which reference the
+                * caller wants (along with the arguments).
                 */
                break;
        default:
@@ -942,7 +1001,26 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
 
                        ret = acpi_get_ref_args(idx == index ? args : NULL,
                                                acpi_fwnode_handle(device),
-                                               &element, end, num_args);
+                                               &element, end, num_args, true);
+                       if (ret < 0)
+                               return ret;
+
+                       if (idx == index)
+                               return 0;
+
+                       break;
+               case ACPI_TYPE_STRING:
+                       ref_fwnode =
+                               acpi_parse_string_ref(fwnode,
+                                                     element->string.pointer);
+                       if (IS_ERR(ref_fwnode))
+                               return PTR_ERR(ref_fwnode);
+
+                       element++;
+
+                       ret = acpi_get_ref_args(idx == index ? args : NULL,
+                                               ref_fwnode, &element, end,
+                                               num_args, false);
                        if (ret < 0)
                                return ret;
 
--
author reviewer
#8 lore raw

Re: [PATCH v8 02/10] ACPI: property: Parse data node string references in properties

Hi Rafael,

Quoted 36 lines · Rafael J. Wysocki
Quoted 8 lines · Sakari Ailus

Add support for parsing property references using strings, besides reference objects that were previously supported. This allows also referencing data nodes which was not possible with reference objects.

Also add pr_fmt() macro to prefix printouts.

While at it, update copyright.

Although I said that it looked good to me, some minor improvements can still be made.

First off, the above changelog is a bit terse.

I think that it would help to provide an example of device properties that would not be parsed properly before the change and can be parsed now.

Quoted 16 lines
---
 drivers/acpi/property.c | 110 ++++++++++++++++++++++++++++++++++------
 1 file changed, 94 insertions(+), 16 deletions(-)
drivers/acpi/property.c +1 -1
index b8d9eb9a433e..08831ffba26c 100644
@@ -2,14 +2,17 @@
 /*
  * ACPI device specific properties support.
  *
- * Copyright (C) 2014, Intel Corporation
+ * Copyright (C) 2014-2023, Intel Corporation
  * All rights reserved.
Quoted 8 lines

drivers/acpi/property.c line 10

Quoted 6 lines
drivers/acpi/property.c +2 -2
  *
  * Authors: Mika Westerberg <[email protected]>
- *          Darren Hart <[email protected]>
- *          Rafael J. Wysocki <[email protected]>
+ *         Darren Hart <[email protected]>
+ *         Rafael J. Wysocki <[email protected]>

I'm not sure if the whitespace change here is really useful.

I did that to address a comment from Andy --- the earlier lines used spaces for indentation.

Quoted 4 lines
Quoted 4 lines
drivers/acpi/property.c +1 -0
+ *         Sakari Ailus <[email protected]>
  */
 
Quoted 16 lines

drivers/acpi/property.c acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode, · line 802

Quoted 12 lines
drivers/acpi/property.c +4 -1
+#define pr_fmt(fmt) "ACPI: " fmt
+
 #include <linux/acpi.h>
 #include <linux/device.h>
 #include <linux/export.h>
@@ -795,7 +798,8 @@ acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode,
 static int acpi_get_ref_args(struct fwnode_reference_args *args,
                             struct fwnode_handle *ref_fwnode,
                             const union acpi_object **element,
-                            const union acpi_object *end, size_t num_args)
+                            const union acpi_object *end, size_t num_args,
+                            bool subnode_string)

The meaning of the new argument isn't really clear. it would be good to somehow help a casual reader of the code to find this out more easily.

I can add comments to v9.

Quoted 73 lines
Quoted 73 lines
drivers/acpi/property.c +43 -8
 {
        u32 nargs = 0, i;
 
@@ -803,13 +807,16 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
         * Find the referred data extension node under the
         * referred device node.
         */
-       for (; *element < end && (*element)->type == ACPI_TYPE_STRING;
-            (*element)++) {
-               const char *child_name = (*element)->string.pointer;
-
-               ref_fwnode = acpi_fwnode_get_named_child_node(ref_fwnode, child_name);
-               if (!ref_fwnode)
-                       return -EINVAL;
+       if (subnode_string) {
+               for (; *element < end && (*element)->type == ACPI_TYPE_STRING;
+                    (*element)++) {
+                       const char *child_name = (*element)->string.pointer;
+
+                       ref_fwnode = acpi_fwnode_get_named_child_node(ref_fwnode,
+                                                                     child_name);
+                       if (!ref_fwnode)
+                               return -EINVAL;
+               }
        }
 
        /*
@@ -820,7 +827,8 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
        for (i = 0; (*element) + i < end && i < num_args; i++) {
                acpi_object_type type = (*element)[i].type;
 
-               if (type == ACPI_TYPE_LOCAL_REFERENCE)
+               if (type == ACPI_TYPE_LOCAL_REFERENCE ||
+                   (!subnode_string && type == ACPI_TYPE_STRING))
                        break;
 
                if (type == ACPI_TYPE_INTEGER)
@@ -844,6 +852,43 @@ static int acpi_get_ref_args(struct fwnode_reference_args *args,
        return 0;
 }
 
+static struct fwnode_handle *
+acpi_parse_string_ref(const struct fwnode_handle *fwnode, const char *refstring)
+{
+       acpi_handle scope, handle;
+       struct acpi_data_node *dn;
+       struct acpi_device *device;
+       acpi_status status;
+
+       if (is_acpi_device_node(fwnode)) {
+               scope = to_acpi_device_node(fwnode)->handle;
+       } else if (is_acpi_data_node(fwnode)) {
+               scope = to_acpi_data_node(fwnode)->handle;
+       } else {
+               pr_debug("bad node type for node %pfw\n", fwnode);
+               return ERR_PTR(-EINVAL);
+       }
+
+       status = acpi_get_handle(scope, refstring, &handle);
+       if (ACPI_FAILURE(status)) {
+               acpi_handle_debug(scope, "can't get handle for %s", refstring);
+               return ERR_PTR(-EINVAL);
+       }
+
+       device = acpi_fetch_acpi_dev(handle);
+       if (device)
+               return acpi_fwnode_handle(device);
+
+       status = acpi_get_data_full(handle, acpi_nondev_subnode_tag,
+                                   (void **)&dn, NULL);
+       if (ACPI_FAILURE(status) || !dn) {
+               acpi_handle_debug(handle, "can't find subnode");
Quoted 8 lines

drivers/acpi/property.c static int acpi_get_ref_args(struct fwnode_reference_args *args, · line 889

Quoted 4 lines
drivers/acpi/property.c +4 -0
+               return ERR_PTR(-EINVAL);
+       }
+
+       return &dn->fwnode;

So on failure this function always returns the same error code. Can it return NULL instead which can be translated into an error code by the caller?

Sure, makes sense.

Quoted 76 lines
Quoted 76 lines
drivers/acpi/property.c +39 -5
+}
+
 /**
  * __acpi_node_get_property_reference - returns handle to the referenced object
  * @fwnode: Firmware node to get the property from
@@ -886,6 +931,7 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
        const union acpi_object *element, *end;
        const union acpi_object *obj;
        const struct acpi_device_data *data;
+       struct fwnode_handle *ref_fwnode;
        struct acpi_device *device;
        int ret, idx = 0;
 
@@ -909,16 +955,29 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
 
                args->fwnode = acpi_fwnode_handle(device);
                args->nargs = 0;
+               return 0;
+       case ACPI_TYPE_STRING:
+               if (index)
+                       return -ENOENT;
+
+               ref_fwnode = acpi_parse_string_ref(fwnode, obj->string.pointer);
+               if (IS_ERR(ref_fwnode))
+                       return PTR_ERR(ref_fwnode);
+
+               args->fwnode = ref_fwnode;
+               args->nargs = 0;
+
                return 0;
        case ACPI_TYPE_PACKAGE:
                /*
                 * If it is not a single reference, then it is a package of
-                * references followed by number of ints as follows:
+                * references, followed by number of ints as follows:
                 *
                 *  Package () { REF, INT, REF, INT, INT }
                 *
-                * The index argument is then used to determine which reference
-                * the caller wants (along with the arguments).
+                * Here, REF may be either a local reference or a string. The
+                * index argument is then used to determine which reference the
+                * caller wants (along with the arguments).
                 */
                break;
        default:
@@ -942,7 +1001,26 @@ int __acpi_node_get_property_reference(const struct fwnode_handle *fwnode,
 
                        ret = acpi_get_ref_args(idx == index ? args : NULL,
                                                acpi_fwnode_handle(device),
-                                               &element, end, num_args);
+                                               &element, end, num_args, true);
+                       if (ret < 0)
+                               return ret;
+
+                       if (idx == index)
+                               return 0;
+
+                       break;
+               case ACPI_TYPE_STRING:
+                       ref_fwnode =
+                               acpi_parse_string_ref(fwnode,
+                                                     element->string.pointer);
+                       if (IS_ERR(ref_fwnode))
+                               return PTR_ERR(ref_fwnode);
+
+                       element++;
+
+                       ret = acpi_get_ref_args(idx == index ? args : NULL,
+                                               ref_fwnode, &element, end,
+                                               num_args, false);
                        if (ret < 0)
                                return ret;
 
--
Signature · 3 lines
Regards,

Sakari Ailus
#9 lore raw

Re: [PATCH v8 02/10] ACPI: property: Parse data node string references in properties

Quoted 73 lines · Sakari Ailus

Hi Rafael,

Quoted 36 lines · Rafael J. Wysocki
Quoted 8 lines · Sakari Ailus

Add support for parsing property references using strings, besides reference objects that were previously supported. This allows also referencing data nodes which was not possible with reference objects.

Also add pr_fmt() macro to prefix printouts.

While at it, update copyright.

Although I said that it looked good to me, some minor improvements can still be made.

First off, the above changelog is a bit terse.

I think that it would help to provide an example of device properties that would not be parsed properly before the change and can be parsed now.

Quoted 16 lines
---
 drivers/acpi/property.c | 110 ++++++++++++++++++++++++++++++++++------
 1 file changed, 94 insertions(+), 16 deletions(-)
drivers/acpi/property.c +1 -1
index b8d9eb9a433e..08831ffba26c 100644
@@ -2,14 +2,17 @@
 /*
  * ACPI device specific properties support.
  *
- * Copyright (C) 2014, Intel Corporation
+ * Copyright (C) 2014-2023, Intel Corporation
  * All rights reserved.
Quoted 8 lines

drivers/acpi/property.c line 10

Quoted 6 lines
drivers/acpi/property.c +2 -2
  *
  * Authors: Mika Westerberg <[email protected]>
- *          Darren Hart <[email protected]>
- *          Rafael J. Wysocki <[email protected]>
+ *         Darren Hart <[email protected]>
+ *         Rafael J. Wysocki <[email protected]>

I'm not sure if the whitespace change here is really useful.

I did that to address a comment from Andy --- the earlier lines used spaces for indentation.

Quoted 4 lines
Quoted 4 lines
drivers/acpi/property.c +1 -0
+ *         Sakari Ailus <[email protected]>
  */
 
Quoted 16 lines

drivers/acpi/property.c acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode, · line 802

Quoted 12 lines
drivers/acpi/property.c +4 -1
+#define pr_fmt(fmt) "ACPI: " fmt
+
 #include <linux/acpi.h>
 #include <linux/device.h>
 #include <linux/export.h>
@@ -795,7 +798,8 @@ acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode,
 static int acpi_get_ref_args(struct fwnode_reference_args *args,
                             struct fwnode_handle *ref_fwnode,
                             const union acpi_object **element,
-                            const union acpi_object *end, size_t num_args)
+                            const union acpi_object *end, size_t num_args,
+                            bool subnode_string)

The meaning of the new argument isn't really clear. it would be good to somehow help a casual reader of the code to find this out more easily.

Quoted 1 line

I can add comments to v9.

If you can send me an example of ASL that will be parsed correctly after this change, but not before, it will help a bit.

author reviewer
#10 lore raw

Re: [PATCH v8 02/10] ACPI: property: Parse data node string references in properties

Hi Rafael,

Quoted 75 lines · Rafael J. Wysocki
Quoted 73 lines · Sakari Ailus

Hi Rafael,

Quoted 36 lines · Rafael J. Wysocki
Quoted 8 lines · Sakari Ailus

Add support for parsing property references using strings, besides reference objects that were previously supported. This allows also referencing data nodes which was not possible with reference objects.

Also add pr_fmt() macro to prefix printouts.

While at it, update copyright.

Although I said that it looked good to me, some minor improvements can still be made.

First off, the above changelog is a bit terse.

I think that it would help to provide an example of device properties that would not be parsed properly before the change and can be parsed now.

Quoted 16 lines
---
 drivers/acpi/property.c | 110 ++++++++++++++++++++++++++++++++++------
 1 file changed, 94 insertions(+), 16 deletions(-)
drivers/acpi/property.c +1 -1
index b8d9eb9a433e..08831ffba26c 100644
@@ -2,14 +2,17 @@
 /*
  * ACPI device specific properties support.
  *
- * Copyright (C) 2014, Intel Corporation
+ * Copyright (C) 2014-2023, Intel Corporation
  * All rights reserved.
Quoted 8 lines

drivers/acpi/property.c line 10

Quoted 6 lines
drivers/acpi/property.c +2 -2
  *
  * Authors: Mika Westerberg <[email protected]>
- *          Darren Hart <[email protected]>
- *          Rafael J. Wysocki <[email protected]>
+ *         Darren Hart <[email protected]>
+ *         Rafael J. Wysocki <[email protected]>

I'm not sure if the whitespace change here is really useful.

I did that to address a comment from Andy --- the earlier lines used spaces for indentation.

Quoted 20 lines
Quoted 10 lines
drivers/acpi/property.c +3 -0
+ *         Sakari Ailus <[email protected]>
  */
 
+#define pr_fmt(fmt) "ACPI: " fmt
+
 #include <linux/acpi.h>
 #include <linux/device.h>
 #include <linux/export.h>

drivers/acpi/property.c acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode, · line 802

Quoted 6 lines
drivers/acpi/property.c +2 -1
@@ -795,7 +798,8 @@ acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode,
 static int acpi_get_ref_args(struct fwnode_reference_args *args,
                             struct fwnode_handle *ref_fwnode,
                             const union acpi_object **element,
-                            const union acpi_object *end, size_t num_args)
+                            const union acpi_object *end, size_t num_args,
+                            bool subnode_string)

The meaning of the new argument isn't really clear. it would be good to somehow help a casual reader of the code to find this out more easily.

Quoted 4 lines
Quoted 1 line

I can add comments to v9.

If you can send me an example of ASL that will be parsed correctly after this change, but not before, it will help a bit.

E.g. this bit from DisCo for Imaging 1.0 (Annex B.1):

	Package () {
	    "mipi-img-flash-leds",
	    Package () {
		"\\_SB.PCI0.I2C2.LEDD.LED0",
		"\\_SB.PCI0.I2C2.LEDD.LED1"
	    },
	},

It's a property with a string reference to an ACPI non-device node, although you can refer to device nodes as well.

You can get the spec from here:
<URL:https://www.mipi.org/mipi-disco-for-imaging-download>.

Signature · 3 lines
Kind regards,

Sakari Ailus
#11 lore raw

Re: [PATCH v8 02/10] ACPI: property: Parse data node string references in properties

Hi Sakari,

Quoted 88 lines · Sakari Ailus
Quoted 72 lines · Rafael J. Wysocki
Quoted 70 lines · Sakari Ailus
Quoted 36 lines · Rafael J. Wysocki
Quoted 8 lines · Sakari Ailus

Add support for parsing property references using strings, besides reference objects that were previously supported. This allows also referencing data nodes which was not possible with reference objects.

Also add pr_fmt() macro to prefix printouts.

While at it, update copyright.

Although I said that it looked good to me, some minor improvements can still be made.

First off, the above changelog is a bit terse.

I think that it would help to provide an example of device properties that would not be parsed properly before the change and can be parsed now.

Quoted 16 lines
---
 drivers/acpi/property.c | 110 ++++++++++++++++++++++++++++++++++------
 1 file changed, 94 insertions(+), 16 deletions(-)
drivers/acpi/property.c +1 -1
index b8d9eb9a433e..08831ffba26c 100644
@@ -2,14 +2,17 @@
 /*
  * ACPI device specific properties support.
  *
- * Copyright (C) 2014, Intel Corporation
+ * Copyright (C) 2014-2023, Intel Corporation
  * All rights reserved.
Quoted 8 lines

drivers/acpi/property.c line 10

Quoted 6 lines
drivers/acpi/property.c +2 -2
  *
  * Authors: Mika Westerberg <[email protected]>
- *          Darren Hart <[email protected]>
- *          Rafael J. Wysocki <[email protected]>
+ *         Darren Hart <[email protected]>
+ *         Rafael J. Wysocki <[email protected]>

I'm not sure if the whitespace change here is really useful.

I did that to address a comment from Andy --- the earlier lines used spaces for indentation.

Quoted 20 lines
Quoted 10 lines
drivers/acpi/property.c +3 -0
+ *         Sakari Ailus <[email protected]>
  */
 
+#define pr_fmt(fmt) "ACPI: " fmt
+
 #include <linux/acpi.h>
 #include <linux/device.h>
 #include <linux/export.h>

drivers/acpi/property.c acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode, · line 802

Quoted 6 lines
drivers/acpi/property.c +2 -1
@@ -795,7 +798,8 @@ acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode,
 static int acpi_get_ref_args(struct fwnode_reference_args *args,
                             struct fwnode_handle *ref_fwnode,
                             const union acpi_object **element,
-                            const union acpi_object *end, size_t num_args)
+                            const union acpi_object *end, size_t num_args,
+                            bool subnode_string)

The meaning of the new argument isn't really clear. it would be good to somehow help a casual reader of the code to find this out more easily.

Quoted 4 lines
Quoted 1 line

I can add comments to v9.

If you can send me an example of ASL that will be parsed correctly after this change, but not before, it will help a bit.

E.g. this bit from DisCo for Imaging 1.0 (Annex B.1):

        Package () {
            "mipi-img-flash-leds",
            Package () {
                "\\_SB.PCI0.I2C2.LEDD.LED0",
                "\\_SB.PCI0.I2C2.LEDD.LED1"
            },
        },
Quoted 2 lines

It's a property with a string reference to an ACPI non-device node, although you can refer to device nodes as well.

This example is missing the definition of LED0 or LED1 from which it would be clear that they are data nodes (or at least one of them is a data node).

Also I'm kind of wondering about the "reference with arguments" part which seems to work differently depending on whether the reference is represented by a string or by a reference object.

Quoted 2 lines

You can get the spec from here:
<URL:https://www.mipi.org/mipi-disco-for-imaging-download>.

Sure, but it alone won't help me much with documenting this code change.

author reviewer
#12 lore raw

Re: [PATCH v8 02/10] ACPI: property: Parse data node string references in properties

Hi Rafael,

Quoted 92 lines · Rafael J. Wysocki

Hi Sakari,

Quoted 88 lines · Sakari Ailus
Quoted 72 lines · Rafael J. Wysocki
Quoted 70 lines · Sakari Ailus
Quoted 36 lines · Rafael J. Wysocki
Quoted 8 lines · Sakari Ailus

Add support for parsing property references using strings, besides reference objects that were previously supported. This allows also referencing data nodes which was not possible with reference objects.

Also add pr_fmt() macro to prefix printouts.

While at it, update copyright.

Although I said that it looked good to me, some minor improvements can still be made.

First off, the above changelog is a bit terse.

I think that it would help to provide an example of device properties that would not be parsed properly before the change and can be parsed now.

Quoted 16 lines
---
 drivers/acpi/property.c | 110 ++++++++++++++++++++++++++++++++++------
 1 file changed, 94 insertions(+), 16 deletions(-)
drivers/acpi/property.c +1 -1
index b8d9eb9a433e..08831ffba26c 100644
@@ -2,14 +2,17 @@
 /*
  * ACPI device specific properties support.
  *
- * Copyright (C) 2014, Intel Corporation
+ * Copyright (C) 2014-2023, Intel Corporation
  * All rights reserved.
Quoted 8 lines

drivers/acpi/property.c line 10

Quoted 6 lines
drivers/acpi/property.c +2 -2
  *
  * Authors: Mika Westerberg <[email protected]>
- *          Darren Hart <[email protected]>
- *          Rafael J. Wysocki <[email protected]>
+ *         Darren Hart <[email protected]>
+ *         Rafael J. Wysocki <[email protected]>

I'm not sure if the whitespace change here is really useful.

I did that to address a comment from Andy --- the earlier lines used spaces for indentation.

Quoted 20 lines
Quoted 10 lines
drivers/acpi/property.c +3 -0
+ *         Sakari Ailus <[email protected]>
  */
 
+#define pr_fmt(fmt) "ACPI: " fmt
+
 #include <linux/acpi.h>
 #include <linux/device.h>
 #include <linux/export.h>

drivers/acpi/property.c acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode, · line 802

Quoted 6 lines
drivers/acpi/property.c +2 -1
@@ -795,7 +798,8 @@ acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode,
 static int acpi_get_ref_args(struct fwnode_reference_args *args,
                             struct fwnode_handle *ref_fwnode,
                             const union acpi_object **element,
-                            const union acpi_object *end, size_t num_args)
+                            const union acpi_object *end, size_t num_args,
+                            bool subnode_string)

The meaning of the new argument isn't really clear. it would be good to somehow help a casual reader of the code to find this out more easily.

Quoted 4 lines
Quoted 1 line

I can add comments to v9.

If you can send me an example of ASL that will be parsed correctly after this change, but not before, it will help a bit.

E.g. this bit from DisCo for Imaging 1.0 (Annex B.1):

        Package () {
            "mipi-img-flash-leds",
            Package () {
                "\\_SB.PCI0.I2C2.LEDD.LED0",
                "\\_SB.PCI0.I2C2.LEDD.LED1"
            },
        },
Quoted 6 lines
Quoted 2 lines

It's a property with a string reference to an ACPI non-device node, although you can refer to device nodes as well.

This example is missing the definition of LED0 or LED1 from which it would be clear that they are data nodes (or at least one of them is a data node).

Ok, perhaps this one could work as a complete example, with a single reference:

	Package ()
        {
	    "mipi-img-flash-leds",  "\\_SB.PCI0.I2C2.LEDD.LED0",
	}

        Device (LEDD)
        {
            Name (_DSD, Package ()  // _DSD: Device-Specific Data
            {
                ToUUID ("dbb8e3e6-5886-4ba6-8795-1319f52a966b"), /* Hierarchical Data Extension */,
                Package ()
                {
                    Package ()
                    {
                        "mipi-img-flash-led-0", 
                        "LED0",
                    }
                },
            })
            Name (LED0, Package ()  // _DSD: Device-Specific Data
            {
                ToUUID ("daffd814-6eba-4d8c-8a91-bc9bbf4aa301") /* Device Properties for _DSD */,
                Package ()
                {
                    Package ()
                    {
                        "mipi-img-max-current", 
                        1000000,
                    }
                }
            })
	}
Quoted 4 lines

Also I'm kind of wondering about the "reference with arguments" part which seems to work differently depending on whether the reference is represented by a string or by a reference object.

Yes. With (device) reference objects, it is possible currently to refer to subnodes with the _DSD data extension child names of those nodes. This is not done with string references as 1) any node can already be referenced so there's no need to and 2) as node references are strings already, it's not possible to distinguish node string references from _DSD data node names. E.g.

	"\\_SB.I2C0.LED0", "LED1"

			   ^ ACPI object name or _DSD data node name?
Quoted 5 lines
Quoted 3 lines

You can get the spec from here:
<URL:https://www.mipi.org/mipi-disco-for-imaging-download>.

Sure, but it alone won't help me much with documenting this code change.

Not as such but the spec is publicly available now.

Signature · 3 lines
Kind regards,

Sakari Ailus
#13 lore raw

Re: [PATCH v8 02/10] ACPI: property: Parse data node string references in properties

Hi Sakari,

Quoted 128 lines · Sakari Ailus

Hi Rafael,

Quoted 90 lines · Rafael J. Wysocki
Quoted 88 lines · Sakari Ailus
Quoted 72 lines · Rafael J. Wysocki
Quoted 70 lines · Sakari Ailus
Quoted 36 lines · Rafael J. Wysocki
Quoted 8 lines · Sakari Ailus

Add support for parsing property references using strings, besides reference objects that were previously supported. This allows also referencing data nodes which was not possible with reference objects.

Also add pr_fmt() macro to prefix printouts.

While at it, update copyright.

Although I said that it looked good to me, some minor improvements can still be made.

First off, the above changelog is a bit terse.

I think that it would help to provide an example of device properties that would not be parsed properly before the change and can be parsed now.

Quoted 16 lines
---
 drivers/acpi/property.c | 110 ++++++++++++++++++++++++++++++++++------
 1 file changed, 94 insertions(+), 16 deletions(-)
drivers/acpi/property.c +1 -1
index b8d9eb9a433e..08831ffba26c 100644
@@ -2,14 +2,17 @@
 /*
  * ACPI device specific properties support.
  *
- * Copyright (C) 2014, Intel Corporation
+ * Copyright (C) 2014-2023, Intel Corporation
  * All rights reserved.
Quoted 8 lines

drivers/acpi/property.c line 10

Quoted 6 lines
drivers/acpi/property.c +2 -2
  *
  * Authors: Mika Westerberg <[email protected]>
- *          Darren Hart <[email protected]>
- *          Rafael J. Wysocki <[email protected]>
+ *         Darren Hart <[email protected]>
+ *         Rafael J. Wysocki <[email protected]>

I'm not sure if the whitespace change here is really useful.

I did that to address a comment from Andy --- the earlier lines used spaces for indentation.

Quoted 20 lines
Quoted 10 lines
drivers/acpi/property.c +3 -0
+ *         Sakari Ailus <[email protected]>
  */
 
+#define pr_fmt(fmt) "ACPI: " fmt
+
 #include <linux/acpi.h>
 #include <linux/device.h>
 #include <linux/export.h>

drivers/acpi/property.c acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode, · line 802

Quoted 6 lines
drivers/acpi/property.c +2 -1
@@ -795,7 +798,8 @@ acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode,
 static int acpi_get_ref_args(struct fwnode_reference_args *args,
                             struct fwnode_handle *ref_fwnode,
                             const union acpi_object **element,
-                            const union acpi_object *end, size_t num_args)
+                            const union acpi_object *end, size_t num_args,
+                            bool subnode_string)

The meaning of the new argument isn't really clear. it would be good to somehow help a casual reader of the code to find this out more easily.

Quoted 4 lines
Quoted 1 line

I can add comments to v9.

If you can send me an example of ASL that will be parsed correctly after this change, but not before, it will help a bit.

E.g. this bit from DisCo for Imaging 1.0 (Annex B.1):

        Package () {
            "mipi-img-flash-leds",
            Package () {
                "\\_SB.PCI0.I2C2.LEDD.LED0",
                "\\_SB.PCI0.I2C2.LEDD.LED1"
            },
        },
Quoted 6 lines
Quoted 2 lines

It's a property with a string reference to an ACPI non-device node, although you can refer to device nodes as well.

This example is missing the definition of LED0 or LED1 from which it would be clear that they are data nodes (or at least one of them is a data node).

Ok, perhaps this one could work as a complete example, with a single reference:

        Package ()
        {
            "mipi-img-flash-leds",  "\\_SB.PCI0.I2C2.LEDD.LED0",
        }

        Device (LEDD)
        {
            Name (_DSD, Package ()  // _DSD: Device-Specific Data
            {
                ToUUID ("dbb8e3e6-5886-4ba6-8795-1319f52a966b"), /* Hierarchical Data Extension */,
                Package ()
                {
                    Package ()
                    {
                        "mipi-img-flash-led-0",
                        "LED0",
                    }
                },
            })
            Name (LED0, Package ()  // _DSD: Device-Specific Data
            {
                ToUUID ("daffd814-6eba-4d8c-8a91-bc9bbf4aa301") /* Device Properties for _DSD */,
                Package ()
                {
Quoted 9 lines
                    Package ()
                    {
                        "mipi-img-max-current",
                        1000000,
                    }
                }
            })
        }

This works, thanks!

Quoted 12 lines
Quoted 4 lines

Also I'm kind of wondering about the "reference with arguments" part which seems to work differently depending on whether the reference is represented by a string or by a reference object.

Yes. With (device) reference objects, it is possible currently to refer to subnodes with the _DSD data extension child names of those nodes. This is not done with string references as 1) any node can already be referenced so there's no need to and 2) as node references are strings already, it's not possible to distinguish node string references from _DSD data node names. E.g.

Quoted 4 lines

"\\_SB.I2C0.LED0", "LED1"

^ ACPI object name or _DSD data node name?

Has this behavior been documented anywhere? Or is there any expectation to see anything like this shipping in production platform firmware?

If any of the above isn't the case, I would be inclined to simply remove this special case and make both the "object reference" and "string" cases work in the same way and if someone needs to refer to a data node, they will just need to use a string (in which case it will be the only option).

author reviewer
#14 lore raw

Re: [PATCH v8 02/10] ACPI: property: Parse data node string references in properties

Hi Rafael,

Quoted 156 lines · Rafael J. Wysocki

Hi Sakari,

Quoted 128 lines · Sakari Ailus

Hi Rafael,

Quoted 90 lines · Rafael J. Wysocki
Quoted 88 lines · Sakari Ailus
Quoted 72 lines · Rafael J. Wysocki
Quoted 70 lines · Sakari Ailus
Quoted 36 lines · Rafael J. Wysocki
Quoted 8 lines · Sakari Ailus

Add support for parsing property references using strings, besides reference objects that were previously supported. This allows also referencing data nodes which was not possible with reference objects.

Also add pr_fmt() macro to prefix printouts.

While at it, update copyright.

Although I said that it looked good to me, some minor improvements can still be made.

First off, the above changelog is a bit terse.

I think that it would help to provide an example of device properties that would not be parsed properly before the change and can be parsed now.

Quoted 16 lines
---
 drivers/acpi/property.c | 110 ++++++++++++++++++++++++++++++++++------
 1 file changed, 94 insertions(+), 16 deletions(-)
drivers/acpi/property.c +1 -1
index b8d9eb9a433e..08831ffba26c 100644
@@ -2,14 +2,17 @@
 /*
  * ACPI device specific properties support.
  *
- * Copyright (C) 2014, Intel Corporation
+ * Copyright (C) 2014-2023, Intel Corporation
  * All rights reserved.
Quoted 8 lines

drivers/acpi/property.c line 10

Quoted 6 lines
drivers/acpi/property.c +2 -2
  *
  * Authors: Mika Westerberg <[email protected]>
- *          Darren Hart <[email protected]>
- *          Rafael J. Wysocki <[email protected]>
+ *         Darren Hart <[email protected]>
+ *         Rafael J. Wysocki <[email protected]>

I'm not sure if the whitespace change here is really useful.

I did that to address a comment from Andy --- the earlier lines used spaces for indentation.

Quoted 20 lines
Quoted 10 lines
drivers/acpi/property.c +3 -0
+ *         Sakari Ailus <[email protected]>
  */
 
+#define pr_fmt(fmt) "ACPI: " fmt
+
 #include <linux/acpi.h>
 #include <linux/device.h>
 #include <linux/export.h>

drivers/acpi/property.c acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode, · line 802

Quoted 6 lines
drivers/acpi/property.c +2 -1
@@ -795,7 +798,8 @@ acpi_fwnode_get_named_child_node(const struct fwnode_handle *fwnode,
 static int acpi_get_ref_args(struct fwnode_reference_args *args,
                             struct fwnode_handle *ref_fwnode,
                             const union acpi_object **element,
-                            const union acpi_object *end, size_t num_args)
+                            const union acpi_object *end, size_t num_args,
+                            bool subnode_string)

The meaning of the new argument isn't really clear. it would be good to somehow help a casual reader of the code to find this out more easily.

Quoted 4 lines
Quoted 1 line

I can add comments to v9.

If you can send me an example of ASL that will be parsed correctly after this change, but not before, it will help a bit.

E.g. this bit from DisCo for Imaging 1.0 (Annex B.1):

        Package () {
            "mipi-img-flash-leds",
            Package () {
                "\\_SB.PCI0.I2C2.LEDD.LED0",
                "\\_SB.PCI0.I2C2.LEDD.LED1"
            },
        },
Quoted 6 lines
Quoted 2 lines

It's a property with a string reference to an ACPI non-device node, although you can refer to device nodes as well.

This example is missing the definition of LED0 or LED1 from which it would be clear that they are data nodes (or at least one of them is a data node).

Ok, perhaps this one could work as a complete example, with a single reference:

        Package ()
        {
            "mipi-img-flash-leds",  "\\_SB.PCI0.I2C2.LEDD.LED0",
        }

        Device (LEDD)
        {
            Name (_DSD, Package ()  // _DSD: Device-Specific Data
            {
                ToUUID ("dbb8e3e6-5886-4ba6-8795-1319f52a966b"), /* Hierarchical Data Extension */,
                Package ()
                {
                    Package ()
                    {
                        "mipi-img-flash-led-0",
                        "LED0",
                    }
                },
            })
            Name (LED0, Package ()  // _DSD: Device-Specific Data
            {
                ToUUID ("daffd814-6eba-4d8c-8a91-bc9bbf4aa301") /* Device Properties for _DSD */,
                Package ()
                {
Quoted 9 lines
                    Package ()
                    {
                        "mipi-img-max-current",
                        1000000,
                    }
                }
            })
        }

This works, thanks!

Quoted 12 lines
Quoted 4 lines

Also I'm kind of wondering about the "reference with arguments" part which seems to work differently depending on whether the reference is represented by a string or by a reference object.

Yes. With (device) reference objects, it is possible currently to refer to subnodes with the _DSD data extension child names of those nodes. This is not done with string references as 1) any node can already be referenced so there's no need to and 2) as node references are strings already, it's not possible to distinguish node string references from _DSD data node names. E.g.

Quoted 8 lines
Quoted 4 lines

"\\_SB.I2C0.LED0", "LED1"

^ ACPI object name or _DSD data node name?

Has this behavior been documented anywhere? Or is there any expectation to see anything like this shipping in production platform firmware?

Good question. Support for this was added by commit 4eb0c3bf5ee52f . AFAIR it was intended to use this in DisCo for Imaging but after review (in a rather liberal sense of the term) it was decided to use string-only references, as in this patch.

I'm not aware of anyone needing this. They've been there for about five years but I'd guess someone would complain if it stops working for them.

Quoted 6 lines

If any of the above isn't the case, I would be inclined to simply remove this special case and make both the "object reference" and "string" cases work in the same way and if someone needs to refer to a data node, they will just need to use a string (in which case it will be the only option).

Works for me.

Signature · 3 lines
Kind regards,

Sakari Ailus
#15 lore raw

Re: [PATCH v8 02/10] ACPI: property: Parse data node string references in properties

Hi Sakari,

Quoted 9 lines · Sakari Ailus

Hi Rafael,

Quoted 5 lines · Rafael J. Wysocki
Quoted 3 lines · Sakari Ailus
Quoted 2 lines · Rafael J. Wysocki

On Mon, May 22, 2023 at 6:28 PM Sakari Ailus
<[email protected]> wrote:

[cut]

Quoted 17 lines
Quoted 17 lines
Quoted 16 lines
Quoted 4 lines

Also I'm kind of wondering about the "reference with arguments" part which seems to work differently depending on whether the reference is represented by a string or by a reference object.

Yes. With (device) reference objects, it is possible currently to refer to subnodes with the _DSD data extension child names of those nodes. This is not done with string references as 1) any node can already be referenced so there's no need to and 2) as node references are strings already, it's not possible to distinguish node string references from _DSD data node names. E.g.

"\\_SB.I2C0.LED0", "LED1"

^ ACPI object name or _DSD data node name?

Quoted 8 lines
Quoted 3 lines

Has this behavior been documented anywhere? Or is there any expectation to see anything like this shipping in production platform firmware?

Good question. Support for this was added by commit 4eb0c3bf5ee52f . AFAIR it was intended to use this in DisCo for Imaging but after review (in a rather liberal sense of the term) it was decided to use string-only references, as in this patch.

It looks like this is sort of a placeholder then.

Quoted 2 lines

I'm not aware of anyone needing this. They've been there for about five years but I'd guess someone would complain if it stops working for them.

It is also quite straightforward to restore if need be.

Quoted 8 lines
Quoted 6 lines

If any of the above isn't the case, I would be inclined to simply remove this special case and make both the "object reference" and "string" cases work in the same way and if someone needs to refer to a data node, they will just need to use a string (in which case it will be the only option).

Works for me.

OK, I'll make this change then.

author reviewer
#16 lore raw

PATCH v8 3/10 ACPI: property: Parse _CRS CSI-2 descriptor

In reply to Sakari Ailus

Parse newly added ACPI _CRS CSI-2 descriptor for CSI-2 and camera configuration, associate it with appropriate devices and allocate memory for software nodes needed to create a DT-like data structure for drivers.

---
 drivers/acpi/Makefile   |   2 +-
 drivers/acpi/internal.h |  27 ++++
 drivers/acpi/mipi.c     | 343 ++++++++++++++++++++++++++++++++++++++++
 drivers/acpi/scan.c     |  50 ++++--
 include/acpi/acpi_bus.h |  11 ++
 5 files changed, 418 insertions(+), 15 deletions(-)
 create mode 100644 drivers/acpi/mipi.c
drivers/acpi/Makefile +1 -1
index a5b649e71ab1..a98fa1bc1554 100644
@@ -37,7 +37,7 @@ acpi-$(CONFIG_ACPI_SLEEP)	+= proc.o
 # ACPI Bus and Device Drivers
 #
 acpi-y				+= bus.o glue.o
-acpi-y				+= scan.o
+acpi-y				+= scan.o mipi.o
 acpi-y				+= resource.o
 acpi-y				+= acpi_processor.o
 acpi-y				+= processor_core.o
drivers/acpi/internal.h +27 -0
index 06ad497067ac..aa5f9c69dbbe 100644
@@ -102,6 +102,24 @@ struct acpi_device_bus_id {
 	struct list_head node;
 };
 
+/* Context for scanning ACPI device nodes for _CRS CSI2 resource descriptors */
+struct acpi_scan_context_csi2 {
+	struct list_head crs_csi2_head;
+	size_t handle_count;
+};
+
+/**
+ * struct acpi_scan_context - Context for scanning ACPI devices
+ * @postponed_head: The list head of the postponed ACPI handles
+ * @device: The first encountered device, typically the root of the scanned tree
+ * @csi2: Context for scanning _CRS CSI2 resource descriptors
+ */
+struct acpi_scan_context {
+	struct list_head postponed_head;
+	struct acpi_device *device;
+	struct acpi_scan_context_csi2 csi2;
+};
+
 void acpi_init_device_object(struct acpi_device *device, acpi_handle handle,
 			     int type, void (*release)(struct device *));
 int acpi_tie_acpi_dev(struct acpi_device *adev);
@@ -282,4 +300,13 @@ void acpi_init_lpit(void);
 static inline void acpi_init_lpit(void) { }
 #endif
 
+/*--------------------------------------------------------------------------
+			ACPI _CRS CSI2 and MIPI DisCo for Imaging conversion
+  -------------------------------------------------------------------------- */
+
+void acpi_crs_csi2_swnodes_del_free(void);
+void acpi_bus_scan_check_crs_csi2(acpi_handle handle, struct acpi_scan_context *ctx);
+void acpi_bus_scan_crs_csi2_release(struct list_head *crs_csi2_handles);
+void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx);
+
 #endif /* _ACPI_INTERNAL_H_ */
drivers/acpi/mipi.c new file +343 -0

343 lines not on this page · raw message

drivers/acpi/scan.c +36 -14

101 lines not on this page · raw message

include/acpi/acpi_bus.h +11 -0
index 57acb895c038..e4e8faac9212 100644
@@ -360,6 +360,17 @@ struct acpi_device_data {
 
 struct acpi_gpio_mapping;
 
+struct acpi_device_software_node_port {
+	unsigned int port_nr;
+};
+
+struct acpi_device_software_nodes {
+	struct acpi_device_software_node_port *ports;
+	struct software_node *nodes;
+	const struct software_node **nodeptrs;
+	unsigned int num_ports;
+};
+
 /* Device */
 struct acpi_device {
 	u32 pld_crc;
Signature · 1 line
2.30.2
#17 lore raw

Re: [PATCH v8 03/10] ACPI: property: Parse _CRS CSI-2 descriptor

Quoted 3 lines · CEST Sakari Ailus

Parse newly added ACPI _CRS CSI-2 descriptor for CSI-2 and camera configuration, associate it with appropriate devices and allocate memory for software nodes needed to create a DT-like data structure for drivers.

It occurred to me, that there would be so many things I would like to change in this patch, so it would be better to create my own version of it, which is appended.

It is based on

https://patchwork.kernel.org/project/linux-acpi/patch/2694293.mvXUDI8C0e@kreacher/

that has just been posted.

IIUC, the idea is to extract the ACPI handle for each resource source in every _CRS CSI-2 resource descriptor and count how many times each handle appears in a CSI-2 context, either because it is referenced from a _CRS CSI-2 resource descriptor (as a "resource source"), or because its device object has CSI-2 resource descriptors in _CRS.

This allows a set of software nodes to be allocated for each of these handles.

If I got that totally wrong, please let me know. Otherwise, I will rework the remaining patches in the series to match this one.

Thanks!
---
From: Rafael J. Wysocki <[email protected]>
Subject: [PATCH] ACPI: scan: Parse _CRS CSI-2 descriptors for all devices

Parse ACPI _CRS CSI-2 resource descriptors defined since ACPI 6.4 (for CSI-2 and camera configuration) for all device objects in the ACPI namespace, find the corresponding "remote" device objects for them and allocate memory for software nodes needed to create a DT-like data structure for drivers (that will be taken care of by subsequent patches).

---
 drivers/acpi/Makefile             |    2 
 drivers/acpi/internal.h           |    7 
 drivers/acpi/mipi-disco-imaging.c |  313 ++++++++++++++++++++++++++++++++++++++
 drivers/acpi/scan.c               |   49 ++++-
 include/acpi/acpi_bus.h           |   11 +
 5 files changed, 372 insertions(+), 10 deletions(-)
 create mode 100644 drivers/acpi/mipi-disco-imaging.c
drivers/acpi/Makefile +1 -1
Index: linux-pm/drivers/acpi/Makefile
===================================================================
@@ -37,7 +37,7 @@ acpi-$(CONFIG_ACPI_SLEEP)	+= proc.o
 # ACPI Bus and Device Drivers
 #
 acpi-y				+= bus.o glue.o
-acpi-y				+= scan.o
+acpi-y				+= scan.o mipi-disco-imaging.o
 acpi-y				+= resource.o
 acpi-y				+= acpi_processor.o
 acpi-y				+= processor_core.o
drivers/acpi/internal.h +7 -0
Index: linux-pm/drivers/acpi/internal.h
===================================================================
@@ -282,4 +282,11 @@ void acpi_init_lpit(void);
 static inline void acpi_init_lpit(void) { }
 #endif
 
+/*--------------------------------------------------------------------------
+		ACPI _CRS CSI-2 and MIPI DisCo for Imaging
+  -------------------------------------------------------------------------- */
+
+void acpi_mipi_check_crs_csi2(acpi_handle handle);
+void acpi_mipi_scan_crs_csi2(void);
+
 #endif /* _ACPI_INTERNAL_H_ */
drivers/acpi/mipi-disco-imaging.c new file +313 -0

313 lines not on this page · raw message

include/acpi/acpi_bus.h +11 -0
Index: linux-pm/include/acpi/acpi_bus.h
===================================================================
@@ -363,6 +363,17 @@ struct acpi_device_data {
 
 struct acpi_gpio_mapping;
 
+struct acpi_device_software_node_port {
+	unsigned int port_nr;
+};
+
+struct acpi_device_software_nodes {
+	struct software_node *nodes;
+	const struct software_node **nodeptrs;
+	struct acpi_device_software_node_port *ports;
+	unsigned int num_ports;
+};
+
 /* Device */
 struct acpi_device {
 	u32 pld_crc;
drivers/acpi/scan.c +40 -9

87 lines not on this page · raw message

author reviewer
#18 lore raw

Re: [PATCH v8 03/10] ACPI: property: Parse _CRS CSI-2 descriptor

Hi Rafael,

Quoted 13 lines · Rafael J. Wysocki
Quoted 3 lines · CEST Sakari Ailus

Parse newly added ACPI _CRS CSI-2 descriptor for CSI-2 and camera configuration, associate it with appropriate devices and allocate memory for software nodes needed to create a DT-like data structure for drivers.

It occurred to me, that there would be so many things I would like to change in this patch, so it would be better to create my own version of it, which is appended.

It is based on

https://patchwork.kernel.org/project/linux-acpi/patch/2694293.mvXUDI8C0e@kreacher/

Quoted 7 lines

that has just been posted.

IIUC, the idea is to extract the ACPI handle for each resource source in every _CRS CSI-2 resource descriptor and count how many times each handle appears in a CSI-2 context, either because it is referenced from a _CRS CSI-2 resource descriptor (as a "resource source"), or because its device object has CSI-2 resource descriptors in _CRS.

Correct.

Quoted 5 lines

This allows a set of software nodes to be allocated for each of these handles.

If I got that totally wrong, please let me know. Otherwise, I will rework the remaining patches in the series to match this one.

It seems about right. I mostly see renames, moving the code around, using the existing dependency list and then parsing sub-tree for _CRS CSI-2 objects right from the bus scan callback.

It seems you've also moved the structs from internal.h to what is now called mipi-disco-imaging.c. They'll be later needed in e.g. scan.c. At least I'd use names that indicate they're related to scanning the bus:
they're not needed after this is done.

I don't have objections to you reworking the rest, but given the number of non-trivial changes, will it work after this? :-) I can also do this, although I would un-do some of the changes in this patch in order to prepare for the rest (such as moving the structs from internal.h).

See e.g. "ACPI: scan: Generate software nodes based on MIPI DisCo for Imaging", I think it's the 6th patch.

...

Quoted 36 lines
drivers/acpi/mipi-disco-imaging.c new file +36 -0
+/**
+ * acpi_mipi_scan_crs_csi2 - Allocate ACPI _CRS CSI-2 software nodes
+ *
+ * Note that this function must be called before any struct acpi_device objects
+ * are bound to any ACPI drivers or scan handlers, so it cannot assume the
+ * existence of struct acpi_device objects for every device present in the ACPI
+ * namespace.
+ *
+ * acpi_scan_lock in scan.c must be held when calling this function.
+ */
+void acpi_mipi_scan_crs_csi2(void)
+{
+	struct crs_csi2 *csi2;
+	LIST_HEAD(aux_list);
+
+	/* Count references to each ACPI handle in the CSI-2 connection graph. */
+	list_for_each_entry(csi2, &acpi_mipi_crs_csi2_list, entry) {
+		struct crs_csi2_connection *conn;
+
+		list_for_each_entry(conn, &csi2->connections, entry) {
+			struct crs_csi2 *remote_csi2;
+			acpi_status status;
+
+			status = acpi_get_data_full(conn->remote_handle,
+						    acpi_mipi_data_tag,
+						    (void **)&remote_csi2,
+						    NULL);
+			if (ACPI_SUCCESS(status) && remote_csi2) {
+				remote_csi2->port_count++;
+				continue;
+			}
+
+			/*
+			 * The "remote" device has no _CRS CSI-2 list entry yet,
+			 * so create one for it and add it to the list.
+			 */

drivers/acpi/mipi-disco-imaging.c line 291

Quoted 10 lines
drivers/acpi/mipi-disco-imaging.c new file +10 -0
+			remote_csi2 = create_crs_csi2_entry(conn->remote_handle);
+			if (!remote_csi2)
+				continue;
+
+			list_add(&remote_csi2->entry, &aux_list);
+		}
+	}
+	list_splice(&aux_list, &acpi_mipi_crs_csi2_list);
+
+	/* Allocate softwware nodes for representing the CSI-2 information. */

"software"

Quoted 5 lines
drivers/acpi/mipi-disco-imaging.c new file +5 -0
+	list_for_each_entry(csi2, &acpi_mipi_crs_csi2_list, entry)
+		alloc_fill_crs_csi2_swnodes(csi2->handle, csi2->port_count);
+
+	acpi_mipi_crs_csi2_release();
+}
Signature · 3 lines
Kind regards,

Sakari Ailus
#19 lore raw

Re: [PATCH v8 03/10] ACPI: property: Parse _CRS CSI-2 descriptor

Hi Sakari,

Quoted 33 lines · Sakari Ailus

Hi Rafael,

Quoted 13 lines · Rafael J. Wysocki
Quoted 3 lines · CEST Sakari Ailus

Parse newly added ACPI _CRS CSI-2 descriptor for CSI-2 and camera configuration, associate it with appropriate devices and allocate memory for software nodes needed to create a DT-like data structure for drivers.

It occurred to me, that there would be so many things I would like to change in this patch, so it would be better to create my own version of it, which is appended.

It is based on

https://patchwork.kernel.org/project/linux-acpi/patch/2694293.mvXUDI8C0e@kreacher/

Quoted 7 lines

that has just been posted.

IIUC, the idea is to extract the ACPI handle for each resource source in every _CRS CSI-2 resource descriptor and count how many times each handle appears in a CSI-2 context, either because it is referenced from a _CRS CSI-2 resource descriptor (as a "resource source"), or because its device object has CSI-2 resource descriptors in _CRS.

Correct.

Quoted 5 lines

This allows a set of software nodes to be allocated for each of these handles.

If I got that totally wrong, please let me know. Otherwise, I will rework the remaining patches in the series to match this one.

Quoted 6 lines

It seems about right. I mostly see renames, moving the code around, using the existing dependency list and then parsing sub-tree for _CRS CSI-2 objects right from the bus scan callback.

It seems you've also moved the structs from internal.h to what is now called mipi-disco-imaging.c.

No, I haven't moved anything in this direction, I've just dropped them.

They can be added in the next patches if needed.

Quoted 6 lines

They'll be later needed in e.g. scan.c. At
least I'd use names that indicate they're related to scanning the bus:
they're not needed after this is done.

I don't have objections to you reworking the rest, but given the number of non-trivial changes, will it work after this? :-)

Probably not right from the start, but after some minor adjustments it should work, unless I've missed something significant.

Quoted 5 lines

I can also do this, although I would un-do some of the changes in this patch in order to prepare for the rest (such as moving the structs from internal.h).

See e.g. "ACPI: scan: Generate software nodes based on MIPI DisCo for Imaging", I think it's the 6th patch.

I will.

Thanks!

author reviewer
#20 lore raw

Re: [PATCH v8 03/10] ACPI: property: Parse _CRS CSI-2 descriptor

Hi Rafael,

Quoted 37 lines · Rafael J. Wysocki

Hi Sakari,

Quoted 33 lines · Sakari Ailus

Hi Rafael,

Quoted 13 lines · Rafael J. Wysocki
Quoted 3 lines · CEST Sakari Ailus

Parse newly added ACPI _CRS CSI-2 descriptor for CSI-2 and camera configuration, associate it with appropriate devices and allocate memory for software nodes needed to create a DT-like data structure for drivers.

It occurred to me, that there would be so many things I would like to change in this patch, so it would be better to create my own version of it, which is appended.

It is based on

https://patchwork.kernel.org/project/linux-acpi/patch/2694293.mvXUDI8C0e@kreacher/

Quoted 7 lines

that has just been posted.

IIUC, the idea is to extract the ACPI handle for each resource source in every _CRS CSI-2 resource descriptor and count how many times each handle appears in a CSI-2 context, either because it is referenced from a _CRS CSI-2 resource descriptor (as a "resource source"), or because its device object has CSI-2 resource descriptors in _CRS.

Correct.

Quoted 5 lines

This allows a set of software nodes to be allocated for each of these handles.

If I got that totally wrong, please let me know. Otherwise, I will rework the remaining patches in the series to match this one.

Quoted 8 lines
Quoted 6 lines

It seems about right. I mostly see renames, moving the code around, using the existing dependency list and then parsing sub-tree for _CRS CSI-2 objects right from the bus scan callback.

It seems you've also moved the structs from internal.h to what is now called mipi-disco-imaging.c.

No, I haven't moved anything in this direction, I've just dropped them.

Ah, I think the struct was added to scan.c by the earlier patch. Indeed, by doing the _CRS CSI2 scanning right from the device scanning callback, it's probably possible to omit these.

Signature · 3 lines
Regards,

Sakari Ailus
author reviewer
#21 lore raw

PATCH v8 4/10 device property: Add SOFTWARE_NODE() macro for defining software nodes

In reply to Sakari Ailus

Add SOFTWARE_NODE() macro in order to make defining software nodes look nicer. This is analogous to different PROPERTY_ENTRY_*() macros for defining properties.

---
 include/linux/property.h | 7 +++++++
 1 file changed, 7 insertions(+)
include/linux/property.h +7 -0
index 0a29db15ff34..b2562f512c49 100644
@@ -477,6 +477,13 @@ struct software_node {
 	const struct property_entry *properties;
 };
 
+#define SOFTWARE_NODE(_name_, _properties_, _parent_)	\
+	(struct software_node) {			\
+		.name = _name_,				\
+		.properties = _properties_,		\
+		.parent = _parent_,			\
+	}
+
 bool is_software_node(const struct fwnode_handle *fwnode);
 const struct software_node *
 to_software_node(const struct fwnode_handle *fwnode);
Signature · 1 line
2.30.2
author reviewer
#22 lore raw

PATCH v8 5/10 ACPI: property: Prepare generating swnodes for ACPI and DisCo for Imaging

In reply to Sakari Ailus

Prepare generating software nodes for information parsed from ACPI _CRS for CSI-2 as well as MIPI DisCo for Imaging spec. The software nodes are compliant with existing ACPI or DT definitions and are parsed by relevant drivers without changes.

---
 drivers/acpi/internal.h |   3 +
 drivers/acpi/mipi.c     | 358 +++++++++++++++++++++++++++++++++++++++-
 drivers/acpi/scan.c     |  21 ++-
 include/acpi/acpi_bus.h |  70 ++++++++
 4 files changed, 449 insertions(+), 3 deletions(-)
drivers/acpi/internal.h +3 -0
index aa5f9c69dbbe..1634a177c75e 100644
@@ -304,9 +304,12 @@ static inline void acpi_init_lpit(void) { }
 			ACPI _CRS CSI2 and MIPI DisCo for Imaging conversion
   -------------------------------------------------------------------------- */
 
+#define MIPI_IMG_PORT_PREFIX "mipi-img-port-"
+
 void acpi_crs_csi2_swnodes_del_free(void);
 void acpi_bus_scan_check_crs_csi2(acpi_handle handle, struct acpi_scan_context *ctx);
 void acpi_bus_scan_crs_csi2_release(struct list_head *crs_csi2_handles);
 void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx);
+void acpi_init_swnodes(struct acpi_device *device);
 
 #endif /* _ACPI_INTERNAL_H_ */
drivers/acpi/mipi.c +357 -1

399 lines not on this page · raw message

drivers/acpi/scan.c +19 -2
index ddf7701a57d0..bf0645854cf1 100644
@@ -449,10 +449,28 @@ static void acpi_free_power_resources_lists(struct acpi_device *device)
 	}
 }
 
+static void acpi_free_swnodes(struct acpi_device *device)
+{
+	struct acpi_device_software_nodes *ads = device->swnodes;
+	struct fwnode_handle *primary;
+
+	if (!ads)
+		return;
+
+	software_node_unregister_node_group(ads->nodeptrs);
+	primary = acpi_fwnode_handle(device);
+	primary->secondary = NULL;
+	kfree(ads->nodes[ACPI_DEVICE_SWNODE_ROOT].name);
+	kfree(ads);
+
+	device->swnodes = NULL;
+}
+
 static void acpi_device_release(struct device *dev)
 {
 	struct acpi_device *acpi_dev = to_acpi_device(dev);
 
+	acpi_free_swnodes(acpi_dev);
 	acpi_free_properties(acpi_dev);
 	acpi_free_pnp_ids(&acpi_dev->pnp);
 	acpi_free_power_resources_lists(acpi_dev);
@@ -2050,8 +2068,7 @@ struct acpi_postponed_handle {
  * Add a given ACPI handle to a list of ACPI objects for which the creation
  * of the device objects is to be postponed.
  */
-static void acpi_bus_handle_postpone(acpi_handle handle,
-				     struct list_head *head)
+static void acpi_bus_handle_postpone(acpi_handle handle, struct list_head *head)
 {
 	struct acpi_postponed_handle *ph;
 
include/acpi/acpi_bus.h +70 -0

91 lines not on this page · raw message

Signature · 1 line
2.30.2
#23 lore raw

Re: [PATCH v8 05/10] ACPI: property: Prepare generating swnodes for ACPI and DisCo for Imaging

Quoted 4 lines · Sakari Ailus

Prepare generating software nodes for information parsed from ACPI _CRS for CSI-2 as well as MIPI DisCo for Imaging spec. The software nodes are compliant with existing ACPI or DT definitions and are parsed by relevant drivers without changes.

There is discussion is ongoing in the previous version of the series for this patch.

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#24 lore raw

Re: [PATCH v8 05/10] ACPI: property: Prepare generating swnodes for ACPI and DisCo for Imaging

In reply to Sakari Ailus
Quoted 170 lines · Sakari Ailus

Prepare generating software nodes for information parsed from ACPI _CRS for CSI-2 as well as MIPI DisCo for Imaging spec. The software nodes are compliant with existing ACPI or DT definitions and are parsed by relevant drivers without changes.

---
 drivers/acpi/internal.h |   3 +
 drivers/acpi/mipi.c     | 358 +++++++++++++++++++++++++++++++++++++++-
 drivers/acpi/scan.c     |  21 ++-
 include/acpi/acpi_bus.h |  70 ++++++++
 4 files changed, 449 insertions(+), 3 deletions(-)
drivers/acpi/internal.h +3 -0
index aa5f9c69dbbe..1634a177c75e 100644
@@ -304,9 +304,12 @@ static inline void acpi_init_lpit(void) { }
                        ACPI _CRS CSI2 and MIPI DisCo for Imaging conversion
   -------------------------------------------------------------------------- */
 
+#define MIPI_IMG_PORT_PREFIX "mipi-img-port-"
+
 void acpi_crs_csi2_swnodes_del_free(void);
 void acpi_bus_scan_check_crs_csi2(acpi_handle handle, struct acpi_scan_context *ctx);
 void acpi_bus_scan_crs_csi2_release(struct list_head *crs_csi2_handles);
 void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx);
+void acpi_init_swnodes(struct acpi_device *device);
 
 #endif /* _ACPI_INTERNAL_H_ */
drivers/acpi/mipi.c +88 -1
index d719c879eb83..03079f78bd2c 100644
@@ -3,7 +3,8 @@
  * MIPI DisCo for Imaging support.
  *
  * Support MIPI DisCo for Imaging by parsing ACPI _CRS CSI2 records and DisCo
- * for Imaging data structures.
+ * for Imaging data structures and generating nodes and properties using
+ * software nodes compliant with DT definitions of the similar scope.
  *
  * Also see <URL:https://www.mipi.org/specifications/mipi-disco-imaging>.
  *
@@ -20,6 +21,8 @@
 #include <linux/sort.h>
 #include <linux/string.h>
 
+#include <media/v4l2-fwnode.h>
+
 #include "internal.h"
 
 /* Temporary ACPI device handle to software node data structure mapping */
@@ -31,6 +34,18 @@ struct crs_csi2_swnodes {
 
 static LIST_HEAD(crs_csi2_swnodes);
 
+/* Obtain pre-allocated software nodes for an ACPI device handle */
+static struct acpi_device_software_nodes *crs_csi2_swnode_get(acpi_handle handle)
+{
+       struct crs_csi2_swnodes *swnodes;
+
+       list_for_each_entry(swnodes, &crs_csi2_swnodes, list)
+               if (swnodes->handle == handle)
+                       return swnodes->ads;
+
+       return NULL;
+}
+
 static void crs_csi2_swnode_del_free(struct crs_csi2_swnodes *swnodes)
 {
        list_del(&swnodes->list);
@@ -166,6 +181,35 @@ struct acpi_handle_ref {
 
 #define NO_CSI2_PORT (UINT_MAX - 1)
 
+/*
+ * Return next free entry in ports array of a software nodes related to an ACPI
+ * device.
+ */
+static unsigned int next_csi2_port_index(struct acpi_device_software_nodes *ads,
+                                        unsigned int port_nr)
+{
+       unsigned int i;
+
+       for (i = 0; i < ads->num_ports; i++) {
+               struct acpi_device_software_node_port *port = &ads->ports[i];
+
+               if (port->port_nr == port_nr)
+                       return i;
+
+               if (port->port_nr == NO_CSI2_PORT) {
+                       port->port_nr = port_nr;
+                       return i;
+               }
+       }
+
+       return NO_CSI2_PORT;
+}
+
+/* Print graph port name into a buffer, return non-zero if failed. */
+#define GRAPH_PORT_NAME(var, num)                                          \
+       (snprintf((var), sizeof(var), SWNODE_GRAPH_PORT_NAME_FMT, (num)) >= \
+        sizeof(var))
+
 static int crs_handle_cmp(const void *__a, const void *__b)
 {
        const struct acpi_handle_ref *a = __a, *b = __b;
@@ -258,6 +302,9 @@ static void acpi_crs_csi2_alloc_fill_swnodes(size_t ports_count, acpi_handle han
                          ports_count);
 }
 
+#define ACPI_CRS_CSI2_PHY_TYPE_C       0
+#define ACPI_CRS_CSI2_PHY_TYPE_D       1
+
 /**
  * acpi_bus_scan_crs_csi2 - Construct software nodes out of ACPI _CRS CSI2
  *                         resource descriptors
@@ -274,6 +321,8 @@ static void acpi_crs_csi2_alloc_fill_swnodes(size_t ports_count, acpi_handle han
  * 3. Allocate memory for swnodes each ACPI device requires later on, and
  *    generate a list of such allocations.
  *
+ * 4. Set up properties for software nodes.
+ *
  * Note that struct acpi_device may not be available yet at this time.
  *
  * acpi_scan_lock in scan.c must be held when calling this function.
@@ -339,5 +388,312 @@ void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx)
                this_count = this->count;
        }
 
+       /*
+        * Allocate and set up necessary software nodes for each device and set
+        * up properties from _CRS CSI2 descriptor.
+        */
+       list_for_each_entry(csi2, &ctx->crs_csi2_head, list) {
+               struct acpi_device_software_nodes *local_swnodes;
+               struct crs_csi2_instance *inst;
+
+               local_swnodes = crs_csi2_swnode_get(csi2->handle);
+               if (WARN_ON_ONCE(!local_swnodes))
+                       continue;
+
+               list_for_each_entry(inst, &csi2->buses, list) {
+                       struct acpi_device_software_nodes *remote_swnodes;
+                       struct acpi_device_software_node_port *local_port;
+                       struct acpi_device_software_node_port *remote_port;
+                       struct software_node *local_node, *remote_node;
+                       unsigned int local_index, remote_index;
+                       unsigned int bus_type;
+
+                       remote_swnodes = crs_csi2_swnode_get(inst->remote_handle);
+                       if (WARN_ON_ONCE(!remote_swnodes))
+                               continue;
+
+                       local_index = next_csi2_port_index(local_swnodes, inst->csi2.local_port_instance);
+                       remote_index = next_csi2_port_index(remote_swnodes, inst->csi2.resource_source.index);
+
+                       if (WARN_ON_ONCE(local_index >= local_swnodes->num_ports) ||
+                           WARN_ON_ONCE(remote_index >= remote_swnodes->num_ports))
+                               goto out_free;
+
+                       switch (inst->csi2.phy_type) {
+                       case ACPI_CRS_CSI2_PHY_TYPE_C:
+                               bus_type = V4L2_FWNODE_BUS_TYPE_CSI2_CPHY;
+                               break;
+                       case ACPI_CRS_CSI2_PHY_TYPE_D:
+                               bus_type = V4L2_FWNODE_BUS_TYPE_CSI2_DPHY;
+                               break;

drivers/acpi/mipi.c void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx) · line 438

Quoted 10 lines
drivers/acpi/mipi.c +10 -0
+                       default:
+                               acpi_handle_info(csi2->handle,
+                                                "ignoring CSI-2 PHY type %u\n",
+                                                inst->csi2.phy_type);
+                               continue;
+                       }
+
+                       local_port = &local_swnodes->ports[local_index];
+                       local_node = &local_swnodes->nodes[ACPI_DEVICE_SWNODE_EP(local_index)];
+                       local_port->remote_ep_ref[0] = SOFTWARE_NODE_REFERENCE(local_node);

This looks odd. Is local_port pointing to its own node as a remote endpont, or am I confused?

drivers/acpi/mipi.c void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx) · line 443

Quoted 5 lines
drivers/acpi/mipi.c +5 -0
+                       local_port->crs_csi2_local = true;
+
+                       remote_port = &remote_swnodes->ports[remote_index];
+                       remote_node = &remote_swnodes->nodes[ACPI_DEVICE_SWNODE_EP(remote_index)];
+                       remote_port->remote_ep_ref[0] = SOFTWARE_NODE_REFERENCE(remote_node);

Analogously here.

author reviewer
#25 lore raw

Re: [PATCH v8 05/10] ACPI: property: Prepare generating swnodes for ACPI and DisCo for Imaging

Hi Rafael,

Thanks for the review.

Quoted 39 lines · Rafael J. Wysocki
Quoted 39 lines
drivers/acpi/mipi.c +39 -0
+       list_for_each_entry(csi2, &ctx->crs_csi2_head, list) {
+               struct acpi_device_software_nodes *local_swnodes;
+               struct crs_csi2_instance *inst;
+
+               local_swnodes = crs_csi2_swnode_get(csi2->handle);
+               if (WARN_ON_ONCE(!local_swnodes))
+                       continue;
+
+               list_for_each_entry(inst, &csi2->buses, list) {
+                       struct acpi_device_software_nodes *remote_swnodes;
+                       struct acpi_device_software_node_port *local_port;
+                       struct acpi_device_software_node_port *remote_port;
+                       struct software_node *local_node, *remote_node;
+                       unsigned int local_index, remote_index;
+                       unsigned int bus_type;
+
+                       remote_swnodes = crs_csi2_swnode_get(inst->remote_handle);
+                       if (WARN_ON_ONCE(!remote_swnodes))
+                               continue;
+
+                       local_index = next_csi2_port_index(local_swnodes, inst->csi2.local_port_instance);
+                       remote_index = next_csi2_port_index(remote_swnodes, inst->csi2.resource_source.index);
+
+                       if (WARN_ON_ONCE(local_index >= local_swnodes->num_ports) ||
+                           WARN_ON_ONCE(remote_index >= remote_swnodes->num_ports))
+                               goto out_free;
+
+                       switch (inst->csi2.phy_type) {
+                       case ACPI_CRS_CSI2_PHY_TYPE_C:
+                               bus_type = V4L2_FWNODE_BUS_TYPE_CSI2_CPHY;
+                               break;
+                       case ACPI_CRS_CSI2_PHY_TYPE_D:
+                               bus_type = V4L2_FWNODE_BUS_TYPE_CSI2_DPHY;
+                               break;
+                       default:
+                               acpi_handle_info(csi2->handle,
+                                                "ignoring CSI-2 PHY type %u\n",
+                                                inst->csi2.phy_type);
+                               continue;
Quoted 8 lines

drivers/acpi/mipi.c void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx) · line 438

Quoted 5 lines
drivers/acpi/mipi.c +5 -0
+                       }
+
+                       local_port = &local_swnodes->ports[local_index];
+                       local_node = &local_swnodes->nodes[ACPI_DEVICE_SWNODE_EP(local_index)];
+                       local_port->remote_ep_ref[0] = SOFTWARE_NODE_REFERENCE(local_node);

This looks odd. Is local_port pointing to its own node as a remote endpont, or am I confused?

This is a reference to a software node that will be, in turn, referenced by the "remote-endpoint" property entry in the remote node. Look for ACPI_DEVICE_SWNODE_EP_REMOTE_EP a few lines below these.

Quoted 8 lines

drivers/acpi/mipi.c void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx) · line 443

Quoted 6 lines
drivers/acpi/mipi.c +5 -0
+                       local_port->crs_csi2_local = true;
+
+                       remote_port = &remote_swnodes->ports[remote_index];
+                       remote_node = &remote_swnodes->nodes[ACPI_DEVICE_SWNODE_EP(remote_index)];
+                       remote_port->remote_ep_ref[0] = SOFTWARE_NODE_REFERENCE(remote_node);

Analogously here.

Signature · 3 lines
Regards,

Sakari Ailus
#26 lore raw

Re: [PATCH v8 05/10] ACPI: property: Prepare generating swnodes for ACPI and DisCo for Imaging

Quoted 54 lines · Sakari Ailus

Hi Rafael,

Thanks for the review.

Quoted 39 lines · Rafael J. Wysocki
Quoted 39 lines
drivers/acpi/mipi.c +39 -0
+       list_for_each_entry(csi2, &ctx->crs_csi2_head, list) {
+               struct acpi_device_software_nodes *local_swnodes;
+               struct crs_csi2_instance *inst;
+
+               local_swnodes = crs_csi2_swnode_get(csi2->handle);
+               if (WARN_ON_ONCE(!local_swnodes))
+                       continue;
+
+               list_for_each_entry(inst, &csi2->buses, list) {
+                       struct acpi_device_software_nodes *remote_swnodes;
+                       struct acpi_device_software_node_port *local_port;
+                       struct acpi_device_software_node_port *remote_port;
+                       struct software_node *local_node, *remote_node;
+                       unsigned int local_index, remote_index;
+                       unsigned int bus_type;
+
+                       remote_swnodes = crs_csi2_swnode_get(inst->remote_handle);
+                       if (WARN_ON_ONCE(!remote_swnodes))
+                               continue;
+
+                       local_index = next_csi2_port_index(local_swnodes, inst->csi2.local_port_instance);
+                       remote_index = next_csi2_port_index(remote_swnodes, inst->csi2.resource_source.index);
+
+                       if (WARN_ON_ONCE(local_index >= local_swnodes->num_ports) ||
+                           WARN_ON_ONCE(remote_index >= remote_swnodes->num_ports))
+                               goto out_free;
+
+                       switch (inst->csi2.phy_type) {
+                       case ACPI_CRS_CSI2_PHY_TYPE_C:
+                               bus_type = V4L2_FWNODE_BUS_TYPE_CSI2_CPHY;
+                               break;
+                       case ACPI_CRS_CSI2_PHY_TYPE_D:
+                               bus_type = V4L2_FWNODE_BUS_TYPE_CSI2_DPHY;
+                               break;
+                       default:
+                               acpi_handle_info(csi2->handle,
+                                                "ignoring CSI-2 PHY type %u\n",
+                                                inst->csi2.phy_type);
+                               continue;
Quoted 8 lines

drivers/acpi/mipi.c void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx) · line 438

Quoted 5 lines
drivers/acpi/mipi.c +5 -0
+                       }
+
+                       local_port = &local_swnodes->ports[local_index];
+                       local_node = &local_swnodes->nodes[ACPI_DEVICE_SWNODE_EP(local_index)];
+                       local_port->remote_ep_ref[0] = SOFTWARE_NODE_REFERENCE(local_node);

This looks odd. Is local_port pointing to its own node as a remote endpont, or am I confused?

Quoted 3 lines

This is a reference to a software node that will be, in turn, referenced by the "remote-endpoint" property entry in the remote node. Look for ACPI_DEVICE_SWNODE_EP_REMOTE_EP a few lines below these.

To be precise, IIUC, it is going to be the "remote-endpoint" value for the remote node.

OK, thanks for the explanation. This isn't exactly straightforward TBH.

#27 lore raw

Re: [PATCH v8 05/10] ACPI: property: Prepare generating swnodes for ACPI and DisCo for Imaging

Quoted 3 lines · Rafael J. Wysocki

On Wed, May 17, 2023 at 1:54 PM Sakari Ailus
<[email protected]> wrote:

[cut]

Quoted 7 lines
Quoted 7 lines
Quoted 6 lines

drivers/acpi/mipi.c void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx) · line 438

Quoted 3 lines
drivers/acpi/mipi.c +3 -0
+                       local_port = &local_swnodes->ports[local_index];
+                       local_node = &local_swnodes->nodes[ACPI_DEVICE_SWNODE_EP(local_index)];
+                       local_port->remote_ep_ref[0] = SOFTWARE_NODE_REFERENCE(local_node);

This looks odd. Is local_port pointing to its own node as a remote endpont, or am I confused?

Quoted 8 lines
Quoted 3 lines

This is a reference to a software node that will be, in turn, referenced by the "remote-endpoint" property entry in the remote node. Look for ACPI_DEVICE_SWNODE_EP_REMOTE_EP a few lines below these.

To be precise, IIUC, it is going to be the "remote-endpoint" value for the remote node.

OK, thanks for the explanation. This isn't exactly straightforward TBH.

So I'd arrange it so that the value of the "remote-endpoint" property in a given node is stored in that node (the value itself being a reference to another endpoint).

author reviewer
#28 lore raw

Re: [PATCH v8 05/10] ACPI: property: Prepare generating swnodes for ACPI and DisCo for Imaging

Quoted 18 lines · Rafael J. Wysocki
Quoted 3 lines · Rafael J. Wysocki

On Wed, May 17, 2023 at 1:54 PM Sakari Ailus
<[email protected]> wrote:

[cut]

Quoted 11 lines
Quoted 10 lines
Quoted 6 lines

drivers/acpi/mipi.c void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx) · line 438

Quoted 3 lines
drivers/acpi/mipi.c +3 -0
+                       local_port = &local_swnodes->ports[local_index];
+                       local_node = &local_swnodes->nodes[ACPI_DEVICE_SWNODE_EP(local_index)];
+                       local_port->remote_ep_ref[0] = SOFTWARE_NODE_REFERENCE(local_node);

This looks odd. Is local_port pointing to its own node as a remote endpont, or am I confused?

This is a reference to a software node that will be, in turn, referenced by the "remote-endpoint" property entry in the remote node. Look for ACPI_DEVICE_SWNODE_EP_REMOTE_EP a few lines below these.

Quoted 8 lines
Quoted 4 lines

To be precise, IIUC, it is going to be the "remote-endpoint" value for the remote node.

OK, thanks for the explanation. This isn't exactly straightforward TBH.

So I'd arrange it so that the value of the "remote-endpoint" property in a given node is stored in that node (the value itself being a reference to another endpoint).

Fine for me.

Signature · 1 line
Sakari Ailus
author reviewer
#29 lore raw

PATCH v8 6/10 ACPI: scan: Generate software nodes based on MIPI DisCo for Imaging

In reply to Sakari Ailus

Generate software nodes for driver use, based on MIPI DisCo for Imaging definitions.

During the (sub-)namespace walk, ACPI device nodes are created but the drivers aren't probed for the devices yet. A convenient way to determine which ACPI devices this applies to is to find a hierarchical data node that begins with "mipi-img-port-". These devices need software nodes that need to be present before probing, and can only be constructed once the related _CRS CSI2 records have been parsed.

---
 drivers/acpi/internal.h | 11 ++++--
 drivers/acpi/power.c    |  2 +-
 drivers/acpi/property.c | 48 +++++++++++++++++--------
 drivers/acpi/scan.c     | 80 +++++++++++++++++++++++++++++++++--------
 4 files changed, 109 insertions(+), 32 deletions(-)
drivers/acpi/internal.h +9 -2
index 1634a177c75e..36799cde281c 100644
@@ -111,17 +111,23 @@ struct acpi_scan_context_csi2 {
 /**
  * struct acpi_scan_context - Context for scanning ACPI devices
  * @postponed_head: The list head of the postponed ACPI handles
+ * @mipi_img_head: The list of ACPI devices with MIPI DisCo for Imaging related
+ *		   properties
  * @device: The first encountered device, typically the root of the scanned tree
  * @csi2: Context for scanning _CRS CSI2 resource descriptors
  */
 struct acpi_scan_context {
 	struct list_head postponed_head;
+	struct list_head mipi_img_head;
 	struct acpi_device *device;
 	struct acpi_scan_context_csi2 csi2;
 };
 
+void acpi_bus_device_postpone(struct acpi_device *device,
+			      struct list_head *head);
 void acpi_init_device_object(struct acpi_device *device, acpi_handle handle,
-			     int type, void (*release)(struct device *));
+			     int type, void (*release)(struct device *),
+			     struct list_head *mipi_img_head);
 int acpi_tie_acpi_dev(struct acpi_device *adev);
 int acpi_device_add(struct acpi_device *device);
 int acpi_device_setup_files(struct acpi_device *dev);
@@ -275,7 +281,8 @@ static inline bool force_storage_d3(void)
   -------------------------------------------------------------------------- */
 #define ACPI_DT_NAMESPACE_HID	"PRP0001"
 
-void acpi_init_properties(struct acpi_device *adev);
+void acpi_init_properties(struct acpi_device *adev,
+			  struct list_head *mipi_img_head);
 void acpi_free_properties(struct acpi_device *adev);
 
 #ifdef CONFIG_X86
drivers/acpi/power.c +1 -1
index 23507d29f000..1096cdbf3d28 100644
@@ -945,7 +945,7 @@ struct acpi_device *acpi_add_power_resource(acpi_handle handle)
 
 	device = &resource->device;
 	acpi_init_device_object(device, handle, ACPI_BUS_TYPE_POWER,
-				acpi_release_power_resource);
+				acpi_release_power_resource, NULL);
 	mutex_init(&resource->resource_lock);
 	INIT_LIST_HEAD(&resource->list_node);
 	INIT_LIST_HEAD(&resource->dependents);
drivers/acpi/property.c +33 -15

104 lines not on this page · raw message

drivers/acpi/scan.c +66 -14

169 lines not on this page · raw message

Signature · 1 line
2.30.2
author reviewer
#30 lore raw

PATCH v8 7/10 ACPI: property: Dig "rotation" property for devices with CSI2 _CRS

In reply to Sakari Ailus

Dig "rotation" property value for devices with _CRS CSI2 resource descriptor. The value comes from _PLD (physical location of device) object, if it exists for the device.

This way camera sensor drivers that know the "rotation" property do not need to care about _PLD on ACPI.

---
 drivers/acpi/mipi.c     | 17 +++++++++++++++++
 include/acpi/acpi_bus.h |  1 +
 2 files changed, 18 insertions(+)
drivers/acpi/mipi.c +17 -0
index 03079f78bd2c..0232cc123ead 100644
@@ -652,10 +652,12 @@ static void init_port_csi2_remote(struct acpi_device *device,
  */
 void acpi_init_swnodes(struct acpi_device *device)
 {
+	struct fwnode_handle *fwnode = acpi_fwnode_handle(device);
 	struct acpi_device_software_nodes *ads;
 	struct acpi_buffer buffer = { .length = ACPI_ALLOCATE_BUFFER };
 	acpi_handle handle = acpi_device_handle(device);
 	struct fwnode_handle *primary;
+	unsigned int prop_index = 0;
 	acpi_status status;
 	unsigned int i;
 	int ret;
@@ -664,6 +666,21 @@ void acpi_init_swnodes(struct acpi_device *device)
 	if (!ads)
 		return;
 
+	/*
+	 * Check if "rotation" property exists and if it doesn't but there's a
+	 * _PLD object, then get the rotation value from there.
+	 */
+	if (!fwnode_property_present(fwnode, "rotation")) {
+		struct acpi_pld_info *pld;
+
+		status = acpi_get_physical_device_location(handle, &pld);
+		if (ACPI_SUCCESS(status)) {
+			ads->dev_props[NEXT_PROPERTY(prop_index, DEV_ROTATION)] =
+				PROPERTY_ENTRY_U32("rotation", pld->rotation * 45U);
+			kfree(pld);
+		}
+	}
+
 	status = acpi_get_name(handle, ACPI_FULL_PATHNAME, &buffer);
 	if (ACPI_FAILURE(status)) {
 		acpi_handle_warn(handle, "cannot get path name\n");
include/acpi/acpi_bus.h +1 -0
index 29d1e4d49d0c..83a89797a122 100644
@@ -361,6 +361,7 @@ struct acpi_device_data {
 struct acpi_gpio_mapping;
 
 enum acpi_device_swnode_dev_props {
+	ACPI_DEVICE_SWNODE_DEV_ROTATION,
 	ACPI_DEVICE_SWNODE_DEV_NUM_OF,
 	ACPI_DEVICE_SWNODE_DEV_NUM_ENTRIES
 };
Signature · 1 line
2.30.2
author reviewer
#31 lore raw

PATCH v8 8/10 ACPI: property: Rename parsed MIPI DisCo for Imaging properties

In reply to Sakari Ailus

MIPI DisCo for Imaging defines properties for sensor-adjacent devices such as EEPROM, LED flash or lens VCM as either device or sub-node references. This is compliant with existing DT definitions apart from property names.

Rename parsed MIPI-defined properties so drivers will have a unified view of them as defined in DT and already parsed by drivers. This can be done in-place as the MIPI-defined property strings are always longer than the DT one. This also results in loss of constness in parser function arguments.

Individual bindings to devices could define the references differently between MIPI DisCo for Imaging and DT, in terms of device or sub-node references. This will still need to be handled in the drivers themselves.

---
 drivers/acpi/internal.h |  1 +
 drivers/acpi/mipi.c     | 36 ++++++++++++++++++++++++++++++++++++
 drivers/acpi/property.c | 22 ++++++++++++----------
 3 files changed, 49 insertions(+), 10 deletions(-)
drivers/acpi/internal.h +1 -0
index 36799cde281c..ef5af9fd9ffa 100644
@@ -318,5 +318,6 @@ void acpi_bus_scan_check_crs_csi2(acpi_handle handle, struct acpi_scan_context *
 void acpi_bus_scan_crs_csi2_release(struct list_head *crs_csi2_handles);
 void acpi_bus_scan_crs_csi2(struct acpi_scan_context_csi2 *ctx);
 void acpi_init_swnodes(struct acpi_device *device);
+void acpi_properties_prepare_mipi(union acpi_object *elements);
 
 #endif /* _ACPI_INTERNAL_H_ */
drivers/acpi/mipi.c +36 -0
index 0232cc123ead..576301b9bb89 100644
@@ -714,3 +714,39 @@ void acpi_init_swnodes(struct acpi_device *device)
 	primary = acpi_fwnode_handle(device);
 	primary->secondary = software_node_fwnode(ads->nodes);
 }
+
+static const struct mipi_disco_prop {
+	const char *mipi_prop;
+	const char *dt_prop;
+} mipi_disco_props[] = {
+	{ "mipi-img-lens-focus",	"lens-focus" },
+	{ "mipi-img-flash-leds",	"flash-leds" },
+	{ "mipi-img-clock-frequency",	"clock-frequency" },
+	{ "mipi-img-led-max-current",	"led-max-microamp" },
+	{ "mipi-img-flash-max-current",	"flash-max-microamp" },
+	{ "mipi-img-flash-max-timeout",	"flash-max-timeout-us" },
+};
+
+/**
+ * acpi_properties_prepare_mipi - Rename MIPI properties as commin DT ones
+ *
+ * @elements: ACPI object containing _DSD properties for a device node
+ *
+ * Renames MIPI-defined properties as common DT ones. The pre-requisite is that
+ * the names of all such MIPI properties are no longer than the corresponding DT
+ * ones.
+ */
+void acpi_properties_prepare_mipi(union acpi_object *elements)
+{
+	char *e0 = elements[0].string.pointer;
+	unsigned int i;
+
+	/* Replace MIPI DisCo for Imaging property names with DT equivalents. */
+	for (i = 0; i < ARRAY_SIZE(mipi_disco_props); i++) {
+		if (!strcmp(mipi_disco_props[i].mipi_prop, e0)) {
+			WARN_ON(strscpy(e0, mipi_disco_props[i].dt_prop,
+					elements[0].string.length) < 0);
+			break;
+		}
+	}
+}
drivers/acpi/property.c +12 -10
index 1892787e73a6..541bda2c118f 100644
@@ -127,7 +127,7 @@ static bool acpi_nondev_subnode_extract(union acpi_object *desc,
 }
 
 static bool acpi_nondev_subnode_data_ok(acpi_handle handle,
-					const union acpi_object *link,
+					union acpi_object *link,
 					struct list_head *list,
 					struct fwnode_handle *parent)
 {
@@ -148,7 +148,7 @@ static bool acpi_nondev_subnode_data_ok(acpi_handle handle,
 }
 
 static bool acpi_nondev_subnode_ok(acpi_handle scope,
-				   const union acpi_object *link,
+				   union acpi_object *link,
 				   struct list_head *list,
 				   struct fwnode_handle *parent)
 {
@@ -292,22 +292,24 @@ static bool acpi_property_value_ok(const union acpi_object *value)
 	return false;
 }
 
-static bool acpi_properties_format_valid(const union acpi_object *properties)
+static bool acpi_properties_prepare(union acpi_object *properties)
 {
-	int i;
+	unsigned int i;
 
 	for (i = 0; i < properties->package.count; i++) {
-		const union acpi_object *property;
+		union acpi_object *property = &properties->package.elements[i];
+		union acpi_object *elements = property->package.elements;
 
-		property = &properties->package.elements[i];
 		/*
 		 * Only two elements allowed, the first one must be a string and
 		 * the second one has to satisfy certain conditions.
 		 */
-		if (property->package.count != 2
-		    || property->package.elements[0].type != ACPI_TYPE_STRING
-		    || !acpi_property_value_ok(&property->package.elements[1]))
+		if (property->package.count != 2 ||
+		    elements[0].type != ACPI_TYPE_STRING ||
+		    !acpi_property_value_ok(&elements[1]))
 			return false;
+
+		acpi_properties_prepare_mipi(elements);
 	}
 	return true;
 }
@@ -539,7 +541,7 @@ static bool acpi_extract_properties(acpi_handle scope, union acpi_object *desc,
 		 * We found the matching GUID. Now validate the format of the
 		 * package immediately following it.
 		 */
-		if (!acpi_properties_format_valid(properties))
+		if (!acpi_properties_prepare(properties))
 			continue;
 
 		acpi_data_add_props(data, (const guid_t *)guid->buffer.pointer,
Signature · 1 line
2.30.2
#32 lore raw

Re: [PATCH v8 08/10] ACPI: property: Rename parsed MIPI DisCo for Imaging properties

Quoted 7 lines · Sakari Ailus

MIPI DisCo for Imaging defines properties for sensor-adjacent devices such as EEPROM, LED flash or lens VCM as either device or sub-node references. This is compliant with existing DT definitions apart from property names.

Rename parsed MIPI-defined properties so drivers will have a unified view of them as defined in DT and already parsed by drivers.

I don't particularly like this idea.

One of the drawbacks is that if somebody doesn't care about DT bindings (for instance, because they will always run on platforms without DT), they won't be able to use the MIPI-defined property names in their code.

I would very much prefer to add a set of DT-defined properties with the same values. The, whoever wants to use the property names from the DT bindings, they will be able to do that, but it will be also possible to use the MIPI-defined ones.

The previous patch adds the "rotation" property to the swnodes set, so I don't see any problems with doing that for the properties in question.

author reviewer
#33 lore raw

Re: [PATCH v8 08/10] ACPI: property: Rename parsed MIPI DisCo for Imaging properties

Hi Rafael,

Quoted 17 lines · Rafael J. Wysocki
Quoted 7 lines · Sakari Ailus

MIPI DisCo for Imaging defines properties for sensor-adjacent devices such as EEPROM, LED flash or lens VCM as either device or sub-node references. This is compliant with existing DT definitions apart from property names.

Rename parsed MIPI-defined properties so drivers will have a unified view of them as defined in DT and already parsed by drivers.

I don't particularly like this idea.

One of the drawbacks is that if somebody doesn't care about DT bindings (for instance, because they will always run on platforms without DT), they won't be able to use the MIPI-defined property names in their code.

Quoted 8 lines

I would very much prefer to add a set of DT-defined properties with the same values. The, whoever wants to use the property names from the DT bindings, they will be able to do that, but it will be also possible to use the MIPI-defined ones.

The previous patch adds the "rotation" property to the swnodes set, so I don't see any problems with doing that for the properties in question.

I don't think this would be a problem really, no, but I question the need to ever access the MIPI specification names in Linux outside this piece of code. Drivers for cameras, lens controllers and LED flashes generally try to avoid being specific to a given firmware interface and the established de facto naming of these properties in the kernel is aligned with Devicetree.

I'd like to see differences only when the functionality differs, otherwise they should be the same. Creating a copy when you can modify it is waste of a bit of memory. On the upside, the object memory could remain const that way.

Signature · 3 lines
Kind regards,

Sakari Ailus
author reviewer
#34 lore raw

PATCH v8 9/10 ACPI: property: Skip MIPI property table without "mipi-img" prefix

In reply to Sakari Ailus

For all _DSD properties, skip going through the MIPI DisCo for Imaging property name substitution table if the property doesn't have "mipi-img-" prefix.

---
 drivers/acpi/mipi.c | 5 +++++
 1 file changed, 5 insertions(+)
drivers/acpi/mipi.c +5 -0
index 576301b9bb89..90ad4b43f008 100644
@@ -715,6 +715,8 @@ void acpi_init_swnodes(struct acpi_device *device)
 	primary->secondary = software_node_fwnode(ads->nodes);
 }
 
+#define MIPI_IMG_PREFIX "mipi-img-"
+
 static const struct mipi_disco_prop {
 	const char *mipi_prop;
 	const char *dt_prop;
@@ -741,6 +743,9 @@ void acpi_properties_prepare_mipi(union acpi_object *elements)
 	char *e0 = elements[0].string.pointer;
 	unsigned int i;
 
+	if (!str_has_prefix(elements[0].string.pointer, MIPI_IMG_PREFIX))
+		return;
+
 	/* Replace MIPI DisCo for Imaging property names with DT equivalents. */
 	for (i = 0; i < ARRAY_SIZE(mipi_disco_props); i++) {
 		if (!strcmp(mipi_disco_props[i].mipi_prop, e0)) {
Signature · 1 line
2.30.2
author reviewer
#35 lore raw

PATCH v8 10/10 ACPI: property: Document _CRS CSI-2 and DisCo for Imaging support

In reply to Sakari Ailus

Document how ACPI _CRS CSI-2 and DisCo for Imaging works. It's non-trivial so such documentation can be useful.

---
 drivers/acpi/mipi.c | 37 +++++++++++++++++++++++++++++++++++++
 1 file changed, 37 insertions(+)
drivers/acpi/mipi.c +37 -0
index 90ad4b43f008..926d39948268 100644
@@ -9,6 +9,43 @@
  * Also see <URL:https://www.mipi.org/specifications/mipi-disco-imaging>.
  *
  * Copyright (C) 2023 Intel Corporation
+ *
+ * _CRS CSI-2 descriptors, as defined starting from ACPI 6.4 [1], contain
+ * information on cross-device CSI-2 bus configuration. The descriptors are
+ * located under transmitter devices, and the receiver devices have no direct
+ * way to access them even if the information in these descriptors is equally
+ * important for receivers. This information is amended with MIPI DisCo for
+ * Imaging [2] specification that defines _DSD data nodes and properties.
+ *
+ * The support for these is based on two-fold approach, firstly renaming
+ * properties where semantics matches and secondly gathering information to
+ * generate properties using information gathered from various sources. The
+ * former is trivial (see acpi_properties_prepare_mipi() at the end of the
+ * file) whereas the latter requires a more elaborate explanation.
+ *
+ * acpi_bus_scan_crs_csi2() scans an ACPI bus for devices with _CRS CSI-2
+ * descriptors and stores them to a linked list. This is done as traversing just
+ * this list is much smaller task than the entire DSDT. This list is then used
+ * to figure out how much memory is needed for swnodes related to a given ACPI
+ * device (handle). Further on, the same function sets the property values for
+ * the properties the values of which are obtained from the _CRS CSI-2
+ * descriptor. The information is stored into another list where the information
+ * can be looked up based on device's acpi_handle as the struct acpi_device
+ * isn't available yet at this point (and could not, as cross-device references
+ * need to be set up before the devices are available for drivers to probe).
+ *
+ * For each struct acpi_device, acpi_init_swnodes() further obtains information
+ * required to find out the values for the rest of the properties needed by
+ * drivers. This includes all port and endpoint properties as the node
+ * structures used by DT graphs and DisCo for Imaging are different. Finally the
+ * function registers software nodes for the device and sets the secondary
+ * pointer for the ACPI device's fwnode.
+ *
+ * Access to data the structures is serialised using acpi_scan_lock in scan.c.
+ *
+ * [1] https://uefi.org/sites/default/files/resources/ACPI_Spec_6_4_Jan22.pdf
+ *
+ * [2] https://www.mipi.org/specifications/mipi-disco-imaging
  */
 
 #include <linux/acpi.h>
Signature · 1 line
2.30.2