[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <8ccfd660-791a-6cf1-2982-b4f09ae6d3e1@linux.intel.com>
Date: Mon, 30 Jun 2025 15:06:38 +0300 (EEST)
From: Ilpo Järvinen <ilpo.jarvinen@...ux.intel.com>
To: "David E. Box" <david.e.box@...ux.intel.com>
cc: LKML <linux-kernel@...r.kernel.org>, platform-driver-x86@...r.kernel.org,
srinivas.pandruvada@...ux.intel.com,
Andy Shevchenko <andriy.shevchenko@...ux.intel.com>, tony.luck@...el.com,
xi.pardee@...ux.intel.com, Hans de Goede <hdegoede@...hat.com>
Subject: Re: [PATCH V2 04/15] platform/x86/intel/vsec: Add device links to
enforce dependencies
On Mon, 16 Jun 2025, David E. Box wrote:
> New Intel VSEC features will have dependencies on other features, requiring
> certain supplier drivers to be probed before their consumers. To enforce
> this dependency ordering, introduce device links using device_link_add(),
> ensuring that suppliers are fully registered before consumers are probed.
>
> - Add device link tracking by storing supplier devices and tracking their
> state.
> - Implement intel_vsec_link_devices() to establish links between suppliers
> and consumers based on feature dependencies.
> - Add get_consumer_dependencies() to retrieve supplier-consumer
> relationships.
> - Modify feature registration logic:
> * Consumers now check that all required suppliers are registered before
> being initialized.
> * suppliers_ready() verifies that all required supplier devices are
> available.
> - Prevent potential null consumer name issue in sysfs:
> - Use dev_set_name() when creating auxiliary devices to ensure a
> unique, non-null consumer name.
> - Update intel_vsec_pci_probe() to loop up to the number of possible
> features or when all devices are registered, whichever comes first.
> - Introduce VSEC_CAP_UNUSED to prevent sub-features (registered via
> exported APIs) from being mistakenly linked.
>
> Signed-off-by: David E. Box <david.e.box@...ux.intel.com>
> ---
>
> Changes in v2:
> - Simply dependency search in get_consumer_dependencies() per comments
> from Ilpo.
> - Add rollback for auxiliary_device_uninit() in intel_vsec_add_aux().
> - In suppliers_ready() clarify that for_each_set_bit() is searching for
> all *ready* suppliers, not all suppliers. If any is not ready and not
> ignored, it immediately returns.
> - In suppliers_ready() check device_link_add() return status.
> - In intel_vsec_probe() uses info->caps consistently.
> - Fix spelling errors and remove unrelated changes.
>
> drivers/platform/x86/intel/vsec.c | 219 ++++++++++++++++++++++++++++--
> include/linux/intel_vsec.h | 28 +++-
> 2 files changed, 232 insertions(+), 15 deletions(-)
>
> diff --git a/drivers/platform/x86/intel/vsec.c b/drivers/platform/x86/intel/vsec.c
> index f01651f498ca..e497b570f814 100644
> --- a/drivers/platform/x86/intel/vsec.c
> +++ b/drivers/platform/x86/intel/vsec.c
> @@ -15,9 +15,11 @@
>
> #include <linux/auxiliary_bus.h>
> #include <linux/bits.h>
> +#include <linux/bitops.h>
> #include <linux/cleanup.h>
> #include <linux/delay.h>
> #include <linux/idr.h>
> +#include <linux/log2.h>
> #include <linux/intel_vsec.h>
> #include <linux/kernel.h>
> #include <linux/module.h>
> @@ -32,8 +34,17 @@ static DEFINE_IDA(intel_vsec_ida);
> static DEFINE_IDA(intel_vsec_sdsi_ida);
> static DEFINE_XARRAY_ALLOC(auxdev_array);
>
> +enum vsec_device_state {
> + STATE_NOT_FOUND,
> + STATE_REGISTERED,
> + STATE_SKIP,
> +};
> +
> struct vsec_priv {
> struct intel_vsec_platform_info *info;
> + struct device *suppliers[VSEC_FEATURE_COUNT];
> + enum vsec_device_state state[VSEC_FEATURE_COUNT];
> + unsigned long found_caps;
> };
>
> static const char *intel_vsec_name(enum intel_vsec_id id)
> @@ -95,6 +106,74 @@ static void intel_vsec_dev_release(struct device *dev)
> kfree(intel_vsec_dev);
> }
>
> +static const struct vsec_feature_dependency *
> +get_consumer_dependencies(struct vsec_priv *priv, int cap_id)
> +{
> + const struct vsec_feature_dependency *deps = priv->info->deps;
> + int consumer_id = priv->info->num_deps;
> +
> + if (!deps)
> + return NULL;
> +
> + while (consumer_id--)
> + if (deps[consumer_id].feature == BIT(cap_id))
> + return &deps[consumer_id];
> +
> + return NULL;
> +}
> +
> +/*
> + * Although pci_device_id table is available in the pdev, this prototype is
> + * necessary because the code using it can be called by an exported API that
> + * might pass a different pdev.
> + */
> +static const struct pci_device_id intel_vsec_pci_ids[];
> +
> +static int intel_vsec_link_devices(struct pci_dev *pdev, struct device *dev,
> + int consumer_id)
> +{
> + const struct vsec_feature_dependency *deps;
> + enum vsec_device_state *state;
> + struct device **suppliers;
> + struct vsec_priv *priv;
> + int supplier_id;
> +
> + if (!consumer_id)
> + return 0;
> +
> + if (!pci_match_id(intel_vsec_pci_ids, pdev))
> + return 0;
> +
> + priv = pci_get_drvdata(pdev);
> + state = priv->state;
> + suppliers = priv->suppliers;
> +
> + priv->suppliers[consumer_id] = dev;
> +
> + deps = get_consumer_dependencies(priv, consumer_id);
> + if (!deps)
> + return 0;
> +
> + for_each_set_bit(supplier_id, &deps->supplier_bitmap, VSEC_FEATURE_COUNT) {
> + struct device_link *link;
> +
> + if (state[supplier_id] != STATE_REGISTERED)
> + continue;
> +
> + if (!suppliers[supplier_id]) {
> + dev_err(dev, "Bad supplier list\n");
> + return -EINVAL;
> + }
> +
> + link = device_link_add(dev, suppliers[supplier_id],
> + DL_FLAG_AUTOPROBE_CONSUMER);
> + if (!link)
> + return -EINVAL;
> + }
> +
> + return 0;
> +}
> +
> int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
> struct intel_vsec_device *intel_vsec_dev,
> const char *name)
> @@ -132,19 +211,37 @@ int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
> return ret;
> }
>
> + /*
> + * Assign a name now to ensure that the device link doesn't contain
> + * a null string for the consumer name. This is a problem when a supplier
> + * supplies more than one consumer and can lead to a duplicate name error
> + * when the link is created in sysfs.
> + */
> + ret = dev_set_name(&auxdev->dev, "%s.%s.%d", KBUILD_MODNAME, auxdev->name,
> + auxdev->id);
> + if (ret)
> + goto cleanup_aux;
> +
> + ret = intel_vsec_link_devices(pdev, &auxdev->dev, intel_vsec_dev->cap_id);
> + if (ret)
> + goto cleanup_aux;
> +
> ret = auxiliary_device_add(auxdev);
> - if (ret < 0) {
> - auxiliary_device_uninit(auxdev);
> - return ret;
> - }
> + if (ret)
> + goto cleanup_aux;
>
> return devm_add_action_or_reset(parent, intel_vsec_remove_aux,
> auxdev);
> +
> +cleanup_aux:
> + auxiliary_device_uninit(auxdev);
> + return ret;
> }
> EXPORT_SYMBOL_NS_GPL(intel_vsec_add_aux, "INTEL_VSEC");
>
> static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *header,
> - struct intel_vsec_platform_info *info)
> + struct intel_vsec_platform_info *info,
> + unsigned long cap_id)
> {
> struct intel_vsec_device __free(kfree) *intel_vsec_dev = NULL;
> struct resource __free(kfree) *res = NULL;
> @@ -211,6 +308,7 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
> intel_vsec_dev->quirks = info->quirks;
> intel_vsec_dev->base_addr = info->base_addr;
> intel_vsec_dev->priv_data = info->priv_data;
> + intel_vsec_dev->cap_id = cap_id;
>
> if (header->id == VSEC_ID_SDSI)
> intel_vsec_dev->ida = &intel_vsec_sdsi_ida;
> @@ -225,6 +323,101 @@ static int intel_vsec_add_dev(struct pci_dev *pdev, struct intel_vsec_header *he
> intel_vsec_name(header->id));
> }
>
> +static bool suppliers_ready(struct vsec_priv *priv,
> + const struct vsec_feature_dependency *consumer_deps,
> + int cap_id)
> +{
> + enum vsec_device_state *state = priv->state;
> + int supplier_id;
> +
> + if (consumer_deps->feature != BIT(cap_id))
> + return false; /* Should not happen */
I don't recall if I asked this already, but does the comment imply this is
an error on the caller side warranting using WARN_ON_ONCE()?
--
i.
> +
> + /*
> + * Verify that all required suppliers have been found. Return false
> + * immediately if any are still missing.
> + */
> + for_each_set_bit(supplier_id, &consumer_deps->supplier_bitmap, VSEC_FEATURE_COUNT) {
> + if (state[supplier_id] == STATE_SKIP)
> + continue;
> +
> + if (state[supplier_id] == STATE_NOT_FOUND)
> + return false;
> + }
> +
> + /*
> + * All suppliers have been found and the consumer is ready to be
> + * registered.
> + */
> + return true;
> +}
> +
> +static int get_cap_id(u32 header_id, unsigned long *cap_id)
> +{
> + switch (header_id) {
> + case VSEC_ID_TELEMETRY:
> + *cap_id = ilog2(VSEC_CAP_TELEMETRY);
> + break;
> + case VSEC_ID_WATCHER:
> + *cap_id = ilog2(VSEC_CAP_WATCHER);
> + break;
> + case VSEC_ID_CRASHLOG:
> + *cap_id = ilog2(VSEC_CAP_CRASHLOG);
> + break;
> + case VSEC_ID_SDSI:
> + *cap_id = ilog2(VSEC_CAP_SDSI);
> + break;
> + case VSEC_ID_TPMI:
> + *cap_id = ilog2(VSEC_CAP_TPMI);
> + break;
> + default:
> + return -EINVAL;
> + }
> +
> + return 0;
> +}
> +
> +static int intel_vsec_register_device(struct pci_dev *pdev,
> + struct intel_vsec_header *header,
> + struct intel_vsec_platform_info *info)
> +{
> + const struct vsec_feature_dependency *consumer_deps;
> + struct vsec_priv *priv;
> + unsigned long cap_id;
> + int ret;
> +
> + ret = get_cap_id(header->id, &cap_id);
> + if (ret)
> + return ret;
> +
> + /*
> + * Only track dependencies for devices probed by the VSEC driver.
> + * For others using the exported APIs, add the device directly.
> + */
> + if (!pci_match_id(intel_vsec_pci_ids, pdev))
> + return intel_vsec_add_dev(pdev, header, info, cap_id);
> +
> + priv = pci_get_drvdata(pdev);
> + if (priv->state[cap_id] == STATE_REGISTERED ||
> + priv->state[cap_id] == STATE_SKIP)
> + return -EEXIST;
> +
> + priv->found_caps |= BIT(cap_id);
> +
> + consumer_deps = get_consumer_dependencies(priv, cap_id);
> + if (!consumer_deps || suppliers_ready(priv, consumer_deps, cap_id)) {
> + ret = intel_vsec_add_dev(pdev, header, info, cap_id);
> + if (ret)
> + priv->state[cap_id] = STATE_SKIP;
> + else
> + priv->state[cap_id] = STATE_REGISTERED;
> +
> + return ret;
> + }
> +
> + return -EAGAIN;
> +}
> +
> static bool intel_vsec_walk_header(struct pci_dev *pdev,
> struct intel_vsec_platform_info *info)
> {
> @@ -233,7 +426,7 @@ static bool intel_vsec_walk_header(struct pci_dev *pdev,
> int ret;
>
> for ( ; *header; header++) {
> - ret = intel_vsec_add_dev(pdev, *header, info);
> + ret = intel_vsec_register_device(pdev, *header, info);
> if (!ret)
> have_devices = true;
> }
> @@ -281,7 +474,7 @@ static bool intel_vsec_walk_dvsec(struct pci_dev *pdev,
> pci_read_config_dword(pdev, pos + PCI_DVSEC_HEADER2, &hdr);
> header.id = PCI_DVSEC_HEADER2_ID(hdr);
>
> - ret = intel_vsec_add_dev(pdev, &header, info);
> + ret = intel_vsec_register_device(pdev, &header, info);
> if (ret)
> continue;
>
> @@ -326,7 +519,7 @@ static bool intel_vsec_walk_vsec(struct pci_dev *pdev,
> header.tbir = INTEL_DVSEC_TABLE_BAR(table);
> header.offset = INTEL_DVSEC_TABLE_OFFSET(table);
>
> - ret = intel_vsec_add_dev(pdev, &header, info);
> + ret = intel_vsec_register_device(pdev, &header, info);
> if (ret)
> continue;
>
> @@ -375,7 +568,7 @@ static int intel_vsec_pci_probe(struct pci_dev *pdev, const struct pci_device_id
> struct intel_vsec_platform_info *info;
> struct vsec_priv *priv;
> bool have_devices = false;
> - int ret;
> + int num_caps, ret;
>
> ret = pcim_enable_device(pdev);
> if (ret)
> @@ -393,7 +586,13 @@ static int intel_vsec_pci_probe(struct pci_dev *pdev, const struct pci_device_id
> priv->info = info;
> pci_set_drvdata(pdev, priv);
>
> - intel_vsec_feature_walk(pdev, &have_devices, info);
> + num_caps = hweight_long(info->caps);
> + while (num_caps--) {
> + intel_vsec_feature_walk(pdev, &have_devices, info);
> +
> + if (priv->found_caps == info->caps)
> + break;
> + }
>
> if (!have_devices)
> return -ENODEV;
> diff --git a/include/linux/intel_vsec.h b/include/linux/intel_vsec.h
> index bc95821f1bfb..71067afaca99 100644
> --- a/include/linux/intel_vsec.h
> +++ b/include/linux/intel_vsec.h
> @@ -5,11 +5,18 @@
> #include <linux/auxiliary_bus.h>
> #include <linux/bits.h>
>
> -#define VSEC_CAP_TELEMETRY BIT(0)
> -#define VSEC_CAP_WATCHER BIT(1)
> -#define VSEC_CAP_CRASHLOG BIT(2)
> -#define VSEC_CAP_SDSI BIT(3)
> -#define VSEC_CAP_TPMI BIT(4)
> +/*
> + * VSEC_CAP_UNUSED is reserved. It exists to prevent zero initialized
> + * intel_vsec devices from being automatically set to a known
> + * capability with ID 0
> + */
> +#define VSEC_CAP_UNUSED BIT(0)
> +#define VSEC_CAP_TELEMETRY BIT(1)
> +#define VSEC_CAP_WATCHER BIT(2)
> +#define VSEC_CAP_CRASHLOG BIT(3)
> +#define VSEC_CAP_SDSI BIT(4)
> +#define VSEC_CAP_TPMI BIT(5)
> +#define VSEC_FEATURE_COUNT 6
>
> /* Intel DVSEC offsets */
> #define INTEL_DVSEC_ENTRIES 0xA
> @@ -81,22 +88,31 @@ struct pmt_callbacks {
> int (*read_telem)(struct pci_dev *pdev, u32 guid, u64 *data, loff_t off, u32 count);
> };
>
> +struct vsec_feature_dependency {
> + unsigned long feature;
> + unsigned long supplier_bitmap;
> +};
> +
> /**
> * struct intel_vsec_platform_info - Platform specific data
> * @parent: parent device in the auxbus chain
> * @headers: list of headers to define the PMT client devices to create
> + * @deps: array of feature dependencies
> * @priv_data: private data, usable by parent devices, currently a callback
> * @caps: bitmask of PMT capabilities for the given headers
> * @quirks: bitmask of VSEC device quirks
> * @base_addr: allow a base address to be specified (rather than derived)
> + * @num_deps: Count feature dependencies
> */
> struct intel_vsec_platform_info {
> struct device *parent;
> struct intel_vsec_header **headers;
> + const struct vsec_feature_dependency *deps;
> void *priv_data;
> unsigned long caps;
> unsigned long quirks;
> u64 base_addr;
> + int num_deps;
> };
>
> /**
> @@ -110,6 +126,7 @@ struct intel_vsec_platform_info {
> * @priv_data: any private data needed
> * @quirks: specified quirks
> * @base_addr: base address of entries (if specified)
> + * @cap_id: the enumerated id of the vsec feature
> */
> struct intel_vsec_device {
> struct auxiliary_device auxdev;
> @@ -122,6 +139,7 @@ struct intel_vsec_device {
> size_t priv_data_size;
> unsigned long quirks;
> u64 base_addr;
> + unsigned long cap_id;
> };
>
> int intel_vsec_add_aux(struct pci_dev *pdev, struct device *parent,
>
Powered by blists - more mailing lists