[PATCH RFC 2/3] PCI: Add per-device debugfs directories

[Wei Huang <wei.huang2@amd.com>]

PCI features that export debug information have no common place to put it.
The only existing user, PTM, creates a "pcie_ptm_<device>" directory in the
debugfs root, which does not scale as more features follow and leaves each
user to invent its own lifetime rules and leaf names.

Add /sys/kernel/debug/pci/<device>/ directories owned by the PCI core.
pci_dev_debugfs_dir() returns the directory of a device, creating it and
the PCI root directory on first use, so devices without debug content do
not get an empty directory. pci_destroy_dev() removes the directory
recursively, so a feature that pins data in a debugfs file needs no removal
hook of its own: debugfs_remove_recursive() returns only once no operation
on those files is still running.

No functional change for existing users; the next patch is the first
consumer.

Signed-off-by: Wei Huang <wei.huang2@amd.com>
---
 drivers/pci/Makefile  |  1 +
 drivers/pci/debugfs.c | 60 +++++++++++++++++++++++++++++++++++++++++++
 drivers/pci/pci.h     | 12 +++++++++
 drivers/pci/remove.c  |  1 +
 include/linux/pci.h   |  3 +++
 5 files changed, 77 insertions(+)
 create mode 100644 drivers/pci/debugfs.c

diff --git a/drivers/pci/Makefile b/drivers/pci/Makefile
index 41ebc3b9a518..dbc841c3e990 100644
--- a/drivers/pci/Makefile
+++ b/drivers/pci/Makefile
@@ -15,6 +15,7 @@ ifdef CONFIG_PCI
 obj-$(CONFIG_PROC_FS)		+= proc.o
 obj-$(CONFIG_SYSFS)		+= pci-sysfs.o slot.o
 obj-$(CONFIG_ACPI)		+= pci-acpi.o
+obj-$(CONFIG_DEBUG_FS)		+= debugfs.o
 obj-$(CONFIG_GENERIC_PCI_IOMAP) += iomap.o
 endif
 
diff --git a/drivers/pci/debugfs.c b/drivers/pci/debugfs.c
new file mode 100644
index 000000000000..682fb614d7cf
--- /dev/null
+++ b/drivers/pci/debugfs.c
@@ -0,0 +1,60 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Per-device debugfs directories for the PCI core
+ *
+ * PCI features that want to export debug information create their files in
+ * the directory returned by pci_dev_debugfs_dir().  The directory itself is
+ * created on first use and is torn down, together with everything below it,
+ * when the device is destroyed.
+ */
+#include <linux/debugfs.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+
+#include "pci.h"
+
+/* Serializes creation of the root and of the per-device directories */
+static DEFINE_MUTEX(pci_debugfs_lock);
+static struct dentry *pci_debugfs_root;
+
+/**
+ * pci_dev_debugfs_dir - Return the debugfs directory of a PCI device
+ * @dev: PCI device
+ *
+ * Return the /sys/kernel/debug/pci/<device> directory of @dev, creating it
+ * and the PCI root directory if they do not exist yet.  Devices that have no
+ * debugfs content never get a directory.
+ *
+ * Return: the dentry of the directory, or an ERR_PTR() on failure.  The
+ * return value only needs to be checked with IS_ERR() if the caller cares;
+ * it can be passed to debugfs_create_*() as is.
+ */
+struct dentry *pci_dev_debugfs_dir(struct pci_dev *dev)
+{
+	guard(mutex)(&pci_debugfs_lock);
+
+	if (!pci_debugfs_root)
+		pci_debugfs_root = debugfs_create_dir("pci", NULL);
+
+	if (!dev->debugfs)
+		dev->debugfs = debugfs_create_dir(pci_name(dev),
+						  pci_debugfs_root);
+
+	return dev->debugfs;
+}
+
+/**
+ * pci_dev_debugfs_remove - Remove the debugfs directory of a PCI device
+ * @dev: PCI device
+ *
+ * Remove the device's debugfs directory and everything below it.  Returns
+ * once no debugfs operation on those files is running any more, so the
+ * owners of the files can drop the data the files refer to afterwards.
+ */
+void pci_dev_debugfs_remove(struct pci_dev *dev)
+{
+	guard(mutex)(&pci_debugfs_lock);
+
+	debugfs_remove_recursive(dev->debugfs);
+	dev->debugfs = NULL;
+}
diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index ba3c3fddddc2..aead99a4ea34 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -1059,6 +1059,18 @@ static inline int pci_resource_num_to_vf_bar(int resno)
 }
 #endif /* CONFIG_PCI_IOV */
 
+#ifdef CONFIG_DEBUG_FS
+struct dentry *pci_dev_debugfs_dir(struct pci_dev *dev);
+void pci_dev_debugfs_remove(struct pci_dev *dev);
+#else
+static inline struct dentry *pci_dev_debugfs_dir(struct pci_dev *dev)
+{
+	return ERR_PTR(-ENODEV);
+}
+
+static inline void pci_dev_debugfs_remove(struct pci_dev *dev) { }
+#endif
+
 #ifdef CONFIG_PCIE_TPH
 void pci_restore_tph_state(struct pci_dev *dev);
 void pci_save_tph_state(struct pci_dev *dev);
diff --git a/drivers/pci/remove.c b/drivers/pci/remove.c
index e711ac1d4e38..e10139117d57 100644
--- a/drivers/pci/remove.c
+++ b/drivers/pci/remove.c
@@ -37,6 +37,7 @@ static void pci_destroy_dev(struct pci_dev *dev)
 	platform_pci_remove_wake(dev);
 	pci_doe_sysfs_teardown(dev);
 	pci_npem_remove(dev);
+	pci_dev_debugfs_remove(dev);
 
 	/*
 	 * While device is in D0 drop the device from TSM link operations
diff --git a/include/linux/pci.h b/include/linux/pci.h
index d31a8d107b1e..251d6eeef087 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -580,6 +580,9 @@ struct pci_dev {
 #endif
 #ifdef CONFIG_PCI_TSM
 	struct pci_tsm *tsm;		/* TSM operation state */
+#endif
+#ifdef CONFIG_DEBUG_FS
+	struct dentry	*debugfs;	/* debugfs directory, if any */
 #endif
 	u16		acs_cap;	/* ACS Capability offset */
 	u16		acs_capabilities; /* ACS Capabilities */
-- 
2.55.0