Merge branch 'pci/bwctrl'

- Add read/modify/write locking for Link Control 2, which is used to manage
  Link speed (Ilpo Järvinen)

- Cache all supported Link speeds for use by the PCIe bandwidth controller
  (Ilpo Järvinen)

- Extract the Link Bandwidth Management Status check into pcie_lbms_seen(),
  where it can be shared between the bandwidth controller and quirks that
  use it to help retrain failed links (Ilpo Järvinen)

- Re-add Link Bandwidth notification support with updates to address the
  reasons it was previously reverted (Alexandru Gagniuc, Ilpo Järvinen)

- Add pcie_set_target_speed() and related functionality to manage PCIe Link
  speed based on thermal constraints (Ilpo Järvinen)

- Add a thermal cooling driver to throttle PCIe Links via the existing
  thermal management framework (Ilpo Järvinen)

- Add a userspace selftest for the PCIe bandwidth controller (Ilpo
  Järvinen)

- Drop duplicate pcie_get_speed_cap(), pcie_get_width_cap() declarations
  (Bjorn Helgaas)

* pci/bwctrl:
  PCI: Drop duplicate pcie_get_speed_cap(), pcie_get_width_cap() declarations
  selftests/pcie_bwctrl: Create selftests
  thermal: Add PCIe cooling driver
  PCI/bwctrl: Add pcie_set_target_speed() to set PCIe Link Speed
  PCI/bwctrl: Re-add BW notification portdrv as PCIe BW controller
  PCI: Abstract LBMS seen check into pcie_lbms_seen()
  PCI: Refactor pcie_update_link_speed()
  PCI: Store all PCIe Supported Link Speeds
  PCI: Protect Link Control 2 Register with RMW locking
  Documentation PCI: Reformat RMW ops documentation

13 files changed, 581 insertions, 48 deletions

diff --git a/drivers/pci/hotplug/pciehp_ctrl.c b/drivers/pci/hotplug/pciehp_ctrl.c
index dcdbfcf404dd..d603a7aa7483 100644
--- a/drivers/pci/hotplug/pciehp_ctrl.c
+++ b/drivers/pci/hotplug/pciehp_ctrl.c
@@ -19,6 +19,8 @@
 #include <linux/types.h>
 #include <linux/pm_runtime.h>
 #include <linux/pci.h>
+
+#include "../pci.h"
 #include "pciehp.h"
 
 /* The following routines constitute the bulk of the
@@ -127,6 +129,9 @@ static void remove_board(struct controller *ctrl, bool safe_removal)
 
 	pciehp_set_indicators(ctrl, PCI_EXP_SLTCTL_PWR_IND_OFF,
 			      INDICATOR_NOOP);
+
+	/* Don't carry LBMS indications across */
+	pcie_reset_lbms_count(ctrl->pcie->port);
 }
 
 static int pciehp_enable_slot(struct controller *ctrl);
diff --git a/drivers/pci/hotplug/pciehp_hpc.c b/drivers/pci/hotplug/pciehp_hpc.c
index 736ad8baa2a5..bb5a8d9f03ad 100644
--- a/drivers/pci/hotplug/pciehp_hpc.c
+++ b/drivers/pci/hotplug/pciehp_hpc.c
@@ -319,7 +319,7 @@ int pciehp_check_link_status(struct controller *ctrl)
 		return -1;
 	}
 
-	pcie_update_link_speed(ctrl->pcie->port->subordinate, lnk_status);
+	__pcie_update_link_speed(ctrl->pcie->port->subordinate, lnk_status);
 
 	if (!found) {
 		ctrl_info(ctrl, "Slot(%s): No device found\n",
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index 7d85c04fbba2..f85f380cdb9b 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -4740,7 +4740,7 @@ int pcie_retrain_link(struct pci_dev *pdev, bool use_lt)
 	 * to track link speed or width changes made by hardware itself
 	 * in attempt to correct unreliable link operation.
 	 */
-	pcie_capability_write_word(pdev, PCI_EXP_LNKSTA, PCI_EXP_LNKSTA_LBMS);
+	pcie_reset_lbms_count(pdev);
 	return rc;
 }
 
@@ -6189,38 +6189,64 @@ u32 pcie_bandwidth_available(struct pci_dev *dev, struct pci_dev **limiting_dev,
 EXPORT_SYMBOL(pcie_bandwidth_available);
 
 /**
- * pcie_get_speed_cap - query for the PCI device's link speed capability
+ * pcie_get_supported_speeds - query Supported Link Speed Vector
  * @dev: PCI device to query
  *
- * Query the PCI device speed capability.  Return the maximum link speed
- * supported by the device.
+ * Query @dev supported link speeds.
+ *
+ * Implementation Note in PCIe r6.0 sec 7.5.3.18 recommends determining
+ * supported link speeds using the Supported Link Speeds Vector in the Link
+ * Capabilities 2 Register (when available).
+ *
+ * Link Capabilities 2 was added in PCIe r3.0, sec 7.8.18.
+ *
+ * Without Link Capabilities 2, i.e., prior to PCIe r3.0, Supported Link
+ * Speeds field in Link Capabilities is used and only 2.5 GT/s and 5.0 GT/s
+ * speeds were defined.
+ *
+ * For @dev without Supported Link Speed Vector, the field is synthesized
+ * from the Max Link Speed field in the Link Capabilities Register.
+ *
+ * Return: Supported Link Speeds Vector (+ reserved 0 at LSB).
  */
-enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
+u8 pcie_get_supported_speeds(struct pci_dev *dev)
 {
 	u32 lnkcap2, lnkcap;
+	u8 speeds;
 
 	/*
-	 * Link Capabilities 2 was added in PCIe r3.0, sec 7.8.18.  The
-	 * implementation note there recommends using the Supported Link
-	 * Speeds Vector in Link Capabilities 2 when supported.
-	 *
-	 * Without Link Capabilities 2, i.e., prior to PCIe r3.0, software
-	 * should use the Supported Link Speeds field in Link Capabilities,
-	 * where only 2.5 GT/s and 5.0 GT/s speeds were defined.
+	 * Speeds retain the reserved 0 at LSB before PCIe Supported Link
+	 * Speeds Vector to allow using SLS Vector bit defines directly.
 	 */
 	pcie_capability_read_dword(dev, PCI_EXP_LNKCAP2, &lnkcap2);
+	speeds = lnkcap2 & PCI_EXP_LNKCAP2_SLS;
 
 	/* PCIe r3.0-compliant */
-	if (lnkcap2)
-		return PCIE_LNKCAP2_SLS2SPEED(lnkcap2);
+	if (speeds)
+		return speeds;
 
 	pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
+
+	/* Synthesize from the Max Link Speed field */
 	if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_5_0GB)
-		return PCIE_SPEED_5_0GT;
+		speeds = PCI_EXP_LNKCAP2_SLS_5_0GB | PCI_EXP_LNKCAP2_SLS_2_5GB;
 	else if ((lnkcap & PCI_EXP_LNKCAP_SLS) == PCI_EXP_LNKCAP_SLS_2_5GB)
-		return PCIE_SPEED_2_5GT;
+		speeds = PCI_EXP_LNKCAP2_SLS_2_5GB;
 
-	return PCI_SPEED_UNKNOWN;
+	return speeds;
+}
+
+/**
+ * pcie_get_speed_cap - query for the PCI device's link speed capability
+ * @dev: PCI device to query
+ *
+ * Query the PCI device speed capability.
+ *
+ * Return: the maximum link speed supported by the device.
+ */
+enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev)
+{
+	return PCIE_LNKCAP2_SLS2SPEED(dev->supported_speeds);
 }
 EXPORT_SYMBOL(pcie_get_speed_cap);
 
diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index 14d00ce45bfa..1d5c519e19b1 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -331,6 +331,17 @@ void pci_disable_bridge_window(struct pci_dev *dev);
 struct pci_bus *pci_bus_get(struct pci_bus *bus);
 void pci_bus_put(struct pci_bus *bus);
 
+#define PCIE_LNKCAP_SLS2SPEED(lnkcap)					\
+({									\
+	((lnkcap) == PCI_EXP_LNKCAP_SLS_64_0GB ? PCIE_SPEED_64_0GT :	\
+	 (lnkcap) == PCI_EXP_LNKCAP_SLS_32_0GB ? PCIE_SPEED_32_0GT :	\
+	 (lnkcap) == PCI_EXP_LNKCAP_SLS_16_0GB ? PCIE_SPEED_16_0GT :	\
+	 (lnkcap) == PCI_EXP_LNKCAP_SLS_8_0GB ? PCIE_SPEED_8_0GT :	\
+	 (lnkcap) == PCI_EXP_LNKCAP_SLS_5_0GB ? PCIE_SPEED_5_0GT :	\
+	 (lnkcap) == PCI_EXP_LNKCAP_SLS_2_5GB ? PCIE_SPEED_2_5GT :	\
+	 PCI_SPEED_UNKNOWN);						\
+})
+
 /* PCIe link information from Link Capabilities 2 */
 #define PCIE_LNKCAP2_SLS2SPEED(lnkcap2) \
 	((lnkcap2) & PCI_EXP_LNKCAP2_SLS_64_0GB ? PCIE_SPEED_64_0GT : \
@@ -341,6 +352,15 @@ void pci_bus_put(struct pci_bus *bus);
 	 (lnkcap2) & PCI_EXP_LNKCAP2_SLS_2_5GB ? PCIE_SPEED_2_5GT : \
 	 PCI_SPEED_UNKNOWN)
 
+#define PCIE_LNKCTL2_TLS2SPEED(lnkctl2) \
+	((lnkctl2) == PCI_EXP_LNKCTL2_TLS_64_0GT ? PCIE_SPEED_64_0GT : \
+	 (lnkctl2) == PCI_EXP_LNKCTL2_TLS_32_0GT ? PCIE_SPEED_32_0GT : \
+	 (lnkctl2) == PCI_EXP_LNKCTL2_TLS_16_0GT ? PCIE_SPEED_16_0GT : \
+	 (lnkctl2) == PCI_EXP_LNKCTL2_TLS_8_0GT ? PCIE_SPEED_8_0GT : \
+	 (lnkctl2) == PCI_EXP_LNKCTL2_TLS_5_0GT ? PCIE_SPEED_5_0GT : \
+	 (lnkctl2) == PCI_EXP_LNKCTL2_TLS_2_5GT ? PCIE_SPEED_2_5GT : \
+	 PCI_SPEED_UNKNOWN)
+
 /* PCIe speed to Mb/s reduced by encoding overhead */
 #define PCIE_SPEED2MBS_ENC(speed) \
 	((speed) == PCIE_SPEED_64_0GT ? 64000*1/1 : \
@@ -373,12 +393,16 @@ static inline int pcie_dev_speed_mbps(enum pci_bus_speed speed)
 	return -EINVAL;
 }
 
+u8 pcie_get_supported_speeds(struct pci_dev *dev);
 const char *pci_speed_string(enum pci_bus_speed speed);
-enum pci_bus_speed pcie_get_speed_cap(struct pci_dev *dev);
-enum pcie_link_width pcie_get_width_cap(struct pci_dev *dev);
 void __pcie_print_link_status(struct pci_dev *dev, bool verbose);
 void pcie_report_downtraining(struct pci_dev *dev);
-void pcie_update_link_speed(struct pci_bus *bus, u16 link_status);
+
+static inline void __pcie_update_link_speed(struct pci_bus *bus, u16 linksta)
+{
+	bus->cur_bus_speed = pcie_link_speed[linksta & PCI_EXP_LNKSTA_CLS];
+}
+void pcie_update_link_speed(struct pci_bus *bus);
 
 /* Single Root I/O Virtualization */
 struct pci_sriov {
@@ -692,6 +716,17 @@ static inline void pcie_set_ecrc_checking(struct pci_dev *dev) { }
 static inline void pcie_ecrc_get_policy(char *str) { }
 #endif
 
+#ifdef CONFIG_PCIEPORTBUS
+void pcie_reset_lbms_count(struct pci_dev *port);
+int pcie_lbms_count(struct pci_dev *port, unsigned long *val);
+#else
+static inline void pcie_reset_lbms_count(struct pci_dev *port) {}
+static inline int pcie_lbms_count(struct pci_dev *port, unsigned long *val)
+{
+	return -EOPNOTSUPP;
+}
+#endif
+
 struct pci_dev_reset_methods {
 	u16 vendor;
 	u16 device;
diff --git a/drivers/pci/pcie/Makefile b/drivers/pci/pcie/Makefile
index 6461aa93fe76..53ccab62314d 100644
--- a/drivers/pci/pcie/Makefile
+++ b/drivers/pci/pcie/Makefile
@@ -4,7 +4,7 @@
 
 pcieportdrv-y			:= portdrv.o rcec.o
 
-obj-$(CONFIG_PCIEPORTBUS)	+= pcieportdrv.o
+obj-$(CONFIG_PCIEPORTBUS)	+= pcieportdrv.o bwctrl.o
 
 obj-y				+= aspm.o
 obj-$(CONFIG_PCIEAER)		+= aer.o err.o
diff --git a/drivers/pci/pcie/bwctrl.c b/drivers/pci/pcie/bwctrl.c
new file mode 100644
index 000000000000..b59cacc740fa
--- /dev/null
+++ b/drivers/pci/pcie/bwctrl.c
@@ -0,0 +1,366 @@
+// SPDX-License-Identifier: GPL-2.0+
+/*
+ * PCIe bandwidth controller
+ *
+ * Author: Alexandru Gagniuc <mr.nuke.me@gmail.com>
+ *
+ * Copyright (C) 2019 Dell Inc
+ * Copyright (C) 2023-2024 Intel Corporation
+ *
+ * The PCIe bandwidth controller provides a way to alter PCIe Link Speeds
+ * and notify the operating system when the Link Width or Speed changes. The
+ * notification capability is required for all Root Ports and Downstream
+ * Ports supporting Link Width wider than x1 and/or multiple Link Speeds.
+ *
+ * This service port driver hooks into the Bandwidth Notification interrupt
+ * watching for changes or links becoming degraded in operation. It updates
+ * the cached Current Link Speed that is exposed to user space through sysfs.
+ */
+
+#define dev_fmt(fmt) "bwctrl: " fmt
+
+#include <linux/atomic.h>
+#include <linux/bitops.h>
+#include <linux/bits.h>
+#include <linux/cleanup.h>
+#include <linux/errno.h>
+#include <linux/interrupt.h>
+#include <linux/mutex.h>
+#include <linux/pci.h>
+#include <linux/pci-bwctrl.h>
+#include <linux/rwsem.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+
+#include "../pci.h"
+#include "portdrv.h"
+
+/**
+ * struct pcie_bwctrl_data - PCIe bandwidth controller
+ * @set_speed_mutex:	Serializes link speed changes
+ * @lbms_count:		Count for LBMS (since last reset)
+ * @cdev:		Thermal cooling device associated with the port
+ */
+struct pcie_bwctrl_data {
+	struct mutex set_speed_mutex;
+	atomic_t lbms_count;
+	struct thermal_cooling_device *cdev;
+};
+
+/*
+ * Prevent port removal during LBMS count accessors and Link Speed changes.
+ *
+ * These have to be differentiated because pcie_bwctrl_change_speed() calls
+ * pcie_retrain_link() which uses LBMS count reset accessor on success
+ * (using just one rwsem triggers "possible recursive locking detected"
+ * warning).
+ */
+static DECLARE_RWSEM(pcie_bwctrl_lbms_rwsem);
+static DECLARE_RWSEM(pcie_bwctrl_setspeed_rwsem);
+
+static bool pcie_valid_speed(enum pci_bus_speed speed)
+{
+	return (speed >= PCIE_SPEED_2_5GT) && (speed <= PCIE_SPEED_64_0GT);
+}
+
+static u16 pci_bus_speed2lnkctl2(enum pci_bus_speed speed)
+{
+	static const u8 speed_conv[] = {
+		[PCIE_SPEED_2_5GT] = PCI_EXP_LNKCTL2_TLS_2_5GT,
+		[PCIE_SPEED_5_0GT] = PCI_EXP_LNKCTL2_TLS_5_0GT,
+		[PCIE_SPEED_8_0GT] = PCI_EXP_LNKCTL2_TLS_8_0GT,
+		[PCIE_SPEED_16_0GT] = PCI_EXP_LNKCTL2_TLS_16_0GT,
+		[PCIE_SPEED_32_0GT] = PCI_EXP_LNKCTL2_TLS_32_0GT,
+		[PCIE_SPEED_64_0GT] = PCI_EXP_LNKCTL2_TLS_64_0GT,
+	};
+
+	if (WARN_ON_ONCE(!pcie_valid_speed(speed)))
+		return 0;
+
+	return speed_conv[speed];
+}
+
+static inline u16 pcie_supported_speeds2target_speed(u8 supported_speeds)
+{
+	return __fls(supported_speeds);
+}
+
+/**
+ * pcie_bwctrl_select_speed - Select Target Link Speed
+ * @port:	PCIe Port
+ * @speed_req:	Requested PCIe Link Speed
+ *
+ * Select Target Link Speed by take into account Supported Link Speeds of
+ * both the Root Port and the Endpoint.
+ *
+ * Return: Target Link Speed (1=2.5GT/s, 2=5GT/s, 3=8GT/s, etc.)
+ */
+static u16 pcie_bwctrl_select_speed(struct pci_dev *port, enum pci_bus_speed speed_req)
+{
+	struct pci_bus *bus = port->subordinate;
+	u8 desired_speeds, supported_speeds;
+	struct pci_dev *dev;
+
+	desired_speeds = GENMASK(pci_bus_speed2lnkctl2(speed_req),
+				 __fls(PCI_EXP_LNKCAP2_SLS_2_5GB));
+
+	supported_speeds = port->supported_speeds;
+	if (bus) {
+		down_read(&pci_bus_sem);
+		dev = list_first_entry_or_null(&bus->devices, struct pci_dev, bus_list);
+		if (dev)
+			supported_speeds &= dev->supported_speeds;
+		up_read(&pci_bus_sem);
+	}
+	if (!supported_speeds)
+		return PCI_EXP_LNKCAP2_SLS_2_5GB;
+
+	return pcie_supported_speeds2target_speed(supported_speeds & desired_speeds);
+}
+
+static int pcie_bwctrl_change_speed(struct pci_dev *port, u16 target_speed, bool use_lt)
+{
+	int ret;
+
+	ret = pcie_capability_clear_and_set_word(port, PCI_EXP_LNKCTL2,
+						 PCI_EXP_LNKCTL2_TLS, target_speed);
+	if (ret != PCIBIOS_SUCCESSFUL)
+		return pcibios_err_to_errno(ret);
+
+	ret = pcie_retrain_link(port, use_lt);
+	if (ret < 0)
+		return ret;
+
+	/*
+	 * Ensure link speed updates also with platforms that have problems
+	 * with notifications.
+	 */
+	if (port->subordinate)
+		pcie_update_link_speed(port->subordinate);
+
+	return 0;
+}
+
+/**
+ * pcie_set_target_speed - Set downstream Link Speed for PCIe Port
+ * @port:	PCIe Port
+ * @speed_req:	Requested PCIe Link Speed
+ * @use_lt:	Wait for the LT or DLLLA bit to detect the end of link training
+ *
+ * Attempt to set PCIe Port Link Speed to @speed_req. @speed_req may be
+ * adjusted downwards to the best speed supported by both the Port and PCIe
+ * Device underneath it.
+ *
+ * Return:
+ * * 0		- on success
+ * * -EINVAL	- @speed_req is not a PCIe Link Speed
+ * * -ENODEV	- @port is not controllable
+ * * -ETIMEDOUT	- changing Link Speed took too long
+ * * -EAGAIN	- Link Speed was changed but @speed_req was not achieved
+ */
+int pcie_set_target_speed(struct pci_dev *port, enum pci_bus_speed speed_req,
+			  bool use_lt)
+{
+	struct pci_bus *bus = port->subordinate;
+	u16 target_speed;
+	int ret;
+
+	if (WARN_ON_ONCE(!pcie_valid_speed(speed_req)))
+		return -EINVAL;
+
+	if (bus && bus->cur_bus_speed == speed_req)
+		return 0;
+
+	target_speed = pcie_bwctrl_select_speed(port, speed_req);
+
+	scoped_guard(rwsem_read, &pcie_bwctrl_setspeed_rwsem) {
+		struct pcie_bwctrl_data *data = port->link_bwctrl;
+
+		/*
+		 * port->link_bwctrl is NULL during initial scan when called
+		 * e.g. from the Target Speed quirk.
+		 */
+		if (data)
+			mutex_lock(&data->set_speed_mutex);
+
+		ret = pcie_bwctrl_change_speed(port, target_speed, use_lt);
+
+		if (data)
+			mutex_unlock(&data->set_speed_mutex);
+	}
+
+	/*
+	 * Despite setting higher speed into the Target Link Speed, empty
+	 * bus won't train to 5GT+ speeds.
+	 */
+	if (!ret && bus && bus->cur_bus_speed != speed_req &&
+	    !list_empty(&bus->devices))
+		ret = -EAGAIN;
+
+	return ret;
+}
+
+static void pcie_bwnotif_enable(struct pcie_device *srv)
+{
+	struct pcie_bwctrl_data *data = srv->port->link_bwctrl;
+	struct pci_dev *port = srv->port;
+	u16 link_status;
+	int ret;
+
+	/* Count LBMS seen so far as one */
+	ret = pcie_capability_read_word(port, PCI_EXP_LNKSTA, &link_status);
+	if (ret == PCIBIOS_SUCCESSFUL && link_status & PCI_EXP_LNKSTA_LBMS)
+		atomic_inc(&data->lbms_count);
+
+	pcie_capability_set_word(port, PCI_EXP_LNKCTL,
+				 PCI_EXP_LNKCTL_LBMIE | PCI_EXP_LNKCTL_LABIE);
+	pcie_capability_write_word(port, PCI_EXP_LNKSTA,
+				   PCI_EXP_LNKSTA_LBMS | PCI_EXP_LNKSTA_LABS);
+
+	/*
+	 * Update after enabling notifications & clearing status bits ensures
+	 * link speed is up to date.
+	 */
+	pcie_update_link_speed(port->subordinate);
+}
+
+static void pcie_bwnotif_disable(struct pci_dev *port)
+{
+	pcie_capability_clear_word(port, PCI_EXP_LNKCTL,
+				   PCI_EXP_LNKCTL_LBMIE | PCI_EXP_LNKCTL_LABIE);
+}
+
+static irqreturn_t pcie_bwnotif_irq(int irq, void *context)
+{
+	struct pcie_device *srv = context;
+	struct pcie_bwctrl_data *data = srv->port->link_bwctrl;
+	struct pci_dev *port = srv->port;
+	u16 link_status, events;
+	int ret;
+
+	ret = pcie_capability_read_word(port, PCI_EXP_LNKSTA, &link_status);
+	if (ret != PCIBIOS_SUCCESSFUL)
+		return IRQ_NONE;
+
+	events = link_status & (PCI_EXP_LNKSTA_LBMS | PCI_EXP_LNKSTA_LABS);
+	if (!events)
+		return IRQ_NONE;
+
+	if (events & PCI_EXP_LNKSTA_LBMS)
+		atomic_inc(&data->lbms_count);
+
+	pcie_capability_write_word(port, PCI_EXP_LNKSTA, events);
+
+	/*
+	 * Interrupts will not be triggered from any further Link Speed
+	 * change until LBMS is cleared by the write. Therefore, re-read the
+	 * speed (inside pcie_update_link_speed()) after LBMS has been
+	 * cleared to avoid missing link speed changes.
+	 */
+	pcie_update_link_speed(port->subordinate);
+
+	return IRQ_HANDLED;
+}
+
+void pcie_reset_lbms_count(struct pci_dev *port)
+{
+	struct pcie_bwctrl_data *data;
+
+	guard(rwsem_read)(&pcie_bwctrl_lbms_rwsem);
+	data = port->link_bwctrl;
+	if (data)
+		atomic_set(&data->lbms_count, 0);
+	else
+		pcie_capability_write_word(port, PCI_EXP_LNKSTA,
+					   PCI_EXP_LNKSTA_LBMS);
+}
+
+int pcie_lbms_count(struct pci_dev *port, unsigned long *val)
+{
+	struct pcie_bwctrl_data *data;
+
+	guard(rwsem_read)(&pcie_bwctrl_lbms_rwsem);
+	data = port->link_bwctrl;
+	if (!data)
+		return -ENOTTY;
+
+	*val = atomic_read(&data->lbms_count);
+
+	return 0;
+}
+
+static int pcie_bwnotif_probe(struct pcie_device *srv)
+{
+	struct pci_dev *port = srv->port;
+	int ret;
+
+	struct pcie_bwctrl_data *data = devm_kzalloc(&srv->device,
+						     sizeof(*data), GFP_KERNEL);
+	if (!data)
+		return -ENOMEM;
+
+	ret = devm_mutex_init(&srv->device, &data->set_speed_mutex);
+	if (ret)
+		return ret;
+
+	ret = devm_request_irq(&srv->device, srv->irq, pcie_bwnotif_irq,
+			       IRQF_SHARED, "PCIe bwctrl", srv);
+	if (ret)
+		return ret;
+
+	scoped_guard(rwsem_write, &pcie_bwctrl_setspeed_rwsem) {
+		scoped_guard(rwsem_write, &pcie_bwctrl_lbms_rwsem) {
+			port->link_bwctrl = no_free_ptr(data);
+			pcie_bwnotif_enable(srv);
+		}
+	}
+
+	pci_dbg(port, "enabled with IRQ %d\n", srv->irq);
+
+	/* Don't fail on errors. Don't leave IS_ERR() "pointer" into ->cdev */
+	port->link_bwctrl->cdev = pcie_cooling_device_register(port);
+	if (IS_ERR(port->link_bwctrl->cdev))
+		port->link_bwctrl->cdev = NULL;
+
+	return 0;
+}
+
+static void pcie_bwnotif_remove(struct pcie_device *srv)
+{
+	struct pcie_bwctrl_data *data = srv->port->link_bwctrl;
+
+	pcie_cooling_device_unregister(data->cdev);
+
+	pcie_bwnotif_disable(srv->port);
+
+	scoped_guard(rwsem_write, &pcie_bwctrl_setspeed_rwsem)
+		scoped_guard(rwsem_write, &pcie_bwctrl_lbms_rwsem)
+			srv->port->link_bwctrl = NULL;
+}
+
+static int pcie_bwnotif_suspend(struct pcie_device *srv)
+{
+	pcie_bwnotif_disable(srv->port);
+	return 0;
+}
+
+static int pcie_bwnotif_resume(struct pcie_device *srv)
+{
+	pcie_bwnotif_enable(srv);
+	return 0;
+}
+
+static struct pcie_port_service_driver pcie_bwctrl_driver = {
+	.name		= "pcie_bwctrl",
+	.port_type	= PCIE_ANY_PORT,
+	.service	= PCIE_PORT_SERVICE_BWCTRL,
+	.probe		= pcie_bwnotif_probe,
+	.suspend	= pcie_bwnotif_suspend,
+	.resume		= pcie_bwnotif_resume,
+	.remove		= pcie_bwnotif_remove,
+};
+
+int __init pcie_bwctrl_init(void)
+{
+	return pcie_port_service_register(&pcie_bwctrl_driver);
+}
diff --git a/drivers/pci/pcie/portdrv.c b/drivers/pci/pcie/portdrv.c
index 6af5e0425872..5e10306b6308 100644
--- a/drivers/pci/pcie/portdrv.c
+++ b/drivers/pci/pcie/portdrv.c
@@ -68,7 +68,7 @@ static int pcie_message_numbers(struct pci_dev *dev, int mask,
 	 */
 
 	if (mask & (PCIE_PORT_SERVICE_PME | PCIE_PORT_SERVICE_HP |
-		    PCIE_PORT_SERVICE_BWNOTIF)) {
+		    PCIE_PORT_SERVICE_BWCTRL)) {
 		pcie_capability_read_word(dev, PCI_EXP_FLAGS, &reg16);
 		*pme = FIELD_GET(PCI_EXP_FLAGS_IRQ, reg16);
 		nvec = *pme + 1;
@@ -150,11 +150,11 @@ static int pcie_port_enable_irq_vec(struct pci_dev *dev, int *irqs, int mask)
 
 	/* PME, hotplug and bandwidth notification share an MSI/MSI-X vector */
 	if (mask & (PCIE_PORT_SERVICE_PME | PCIE_PORT_SERVICE_HP |
-		    PCIE_PORT_SERVICE_BWNOTIF)) {
+		    PCIE_PORT_SERVICE_BWCTRL)) {
 		pcie_irq = pci_irq_vector(dev, pme);
 		irqs[PCIE_PORT_SERVICE_PME_SHIFT] = pcie_irq;
 		irqs[PCIE_PORT_SERVICE_HP_SHIFT] = pcie_irq;
-		irqs[PCIE_PORT_SERVICE_BWNOTIF_SHIFT] = pcie_irq;
+		irqs[PCIE_PORT_SERVICE_BWCTRL_SHIFT] = pcie_irq;
 	}
 
 	if (mask & PCIE_PORT_SERVICE_AER)
@@ -271,7 +271,7 @@ static int get_port_device_capability(struct pci_dev *dev)
 
 		pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &linkcap);
 		if (linkcap & PCI_EXP_LNKCAP_LBNC)
-			services |= PCIE_PORT_SERVICE_BWNOTIF;
+			services |= PCIE_PORT_SERVICE_BWCTRL;
 	}
 
 	return services;
@@ -828,6 +828,7 @@ static void __init pcie_init_services(void)
 	pcie_aer_init();
 	pcie_pme_init();
 	pcie_dpc_init();
+	pcie_bwctrl_init();
 	pcie_hp_init();
 }
 
diff --git a/drivers/pci/pcie/portdrv.h b/drivers/pci/pcie/portdrv.h
index 12c89ea0313b..bd29d1cc7b8b 100644
--- a/drivers/pci/pcie/portdrv.h
+++ b/drivers/pci/pcie/portdrv.h
@@ -20,8 +20,8 @@
 #define PCIE_PORT_SERVICE_HP		(1 << PCIE_PORT_SERVICE_HP_SHIFT)
 #define PCIE_PORT_SERVICE_DPC_SHIFT	3	/* Downstream Port Containment */
 #define PCIE_PORT_SERVICE_DPC		(1 << PCIE_PORT_SERVICE_DPC_SHIFT)
-#define PCIE_PORT_SERVICE_BWNOTIF_SHIFT	4	/* Bandwidth notification */
-#define PCIE_PORT_SERVICE_BWNOTIF	(1 << PCIE_PORT_SERVICE_BWNOTIF_SHIFT)
+#define PCIE_PORT_SERVICE_BWCTRL_SHIFT	4	/* Bandwidth Controller (notifications) */
+#define PCIE_PORT_SERVICE_BWCTRL	(1 << PCIE_PORT_SERVICE_BWCTRL_SHIFT)
 
 #define PCIE_PORT_DEVICE_MAXSERVICES   5
 
@@ -51,6 +51,8 @@ int pcie_dpc_init(void);
 static inline int pcie_dpc_init(void) { return 0; }
 #endif
 
+int pcie_bwctrl_init(void);
+
 /* Port Type */
 #define PCIE_ANY_PORT			(~0)
 
diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index 4f68414c3086..c138daf78961 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -742,9 +742,13 @@ const char *pci_speed_string(enum pci_bus_speed speed)
 }
 EXPORT_SYMBOL_GPL(pci_speed_string);
 
-void pcie_update_link_speed(struct pci_bus *bus, u16 linksta)
+void pcie_update_link_speed(struct pci_bus *bus)
 {
-	bus->cur_bus_speed = pcie_link_speed[linksta & PCI_EXP_LNKSTA_CLS];
+	struct pci_dev *bridge = bus->self;
+	u16 linksta;
+
+	pcie_capability_read_word(bridge, PCI_EXP_LNKSTA, &linksta);
+	__pcie_update_link_speed(bus, linksta);
 }
 EXPORT_SYMBOL_GPL(pcie_update_link_speed);
 
@@ -827,13 +831,11 @@ static void pci_set_bus_speed(struct pci_bus *bus)
 
 	if (pci_is_pcie(bridge)) {
 		u32 linkcap;
-		u16 linksta;
 
 		pcie_capability_read_dword(bridge, PCI_EXP_LNKCAP, &linkcap);
 		bus->max_bus_speed = pcie_link_speed[linkcap & PCI_EXP_LNKCAP_SLS];
 
-		pcie_capability_read_word(bridge, PCI_EXP_LNKSTA, &linksta);
-		pcie_update_link_speed(bus, linksta);
+		pcie_update_link_speed(bus);
 	}
 }
 
@@ -1947,6 +1949,9 @@ int pci_setup_device(struct pci_dev *dev)
 
 	set_pcie_untrusted(dev);
 
+	if (pci_is_pcie(dev))
+		dev->supported_speeds = pcie_get_supported_speeds(dev);
+
 	/* "Unknown power state" */
 	dev->current_state = PCI_UNKNOWN;
 
diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
index dccb60c1d9cc..dcf1c86a5488 100644
--- a/drivers/pci/quirks.c
+++ b/drivers/pci/quirks.c
@@ -33,6 +33,18 @@
 #include <linux/switchtec.h>
 #include "pci.h"
 
+static bool pcie_lbms_seen(struct pci_dev *dev, u16 lnksta)
+{
+	unsigned long count;
+	int ret;
+
+	ret = pcie_lbms_count(dev, &count);
+	if (ret < 0)
+		return lnksta & PCI_EXP_LNKSTA_LBMS;
+
+	return count > 0;
+}
+
 /*
  * Retrain the link of a downstream PCIe port by hand if necessary.
  *
@@ -96,22 +108,16 @@ int pcie_failed_link_retrain(struct pci_dev *dev)
 
 	pcie_capability_read_word(dev, PCI_EXP_LNKCTL2, &lnkctl2);
 	pcie_capability_read_word(dev, PCI_EXP_LNKSTA, &lnksta);
-	if ((lnksta & (PCI_EXP_LNKSTA_LBMS | PCI_EXP_LNKSTA_DLLLA)) ==
-	    PCI_EXP_LNKSTA_LBMS) {
+	if (!(lnksta & PCI_EXP_LNKSTA_DLLLA) && pcie_lbms_seen(dev, lnksta)) {
 		u16 oldlnkctl2 = lnkctl2;
 
 		pci_info(dev, "broken device, retraining non-functional downstream link at 2.5GT/s\n");
 
-		lnkctl2 &= ~PCI_EXP_LNKCTL2_TLS;
-		lnkctl2 |= PCI_EXP_LNKCTL2_TLS_2_5GT;
-		pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, lnkctl2);
-
-		ret = pcie_retrain_link(dev, false);
+		ret = pcie_set_target_speed(dev, PCIE_SPEED_2_5GT, false);
 		if (ret) {
 			pci_info(dev, "retraining failed\n");
-			pcie_capability_write_word(dev, PCI_EXP_LNKCTL2,
-						   oldlnkctl2);
-			pcie_retrain_link(dev, true);
+			pcie_set_target_speed(dev, PCIE_LNKCTL2_TLS2SPEED(oldlnkctl2),
+					      true);
 			return ret;
 		}
 
@@ -125,11 +131,7 @@ int pcie_failed_link_retrain(struct pci_dev *dev)
 
 		pci_info(dev, "removing 2.5GT/s downstream link speed restriction\n");
 		pcie_capability_read_dword(dev, PCI_EXP_LNKCAP, &lnkcap);
-		lnkctl2 &= ~PCI_EXP_LNKCTL2_TLS;
-		lnkctl2 |= lnkcap & PCI_EXP_LNKCAP_SLS;
-		pcie_capability_write_word(dev, PCI_EXP_LNKCTL2, lnkctl2);
-
-		ret = pcie_retrain_link(dev, false);
+		ret = pcie_set_target_speed(dev, PCIE_LNKCAP_SLS2SPEED(lnkcap), false);
 		if (ret) {
 			pci_info(dev, "retraining failed\n");
 			return ret;
diff --git a/drivers/thermal/Kconfig b/drivers/thermal/Kconfig
index 61e7ae524b1f..d3f9686e26e7 100644
--- a/drivers/thermal/Kconfig
+++ b/drivers/thermal/Kconfig
@@ -220,6 +220,15 @@ config DEVFREQ_THERMAL
 
 	  If you want this support, you should say Y here.
 
+config PCIE_THERMAL
+	bool "PCIe cooling support"
+	depends on PCIEPORTBUS
+	help
+	  This implements PCIe cooling mechanism through bandwidth reduction
+	  for PCIe devices.
+
+	  If you want this support, you should say Y here.
+
 config THERMAL_EMULATION
 	bool "Thermal emulation mode support"
 	help
diff --git a/drivers/thermal/Makefile b/drivers/thermal/Makefile
index 41c4d56beb40..210c16c91461 100644
--- a/drivers/thermal/Makefile
+++ b/drivers/thermal/Makefile
@@ -31,6 +31,8 @@ thermal_sys-$(CONFIG_CPU_IDLE_THERMAL)	+= cpuidle_cooling.o
 # devfreq cooling
 thermal_sys-$(CONFIG_DEVFREQ_THERMAL) += devfreq_cooling.o
 
+thermal_sys-$(CONFIG_PCIE_THERMAL) += pcie_cooling.o
+
 obj-$(CONFIG_K3_THERMAL)	+= k3_bandgap.o k3_j72xx_bandgap.o
 # platform thermal drivers
 obj-y				+= broadcom/
diff --git a/drivers/thermal/pcie_cooling.c b/drivers/thermal/pcie_cooling.c
new file mode 100644
index 000000000000..a876d64f1582
--- /dev/null
+++ b/drivers/thermal/pcie_cooling.c
@@ -0,0 +1,80 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * PCIe cooling device
+ *
+ * Copyright (C) 2023-2024 Intel Corporation
+ */
+
+#include <linux/build_bug.h>
+#include <linux/cleanup.h>
+#include <linux/err.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/pci-bwctrl.h>
+#include <linux/slab.h>
+#include <linux/sprintf.h>
+#include <linux/thermal.h>
+
+#define COOLING_DEV_TYPE_PREFIX		"PCIe_Port_Link_Speed_"
+
+static int pcie_cooling_get_max_level(struct thermal_cooling_device *cdev, unsigned long *state)
+{
+	struct pci_dev *port = cdev->devdata;
+
+	/* cooling state 0 is same as the maximum PCIe speed */
+	*state = port->subordinate->max_bus_speed - PCIE_SPEED_2_5GT;
+
+	return 0;
+}
+
+static int pcie_cooling_get_cur_level(struct thermal_cooling_device *cdev, unsigned long *state)
+{
+	struct pci_dev *port = cdev->devdata;
+
+	/* cooling state 0 is same as the maximum PCIe speed */
+	*state = cdev->max_state - (port->subordinate->cur_bus_speed - PCIE_SPEED_2_5GT);
+
+	return 0;
+}
+
+static int pcie_cooling_set_cur_level(struct thermal_cooling_device *cdev, unsigned long state)
+{
+	struct pci_dev *port = cdev->devdata;
+	enum pci_bus_speed speed;
+
+	/* cooling state 0 is same as the maximum PCIe speed */
+	speed = (cdev->max_state - state) + PCIE_SPEED_2_5GT;
+
+	return pcie_set_target_speed(port, speed, true);
+}
+
+static struct thermal_cooling_device_ops pcie_cooling_ops = {
+	.get_max_state = pcie_cooling_get_max_level,
+	.get_cur_state = pcie_cooling_get_cur_level,
+	.set_cur_state = pcie_cooling_set_cur_level,
+};
+
+struct thermal_cooling_device *pcie_cooling_device_register(struct pci_dev *port)
+{
+	char *name __free(kfree) =
+		kasprintf(GFP_KERNEL, COOLING_DEV_TYPE_PREFIX "%s", pci_name(port));
+	if (!name)
+		return ERR_PTR(-ENOMEM);
+
+	return thermal_cooling_device_register(name, port, &pcie_cooling_ops);
+}
+
+void pcie_cooling_device_unregister(struct thermal_cooling_device *cdev)
+{
+	thermal_cooling_device_unregister(cdev);
+}
+
+/* For bus_speed <-> state arithmetic */
+static_assert(PCIE_SPEED_2_5GT + 1 == PCIE_SPEED_5_0GT);
+static_assert(PCIE_SPEED_5_0GT + 1 == PCIE_SPEED_8_0GT);
+static_assert(PCIE_SPEED_8_0GT + 1 == PCIE_SPEED_16_0GT);
+static_assert(PCIE_SPEED_16_0GT + 1 == PCIE_SPEED_32_0GT);
+static_assert(PCIE_SPEED_32_0GT + 1 == PCIE_SPEED_64_0GT);
+
+MODULE_AUTHOR("Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>");
+MODULE_DESCRIPTION("PCIe cooling driver");