thunderbolt: Fix NULL dereference in tb_remove_work()grafted

There is a slight race between tb_remove_work() and tb_domain_remove()
which leads to dereferencing a NULL tb->root_switch pointer inside
tb_free_unplugged_xdomains():

        Thread A                            Thread B

  tb_remove_work()
                                    tb_domain_remove()
                                      mutex_lock(&tb->lock)
                                      tb_stop()
                                        /* doesn't cancel a running callback */
                                        cancel_delayed_work(&tcm->remove_work)
                                        ...
                                        tb_switch_remove(tb->root_switch)
                                        tb->root_switch = NULL
                                      mutex_unlock(&tb->lock)
    mutex_lock(&tb->lock)
    ...
    /* without checking ->root_switch */
    tb_free_unplugged_xdomains(tb->root_switch)
    mutex_unlock(&tb->lock)

Commit a8937f35cf39 ("thunderbolt: Remove XDomain from the bus without
holding tb->lock") doesn't seem right to move tb_free_unplugged_xdomains()
out of the &tb->lock section and the check for tb->root_switch, in
particular.  It states:

  For this reason separate removing the XDomain from the topology data
  structures (where we need the lock) from unregistering the device from
  the bus (where remove callbacks of the drivers are being called).

tb_free_unplugged_xdomains() belongs to the former group of functions
requiring the lock.  And it also calls tb_xdomain_remove() which should
only be called with &tb->lock held.

Found by Linux Verification Center (linuxtesting.org) with Svace static
analysis tool.

Fixes: a8937f35cf39 ("thunderbolt: Remove XDomain from the bus without holding tb->lock")
Cc: stable@vger.kernel.org
Signed-off-by: Fedor Pchelkin <pchelkin@ispras.ru>
Signed-off-by: Mika Westerberg <mika.westerberg@linux.intel.com>

1 files changed, 385 insertions, 0 deletions

diff --git a/drivers/clocksource/timer-nxp-pit.c b/drivers/clocksource/timer-nxp-pit.c
new file mode 100644
index 000000000..2f70d1d5e
--- /dev/null
+++ b/drivers/clocksource/timer-nxp-pit.c
@@ -0,0 +1,385 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright 2012-2013 Freescale Semiconductor, Inc.
+ * Copyright 2018,2021-2025 NXP
+ */
+#include <linux/interrupt.h>
+#include <linux/clockchips.h>
+#include <linux/cpuhotplug.h>
+#include <linux/clk.h>
+#include <linux/of_address.h>
+#include <linux/of_irq.h>
+#include <linux/sched_clock.h>
+#include <linux/platform_device.h>
+
+/*
+ * Each pit takes 0x10 Bytes register space
+ */
+#define PIT0_OFFSET	0x100
+#define PIT_CH(n)       (PIT0_OFFSET + 0x10 * (n))
+
+#define PITMCR(__base)	(__base)
+
+#define PITMCR_FRZ	BIT(0)
+#define PITMCR_MDIS	BIT(1)
+
+#define PITLDVAL(__base)	(__base)
+#define PITTCTRL(__base)	((__base) + 0x08)
+
+#define PITCVAL_OFFSET	0x04
+#define PITCVAL(__base)	((__base) + 0x04)
+
+#define PITTCTRL_TEN			BIT(0)
+#define PITTCTRL_TIE			BIT(1)
+
+#define PITTFLG(__base)	((__base) + 0x0c)
+
+#define PITTFLG_TIF			BIT(0)
+
+struct pit_timer {
+	void __iomem *clksrc_base;
+	void __iomem *clkevt_base;
+	struct clock_event_device ced;
+	struct clocksource cs;
+	int rate;
+};
+
+struct pit_timer_data {
+	int max_pit_instances;
+};
+
+static DEFINE_PER_CPU(struct pit_timer *, pit_timers);
+
+/*
+ * Global structure for multiple PITs initialization
+ */
+static int pit_instances;
+static int max_pit_instances = 1;
+
+static void __iomem *sched_clock_base;
+
+static inline struct pit_timer *ced_to_pit(struct clock_event_device *ced)
+{
+	return container_of(ced, struct pit_timer, ced);
+}
+
+static inline struct pit_timer *cs_to_pit(struct clocksource *cs)
+{
+	return container_of(cs, struct pit_timer, cs);
+}
+
+static inline void pit_module_enable(void __iomem *base)
+{
+	writel(0, PITMCR(base));
+}
+
+static inline void pit_module_disable(void __iomem *base)
+{
+	writel(PITMCR_MDIS, PITMCR(base));
+}
+
+static inline void pit_timer_enable(void __iomem *base, bool tie)
+{
+	u32 val = PITTCTRL_TEN | (tie ? PITTCTRL_TIE : 0);
+
+	writel(val, PITTCTRL(base));
+}
+
+static inline void pit_timer_disable(void __iomem *base)
+{
+	writel(0, PITTCTRL(base));
+}
+
+static inline void pit_timer_set_counter(void __iomem *base, unsigned int cnt)
+{
+	writel(cnt, PITLDVAL(base));
+}
+
+static inline void pit_timer_irqack(struct pit_timer *pit)
+{
+	writel(PITTFLG_TIF, PITTFLG(pit->clkevt_base));
+}
+
+static u64 notrace pit_read_sched_clock(void)
+{
+	return ~readl(sched_clock_base);
+}
+
+static u64 pit_timer_clocksource_read(struct clocksource *cs)
+{
+	struct pit_timer *pit = cs_to_pit(cs);
+
+	return (u64)~readl(PITCVAL(pit->clksrc_base));
+}
+
+static int pit_clocksource_init(struct pit_timer *pit, const char *name,
+				void __iomem *base, unsigned long rate)
+{
+	/*
+	 * The channels 0 and 1 can be chained to build a 64-bit
+	 * timer. Let's use the channel 2 as a clocksource and leave
+	 * the channels 0 and 1 unused for anyone else who needs them
+	 */
+	pit->clksrc_base = base + PIT_CH(2);
+	pit->cs.name = name;
+	pit->cs.rating = 300;
+	pit->cs.read = pit_timer_clocksource_read;
+	pit->cs.mask = CLOCKSOURCE_MASK(32);
+	pit->cs.flags = CLOCK_SOURCE_IS_CONTINUOUS;
+
+	/* set the max load value and start the clock source counter */
+	pit_timer_disable(pit->clksrc_base);
+	pit_timer_set_counter(pit->clksrc_base, ~0);
+	pit_timer_enable(pit->clksrc_base, 0);
+
+	sched_clock_base = pit->clksrc_base + PITCVAL_OFFSET;
+	sched_clock_register(pit_read_sched_clock, 32, rate);
+
+	return clocksource_register_hz(&pit->cs, rate);
+}
+
+static int pit_set_next_event(unsigned long delta, struct clock_event_device *ced)
+{
+	struct pit_timer *pit = ced_to_pit(ced);
+
+	/*
+	 * set a new value to PITLDVAL register will not restart the timer,
+	 * to abort the current cycle and start a timer period with the new
+	 * value, the timer must be disabled and enabled again.
+	 * and the PITLAVAL should be set to delta minus one according to pit
+	 * hardware requirement.
+	 */
+	pit_timer_disable(pit->clkevt_base);
+	pit_timer_set_counter(pit->clkevt_base, delta - 1);
+	pit_timer_enable(pit->clkevt_base, true);
+
+	return 0;
+}
+
+static int pit_shutdown(struct clock_event_device *ced)
+{
+	struct pit_timer *pit = ced_to_pit(ced);
+
+	pit_timer_disable(pit->clkevt_base);
+
+	return 0;
+}
+
+static int pit_set_periodic(struct clock_event_device *ced)
+{
+	struct pit_timer *pit = ced_to_pit(ced);
+
+	pit_set_next_event(pit->rate / HZ, ced);
+
+	return 0;
+}
+
+static irqreturn_t pit_timer_interrupt(int irq, void *dev_id)
+{
+	struct clock_event_device *ced = dev_id;
+	struct pit_timer *pit = ced_to_pit(ced);
+
+	pit_timer_irqack(pit);
+
+	/*
+	 * pit hardware doesn't support oneshot, it will generate an interrupt
+	 * and reload the counter value from PITLDVAL when PITCVAL reach zero,
+	 * and start the counter again. So software need to disable the timer
+	 * to stop the counter loop in ONESHOT mode.
+	 */
+	if (likely(clockevent_state_oneshot(ced)))
+		pit_timer_disable(pit->clkevt_base);
+
+	ced->event_handler(ced);
+
+	return IRQ_HANDLED;
+}
+
+static int pit_clockevent_per_cpu_init(struct pit_timer *pit, const char *name,
+				       void __iomem *base, unsigned long rate,
+				       int irq, unsigned int cpu)
+{
+	int ret;
+
+	/*
+	 * The channels 0 and 1 can be chained to build a 64-bit
+	 * timer. Let's use the channel 3 as a clockevent and leave
+	 * the channels 0 and 1 unused for anyone else who needs them
+	 */
+	pit->clkevt_base = base + PIT_CH(3);
+	pit->rate = rate;
+
+	pit_timer_disable(pit->clkevt_base);
+
+	pit_timer_irqack(pit);
+
+	ret = request_irq(irq, pit_timer_interrupt, IRQF_TIMER | IRQF_NOBALANCING,
+			  name, &pit->ced);
+	if (ret)
+		return ret;
+
+	pit->ced.cpumask = cpumask_of(cpu);
+	pit->ced.irq = irq;
+
+	pit->ced.name = name;
+	pit->ced.features = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT;
+	pit->ced.set_state_shutdown = pit_shutdown;
+	pit->ced.set_state_periodic = pit_set_periodic;
+	pit->ced.set_next_event	= pit_set_next_event;
+	pit->ced.rating	= 300;
+
+	per_cpu(pit_timers, cpu) = pit;
+
+	return 0;
+}
+
+static void pit_clockevent_per_cpu_exit(struct pit_timer *pit, unsigned int cpu)
+{
+	pit_timer_disable(pit->clkevt_base);
+	free_irq(pit->ced.irq, &pit->ced);
+	per_cpu(pit_timers, cpu) = NULL;
+}
+
+static int pit_clockevent_starting_cpu(unsigned int cpu)
+{
+	struct pit_timer *pit = per_cpu(pit_timers, cpu);
+	int ret;
+
+	if (!pit)
+		return 0;
+
+	ret = irq_force_affinity(pit->ced.irq, cpumask_of(cpu));
+	if (ret) {
+		pit_clockevent_per_cpu_exit(pit, cpu);
+		return ret;
+	}
+
+	/*
+	 * The value for the LDVAL register trigger is calculated as:
+	 * LDVAL trigger = (period / clock period) - 1
+	 * The pit is a 32-bit down count timer, when the counter value
+	 * reaches 0, it will generate an interrupt, thus the minimal
+	 * LDVAL trigger value is 1. And then the min_delta is
+	 * minimal LDVAL trigger value + 1, and the max_delta is full 32-bit.
+	 */
+	clockevents_config_and_register(&pit->ced, pit->rate, 2, 0xffffffff);
+
+	return 0;
+}
+
+static int pit_timer_init(struct device_node *np)
+{
+	struct pit_timer *pit;
+	struct clk *pit_clk;
+	void __iomem *timer_base;
+	const char *name = of_node_full_name(np);
+	unsigned long clk_rate;
+	int irq, ret;
+
+	pit = kzalloc_obj(*pit);
+	if (!pit)
+		return -ENOMEM;
+
+	ret = -ENXIO;
+	timer_base = of_iomap(np, 0);
+	if (!timer_base) {
+		pr_err("Failed to iomap\n");
+		goto out_kfree;
+	}
+
+	ret = -EINVAL;
+	irq = irq_of_parse_and_map(np, 0);
+	if (irq <= 0) {
+		pr_err("Failed to irq_of_parse_and_map\n");
+		goto out_iounmap;
+	}
+
+	pit_clk = of_clk_get(np, 0);
+	if (IS_ERR(pit_clk)) {
+		ret = PTR_ERR(pit_clk);
+		goto out_irq_dispose_mapping;
+	}
+
+	ret = clk_prepare_enable(pit_clk);
+	if (ret)
+		goto out_clk_put;
+
+	clk_rate = clk_get_rate(pit_clk);
+
+	pit_module_disable(timer_base);
+
+	ret = pit_clocksource_init(pit, name, timer_base, clk_rate);
+	if (ret) {
+		pr_err("Failed to initialize clocksource '%pOF'\n", np);
+		goto out_pit_module_disable;
+	}
+
+	ret = pit_clockevent_per_cpu_init(pit, name, timer_base, clk_rate, irq, pit_instances);
+	if (ret) {
+		pr_err("Failed to initialize clockevent '%pOF'\n", np);
+		goto out_pit_clocksource_unregister;
+	}
+
+	/* enable the pit module */
+	pit_module_enable(timer_base);
+
+	pit_instances++;
+
+	if (pit_instances == max_pit_instances) {
+		ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "PIT timer:starting",
+					pit_clockevent_starting_cpu, NULL);
+		if (ret < 0) {
+			pit_clockevent_per_cpu_exit(pit, pit_instances);
+			goto out_pit_clocksource_unregister;
+		}
+	}
+
+	return 0;
+
+out_pit_clocksource_unregister:
+	clocksource_unregister(&pit->cs);
+out_pit_module_disable:
+	pit_module_disable(timer_base);
+	clk_disable_unprepare(pit_clk);
+out_clk_put:
+	clk_put(pit_clk);
+out_irq_dispose_mapping:
+	irq_dispose_mapping(irq);
+out_iounmap:
+	iounmap(timer_base);
+out_kfree:
+	kfree(pit);
+
+	return ret;
+}
+
+static int pit_timer_probe(struct platform_device *pdev)
+{
+	const struct pit_timer_data *pit_timer_data;
+
+	pit_timer_data = of_device_get_match_data(&pdev->dev);
+	if (pit_timer_data)
+		max_pit_instances = pit_timer_data->max_pit_instances;
+
+	return pit_timer_init(pdev->dev.of_node);
+}
+
+static struct pit_timer_data s32g2_data = { .max_pit_instances = 2 };
+
+static const struct of_device_id pit_timer_of_match[] = {
+	{ .compatible = "nxp,s32g2-pit", .data = &s32g2_data },
+	{ }
+};
+MODULE_DEVICE_TABLE(of, pit_timer_of_match);
+
+static struct platform_driver nxp_pit_driver = {
+	.driver = {
+		.name = "nxp-pit",
+		.of_match_table = pit_timer_of_match,
+		.suppress_bind_attrs = true,
+	},
+	.probe = pit_timer_probe,
+};
+builtin_platform_driver(nxp_pit_driver);
+
+TIMER_OF_DECLARE(vf610, "fsl,vf610-pit", pit_timer_init);