* [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
@ 2022-10-18 13:04 ` Pierre Gondois
0 siblings, 0 replies; 12+ messages in thread
From: Pierre Gondois @ 2022-10-18 13:04 UTC (permalink / raw)
To: linux-kernel
Cc: Pierre Gondois, James Morse, Will Deacon, Huacai Chen,
John Garry, Shaokun Zhang, WANG Xuerui, Qi Liu, Bharat Bhushan,
Huang Ying, Bibo Mao, Jason A. Donenfeld, Daniel Lezcano,
linux-arm-kernel
On an Ampere Altra,
Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
Ampere Altra triggers:
[ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
[ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
[ 15.683157] preempt_count: 0, expected: 0
[ 15.683159] RCU nest depth: 0, expected: 0
[ 15.683163] 3 locks held by cpuhp/0/24:
[ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
[ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
[ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
[ 15.683224] irq event stamp: 36
[ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
[ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
[ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
[ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
[ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
[ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
[ 15.683268] Call trace:
[ 15.683271] dump_backtrace+0x114/0x120
[ 15.683277] show_stack+0x20/0x70
[ 15.683279] dump_stack_lvl+0x9c/0xd8
[ 15.683288] dump_stack+0x18/0x34
[ 15.683289] __might_resched+0x188/0x228
[ 15.683292] rt_spin_lock+0x70/0x120
[ 15.683301] sdei_cpuhp_up+0x3c/0x130
[ 15.683303] cpuhp_invoke_callback+0x250/0xf08
[ 15.683305] cpuhp_thread_fun+0x120/0x248
[ 15.683308] smpboot_thread_fn+0x280/0x320
[ 15.683315] kthread+0x130/0x140
[ 15.683321] ret_from_fork+0x10/0x20
sdei_cpuhp_up() is called in the STARTING hotplug section,
which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
state to the _ONLINE section to execute the cpuhp cb with
preemption enabled.
Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
calling CPU. It is checked that preemption is disabled for them.
_ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
Preemption is enabled in those threads, but their cpumask is limited
to 1 CPU.
Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
don't trigger them. This means that no check will be done for some
cases, e.g. sdei_mask_local_cpu() invocations.
Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
which acts on the calling CPU.
Suggested-by: James Morse <james.morse@arm.com>
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
---
drivers/firmware/arm_sdei.c | 28 ++++++++++++++--------------
include/linux/cpuhotplug.h | 2 +-
2 files changed, 15 insertions(+), 15 deletions(-)
diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
index 1e1a51510e83..9b03e164a37a 100644
--- a/drivers/firmware/arm_sdei.c
+++ b/drivers/firmware/arm_sdei.c
@@ -301,8 +301,6 @@ int sdei_mask_local_cpu(void)
{
int err;
- WARN_ON_ONCE(preemptible());
-
err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_MASK, 0, 0, 0, 0, 0, NULL);
if (err && err != -EIO) {
pr_warn_once("failed to mask CPU[%u]: %d\n",
@@ -315,6 +313,7 @@ int sdei_mask_local_cpu(void)
static void _ipi_mask_cpu(void *ignored)
{
+ WARN_ON_ONCE(preemptible());
sdei_mask_local_cpu();
}
@@ -322,8 +321,6 @@ int sdei_unmask_local_cpu(void)
{
int err;
- WARN_ON_ONCE(preemptible());
-
err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_UNMASK, 0, 0, 0, 0, 0, NULL);
if (err && err != -EIO) {
pr_warn_once("failed to unmask CPU[%u]: %d\n",
@@ -336,6 +333,7 @@ int sdei_unmask_local_cpu(void)
static void _ipi_unmask_cpu(void *ignored)
{
+ WARN_ON_ONCE(preemptible());
sdei_unmask_local_cpu();
}
@@ -343,6 +341,8 @@ static void _ipi_private_reset(void *ignored)
{
int err;
+ WARN_ON_ONCE(preemptible());
+
err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PRIVATE_RESET, 0, 0, 0, 0, 0,
NULL);
if (err && err != -EIO)
@@ -389,8 +389,6 @@ static void _local_event_enable(void *data)
int err;
struct sdei_crosscall_args *arg = data;
- WARN_ON_ONCE(preemptible());
-
err = sdei_api_event_enable(arg->event->event_num);
sdei_cross_call_return(arg, err);
@@ -401,6 +399,8 @@ int sdei_event_enable(u32 event_num)
int err = -EINVAL;
struct sdei_event *event;
+ WARN_ON_ONCE(preemptible());
+
mutex_lock(&sdei_events_lock);
event = sdei_event_find(event_num);
if (!event) {
@@ -479,8 +479,6 @@ static void _local_event_unregister(void *data)
int err;
struct sdei_crosscall_args *arg = data;
- WARN_ON_ONCE(preemptible());
-
err = sdei_api_event_unregister(arg->event->event_num);
sdei_cross_call_return(arg, err);
@@ -492,6 +490,7 @@ int sdei_event_unregister(u32 event_num)
struct sdei_event *event;
WARN_ON(in_nmi());
+ WARN_ON_ONCE(preemptible());
mutex_lock(&sdei_events_lock);
event = sdei_event_find(event_num);
@@ -561,8 +560,6 @@ static void _local_event_register(void *data)
struct sdei_registered_event *reg;
struct sdei_crosscall_args *arg = data;
- WARN_ON(preemptible());
-
reg = per_cpu_ptr(arg->event->private_registered, smp_processor_id());
err = sdei_api_event_register(arg->event->event_num, sdei_entry_point,
reg, 0, 0);
@@ -576,6 +573,7 @@ int sdei_event_register(u32 event_num, sdei_event_callback *cb, void *arg)
struct sdei_event *event;
WARN_ON(in_nmi());
+ WARN_ON_ONCE(preemptible());
mutex_lock(&sdei_events_lock);
if (sdei_event_find(event_num)) {
@@ -717,6 +715,8 @@ static int sdei_pm_notifier(struct notifier_block *nb, unsigned long action,
{
int rv;
+ WARN_ON_ONCE(preemptible());
+
switch (action) {
case CPU_PM_ENTER:
rv = sdei_mask_local_cpu();
@@ -765,7 +765,7 @@ static int sdei_device_freeze(struct device *dev)
int err;
/* unregister private events */
- cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
+ cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
err = sdei_unregister_shared();
if (err)
@@ -786,7 +786,7 @@ static int sdei_device_thaw(struct device *dev)
return err;
}
- err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
+ err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
&sdei_cpuhp_up, &sdei_cpuhp_down);
if (err)
pr_warn("Failed to re-register CPU hotplug notifier...\n");
@@ -823,7 +823,7 @@ static int sdei_reboot_notifier(struct notifier_block *nb, unsigned long action,
* We are going to reset the interface, after this there is no point
* doing work when we take CPUs offline.
*/
- cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
+ cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
sdei_platform_reset();
@@ -1003,7 +1003,7 @@ static int sdei_probe(struct platform_device *pdev)
goto remove_cpupm;
}
- err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
+ err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
&sdei_cpuhp_up, &sdei_cpuhp_down);
if (err) {
pr_warn("Failed to register CPU hotplug notifier...\n");
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index f61447913db9..63b89e00a4c3 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -161,7 +161,6 @@ enum cpuhp_state {
CPUHP_AP_PERF_X86_CSTATE_STARTING,
CPUHP_AP_PERF_XTENSA_STARTING,
CPUHP_AP_MIPS_OP_LOONGSON3_STARTING,
- CPUHP_AP_ARM_SDEI_STARTING,
CPUHP_AP_ARM_VFP_STARTING,
CPUHP_AP_ARM64_DEBUG_MONITORS_STARTING,
CPUHP_AP_PERF_ARM_HW_BREAKPOINT_STARTING,
@@ -209,6 +208,7 @@ enum cpuhp_state {
CPUHP_AP_IRQ_AFFINITY_ONLINE,
CPUHP_AP_BLK_MQ_ONLINE,
CPUHP_AP_ARM_MVEBU_SYNC_CLOCKS,
+ CPUHP_AP_ARM_SDEI_ONLINE,
CPUHP_AP_X86_INTEL_EPB_ONLINE,
CPUHP_AP_PERF_ONLINE,
CPUHP_AP_PERF_X86_ONLINE,
--
2.25.1
^ permalink raw reply related [flat|nested] 12+ messages in thread
* [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
@ 2022-10-18 13:04 ` Pierre Gondois
0 siblings, 0 replies; 12+ messages in thread
From: Pierre Gondois @ 2022-10-18 13:04 UTC (permalink / raw)
To: linux-kernel
Cc: Pierre Gondois, James Morse, Will Deacon, Huacai Chen,
John Garry, Shaokun Zhang, WANG Xuerui, Qi Liu, Bharat Bhushan,
Huang Ying, Bibo Mao, Jason A. Donenfeld, Daniel Lezcano,
linux-arm-kernel
On an Ampere Altra,
Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
Ampere Altra triggers:
[ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
[ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
[ 15.683157] preempt_count: 0, expected: 0
[ 15.683159] RCU nest depth: 0, expected: 0
[ 15.683163] 3 locks held by cpuhp/0/24:
[ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
[ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
[ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
[ 15.683224] irq event stamp: 36
[ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
[ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
[ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
[ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
[ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
[ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
[ 15.683268] Call trace:
[ 15.683271] dump_backtrace+0x114/0x120
[ 15.683277] show_stack+0x20/0x70
[ 15.683279] dump_stack_lvl+0x9c/0xd8
[ 15.683288] dump_stack+0x18/0x34
[ 15.683289] __might_resched+0x188/0x228
[ 15.683292] rt_spin_lock+0x70/0x120
[ 15.683301] sdei_cpuhp_up+0x3c/0x130
[ 15.683303] cpuhp_invoke_callback+0x250/0xf08
[ 15.683305] cpuhp_thread_fun+0x120/0x248
[ 15.683308] smpboot_thread_fn+0x280/0x320
[ 15.683315] kthread+0x130/0x140
[ 15.683321] ret_from_fork+0x10/0x20
sdei_cpuhp_up() is called in the STARTING hotplug section,
which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
state to the _ONLINE section to execute the cpuhp cb with
preemption enabled.
Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
calling CPU. It is checked that preemption is disabled for them.
_ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
Preemption is enabled in those threads, but their cpumask is limited
to 1 CPU.
Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
don't trigger them. This means that no check will be done for some
cases, e.g. sdei_mask_local_cpu() invocations.
Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
which acts on the calling CPU.
Suggested-by: James Morse <james.morse@arm.com>
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
---
drivers/firmware/arm_sdei.c | 28 ++++++++++++++--------------
include/linux/cpuhotplug.h | 2 +-
2 files changed, 15 insertions(+), 15 deletions(-)
diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
index 1e1a51510e83..9b03e164a37a 100644
--- a/drivers/firmware/arm_sdei.c
+++ b/drivers/firmware/arm_sdei.c
@@ -301,8 +301,6 @@ int sdei_mask_local_cpu(void)
{
int err;
- WARN_ON_ONCE(preemptible());
-
err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_MASK, 0, 0, 0, 0, 0, NULL);
if (err && err != -EIO) {
pr_warn_once("failed to mask CPU[%u]: %d\n",
@@ -315,6 +313,7 @@ int sdei_mask_local_cpu(void)
static void _ipi_mask_cpu(void *ignored)
{
+ WARN_ON_ONCE(preemptible());
sdei_mask_local_cpu();
}
@@ -322,8 +321,6 @@ int sdei_unmask_local_cpu(void)
{
int err;
- WARN_ON_ONCE(preemptible());
-
err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_UNMASK, 0, 0, 0, 0, 0, NULL);
if (err && err != -EIO) {
pr_warn_once("failed to unmask CPU[%u]: %d\n",
@@ -336,6 +333,7 @@ int sdei_unmask_local_cpu(void)
static void _ipi_unmask_cpu(void *ignored)
{
+ WARN_ON_ONCE(preemptible());
sdei_unmask_local_cpu();
}
@@ -343,6 +341,8 @@ static void _ipi_private_reset(void *ignored)
{
int err;
+ WARN_ON_ONCE(preemptible());
+
err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PRIVATE_RESET, 0, 0, 0, 0, 0,
NULL);
if (err && err != -EIO)
@@ -389,8 +389,6 @@ static void _local_event_enable(void *data)
int err;
struct sdei_crosscall_args *arg = data;
- WARN_ON_ONCE(preemptible());
-
err = sdei_api_event_enable(arg->event->event_num);
sdei_cross_call_return(arg, err);
@@ -401,6 +399,8 @@ int sdei_event_enable(u32 event_num)
int err = -EINVAL;
struct sdei_event *event;
+ WARN_ON_ONCE(preemptible());
+
mutex_lock(&sdei_events_lock);
event = sdei_event_find(event_num);
if (!event) {
@@ -479,8 +479,6 @@ static void _local_event_unregister(void *data)
int err;
struct sdei_crosscall_args *arg = data;
- WARN_ON_ONCE(preemptible());
-
err = sdei_api_event_unregister(arg->event->event_num);
sdei_cross_call_return(arg, err);
@@ -492,6 +490,7 @@ int sdei_event_unregister(u32 event_num)
struct sdei_event *event;
WARN_ON(in_nmi());
+ WARN_ON_ONCE(preemptible());
mutex_lock(&sdei_events_lock);
event = sdei_event_find(event_num);
@@ -561,8 +560,6 @@ static void _local_event_register(void *data)
struct sdei_registered_event *reg;
struct sdei_crosscall_args *arg = data;
- WARN_ON(preemptible());
-
reg = per_cpu_ptr(arg->event->private_registered, smp_processor_id());
err = sdei_api_event_register(arg->event->event_num, sdei_entry_point,
reg, 0, 0);
@@ -576,6 +573,7 @@ int sdei_event_register(u32 event_num, sdei_event_callback *cb, void *arg)
struct sdei_event *event;
WARN_ON(in_nmi());
+ WARN_ON_ONCE(preemptible());
mutex_lock(&sdei_events_lock);
if (sdei_event_find(event_num)) {
@@ -717,6 +715,8 @@ static int sdei_pm_notifier(struct notifier_block *nb, unsigned long action,
{
int rv;
+ WARN_ON_ONCE(preemptible());
+
switch (action) {
case CPU_PM_ENTER:
rv = sdei_mask_local_cpu();
@@ -765,7 +765,7 @@ static int sdei_device_freeze(struct device *dev)
int err;
/* unregister private events */
- cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
+ cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
err = sdei_unregister_shared();
if (err)
@@ -786,7 +786,7 @@ static int sdei_device_thaw(struct device *dev)
return err;
}
- err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
+ err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
&sdei_cpuhp_up, &sdei_cpuhp_down);
if (err)
pr_warn("Failed to re-register CPU hotplug notifier...\n");
@@ -823,7 +823,7 @@ static int sdei_reboot_notifier(struct notifier_block *nb, unsigned long action,
* We are going to reset the interface, after this there is no point
* doing work when we take CPUs offline.
*/
- cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
+ cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
sdei_platform_reset();
@@ -1003,7 +1003,7 @@ static int sdei_probe(struct platform_device *pdev)
goto remove_cpupm;
}
- err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
+ err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
&sdei_cpuhp_up, &sdei_cpuhp_down);
if (err) {
pr_warn("Failed to register CPU hotplug notifier...\n");
diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
index f61447913db9..63b89e00a4c3 100644
--- a/include/linux/cpuhotplug.h
+++ b/include/linux/cpuhotplug.h
@@ -161,7 +161,6 @@ enum cpuhp_state {
CPUHP_AP_PERF_X86_CSTATE_STARTING,
CPUHP_AP_PERF_XTENSA_STARTING,
CPUHP_AP_MIPS_OP_LOONGSON3_STARTING,
- CPUHP_AP_ARM_SDEI_STARTING,
CPUHP_AP_ARM_VFP_STARTING,
CPUHP_AP_ARM64_DEBUG_MONITORS_STARTING,
CPUHP_AP_PERF_ARM_HW_BREAKPOINT_STARTING,
@@ -209,6 +208,7 @@ enum cpuhp_state {
CPUHP_AP_IRQ_AFFINITY_ONLINE,
CPUHP_AP_BLK_MQ_ONLINE,
CPUHP_AP_ARM_MVEBU_SYNC_CLOCKS,
+ CPUHP_AP_ARM_SDEI_ONLINE,
CPUHP_AP_X86_INTEL_EPB_ONLINE,
CPUHP_AP_PERF_ONLINE,
CPUHP_AP_PERF_X86_ONLINE,
--
2.25.1
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply related [flat|nested] 12+ messages in thread
* Re: [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
2022-10-18 13:04 ` Pierre Gondois
@ 2022-10-18 13:07 ` Pierre Gondois
-1 siblings, 0 replies; 12+ messages in thread
From: Pierre Gondois @ 2022-10-18 13:07 UTC (permalink / raw)
To: linux-kernel
Cc: James Morse, Will Deacon, Huacai Chen, John Garry, Shaokun Zhang,
WANG Xuerui, Qi Liu, Bharat Bhushan, Huang Ying, Bibo Mao,
Jason A. Donenfeld, Daniel Lezcano, linux-arm-kernel
The changelog was removed:
v2:
- Don't use a raw spin_lock, but move the CPUHP_AP_ARM_SDEI_
cpuhp state to the _ONLINE section.
On 10/18/22 15:04, Pierre Gondois wrote:
> On an Ampere Altra,
> Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
> Ampere Altra triggers:
> [ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
> [ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
> [ 15.683157] preempt_count: 0, expected: 0
> [ 15.683159] RCU nest depth: 0, expected: 0
> [ 15.683163] 3 locks held by cpuhp/0/24:
> [ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
> [ 15.683224] irq event stamp: 36
> [ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
> [ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
> [ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
> [ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
> [ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
> [ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
> [ 15.683268] Call trace:
> [ 15.683271] dump_backtrace+0x114/0x120
> [ 15.683277] show_stack+0x20/0x70
> [ 15.683279] dump_stack_lvl+0x9c/0xd8
> [ 15.683288] dump_stack+0x18/0x34
> [ 15.683289] __might_resched+0x188/0x228
> [ 15.683292] rt_spin_lock+0x70/0x120
> [ 15.683301] sdei_cpuhp_up+0x3c/0x130
> [ 15.683303] cpuhp_invoke_callback+0x250/0xf08
> [ 15.683305] cpuhp_thread_fun+0x120/0x248
> [ 15.683308] smpboot_thread_fn+0x280/0x320
> [ 15.683315] kthread+0x130/0x140
> [ 15.683321] ret_from_fork+0x10/0x20
>
> sdei_cpuhp_up() is called in the STARTING hotplug section,
> which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
> state to the _ONLINE section to execute the cpuhp cb with
> preemption enabled.
>
> Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
> calling CPU. It is checked that preemption is disabled for them.
> _ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
> Preemption is enabled in those threads, but their cpumask is limited
> to 1 CPU.
> Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
> don't trigger them. This means that no check will be done for some
> cases, e.g. sdei_mask_local_cpu() invocations.
>
> Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
> which acts on the calling CPU.
>
> Suggested-by: James Morse <james.morse@arm.com>
> Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
> ---
> drivers/firmware/arm_sdei.c | 28 ++++++++++++++--------------
> include/linux/cpuhotplug.h | 2 +-
> 2 files changed, 15 insertions(+), 15 deletions(-)
>
> diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
> index 1e1a51510e83..9b03e164a37a 100644
> --- a/drivers/firmware/arm_sdei.c
> +++ b/drivers/firmware/arm_sdei.c
> @@ -301,8 +301,6 @@ int sdei_mask_local_cpu(void)
> {
> int err;
>
> - WARN_ON_ONCE(preemptible());
> -
> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_MASK, 0, 0, 0, 0, 0, NULL);
> if (err && err != -EIO) {
> pr_warn_once("failed to mask CPU[%u]: %d\n",
> @@ -315,6 +313,7 @@ int sdei_mask_local_cpu(void)
>
> static void _ipi_mask_cpu(void *ignored)
> {
> + WARN_ON_ONCE(preemptible());
> sdei_mask_local_cpu();
> }
>
> @@ -322,8 +321,6 @@ int sdei_unmask_local_cpu(void)
> {
> int err;
>
> - WARN_ON_ONCE(preemptible());
> -
> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_UNMASK, 0, 0, 0, 0, 0, NULL);
> if (err && err != -EIO) {
> pr_warn_once("failed to unmask CPU[%u]: %d\n",
> @@ -336,6 +333,7 @@ int sdei_unmask_local_cpu(void)
>
> static void _ipi_unmask_cpu(void *ignored)
> {
> + WARN_ON_ONCE(preemptible());
> sdei_unmask_local_cpu();
> }
>
> @@ -343,6 +341,8 @@ static void _ipi_private_reset(void *ignored)
> {
> int err;
>
> + WARN_ON_ONCE(preemptible());
> +
> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PRIVATE_RESET, 0, 0, 0, 0, 0,
> NULL);
> if (err && err != -EIO)
> @@ -389,8 +389,6 @@ static void _local_event_enable(void *data)
> int err;
> struct sdei_crosscall_args *arg = data;
>
> - WARN_ON_ONCE(preemptible());
> -
> err = sdei_api_event_enable(arg->event->event_num);
>
> sdei_cross_call_return(arg, err);
> @@ -401,6 +399,8 @@ int sdei_event_enable(u32 event_num)
> int err = -EINVAL;
> struct sdei_event *event;
>
> + WARN_ON_ONCE(preemptible());
> +
> mutex_lock(&sdei_events_lock);
> event = sdei_event_find(event_num);
> if (!event) {
> @@ -479,8 +479,6 @@ static void _local_event_unregister(void *data)
> int err;
> struct sdei_crosscall_args *arg = data;
>
> - WARN_ON_ONCE(preemptible());
> -
> err = sdei_api_event_unregister(arg->event->event_num);
>
> sdei_cross_call_return(arg, err);
> @@ -492,6 +490,7 @@ int sdei_event_unregister(u32 event_num)
> struct sdei_event *event;
>
> WARN_ON(in_nmi());
> + WARN_ON_ONCE(preemptible());
>
> mutex_lock(&sdei_events_lock);
> event = sdei_event_find(event_num);
> @@ -561,8 +560,6 @@ static void _local_event_register(void *data)
> struct sdei_registered_event *reg;
> struct sdei_crosscall_args *arg = data;
>
> - WARN_ON(preemptible());
> -
> reg = per_cpu_ptr(arg->event->private_registered, smp_processor_id());
> err = sdei_api_event_register(arg->event->event_num, sdei_entry_point,
> reg, 0, 0);
> @@ -576,6 +573,7 @@ int sdei_event_register(u32 event_num, sdei_event_callback *cb, void *arg)
> struct sdei_event *event;
>
> WARN_ON(in_nmi());
> + WARN_ON_ONCE(preemptible());
>
> mutex_lock(&sdei_events_lock);
> if (sdei_event_find(event_num)) {
> @@ -717,6 +715,8 @@ static int sdei_pm_notifier(struct notifier_block *nb, unsigned long action,
> {
> int rv;
>
> + WARN_ON_ONCE(preemptible());
> +
> switch (action) {
> case CPU_PM_ENTER:
> rv = sdei_mask_local_cpu();
> @@ -765,7 +765,7 @@ static int sdei_device_freeze(struct device *dev)
> int err;
>
> /* unregister private events */
> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
>
> err = sdei_unregister_shared();
> if (err)
> @@ -786,7 +786,7 @@ static int sdei_device_thaw(struct device *dev)
> return err;
> }
>
> - err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
> + err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
> &sdei_cpuhp_up, &sdei_cpuhp_down);
> if (err)
> pr_warn("Failed to re-register CPU hotplug notifier...\n");
> @@ -823,7 +823,7 @@ static int sdei_reboot_notifier(struct notifier_block *nb, unsigned long action,
> * We are going to reset the interface, after this there is no point
> * doing work when we take CPUs offline.
> */
> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
>
> sdei_platform_reset();
>
> @@ -1003,7 +1003,7 @@ static int sdei_probe(struct platform_device *pdev)
> goto remove_cpupm;
> }
>
> - err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
> + err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
> &sdei_cpuhp_up, &sdei_cpuhp_down);
> if (err) {
> pr_warn("Failed to register CPU hotplug notifier...\n");
> diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
> index f61447913db9..63b89e00a4c3 100644
> --- a/include/linux/cpuhotplug.h
> +++ b/include/linux/cpuhotplug.h
> @@ -161,7 +161,6 @@ enum cpuhp_state {
> CPUHP_AP_PERF_X86_CSTATE_STARTING,
> CPUHP_AP_PERF_XTENSA_STARTING,
> CPUHP_AP_MIPS_OP_LOONGSON3_STARTING,
> - CPUHP_AP_ARM_SDEI_STARTING,
> CPUHP_AP_ARM_VFP_STARTING,
> CPUHP_AP_ARM64_DEBUG_MONITORS_STARTING,
> CPUHP_AP_PERF_ARM_HW_BREAKPOINT_STARTING,
> @@ -209,6 +208,7 @@ enum cpuhp_state {
> CPUHP_AP_IRQ_AFFINITY_ONLINE,
> CPUHP_AP_BLK_MQ_ONLINE,
> CPUHP_AP_ARM_MVEBU_SYNC_CLOCKS,
> + CPUHP_AP_ARM_SDEI_ONLINE,
> CPUHP_AP_X86_INTEL_EPB_ONLINE,
> CPUHP_AP_PERF_ONLINE,
> CPUHP_AP_PERF_X86_ONLINE,
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
@ 2022-10-18 13:07 ` Pierre Gondois
0 siblings, 0 replies; 12+ messages in thread
From: Pierre Gondois @ 2022-10-18 13:07 UTC (permalink / raw)
To: linux-kernel
Cc: James Morse, Will Deacon, Huacai Chen, John Garry, Shaokun Zhang,
WANG Xuerui, Qi Liu, Bharat Bhushan, Huang Ying, Bibo Mao,
Jason A. Donenfeld, Daniel Lezcano, linux-arm-kernel
The changelog was removed:
v2:
- Don't use a raw spin_lock, but move the CPUHP_AP_ARM_SDEI_
cpuhp state to the _ONLINE section.
On 10/18/22 15:04, Pierre Gondois wrote:
> On an Ampere Altra,
> Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
> Ampere Altra triggers:
> [ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
> [ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
> [ 15.683157] preempt_count: 0, expected: 0
> [ 15.683159] RCU nest depth: 0, expected: 0
> [ 15.683163] 3 locks held by cpuhp/0/24:
> [ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
> [ 15.683224] irq event stamp: 36
> [ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
> [ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
> [ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
> [ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
> [ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
> [ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
> [ 15.683268] Call trace:
> [ 15.683271] dump_backtrace+0x114/0x120
> [ 15.683277] show_stack+0x20/0x70
> [ 15.683279] dump_stack_lvl+0x9c/0xd8
> [ 15.683288] dump_stack+0x18/0x34
> [ 15.683289] __might_resched+0x188/0x228
> [ 15.683292] rt_spin_lock+0x70/0x120
> [ 15.683301] sdei_cpuhp_up+0x3c/0x130
> [ 15.683303] cpuhp_invoke_callback+0x250/0xf08
> [ 15.683305] cpuhp_thread_fun+0x120/0x248
> [ 15.683308] smpboot_thread_fn+0x280/0x320
> [ 15.683315] kthread+0x130/0x140
> [ 15.683321] ret_from_fork+0x10/0x20
>
> sdei_cpuhp_up() is called in the STARTING hotplug section,
> which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
> state to the _ONLINE section to execute the cpuhp cb with
> preemption enabled.
>
> Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
> calling CPU. It is checked that preemption is disabled for them.
> _ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
> Preemption is enabled in those threads, but their cpumask is limited
> to 1 CPU.
> Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
> don't trigger them. This means that no check will be done for some
> cases, e.g. sdei_mask_local_cpu() invocations.
>
> Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
> which acts on the calling CPU.
>
> Suggested-by: James Morse <james.morse@arm.com>
> Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
> ---
> drivers/firmware/arm_sdei.c | 28 ++++++++++++++--------------
> include/linux/cpuhotplug.h | 2 +-
> 2 files changed, 15 insertions(+), 15 deletions(-)
>
> diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
> index 1e1a51510e83..9b03e164a37a 100644
> --- a/drivers/firmware/arm_sdei.c
> +++ b/drivers/firmware/arm_sdei.c
> @@ -301,8 +301,6 @@ int sdei_mask_local_cpu(void)
> {
> int err;
>
> - WARN_ON_ONCE(preemptible());
> -
> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_MASK, 0, 0, 0, 0, 0, NULL);
> if (err && err != -EIO) {
> pr_warn_once("failed to mask CPU[%u]: %d\n",
> @@ -315,6 +313,7 @@ int sdei_mask_local_cpu(void)
>
> static void _ipi_mask_cpu(void *ignored)
> {
> + WARN_ON_ONCE(preemptible());
> sdei_mask_local_cpu();
> }
>
> @@ -322,8 +321,6 @@ int sdei_unmask_local_cpu(void)
> {
> int err;
>
> - WARN_ON_ONCE(preemptible());
> -
> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_UNMASK, 0, 0, 0, 0, 0, NULL);
> if (err && err != -EIO) {
> pr_warn_once("failed to unmask CPU[%u]: %d\n",
> @@ -336,6 +333,7 @@ int sdei_unmask_local_cpu(void)
>
> static void _ipi_unmask_cpu(void *ignored)
> {
> + WARN_ON_ONCE(preemptible());
> sdei_unmask_local_cpu();
> }
>
> @@ -343,6 +341,8 @@ static void _ipi_private_reset(void *ignored)
> {
> int err;
>
> + WARN_ON_ONCE(preemptible());
> +
> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PRIVATE_RESET, 0, 0, 0, 0, 0,
> NULL);
> if (err && err != -EIO)
> @@ -389,8 +389,6 @@ static void _local_event_enable(void *data)
> int err;
> struct sdei_crosscall_args *arg = data;
>
> - WARN_ON_ONCE(preemptible());
> -
> err = sdei_api_event_enable(arg->event->event_num);
>
> sdei_cross_call_return(arg, err);
> @@ -401,6 +399,8 @@ int sdei_event_enable(u32 event_num)
> int err = -EINVAL;
> struct sdei_event *event;
>
> + WARN_ON_ONCE(preemptible());
> +
> mutex_lock(&sdei_events_lock);
> event = sdei_event_find(event_num);
> if (!event) {
> @@ -479,8 +479,6 @@ static void _local_event_unregister(void *data)
> int err;
> struct sdei_crosscall_args *arg = data;
>
> - WARN_ON_ONCE(preemptible());
> -
> err = sdei_api_event_unregister(arg->event->event_num);
>
> sdei_cross_call_return(arg, err);
> @@ -492,6 +490,7 @@ int sdei_event_unregister(u32 event_num)
> struct sdei_event *event;
>
> WARN_ON(in_nmi());
> + WARN_ON_ONCE(preemptible());
>
> mutex_lock(&sdei_events_lock);
> event = sdei_event_find(event_num);
> @@ -561,8 +560,6 @@ static void _local_event_register(void *data)
> struct sdei_registered_event *reg;
> struct sdei_crosscall_args *arg = data;
>
> - WARN_ON(preemptible());
> -
> reg = per_cpu_ptr(arg->event->private_registered, smp_processor_id());
> err = sdei_api_event_register(arg->event->event_num, sdei_entry_point,
> reg, 0, 0);
> @@ -576,6 +573,7 @@ int sdei_event_register(u32 event_num, sdei_event_callback *cb, void *arg)
> struct sdei_event *event;
>
> WARN_ON(in_nmi());
> + WARN_ON_ONCE(preemptible());
>
> mutex_lock(&sdei_events_lock);
> if (sdei_event_find(event_num)) {
> @@ -717,6 +715,8 @@ static int sdei_pm_notifier(struct notifier_block *nb, unsigned long action,
> {
> int rv;
>
> + WARN_ON_ONCE(preemptible());
> +
> switch (action) {
> case CPU_PM_ENTER:
> rv = sdei_mask_local_cpu();
> @@ -765,7 +765,7 @@ static int sdei_device_freeze(struct device *dev)
> int err;
>
> /* unregister private events */
> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
>
> err = sdei_unregister_shared();
> if (err)
> @@ -786,7 +786,7 @@ static int sdei_device_thaw(struct device *dev)
> return err;
> }
>
> - err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
> + err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
> &sdei_cpuhp_up, &sdei_cpuhp_down);
> if (err)
> pr_warn("Failed to re-register CPU hotplug notifier...\n");
> @@ -823,7 +823,7 @@ static int sdei_reboot_notifier(struct notifier_block *nb, unsigned long action,
> * We are going to reset the interface, after this there is no point
> * doing work when we take CPUs offline.
> */
> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
>
> sdei_platform_reset();
>
> @@ -1003,7 +1003,7 @@ static int sdei_probe(struct platform_device *pdev)
> goto remove_cpupm;
> }
>
> - err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
> + err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
> &sdei_cpuhp_up, &sdei_cpuhp_down);
> if (err) {
> pr_warn("Failed to register CPU hotplug notifier...\n");
> diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
> index f61447913db9..63b89e00a4c3 100644
> --- a/include/linux/cpuhotplug.h
> +++ b/include/linux/cpuhotplug.h
> @@ -161,7 +161,6 @@ enum cpuhp_state {
> CPUHP_AP_PERF_X86_CSTATE_STARTING,
> CPUHP_AP_PERF_XTENSA_STARTING,
> CPUHP_AP_MIPS_OP_LOONGSON3_STARTING,
> - CPUHP_AP_ARM_SDEI_STARTING,
> CPUHP_AP_ARM_VFP_STARTING,
> CPUHP_AP_ARM64_DEBUG_MONITORS_STARTING,
> CPUHP_AP_PERF_ARM_HW_BREAKPOINT_STARTING,
> @@ -209,6 +208,7 @@ enum cpuhp_state {
> CPUHP_AP_IRQ_AFFINITY_ONLINE,
> CPUHP_AP_BLK_MQ_ONLINE,
> CPUHP_AP_ARM_MVEBU_SYNC_CLOCKS,
> + CPUHP_AP_ARM_SDEI_ONLINE,
> CPUHP_AP_X86_INTEL_EPB_ONLINE,
> CPUHP_AP_PERF_ONLINE,
> CPUHP_AP_PERF_X86_ONLINE,
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
2022-10-18 13:04 ` Pierre Gondois
@ 2022-11-07 16:14 ` Will Deacon
-1 siblings, 0 replies; 12+ messages in thread
From: Will Deacon @ 2022-11-07 16:14 UTC (permalink / raw)
To: Pierre Gondois
Cc: linux-kernel, James Morse, Huacai Chen, John Garry,
Shaokun Zhang, WANG Xuerui, Qi Liu, Bharat Bhushan, Huang Ying,
Bibo Mao, Jason A. Donenfeld, Daniel Lezcano, linux-arm-kernel
On Tue, Oct 18, 2022 at 03:04:53PM +0200, Pierre Gondois wrote:
> On an Ampere Altra,
> Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
> Ampere Altra triggers:
> [ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
> [ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
> [ 15.683157] preempt_count: 0, expected: 0
> [ 15.683159] RCU nest depth: 0, expected: 0
> [ 15.683163] 3 locks held by cpuhp/0/24:
> [ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
> [ 15.683224] irq event stamp: 36
> [ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
> [ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
> [ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
> [ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
> [ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
> [ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
> [ 15.683268] Call trace:
> [ 15.683271] dump_backtrace+0x114/0x120
> [ 15.683277] show_stack+0x20/0x70
> [ 15.683279] dump_stack_lvl+0x9c/0xd8
> [ 15.683288] dump_stack+0x18/0x34
> [ 15.683289] __might_resched+0x188/0x228
> [ 15.683292] rt_spin_lock+0x70/0x120
> [ 15.683301] sdei_cpuhp_up+0x3c/0x130
> [ 15.683303] cpuhp_invoke_callback+0x250/0xf08
> [ 15.683305] cpuhp_thread_fun+0x120/0x248
> [ 15.683308] smpboot_thread_fn+0x280/0x320
> [ 15.683315] kthread+0x130/0x140
> [ 15.683321] ret_from_fork+0x10/0x20
>
> sdei_cpuhp_up() is called in the STARTING hotplug section,
> which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
> state to the _ONLINE section to execute the cpuhp cb with
> preemption enabled.
>
> Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
> calling CPU. It is checked that preemption is disabled for them.
> _ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
> Preemption is enabled in those threads, but their cpumask is limited
> to 1 CPU.
> Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
> don't trigger them. This means that no check will be done for some
> cases, e.g. sdei_mask_local_cpu() invocations.
>
> Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
> which acts on the calling CPU.
>
> Suggested-by: James Morse <james.morse@arm.com>
> Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
> ---
> drivers/firmware/arm_sdei.c | 28 ++++++++++++++--------------
> include/linux/cpuhotplug.h | 2 +-
> 2 files changed, 15 insertions(+), 15 deletions(-)
James -- are you taking care of this one, or should I queue it via arm64?
Will
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
@ 2022-11-07 16:14 ` Will Deacon
0 siblings, 0 replies; 12+ messages in thread
From: Will Deacon @ 2022-11-07 16:14 UTC (permalink / raw)
To: Pierre Gondois
Cc: linux-kernel, James Morse, Huacai Chen, John Garry,
Shaokun Zhang, WANG Xuerui, Qi Liu, Bharat Bhushan, Huang Ying,
Bibo Mao, Jason A. Donenfeld, Daniel Lezcano, linux-arm-kernel
On Tue, Oct 18, 2022 at 03:04:53PM +0200, Pierre Gondois wrote:
> On an Ampere Altra,
> Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
> Ampere Altra triggers:
> [ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
> [ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
> [ 15.683157] preempt_count: 0, expected: 0
> [ 15.683159] RCU nest depth: 0, expected: 0
> [ 15.683163] 3 locks held by cpuhp/0/24:
> [ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
> [ 15.683224] irq event stamp: 36
> [ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
> [ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
> [ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
> [ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
> [ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
> [ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
> [ 15.683268] Call trace:
> [ 15.683271] dump_backtrace+0x114/0x120
> [ 15.683277] show_stack+0x20/0x70
> [ 15.683279] dump_stack_lvl+0x9c/0xd8
> [ 15.683288] dump_stack+0x18/0x34
> [ 15.683289] __might_resched+0x188/0x228
> [ 15.683292] rt_spin_lock+0x70/0x120
> [ 15.683301] sdei_cpuhp_up+0x3c/0x130
> [ 15.683303] cpuhp_invoke_callback+0x250/0xf08
> [ 15.683305] cpuhp_thread_fun+0x120/0x248
> [ 15.683308] smpboot_thread_fn+0x280/0x320
> [ 15.683315] kthread+0x130/0x140
> [ 15.683321] ret_from_fork+0x10/0x20
>
> sdei_cpuhp_up() is called in the STARTING hotplug section,
> which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
> state to the _ONLINE section to execute the cpuhp cb with
> preemption enabled.
>
> Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
> calling CPU. It is checked that preemption is disabled for them.
> _ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
> Preemption is enabled in those threads, but their cpumask is limited
> to 1 CPU.
> Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
> don't trigger them. This means that no check will be done for some
> cases, e.g. sdei_mask_local_cpu() invocations.
>
> Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
> which acts on the calling CPU.
>
> Suggested-by: James Morse <james.morse@arm.com>
> Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
> ---
> drivers/firmware/arm_sdei.c | 28 ++++++++++++++--------------
> include/linux/cpuhotplug.h | 2 +-
> 2 files changed, 15 insertions(+), 15 deletions(-)
James -- are you taking care of this one, or should I queue it via arm64?
Will
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
2022-10-18 13:07 ` Pierre Gondois
@ 2023-02-01 16:23 ` Pierre Gondois
-1 siblings, 0 replies; 12+ messages in thread
From: Pierre Gondois @ 2023-02-01 16:23 UTC (permalink / raw)
To: linux-kernel
Cc: James Morse, Will Deacon, Huacai Chen, John Garry, Shaokun Zhang,
WANG Xuerui, Qi Liu, Bharat Bhushan, Huang Ying, Bibo Mao,
Jason A. Donenfeld, Daniel Lezcano, linux-arm-kernel
Hello James,
Just a ping in case this was forgotten,
Regards,
Pierre
On 10/18/22 15:07, Pierre Gondois wrote:
> The changelog was removed:
>
> v2:
> - Don't use a raw spin_lock, but move the CPUHP_AP_ARM_SDEI_
> cpuhp state to the _ONLINE section.
>
>
> On 10/18/22 15:04, Pierre Gondois wrote:
>> On an Ampere Altra,
>> Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
>> Ampere Altra triggers:
>> [ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
>> [ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
>> [ 15.683157] preempt_count: 0, expected: 0
>> [ 15.683159] RCU nest depth: 0, expected: 0
>> [ 15.683163] 3 locks held by cpuhp/0/24:
>> [ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
>> [ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
>> [ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
>> [ 15.683224] irq event stamp: 36
>> [ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
>> [ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
>> [ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
>> [ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
>> [ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
>> [ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
>> [ 15.683268] Call trace:
>> [ 15.683271] dump_backtrace+0x114/0x120
>> [ 15.683277] show_stack+0x20/0x70
>> [ 15.683279] dump_stack_lvl+0x9c/0xd8
>> [ 15.683288] dump_stack+0x18/0x34
>> [ 15.683289] __might_resched+0x188/0x228
>> [ 15.683292] rt_spin_lock+0x70/0x120
>> [ 15.683301] sdei_cpuhp_up+0x3c/0x130
>> [ 15.683303] cpuhp_invoke_callback+0x250/0xf08
>> [ 15.683305] cpuhp_thread_fun+0x120/0x248
>> [ 15.683308] smpboot_thread_fn+0x280/0x320
>> [ 15.683315] kthread+0x130/0x140
>> [ 15.683321] ret_from_fork+0x10/0x20
>>
>> sdei_cpuhp_up() is called in the STARTING hotplug section,
>> which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
>> state to the _ONLINE section to execute the cpuhp cb with
>> preemption enabled.
>>
>> Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
>> calling CPU. It is checked that preemption is disabled for them.
>> _ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
>> Preemption is enabled in those threads, but their cpumask is limited
>> to 1 CPU.
>> Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
>> don't trigger them. This means that no check will be done for some
>> cases, e.g. sdei_mask_local_cpu() invocations.
>>
>> Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
>> which acts on the calling CPU.
>>
>> Suggested-by: James Morse <james.morse@arm.com>
>> Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
>> ---
>> drivers/firmware/arm_sdei.c | 28 ++++++++++++++--------------
>> include/linux/cpuhotplug.h | 2 +-
>> 2 files changed, 15 insertions(+), 15 deletions(-)
>>
>> diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
>> index 1e1a51510e83..9b03e164a37a 100644
>> --- a/drivers/firmware/arm_sdei.c
>> +++ b/drivers/firmware/arm_sdei.c
>> @@ -301,8 +301,6 @@ int sdei_mask_local_cpu(void)
>> {
>> int err;
>>
>> - WARN_ON_ONCE(preemptible());
>> -
>> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_MASK, 0, 0, 0, 0, 0, NULL);
>> if (err && err != -EIO) {
>> pr_warn_once("failed to mask CPU[%u]: %d\n",
>> @@ -315,6 +313,7 @@ int sdei_mask_local_cpu(void)
>>
>> static void _ipi_mask_cpu(void *ignored)
>> {
>> + WARN_ON_ONCE(preemptible());
>> sdei_mask_local_cpu();
>> }
>>
>> @@ -322,8 +321,6 @@ int sdei_unmask_local_cpu(void)
>> {
>> int err;
>>
>> - WARN_ON_ONCE(preemptible());
>> -
>> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_UNMASK, 0, 0, 0, 0, 0, NULL);
>> if (err && err != -EIO) {
>> pr_warn_once("failed to unmask CPU[%u]: %d\n",
>> @@ -336,6 +333,7 @@ int sdei_unmask_local_cpu(void)
>>
>> static void _ipi_unmask_cpu(void *ignored)
>> {
>> + WARN_ON_ONCE(preemptible());
>> sdei_unmask_local_cpu();
>> }
>>
>> @@ -343,6 +341,8 @@ static void _ipi_private_reset(void *ignored)
>> {
>> int err;
>>
>> + WARN_ON_ONCE(preemptible());
>> +
>> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PRIVATE_RESET, 0, 0, 0, 0, 0,
>> NULL);
>> if (err && err != -EIO)
>> @@ -389,8 +389,6 @@ static void _local_event_enable(void *data)
>> int err;
>> struct sdei_crosscall_args *arg = data;
>>
>> - WARN_ON_ONCE(preemptible());
>> -
>> err = sdei_api_event_enable(arg->event->event_num);
>>
>> sdei_cross_call_return(arg, err);
>> @@ -401,6 +399,8 @@ int sdei_event_enable(u32 event_num)
>> int err = -EINVAL;
>> struct sdei_event *event;
>>
>> + WARN_ON_ONCE(preemptible());
>> +
>> mutex_lock(&sdei_events_lock);
>> event = sdei_event_find(event_num);
>> if (!event) {
>> @@ -479,8 +479,6 @@ static void _local_event_unregister(void *data)
>> int err;
>> struct sdei_crosscall_args *arg = data;
>>
>> - WARN_ON_ONCE(preemptible());
>> -
>> err = sdei_api_event_unregister(arg->event->event_num);
>>
>> sdei_cross_call_return(arg, err);
>> @@ -492,6 +490,7 @@ int sdei_event_unregister(u32 event_num)
>> struct sdei_event *event;
>>
>> WARN_ON(in_nmi());
>> + WARN_ON_ONCE(preemptible());
>>
>> mutex_lock(&sdei_events_lock);
>> event = sdei_event_find(event_num);
>> @@ -561,8 +560,6 @@ static void _local_event_register(void *data)
>> struct sdei_registered_event *reg;
>> struct sdei_crosscall_args *arg = data;
>>
>> - WARN_ON(preemptible());
>> -
>> reg = per_cpu_ptr(arg->event->private_registered, smp_processor_id());
>> err = sdei_api_event_register(arg->event->event_num, sdei_entry_point,
>> reg, 0, 0);
>> @@ -576,6 +573,7 @@ int sdei_event_register(u32 event_num, sdei_event_callback *cb, void *arg)
>> struct sdei_event *event;
>>
>> WARN_ON(in_nmi());
>> + WARN_ON_ONCE(preemptible());
>>
>> mutex_lock(&sdei_events_lock);
>> if (sdei_event_find(event_num)) {
>> @@ -717,6 +715,8 @@ static int sdei_pm_notifier(struct notifier_block *nb, unsigned long action,
>> {
>> int rv;
>>
>> + WARN_ON_ONCE(preemptible());
>> +
>> switch (action) {
>> case CPU_PM_ENTER:
>> rv = sdei_mask_local_cpu();
>> @@ -765,7 +765,7 @@ static int sdei_device_freeze(struct device *dev)
>> int err;
>>
>> /* unregister private events */
>> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
>> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
>>
>> err = sdei_unregister_shared();
>> if (err)
>> @@ -786,7 +786,7 @@ static int sdei_device_thaw(struct device *dev)
>> return err;
>> }
>>
>> - err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
>> + err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
>> &sdei_cpuhp_up, &sdei_cpuhp_down);
>> if (err)
>> pr_warn("Failed to re-register CPU hotplug notifier...\n");
>> @@ -823,7 +823,7 @@ static int sdei_reboot_notifier(struct notifier_block *nb, unsigned long action,
>> * We are going to reset the interface, after this there is no point
>> * doing work when we take CPUs offline.
>> */
>> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
>> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
>>
>> sdei_platform_reset();
>>
>> @@ -1003,7 +1003,7 @@ static int sdei_probe(struct platform_device *pdev)
>> goto remove_cpupm;
>> }
>>
>> - err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
>> + err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
>> &sdei_cpuhp_up, &sdei_cpuhp_down);
>> if (err) {
>> pr_warn("Failed to register CPU hotplug notifier...\n");
>> diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
>> index f61447913db9..63b89e00a4c3 100644
>> --- a/include/linux/cpuhotplug.h
>> +++ b/include/linux/cpuhotplug.h
>> @@ -161,7 +161,6 @@ enum cpuhp_state {
>> CPUHP_AP_PERF_X86_CSTATE_STARTING,
>> CPUHP_AP_PERF_XTENSA_STARTING,
>> CPUHP_AP_MIPS_OP_LOONGSON3_STARTING,
>> - CPUHP_AP_ARM_SDEI_STARTING,
>> CPUHP_AP_ARM_VFP_STARTING,
>> CPUHP_AP_ARM64_DEBUG_MONITORS_STARTING,
>> CPUHP_AP_PERF_ARM_HW_BREAKPOINT_STARTING,
>> @@ -209,6 +208,7 @@ enum cpuhp_state {
>> CPUHP_AP_IRQ_AFFINITY_ONLINE,
>> CPUHP_AP_BLK_MQ_ONLINE,
>> CPUHP_AP_ARM_MVEBU_SYNC_CLOCKS,
>> + CPUHP_AP_ARM_SDEI_ONLINE,
>> CPUHP_AP_X86_INTEL_EPB_ONLINE,
>> CPUHP_AP_PERF_ONLINE,
>> CPUHP_AP_PERF_X86_ONLINE,
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
@ 2023-02-01 16:23 ` Pierre Gondois
0 siblings, 0 replies; 12+ messages in thread
From: Pierre Gondois @ 2023-02-01 16:23 UTC (permalink / raw)
To: linux-kernel
Cc: James Morse, Will Deacon, Huacai Chen, John Garry, Shaokun Zhang,
WANG Xuerui, Qi Liu, Bharat Bhushan, Huang Ying, Bibo Mao,
Jason A. Donenfeld, Daniel Lezcano, linux-arm-kernel
Hello James,
Just a ping in case this was forgotten,
Regards,
Pierre
On 10/18/22 15:07, Pierre Gondois wrote:
> The changelog was removed:
>
> v2:
> - Don't use a raw spin_lock, but move the CPUHP_AP_ARM_SDEI_
> cpuhp state to the _ONLINE section.
>
>
> On 10/18/22 15:04, Pierre Gondois wrote:
>> On an Ampere Altra,
>> Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
>> Ampere Altra triggers:
>> [ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
>> [ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
>> [ 15.683157] preempt_count: 0, expected: 0
>> [ 15.683159] RCU nest depth: 0, expected: 0
>> [ 15.683163] 3 locks held by cpuhp/0/24:
>> [ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
>> [ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
>> [ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
>> [ 15.683224] irq event stamp: 36
>> [ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
>> [ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
>> [ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
>> [ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
>> [ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
>> [ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
>> [ 15.683268] Call trace:
>> [ 15.683271] dump_backtrace+0x114/0x120
>> [ 15.683277] show_stack+0x20/0x70
>> [ 15.683279] dump_stack_lvl+0x9c/0xd8
>> [ 15.683288] dump_stack+0x18/0x34
>> [ 15.683289] __might_resched+0x188/0x228
>> [ 15.683292] rt_spin_lock+0x70/0x120
>> [ 15.683301] sdei_cpuhp_up+0x3c/0x130
>> [ 15.683303] cpuhp_invoke_callback+0x250/0xf08
>> [ 15.683305] cpuhp_thread_fun+0x120/0x248
>> [ 15.683308] smpboot_thread_fn+0x280/0x320
>> [ 15.683315] kthread+0x130/0x140
>> [ 15.683321] ret_from_fork+0x10/0x20
>>
>> sdei_cpuhp_up() is called in the STARTING hotplug section,
>> which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
>> state to the _ONLINE section to execute the cpuhp cb with
>> preemption enabled.
>>
>> Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
>> calling CPU. It is checked that preemption is disabled for them.
>> _ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
>> Preemption is enabled in those threads, but their cpumask is limited
>> to 1 CPU.
>> Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
>> don't trigger them. This means that no check will be done for some
>> cases, e.g. sdei_mask_local_cpu() invocations.
>>
>> Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
>> which acts on the calling CPU.
>>
>> Suggested-by: James Morse <james.morse@arm.com>
>> Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
>> ---
>> drivers/firmware/arm_sdei.c | 28 ++++++++++++++--------------
>> include/linux/cpuhotplug.h | 2 +-
>> 2 files changed, 15 insertions(+), 15 deletions(-)
>>
>> diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
>> index 1e1a51510e83..9b03e164a37a 100644
>> --- a/drivers/firmware/arm_sdei.c
>> +++ b/drivers/firmware/arm_sdei.c
>> @@ -301,8 +301,6 @@ int sdei_mask_local_cpu(void)
>> {
>> int err;
>>
>> - WARN_ON_ONCE(preemptible());
>> -
>> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_MASK, 0, 0, 0, 0, 0, NULL);
>> if (err && err != -EIO) {
>> pr_warn_once("failed to mask CPU[%u]: %d\n",
>> @@ -315,6 +313,7 @@ int sdei_mask_local_cpu(void)
>>
>> static void _ipi_mask_cpu(void *ignored)
>> {
>> + WARN_ON_ONCE(preemptible());
>> sdei_mask_local_cpu();
>> }
>>
>> @@ -322,8 +321,6 @@ int sdei_unmask_local_cpu(void)
>> {
>> int err;
>>
>> - WARN_ON_ONCE(preemptible());
>> -
>> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PE_UNMASK, 0, 0, 0, 0, 0, NULL);
>> if (err && err != -EIO) {
>> pr_warn_once("failed to unmask CPU[%u]: %d\n",
>> @@ -336,6 +333,7 @@ int sdei_unmask_local_cpu(void)
>>
>> static void _ipi_unmask_cpu(void *ignored)
>> {
>> + WARN_ON_ONCE(preemptible());
>> sdei_unmask_local_cpu();
>> }
>>
>> @@ -343,6 +341,8 @@ static void _ipi_private_reset(void *ignored)
>> {
>> int err;
>>
>> + WARN_ON_ONCE(preemptible());
>> +
>> err = invoke_sdei_fn(SDEI_1_0_FN_SDEI_PRIVATE_RESET, 0, 0, 0, 0, 0,
>> NULL);
>> if (err && err != -EIO)
>> @@ -389,8 +389,6 @@ static void _local_event_enable(void *data)
>> int err;
>> struct sdei_crosscall_args *arg = data;
>>
>> - WARN_ON_ONCE(preemptible());
>> -
>> err = sdei_api_event_enable(arg->event->event_num);
>>
>> sdei_cross_call_return(arg, err);
>> @@ -401,6 +399,8 @@ int sdei_event_enable(u32 event_num)
>> int err = -EINVAL;
>> struct sdei_event *event;
>>
>> + WARN_ON_ONCE(preemptible());
>> +
>> mutex_lock(&sdei_events_lock);
>> event = sdei_event_find(event_num);
>> if (!event) {
>> @@ -479,8 +479,6 @@ static void _local_event_unregister(void *data)
>> int err;
>> struct sdei_crosscall_args *arg = data;
>>
>> - WARN_ON_ONCE(preemptible());
>> -
>> err = sdei_api_event_unregister(arg->event->event_num);
>>
>> sdei_cross_call_return(arg, err);
>> @@ -492,6 +490,7 @@ int sdei_event_unregister(u32 event_num)
>> struct sdei_event *event;
>>
>> WARN_ON(in_nmi());
>> + WARN_ON_ONCE(preemptible());
>>
>> mutex_lock(&sdei_events_lock);
>> event = sdei_event_find(event_num);
>> @@ -561,8 +560,6 @@ static void _local_event_register(void *data)
>> struct sdei_registered_event *reg;
>> struct sdei_crosscall_args *arg = data;
>>
>> - WARN_ON(preemptible());
>> -
>> reg = per_cpu_ptr(arg->event->private_registered, smp_processor_id());
>> err = sdei_api_event_register(arg->event->event_num, sdei_entry_point,
>> reg, 0, 0);
>> @@ -576,6 +573,7 @@ int sdei_event_register(u32 event_num, sdei_event_callback *cb, void *arg)
>> struct sdei_event *event;
>>
>> WARN_ON(in_nmi());
>> + WARN_ON_ONCE(preemptible());
>>
>> mutex_lock(&sdei_events_lock);
>> if (sdei_event_find(event_num)) {
>> @@ -717,6 +715,8 @@ static int sdei_pm_notifier(struct notifier_block *nb, unsigned long action,
>> {
>> int rv;
>>
>> + WARN_ON_ONCE(preemptible());
>> +
>> switch (action) {
>> case CPU_PM_ENTER:
>> rv = sdei_mask_local_cpu();
>> @@ -765,7 +765,7 @@ static int sdei_device_freeze(struct device *dev)
>> int err;
>>
>> /* unregister private events */
>> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
>> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
>>
>> err = sdei_unregister_shared();
>> if (err)
>> @@ -786,7 +786,7 @@ static int sdei_device_thaw(struct device *dev)
>> return err;
>> }
>>
>> - err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
>> + err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
>> &sdei_cpuhp_up, &sdei_cpuhp_down);
>> if (err)
>> pr_warn("Failed to re-register CPU hotplug notifier...\n");
>> @@ -823,7 +823,7 @@ static int sdei_reboot_notifier(struct notifier_block *nb, unsigned long action,
>> * We are going to reset the interface, after this there is no point
>> * doing work when we take CPUs offline.
>> */
>> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
>> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
>>
>> sdei_platform_reset();
>>
>> @@ -1003,7 +1003,7 @@ static int sdei_probe(struct platform_device *pdev)
>> goto remove_cpupm;
>> }
>>
>> - err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_STARTING, "SDEI",
>> + err = cpuhp_setup_state(CPUHP_AP_ARM_SDEI_ONLINE, "SDEI",
>> &sdei_cpuhp_up, &sdei_cpuhp_down);
>> if (err) {
>> pr_warn("Failed to register CPU hotplug notifier...\n");
>> diff --git a/include/linux/cpuhotplug.h b/include/linux/cpuhotplug.h
>> index f61447913db9..63b89e00a4c3 100644
>> --- a/include/linux/cpuhotplug.h
>> +++ b/include/linux/cpuhotplug.h
>> @@ -161,7 +161,6 @@ enum cpuhp_state {
>> CPUHP_AP_PERF_X86_CSTATE_STARTING,
>> CPUHP_AP_PERF_XTENSA_STARTING,
>> CPUHP_AP_MIPS_OP_LOONGSON3_STARTING,
>> - CPUHP_AP_ARM_SDEI_STARTING,
>> CPUHP_AP_ARM_VFP_STARTING,
>> CPUHP_AP_ARM64_DEBUG_MONITORS_STARTING,
>> CPUHP_AP_PERF_ARM_HW_BREAKPOINT_STARTING,
>> @@ -209,6 +208,7 @@ enum cpuhp_state {
>> CPUHP_AP_IRQ_AFFINITY_ONLINE,
>> CPUHP_AP_BLK_MQ_ONLINE,
>> CPUHP_AP_ARM_MVEBU_SYNC_CLOCKS,
>> + CPUHP_AP_ARM_SDEI_ONLINE,
>> CPUHP_AP_X86_INTEL_EPB_ONLINE,
>> CPUHP_AP_PERF_ONLINE,
>> CPUHP_AP_PERF_X86_ONLINE,
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
2022-10-18 13:04 ` Pierre Gondois
@ 2023-02-10 17:42 ` James Morse
-1 siblings, 0 replies; 12+ messages in thread
From: James Morse @ 2023-02-10 17:42 UTC (permalink / raw)
To: Pierre Gondois, linux-kernel
Cc: Will Deacon, Huacai Chen, John Garry, Shaokun Zhang, WANG Xuerui,
Qi Liu, Bharat Bhushan, Huang Ying, Bibo Mao, Jason A. Donenfeld,
Daniel Lezcano, linux-arm-kernel
Hi Pierre,
Sorry its taken so long for me to catch up with this ...
On 18/10/2022 14:04, Pierre Gondois wrote:
> On an Ampere Altra,
> Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
> Ampere Altra triggers:
> [ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
> [ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
> [ 15.683157] preempt_count: 0, expected: 0
> [ 15.683159] RCU nest depth: 0, expected: 0
> [ 15.683163] 3 locks held by cpuhp/0/24:
> [ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
> [ 15.683224] irq event stamp: 36
> [ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
> [ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
> [ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
> [ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
> [ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
> [ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
> [ 15.683268] Call trace:
> [ 15.683271] dump_backtrace+0x114/0x120
> [ 15.683277] show_stack+0x20/0x70
> [ 15.683279] dump_stack_lvl+0x9c/0xd8
> [ 15.683288] dump_stack+0x18/0x34
> [ 15.683289] __might_resched+0x188/0x228
> [ 15.683292] rt_spin_lock+0x70/0x120
> [ 15.683301] sdei_cpuhp_up+0x3c/0x130
> [ 15.683303] cpuhp_invoke_callback+0x250/0xf08
> [ 15.683305] cpuhp_thread_fun+0x120/0x248
> [ 15.683308] smpboot_thread_fn+0x280/0x320
> [ 15.683315] kthread+0x130/0x140
> [ 15.683321] ret_from_fork+0x10/0x20
> sdei_cpuhp_up() is called in the STARTING hotplug section,
> which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
> state to the _ONLINE section to execute the cpuhp cb with
> preemption enabled.
The background to this is SDEI got its own cpuhp slot because 'perf NMI' support
was one of the use-cases, but this got superseded by pNMI. Without an interaction with
perf, the slot doesn't need to be that early.
> Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
> calling CPU. It is checked that preemption is disabled for them.
> _ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
> Preemption is enabled in those threads, but their cpumask is limited
> to 1 CPU.
> Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
> don't trigger them. This means that no check will be done for some
> cases, e.g. sdei_mask_local_cpu() invocations.
(these things are documentation anyway)
> Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
> which acts on the calling CPU.
> diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
> index 1e1a51510e83..9b03e164a37a 100644
> --- a/drivers/firmware/arm_sdei.c
> +++ b/drivers/firmware/arm_sdei.c
> @@ -401,6 +399,8 @@ int sdei_event_enable(u32 event_num)
> int err = -EINVAL;
> struct sdei_event *event;
>
> + WARN_ON_ONCE(preemptible());
> +
> mutex_lock(&sdei_events_lock);
> event = sdei_event_find(event_num);
> if (!event) {
This doesn't look right. How can this code take a mutex if its in a non-preemptable context?
> @@ -492,6 +490,7 @@ int sdei_event_unregister(u32 event_num)
> struct sdei_event *event;
>
> WARN_ON(in_nmi());
> + WARN_ON_ONCE(preemptible());
>
> mutex_lock(&sdei_events_lock);
> event = sdei_event_find(event_num);
Same again.
> @@ -576,6 +573,7 @@ int sdei_event_register(u32 event_num, sdei_event_callback *cb, void *arg)
> struct sdei_event *event;
>
> WARN_ON(in_nmi());
> + WARN_ON_ONCE(preemptible());
>
> mutex_lock(&sdei_events_lock);
> if (sdei_event_find(event_num)) {
Same again.
I think you've copied these here because they called _local_event_unregister(), but they
did that via on_each_cpu(), which is what made _local_event_unregister() non-preemtable.
You can just remove these three, the original warnings were mostly documentation, and to
catch myself out. (before RT moved the goal posts).
With that:
Reviewed-by: James Morse <james.morse@arm.com>
(could you trim the [timestamps] out of the commit log?)
> @@ -765,7 +765,7 @@ static int sdei_device_freeze(struct device *dev)
> int err;
>
> /* unregister private events */
> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
Is there any mileage in making this CPUHP_AP_ONLINE_DYN ?
Perf really was the only reason that this needed to happen in any particular order.
Thanks,
James
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
@ 2023-02-10 17:42 ` James Morse
0 siblings, 0 replies; 12+ messages in thread
From: James Morse @ 2023-02-10 17:42 UTC (permalink / raw)
To: Pierre Gondois, linux-kernel
Cc: Will Deacon, Huacai Chen, John Garry, Shaokun Zhang, WANG Xuerui,
Qi Liu, Bharat Bhushan, Huang Ying, Bibo Mao, Jason A. Donenfeld,
Daniel Lezcano, linux-arm-kernel
Hi Pierre,
Sorry its taken so long for me to catch up with this ...
On 18/10/2022 14:04, Pierre Gondois wrote:
> On an Ampere Altra,
> Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
> Ampere Altra triggers:
> [ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
> [ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
> [ 15.683157] preempt_count: 0, expected: 0
> [ 15.683159] RCU nest depth: 0, expected: 0
> [ 15.683163] 3 locks held by cpuhp/0/24:
> [ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
> [ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
> [ 15.683224] irq event stamp: 36
> [ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
> [ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
> [ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
> [ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
> [ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
> [ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
> [ 15.683268] Call trace:
> [ 15.683271] dump_backtrace+0x114/0x120
> [ 15.683277] show_stack+0x20/0x70
> [ 15.683279] dump_stack_lvl+0x9c/0xd8
> [ 15.683288] dump_stack+0x18/0x34
> [ 15.683289] __might_resched+0x188/0x228
> [ 15.683292] rt_spin_lock+0x70/0x120
> [ 15.683301] sdei_cpuhp_up+0x3c/0x130
> [ 15.683303] cpuhp_invoke_callback+0x250/0xf08
> [ 15.683305] cpuhp_thread_fun+0x120/0x248
> [ 15.683308] smpboot_thread_fn+0x280/0x320
> [ 15.683315] kthread+0x130/0x140
> [ 15.683321] ret_from_fork+0x10/0x20
> sdei_cpuhp_up() is called in the STARTING hotplug section,
> which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
> state to the _ONLINE section to execute the cpuhp cb with
> preemption enabled.
The background to this is SDEI got its own cpuhp slot because 'perf NMI' support
was one of the use-cases, but this got superseded by pNMI. Without an interaction with
perf, the slot doesn't need to be that early.
> Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
> calling CPU. It is checked that preemption is disabled for them.
> _ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
> Preemption is enabled in those threads, but their cpumask is limited
> to 1 CPU.
> Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
> don't trigger them. This means that no check will be done for some
> cases, e.g. sdei_mask_local_cpu() invocations.
(these things are documentation anyway)
> Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
> which acts on the calling CPU.
> diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
> index 1e1a51510e83..9b03e164a37a 100644
> --- a/drivers/firmware/arm_sdei.c
> +++ b/drivers/firmware/arm_sdei.c
> @@ -401,6 +399,8 @@ int sdei_event_enable(u32 event_num)
> int err = -EINVAL;
> struct sdei_event *event;
>
> + WARN_ON_ONCE(preemptible());
> +
> mutex_lock(&sdei_events_lock);
> event = sdei_event_find(event_num);
> if (!event) {
This doesn't look right. How can this code take a mutex if its in a non-preemptable context?
> @@ -492,6 +490,7 @@ int sdei_event_unregister(u32 event_num)
> struct sdei_event *event;
>
> WARN_ON(in_nmi());
> + WARN_ON_ONCE(preemptible());
>
> mutex_lock(&sdei_events_lock);
> event = sdei_event_find(event_num);
Same again.
> @@ -576,6 +573,7 @@ int sdei_event_register(u32 event_num, sdei_event_callback *cb, void *arg)
> struct sdei_event *event;
>
> WARN_ON(in_nmi());
> + WARN_ON_ONCE(preemptible());
>
> mutex_lock(&sdei_events_lock);
> if (sdei_event_find(event_num)) {
Same again.
I think you've copied these here because they called _local_event_unregister(), but they
did that via on_each_cpu(), which is what made _local_event_unregister() non-preemtable.
You can just remove these three, the original warnings were mostly documentation, and to
catch myself out. (before RT moved the goal posts).
With that:
Reviewed-by: James Morse <james.morse@arm.com>
(could you trim the [timestamps] out of the commit log?)
> @@ -765,7 +765,7 @@ static int sdei_device_freeze(struct device *dev)
> int err;
>
> /* unregister private events */
> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
Is there any mileage in making this CPUHP_AP_ONLINE_DYN ?
Perf really was the only reason that this needed to happen in any particular order.
Thanks,
James
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
2023-02-10 17:42 ` James Morse
@ 2023-02-15 15:55 ` Pierre Gondois
-1 siblings, 0 replies; 12+ messages in thread
From: Pierre Gondois @ 2023-02-15 15:55 UTC (permalink / raw)
To: James Morse, linux-kernel
Cc: Will Deacon, Huacai Chen, John Garry, Shaokun Zhang, WANG Xuerui,
Qi Liu, Bharat Bhushan, Huang Ying, Bibo Mao, Jason A. Donenfeld,
Daniel Lezcano, linux-arm-kernel
Hello James,
On 2/10/23 18:42, James Morse wrote:
> Hi Pierre,
>
> Sorry its taken so long for me to catch up with this ...
No worries.
>
> On 18/10/2022 14:04, Pierre Gondois wrote:
>> On an Ampere Altra,
>> Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
>> Ampere Altra triggers:
>> [ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
>> [ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
>> [ 15.683157] preempt_count: 0, expected: 0
>> [ 15.683159] RCU nest depth: 0, expected: 0
>> [ 15.683163] 3 locks held by cpuhp/0/24:
>> [ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
>> [ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
>> [ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
>> [ 15.683224] irq event stamp: 36
>> [ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
>> [ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
>> [ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
>> [ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
>> [ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
>> [ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
>> [ 15.683268] Call trace:
>> [ 15.683271] dump_backtrace+0x114/0x120
>> [ 15.683277] show_stack+0x20/0x70
>> [ 15.683279] dump_stack_lvl+0x9c/0xd8
>> [ 15.683288] dump_stack+0x18/0x34
>> [ 15.683289] __might_resched+0x188/0x228
>> [ 15.683292] rt_spin_lock+0x70/0x120
>> [ 15.683301] sdei_cpuhp_up+0x3c/0x130
>> [ 15.683303] cpuhp_invoke_callback+0x250/0xf08
>> [ 15.683305] cpuhp_thread_fun+0x120/0x248
>> [ 15.683308] smpboot_thread_fn+0x280/0x320
>> [ 15.683315] kthread+0x130/0x140
>> [ 15.683321] ret_from_fork+0x10/0x20
>
>> sdei_cpuhp_up() is called in the STARTING hotplug section,
>> which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
>> state to the _ONLINE section to execute the cpuhp cb with
>> preemption enabled.
>
> The background to this is SDEI got its own cpuhp slot because 'perf NMI' support
> was one of the use-cases, but this got superseded by pNMI. Without an interaction with
> perf, the slot doesn't need to be that early.
>
>
>> Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
>> calling CPU. It is checked that preemption is disabled for them.
>> _ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
>> Preemption is enabled in those threads, but their cpumask is limited
>> to 1 CPU.
>
>> Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
>> don't trigger them. This means that no check will be done for some
>> cases, e.g. sdei_mask_local_cpu() invocations.
>
> (these things are documentation anyway)
>
>
>> Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
>> which acts on the calling CPU.
>
>> diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
>> index 1e1a51510e83..9b03e164a37a 100644
>> --- a/drivers/firmware/arm_sdei.c
>> +++ b/drivers/firmware/arm_sdei.c
>
>> @@ -401,6 +399,8 @@ int sdei_event_enable(u32 event_num)
>> int err = -EINVAL;
>> struct sdei_event *event;
>>
>> + WARN_ON_ONCE(preemptible());
>> +
>> mutex_lock(&sdei_events_lock);
>> event = sdei_event_find(event_num);
>> if (!event) {
>
> This doesn't look right. How can this code take a mutex if its in a non-preemptable context?
>
>> @@ -492,6 +490,7 @@ int sdei_event_unregister(u32 event_num)
>> struct sdei_event *event;
>>
>> WARN_ON(in_nmi());
>> + WARN_ON_ONCE(preemptible());
>>
>> mutex_lock(&sdei_events_lock);
>> event = sdei_event_find(event_num);
>
> Same again.
>
>> @@ -576,6 +573,7 @@ int sdei_event_register(u32 event_num, sdei_event_callback *cb, void *arg)
>> struct sdei_event *event;
>>
>> WARN_ON(in_nmi());
>> + WARN_ON_ONCE(preemptible());
>>
>> mutex_lock(&sdei_events_lock);
>> if (sdei_event_find(event_num)) {
>
> Same again.
>
> I think you've copied these here because they called _local_event_unregister(), but they
> did that via on_each_cpu(), which is what made _local_event_unregister() non-preemtable.
Yes right, these 3 warnings don't really make sense. They will be removed.
>
> You can just remove these three, the original warnings were mostly documentation, and to
> catch myself out. (before RT moved the goal posts).
>
> With that:
> Reviewed-by: James Morse <james.morse@arm.com>
>
> (could you trim the [timestamps] out of the commit log?)
Yes sure.
>
>
>> @@ -765,7 +765,7 @@ static int sdei_device_freeze(struct device *dev)
>> int err;
>>
>> /* unregister private events */
>> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
>> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
>
> Is there any mileage in making this CPUHP_AP_ONLINE_DYN ?
> Perf really was the only reason that this needed to happen in any particular order.
Ok I will do that.
Thanks for the review,
Regards,
Pierre
>
>
> Thanks,
>
> James
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG
@ 2023-02-15 15:55 ` Pierre Gondois
0 siblings, 0 replies; 12+ messages in thread
From: Pierre Gondois @ 2023-02-15 15:55 UTC (permalink / raw)
To: James Morse, linux-kernel
Cc: Will Deacon, Huacai Chen, John Garry, Shaokun Zhang, WANG Xuerui,
Qi Liu, Bharat Bhushan, Huang Ying, Bibo Mao, Jason A. Donenfeld,
Daniel Lezcano, linux-arm-kernel
Hello James,
On 2/10/23 18:42, James Morse wrote:
> Hi Pierre,
>
> Sorry its taken so long for me to catch up with this ...
No worries.
>
> On 18/10/2022 14:04, Pierre Gondois wrote:
>> On an Ampere Altra,
>> Running a preemp_rt kernel based on v5.19-rc3-rt5 on an
>> Ampere Altra triggers:
>> [ 15.683141] BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:46
>> [ 15.683154] in_atomic(): 0, irqs_disabled(): 128, non_block: 0, pid: 24, name: cpuhp/0
>> [ 15.683157] preempt_count: 0, expected: 0
>> [ 15.683159] RCU nest depth: 0, expected: 0
>> [ 15.683163] 3 locks held by cpuhp/0/24:
>> [ 15.683167] #0: ffffda30217c70d0 (cpu_hotplug_lock){++++}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
>> [ 15.683201] #1: ffffda30217c7120 (cpuhp_state-up){+.+.}-{0:0}, at: cpuhp_thread_fun+0x5c/0x248
>> [ 15.683205] #2: ffffda3021c711f0 (sdei_list_lock){....}-{3:3}, at: sdei_cpuhp_up+0x3c/0x130
>> [ 15.683224] irq event stamp: 36
>> [ 15.683226] hardirqs last enabled at (35): [<ffffda301e85b7bc>] finish_task_switch+0xb4/0x2b0
>> [ 15.683236] hardirqs last disabled at (36): [<ffffda301e812fec>] cpuhp_thread_fun+0x21c/0x248
>> [ 15.683238] softirqs last enabled at (0): [<ffffda301e80b184>] copy_process+0x63c/0x1ac0
>> [ 15.683245] softirqs last disabled at (0): [<0000000000000000>] 0x0
>> [ 15.683258] CPU: 0 PID: 24 Comm: cpuhp/0 Not tainted 5.19.0-rc3-rt5-[...]
>> [ 15.683265] Hardware name: WIWYNN Mt.Jade Server System B81.03001.0005/Mt.Jade Motherboard, BIOS 1.08.20220218 (SCP: 1.08.20220218) 2022/02/18
>> [ 15.683268] Call trace:
>> [ 15.683271] dump_backtrace+0x114/0x120
>> [ 15.683277] show_stack+0x20/0x70
>> [ 15.683279] dump_stack_lvl+0x9c/0xd8
>> [ 15.683288] dump_stack+0x18/0x34
>> [ 15.683289] __might_resched+0x188/0x228
>> [ 15.683292] rt_spin_lock+0x70/0x120
>> [ 15.683301] sdei_cpuhp_up+0x3c/0x130
>> [ 15.683303] cpuhp_invoke_callback+0x250/0xf08
>> [ 15.683305] cpuhp_thread_fun+0x120/0x248
>> [ 15.683308] smpboot_thread_fn+0x280/0x320
>> [ 15.683315] kthread+0x130/0x140
>> [ 15.683321] ret_from_fork+0x10/0x20
>
>> sdei_cpuhp_up() is called in the STARTING hotplug section,
>> which runs whith interrupts disabled. Move CPUHP_AP_ARM_SDEI_
>> state to the _ONLINE section to execute the cpuhp cb with
>> preemption enabled.
>
> The background to this is SDEI got its own cpuhp slot because 'perf NMI' support
> was one of the use-cases, but this got superseded by pNMI. Without an interaction with
> perf, the slot doesn't need to be that early.
>
>
>> Some SDEI calls (e.g. SDEI_1_0_FN_SDEI_PE_MASK) take actions on the
>> calling CPU. It is checked that preemption is disabled for them.
>> _ONLINE cpuhp cb are executed in the 'per CPU hotplug thread'.
>> Preemption is enabled in those threads, but their cpumask is limited
>> to 1 CPU.
>
>> Move 'WARN_ON_ONCE(preemptible())' statements so that SDEI cpuhp cb
>> don't trigger them. This means that no check will be done for some
>> cases, e.g. sdei_mask_local_cpu() invocations.
>
> (these things are documentation anyway)
>
>
>> Also add a check for the SDEI_1_0_FN_SDEI_PRIVATE_RESET SDEI call
>> which acts on the calling CPU.
>
>> diff --git a/drivers/firmware/arm_sdei.c b/drivers/firmware/arm_sdei.c
>> index 1e1a51510e83..9b03e164a37a 100644
>> --- a/drivers/firmware/arm_sdei.c
>> +++ b/drivers/firmware/arm_sdei.c
>
>> @@ -401,6 +399,8 @@ int sdei_event_enable(u32 event_num)
>> int err = -EINVAL;
>> struct sdei_event *event;
>>
>> + WARN_ON_ONCE(preemptible());
>> +
>> mutex_lock(&sdei_events_lock);
>> event = sdei_event_find(event_num);
>> if (!event) {
>
> This doesn't look right. How can this code take a mutex if its in a non-preemptable context?
>
>> @@ -492,6 +490,7 @@ int sdei_event_unregister(u32 event_num)
>> struct sdei_event *event;
>>
>> WARN_ON(in_nmi());
>> + WARN_ON_ONCE(preemptible());
>>
>> mutex_lock(&sdei_events_lock);
>> event = sdei_event_find(event_num);
>
> Same again.
>
>> @@ -576,6 +573,7 @@ int sdei_event_register(u32 event_num, sdei_event_callback *cb, void *arg)
>> struct sdei_event *event;
>>
>> WARN_ON(in_nmi());
>> + WARN_ON_ONCE(preemptible());
>>
>> mutex_lock(&sdei_events_lock);
>> if (sdei_event_find(event_num)) {
>
> Same again.
>
> I think you've copied these here because they called _local_event_unregister(), but they
> did that via on_each_cpu(), which is what made _local_event_unregister() non-preemtable.
Yes right, these 3 warnings don't really make sense. They will be removed.
>
> You can just remove these three, the original warnings were mostly documentation, and to
> catch myself out. (before RT moved the goal posts).
>
> With that:
> Reviewed-by: James Morse <james.morse@arm.com>
>
> (could you trim the [timestamps] out of the commit log?)
Yes sure.
>
>
>> @@ -765,7 +765,7 @@ static int sdei_device_freeze(struct device *dev)
>> int err;
>>
>> /* unregister private events */
>> - cpuhp_remove_state(CPUHP_AP_ARM_SDEI_STARTING);
>> + cpuhp_remove_state(CPUHP_AP_ARM_SDEI_ONLINE);
>
> Is there any mileage in making this CPUHP_AP_ONLINE_DYN ?
> Perf really was the only reason that this needed to happen in any particular order.
Ok I will do that.
Thanks for the review,
Regards,
Pierre
>
>
> Thanks,
>
> James
_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2023-02-15 15:56 UTC | newest]
Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-10-18 13:04 [PATCH v2] firmware: arm_sdei: Fix sleep from invalid context BUG Pierre Gondois
2022-10-18 13:04 ` Pierre Gondois
2022-10-18 13:07 ` Pierre Gondois
2022-10-18 13:07 ` Pierre Gondois
2023-02-01 16:23 ` Pierre Gondois
2023-02-01 16:23 ` Pierre Gondois
2022-11-07 16:14 ` Will Deacon
2022-11-07 16:14 ` Will Deacon
2023-02-10 17:42 ` James Morse
2023-02-10 17:42 ` James Morse
2023-02-15 15:55 ` Pierre Gondois
2023-02-15 15:55 ` Pierre Gondois
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.