Message-ID: <b91d36b4-4dcc-7f1e-6125-1641fdc41d4f@arm.com>
Date:   Mon, 14 Aug 2023 23:57:25 +0100
From:   Suzuki K Poulose <suzuki.poulose@....com>
To:     hejunhao <hejunhao3@...wei.com>, mike.leach@...aro.org,
        leo.yan@...aro.org, anshuman.khandual@....com,
        jonathan.cameron@...wei.com
Cc:     coresight@...ts.linaro.org, linux-arm-kernel@...ts.infradead.org,
        linux-kernel@...r.kernel.org, yangyicong@...wei.com,
        prime.zeng@...ilicon.com
Subject: Re: [PATCH 1/2] coresight: trbe: Fix TRBE potential sleep in atomic
 context

On 14/08/2023 14:32, hejunhao wrote:
> Hi Suzuki
> 
> 
> On 2023/8/14 18:34, Suzuki K Poulose wrote:
>> Hi Junhao
>>
>> On 14/08/2023 10:38, Junhao He wrote:
>>> smp_call_function_single() will allocate an IPI interrupt vector to
>>> the target processor and send a function call request to the interrupt
>>> vector. After the target processor receives the IPI interrupt, it will
>>> execute arm_trbe_remove_coresight_cpu() call request in the interrupt
>>> handler.
>>>
>>> According to the device_unregister() stack information, if other process
>>> is useing the device, the down_write() may sleep, and trigger deadlocks
>>> or unexpected errors.
>>>
>>>    arm_trbe_remove_coresight_cpu
>>>      coresight_unregister
>>>        device_unregister
>>>          device_del
>>>            kobject_del
>>>              __kobject_del
>>>                sysfs_remove_dir
>>>                  kernfs_remove
>>>                    down_write ---------> it may sleep
>>>
>>> Add a helper arm_trbe_disable_cpu() to disable TRBE precpu irq and reset
>>> per TRBE.
>>> Simply call arm_trbe_remove_coresight_cpu() directly without useing the
>>> smp_call_function_single(), which is the same as registering the TRBE
>>> coresight device.
>>>
>>> Fixes: 3fbf7f011f24 ("coresight: sink: Add TRBE driver")
>>> Signed-off-by: Junhao He <hejunhao3@...wei.com>
>>> ---
>>>   drivers/hwtracing/coresight/coresight-trbe.c | 35 +++++++++++---------
>>>   1 file changed, 20 insertions(+), 15 deletions(-)
>>>
>>> diff --git a/drivers/hwtracing/coresight/coresight-trbe.c 
>>> b/drivers/hwtracing/coresight/coresight-trbe.c
>>> index 7720619909d6..ce1e6f537b8d 100644
>>> --- a/drivers/hwtracing/coresight/coresight-trbe.c
>>> +++ b/drivers/hwtracing/coresight/coresight-trbe.c
>>> @@ -1225,6 +1225,17 @@ static void arm_trbe_enable_cpu(void *info)
>>>       enable_percpu_irq(drvdata->irq, IRQ_TYPE_NONE);
>>>   }
>>>   +static void arm_trbe_disable_cpu(void *info)
>>> +{
>>> +    struct trbe_drvdata *drvdata = info;
>>> +    struct trbe_cpudata *cpudata = this_cpu_ptr(drvdata->cpudata);
>>> +
>>> +    disable_percpu_irq(drvdata->irq);
>>> +    trbe_reset_local(cpudata);
>>> +    cpudata->drvdata = NULL;
>>> +}
>>> +
>>> +
>>>   static void arm_trbe_register_coresight_cpu(struct trbe_drvdata 
>>> *drvdata, int cpu)
>>>   {
>>>       struct trbe_cpudata *cpudata = per_cpu_ptr(drvdata->cpudata, cpu);
>>> @@ -1326,18 +1337,12 @@ static void arm_trbe_probe_cpu(void *info)
>>>       cpumask_clear_cpu(cpu, &drvdata->supported_cpus);
>>>   }
>>>   -static void arm_trbe_remove_coresight_cpu(void *info)
>>> +static void arm_trbe_remove_coresight_cpu(struct trbe_drvdata 
>>> *drvdata, int cpu)
>>>   {
>>> -    int cpu = smp_processor_id();
>>> -    struct trbe_drvdata *drvdata = info;
>>> -    struct trbe_cpudata *cpudata = per_cpu_ptr(drvdata->cpudata, cpu);
>>>       struct coresight_device *trbe_csdev = 
>>> coresight_get_percpu_sink(cpu);
>>>   -    disable_percpu_irq(drvdata->irq);
>>> -    trbe_reset_local(cpudata);
>>>       if (trbe_csdev) {
>>>           coresight_unregister(trbe_csdev);
>>> -        cpudata->drvdata = NULL;
>>>           coresight_set_percpu_sink(cpu, NULL);
>>
>> I am a bit concerned about "resetting" the sink from a different CPU.
>> Could we instead, schedule a delayed work to unregister the trbe_csdev?
> 
> Yes, I will try to do that.
> Sorry for my following questions.
> As you mean, do we need to take the same care when setting the percpu sink
> in the register trbe_csdev ?

Apologies, having taken another look, we set the percpu_sink for
a cpu outside smp_call_function(). So, I think your patch is fine.


> 
> Best regards,
> Junhao.
> 
>>
>>
>>>       }
>>>   }
>>> @@ -1366,8 +1371,12 @@ static int arm_trbe_remove_coresight(struct 
>>> trbe_drvdata *drvdata)
>>>   {
>>>       int cpu;
>>>   -    for_each_cpu(cpu, &drvdata->supported_cpus)
>>> -        smp_call_function_single(cpu, arm_trbe_remove_coresight_cpu, 
>>> drvdata, 1);
>>> +    for_each_cpu(cpu, &drvdata->supported_cpus) {
>>> +        if (cpumask_test_cpu(cpu, &drvdata->supported_cpus))
>>> +            smp_call_function_single(cpu, arm_trbe_disable_cpu, 
>>> drvdata, 1);
>>> +        if (cpumask_test_cpu(cpu, &drvdata->supported_cpus))
>>> +            arm_trbe_remove_coresight_cpu(drvdata, cpu);

Do we need to test the cpu here in both places ? We already check that
in the loop entry. The reason why we repeat the check during the probe,
is to skip any CPUs that may have a TRBE not accessible.

Suzuki


>>> +    }
>>>       free_percpu(drvdata->cpudata);
>>>       return 0;
>>>   }
>>> @@ -1406,12 +1415,8 @@ static int arm_trbe_cpu_teardown(unsigned int 
>>> cpu, struct hlist_node *node)
>>>   {
>>>       struct trbe_drvdata *drvdata = hlist_entry_safe(node, struct 
>>> trbe_drvdata, hotplug_node);
>>>   -    if (cpumask_test_cpu(cpu, &drvdata->supported_cpus)) {
>>> -        struct trbe_cpudata *cpudata = per_cpu_ptr(drvdata->cpudata, 
>>> cpu);
>>> -
>>> -        disable_percpu_irq(drvdata->irq);
>>> -        trbe_reset_local(cpudata);
>>> -    }
>>> +    if (cpumask_test_cpu(cpu, &drvdata->supported_cpus))
>>> +        arm_trbe_disable_cpu(drvdata);
>>>       return 0;
>>>   }
>>
>>
>> .
>>
> 

Powered by blists - more mailing lists