Re: [GIT PULL] Urgent ACPI support fix for v6.18

["lihuisong (C)" <lihuisong@huawei.com>]

在 2025/11/28 3:03, Longia, Amandeep Kaur 写道:
> Hi all,
>
> On 11/27/2025 7:39 AM, lihuisong (C) wrote:
>> Hello Rafael,
>>
>> 在 2025/11/26 20:54, Rafael J. Wysocki 写道:
>>> Hi Linus,
>>>
>>> Please pull from the tag
>>>
>>> git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm.git \
>>>   acpi-6.18-rc8
>>>
>>> with top-most commit 43ff36c4a5a574ee83b4b0d3f3d74f09a3a8c2d3
>>>
>>>   Revert "ACPI: processor: idle: Optimize ACPI idle driver 
>>> registration"
>>>
>>> on top of commit ac3fd01e4c1efce8f2c054cdeb2ddd2fc0fb150d
>>>
>>>   Linux 6.18-rc7
>>>
>>> to receive an urgent ACPI support fix for 6.18.
>>>
>>> This reverts a commit that attempted to make the code in the ACPI
>>> processor driver more straightforward, but it turned out to cause
>>> the kernel to crash on at least one system, along with some further
>>> cleanups on top of it.
>> I just found that "ACPI: processor: idle: Optimize ACPI idle driver 
>> registration" depends on the change
>> about cpuhp_setup_state in the commit [1]. Or many CPUs don't create 
>> cpuidle directory.
>> What is the crash? Do you have releated trace?
>>
>> [1] https://lore.kernel.org/all/20240529133446.28446-2- 
>> Jonathan.Cameron@huawei.com/
>
> We have observed the same issue. After booting the system with the 
> latest kernel, the sysfs path /sys/devices/system/cpu/cpu*/cpuidle 
> does not exist. Bisecting between v6.18-rc7 (good) and master (bad) 
> led us to the following commit:
>
> 43ff36c4a5a574ee83b4b0d3f3d74f09a3a8c2d3 Revert "ACPI: processor: 
> idle: Optimize ACPI idle driver registration"
>
>
Hello Amandeep,

I guess that your kernel doesn't have the following patch [1] if you 
also encountered this issue.

The __acpi_processor_start() of all CPUs are called when run 
driver_register(&acpi_processor_driver).
But acpi_processor_register_idle_driver() is called after 
driver_register(&acpi_processor_driver).
As a result, the acpi_processor_power_init() would not register cpuidle 
device and create these directory.

I tested it ok on my platform under the following modification.
Can you test it again using following modification based on kernel 
without revert any ACPI idle patch?
  I am looking forward to your reply. Thanks.
-->
diff --git a/drivers/acpi/processor_driver.c 
b/drivers/acpi/processor_driver.c
index 4b906b56186a..b56a2a228693 100644
--- a/drivers/acpi/processor_driver.c
+++ b/drivers/acpi/processor_driver.c
@@ -268,12 +268,12 @@ static int __init acpi_processor_driver_init(void)
                 acpi_processor_ignore_ppc_init();
         }

+       acpi_processor_register_idle_driver();
+
         result = driver_register(&acpi_processor_driver);
         if (result < 0)
                 return result;

-       acpi_processor_register_idle_driver();
-
         result = cpuhp_setup_state_nocalls(CPUHP_AP_ONLINE_DYN,
                                            "acpi/cpu-drv:online",
                                            acpi_soft_cpu_online, NULL);
diff --git a/drivers/acpi/processor_idle.c b/drivers/acpi/processor_idle.c
index 66d8df5a37a6..a34983abb0b9 100644
--- a/drivers/acpi/processor_idle.c
+++ b/drivers/acpi/processor_idle.c
@@ -1404,6 +1404,7 @@ void acpi_processor_register_idle_driver(void)
                 if (!pr)
                         continue;

+               acpi_processor_cstate_first_run_checks();
                 ret = acpi_processor_get_power_info(pr);
                 if (!ret) {
                         pr->flags.power_setup_done = 1;

[1] 
https://lore.kernel.org/all/20240529133446.28446-2-Jonathan.Cameron@huawei.com/