Re: [PATCH] nvdimm: nvdimm_bus_register: Avoid adding device to the unregistered bus

["lizhijian@fujitsu.com" <lizhijian@fujitsu.com>]

On 16/03/2023 23:54, Dan Williams wrote:
> Li Zhijian wrote:
>> nvdimm_bus_register() could be called from other modules, such as nfit,
>> but it can only be called after the nvdimm_bus_type is registered.
>>
>>   BUG: kernel NULL pointer dereference, address: 0000000000000098
>>   #PF: supervisor read access in kernel mode
>>   #PF: error_code(0x0000) - not-present page
>>   PGD 0 P4D 0
>>   Oops: 0000 [#1] PREEMPT SMP PTI
>>   CPU: 0 PID: 117 Comm: systemd-udevd Not tainted 6.2.0-rc6-pmem+ #97
>>   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
>>   RIP: 0010:bus_add_device+0x58/0x150
>>   Call Trace:
>>    <TASK>
>>    device_add+0x3ac/0x980
>>    nvdimm_bus_register+0x16d/0x1d0
>>    acpi_nfit_init+0xb72/0x1f90 [nfit]
>>    acpi_nfit_add+0x1d5/0x200 [nfit]
>>    acpi_device_probe+0x45/0x160
> 
> Can you explain a bit more how to hit this crash? This has not been a
> problem historically and the explanation above makes it sound like this
> is a theoretical issue.
> 

Dan,

Configure the kconfig with ACPI_NFIT [=m] && LIBNVDIMM [=y], and add extra kernel booting parameter
'initcall_blacklist=libnvdimm_init'. Then kernel panic!
Theoretically, it will also happen if nvdimm_bus_register() failed.


For kdump purpose[1], we need to disable libnvdimm driver to ensure metadata in pmem will not be updated again in kdump kernel
[1]https://lore.kernel.org/linux-mm/3c752fc2-b6a0-2975-ffec-dba3edcf4155@fujitsu.com/T/

Thanks
Zhijian

> libnvdimm_init() *should* be done before the nfit driver can attempt
> nvdimm_bus_register(). So, something else is broken if
> nvdimm_bus_register() can be called before libnvdimm_init(), or after
> libnvdimm_exit().