[PATCH v1 07/11] fs/proc/vmcore: introduce PROC_VMCORE_DEVICE_RAM to detect device RAM ranges in 2nd kernel

[David Hildenbrand]

On 22.11.24 08:31, Baoquan He wrote:
> On 10/25/24 at 05:11pm, David Hildenbrand wrote:
> ......snip...
>> diff --git a/fs/proc/vmcore.c b/fs/proc/vmcore.c
>> index 3e90416ee54e..c332a9a4920b 100644
>> --- a/fs/proc/vmcore.c
>> +++ b/fs/proc/vmcore.c
>> @@ -69,6 +69,8 @@ static LIST_HEAD(vmcore_cb_list);
>>   /* Whether the vmcore has been opened once. */
>>   static bool vmcore_opened;
>>   
>> +static void vmcore_process_device_ram(struct vmcore_cb *cb);
>> +
>>   void register_vmcore_cb(struct vmcore_cb *cb)
>>   {
>>   	INIT_LIST_HEAD(&cb->next);
>> @@ -80,6 +82,8 @@ void register_vmcore_cb(struct vmcore_cb *cb)
>>   	 */
>>   	if (vmcore_opened)
>>   		pr_warn_once("Unexpected vmcore callback registration\n");
>> +	else if (cb->get_device_ram)
>> +		vmcore_process_device_ram(cb);
> 
> Global variable 'vmcore_opened' is used to indicate if /proc/vmcore is
> opened. With &vmcore_mutex, we don't need to worry about concurrent
> opening and modification. However, if people just open /proc/vmcore and
> close it after checking, then s390 will miss the vmcore dumping, is it
> acceptable?

See my reply to your other mail (patch #3).

> 
>>   	mutex_unlock(&vmcore_mutex);
>>   }
>>   EXPORT_SYMBOL_GPL(register_vmcore_cb);
>> @@ -1511,6 +1515,158 @@ int vmcore_add_device_dump(struct vmcoredd_data *data)
> ......
>> +
>> +static void vmcore_process_device_ram(struct vmcore_cb *cb)
>> +{
>> +	unsigned char *e_ident = (unsigned char *)elfcorebuf;
>> +	struct vmcore_mem_node *first, *m;
>> +	LIST_HEAD(list);
>> +	int count;
>> +
>> +	if (cb->get_device_ram(cb, &list)) {
>> +		pr_err("Kdump: obtaining device ram ranges failed\n");
>> +		return;
>> +	}
>> +	count = list_count_nodes(&list);
>> +	if (!count)
>> +		return;
>> +
>> +	/* We only support Elf64 dumps for now. */
>> +	if (WARN_ON_ONCE(e_ident[EI_CLASS] != ELFCLASS64)) {
>> +		pr_err("Kdump: device ram ranges only support Elf64\n");
>> +		goto out_free;
>> +	}
> 
> Only supporting Elf64 dumps seems to be a basic checking, do we need
> to put it at the beginning of function? Otherwise, we spend efforts to
> call cb->get_device_ram(), then fail.

The idea was that if there is nothing to add, then the elf class doesn't 
matter. But yes, I can move this further up.

Thanks!

-- 
Cheers,

David / dhildenb