Re: HTLB mapping for drivers. Driver example

[Alexey Korolev <akorolex@gmail.com>]

Mel,


>
> This seems a lot of burden to put on a device driver, particularly with
> respect to the reservations.

Thanks a lot for review you did. That is right.  I don't like this
burden as well.
>
>> File operations of /dev/hpage_map do the following:
>>
>> In file open we  associate mappings of /dev/xxx with the file on hugetlbfs (like it is done in ipc/shm.c)
>>       file->f_mapping = h_file->f_mapping;
>>
>> In get_unmapped_area we should tell about addressing constraints in case of huge pages by calling hugetlbfs procedures. (as in ipc/shm.c)
>>       return get_unmapped_area(h_file, addr, len, pgoff, flags);
>>
>> We need to let hugetlbfs do architecture specific operations with mapping in mmap call. This driver does not reserve any memory for private mappings
>> so driver requests reservation from hugetlbfs. (Actually driver can do this as well but it will make it more complex)
>>
>> The exit procedure:
>> * removes memory from page cache
>> * deletes file on hugetlbfs vfs mount
>> *  free pages
>>
>> Application example is not shown here but it is very simple. It does the following: open file /dev/hpage_map, mmap a region, read/write memory, unmap file, close file.
>>
>
> For the use-model you have in mind, could you look at Eric Munson's patches
> and determine if the target application would have been happy to call the
> following please?
>
> mmap(0, len, prot, MAP_ANONYMOUS|MAP_HUGETLB, 0, 0)
>
Hmm. But how can I at least identify which driver this call is addressed to?

Thanks,
Alexey

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>