On 7/9/20 4:45 PM, Song Bao Hua (Barry Song) wrote: > > >> -----Original Message----- >> From: Mike Kravetz [mailto:[email protected]] >> Sent: Friday, July 10, 2020 6:58 AM >> To: Song Bao Hua (Barry Song) <[email protected]>; Roman >> Gushchin <[email protected]> >> Cc: Andrew Morton <[email protected]>; [email protected]; >> [email protected]; Linuxarm <[email protected]>; Jonathan >> Cameron <[email protected]> >> Subject: Re: [PATCH v3] mm/hugetlb: avoid hardcoding while checking if cma >> is enable >> >> Looks like this produced a warning in linux-next. I suspect it is due to the >> combination CONFIG_HUGETLB_PAGE && !CONFIG_CMA. >> >> Instead of adding the routine hugetlb_cma_enabled() to scan the hugetlb_cma >> array, could we just use a boolean as follows? It can simply be set in >> hugetlb_cma_reserve when we reserve CMA. > > Maybe just use hugetlb_cma_size? If hugetlb_cma_size is not 0, someone is > trying to use > cma, then bootmem for gigantic pages will be totally ignored according to > discussion here: > https://lkml.org/lkml/2020/7/8/1288 > > if somebody sets a wrong hugetlb_cma_size which causes that cma is not > reserved. > It is the fault of users? We just need to document hugetlb_cma will overwrite > bootmem > reservations? >
Yes, I think using hugetlb_cma_size would be sufficient. If someone specifies hugetlb_cma=<N> and hugepagesz=<gigantic_page_size> hugepages=<X> that is wrong. I don't think we need to worry about 'falling back' to preallocating X gigantic pages if N is a bad value. Or, even if the arch does not support cma allocation. I am working on a patch to check this earlier in command processing. That will make this check unnecessary. However, that patch is based on new command line processing code only in 5.8. So, I think we still need to do this so that it can be backported to stable. -- Mike Kravetz
