Re: [RFC PATCH 3/3] powerpc/mm/hugetlb: Don't enable HugeTLB if we don't have a page table cache

["Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>]

On 5/17/19 9:29 AM, Aneesh Kumar K.V wrote:
> On 5/16/19 8:17 PM, Michael Ellerman wrote:
>> "Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com> writes:
>>> This makes sure we don't enable HugeTLB if the cache is not configured.
>>> I am still not sure about this. IMHO hugetlb support should be a 
>>> hardware
>>> support derivative and any cache allocation failure should be handled 
>>> as I did
>>> in the earlier patch. But then if we were not able to create hugetlb 
>>> page table
>>> cache, we can as well declare hugetlb support disabled thereby 
>>> avoiding calling
>>> into allocation routines.
>>>
>>> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.ibm.com>
>>> ---
>>>   arch/powerpc/mm/hugetlbpage.c | 11 +++++++++--
>>>   1 file changed, 9 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/arch/powerpc/mm/hugetlbpage.c 
>>> b/arch/powerpc/mm/hugetlbpage.c
>>> index ee16a3fb788a..4bf8bc659cc7 100644
>>> --- a/arch/powerpc/mm/hugetlbpage.c
>>> +++ b/arch/powerpc/mm/hugetlbpage.c
>>> @@ -602,6 +602,7 @@ __setup("hugepagesz=", hugepage_setup_sz);
>>>   static int __init hugetlbpage_init(void)
>>>   {
>>>       int psize;
>>> +    bool configured = false;
>>
>> Where's my reverse Christmas tree! :)
> 
> Will fix that :)
> 
>>
>>>       if (hugetlb_disabled) {
>>>           pr_info("HugeTLB support is disabled!\n");
>>> @@ -651,10 +652,16 @@ static int __init hugetlbpage_init(void)
>>>               pgtable_cache_add(pdshift - shift);
>>>           else if (IS_ENABLED(CONFIG_PPC_FSL_BOOK3E) || 
>>> IS_ENABLED(CONFIG_PPC_8xx))
>>>               pgtable_cache_add(PTE_T_ORDER);
>>> +
>>> +        if (!configured)
>>> +            configured = true;
>>
>> I'd just not worry about the if.
>>
>>>       }
>>> -    if (IS_ENABLED(CONFIG_HUGETLB_PAGE_SIZE_VARIABLE))
>>> -        hugetlbpage_init_default();
>>> +    if (configured) {
>>> +        if (IS_ENABLED(CONFIG_HUGETLB_PAGE_SIZE_VARIABLE))
>>> +            hugetlbpage_init_default();
>>> +    } else
>>> +        pr_info("Disabling HugeTLB");
>>
>> We're not actually doing anything to disable it in the
>> CONFIG_HUGETLB_PAGE_SIZE_VARIABLE=n case, but I guess the print is still
>> correct because we didn't enable a size in the for loop above?
>>
>> Can we make it a bit more explicit? Maybe like:
>>
>>    "Disabling HugeTLB, no usable page sizes found."
>>
> 
> That would confuse when they find in the dmesg
> 
> [    0.000000] hash-mmu: Page sizes from device-tree:
> [    0.000000] hash-mmu: base_shift=12: shift=12, sllp=0x0000, 
> avpnm=0x00000000, tlbiel=1, penc=0
> [    0.000000] hash-mmu: base_shift=12: shift=16, sllp=0x0000, 
> avpnm=0x00000000, tlbiel=1, penc=7
> [    0.000000] hash-mmu: base_shift=12: shift=24, sllp=0x0000, 
> avpnm=0x00000000, tlbiel=1, penc=56
> [    0.000000] hash-mmu: base_shift=16: shift=16, sllp=0x0110, 
> avpnm=0x00000000, tlbiel=1, penc=1
> [    0.000000] hash-mmu: base_shift=16: shift=24, sllp=0x0110, 
> avpnm=0x00000000, tlbiel=1, penc=8
> [    0.000000] hash-mmu: base_shift=24: shift=24, sllp=0x0100, 
> avpnm=0x00000001, tlbiel=0, penc=0
> [    0.000000] hash-mmu: base_shift=34: shift=34, sllp=0x0120, 
> avpnm=0x000007ff, tlbiel=0, penc=3

There is another failure condition which i am not sure how to handle 
with the pagetable cache creation failures. With above, if we had kernel 
command line hugepagesz=x hugepages=y, and if that x is a gigantic 
hugepage, we will allocate those pages early even if we don't support 
hugetlb because we failed to create page table cache.

I am not sure whether we should handle that error gracefully?

-aneesh