Re: [PATCH v4] hugetlbfs: Extend the definition of hugepages parameter to support node allocation

[Mike Rapoport]

Hi Mike,

On Wed, Sep 15, 2021 at 03:05:41PM -0700, Mike Kravetz wrote:
> Now, really CC'ing Mike, and sorry for misspelling your name
> 
> On 9/15/21 3:03 PM, Mike Kravetz wrote:
> > On 9/15/21 6:11 AM, zhenguo yao wrote:
> >> Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> 于2021年9月15日周三 上午11:50写道：
> >>>
> >>> On Thu,  9 Sep 2021 22:16:55 +0800 yaozhenguo <yaozhenguo1@xxxxxxxxx> wrote:
> >>>
> >>>> We can specify the number of hugepages to allocate at boot. But the
> >>>> hugepages is balanced in all nodes at present. In some scenarios,
> >>>> we only need hugepages in one node. For example: DPDK needs hugepages
> >>>> which are in the same node as NIC. if DPDK needs four hugepages of 1G
> >>>> size in node1 and system has 16 numa nodes. We must reserve 64 hugepages
> >>>> in kernel cmdline. But, only four hugepages are used. The others should
> >>>> be free after boot. If the system memory is low(for example: 64G), it will
> >>>> be an impossible task. So, Extending hugepages parameter to support
> >>>> specifying hugepages at a specific node.
> >>>> For example add following parameter:
> >>>>
> >>>> hugepagesz=1G hugepages=0:1,1:3
> >>>>
> >>>> It will allocate 1 hugepage in node0 and 3 hugepages in node1.
> >>>>
> >>>> ...
> >>>>
> >>>> @@ -2842,10 +2843,75 @@ static void __init gather_bootmem_prealloc(void)
> >>>>       }
> >>>>  }
> >>>>
> >>>> +static void __init hugetlb_hstate_alloc_pages_onenode(struct hstate *h, int nid)
> >>>> +{
> >>>> +     unsigned long i;
> >>>> +     char buf[32];
> >>>> +
> >>>> +     for (i = 0; i < h->max_huge_pages_node[nid]; ++i) {
> >>>> +             if (hstate_is_gigantic(h)) {
> >>>> +                     struct huge_bootmem_page *m;
> >>>> +                     void *addr;
> >>>> +
> >>>> +                     addr = memblock_alloc_try_nid_raw(
> >>>> +                                     huge_page_size(h), huge_page_size(h),
> >>>> +                                     0, MEMBLOCK_ALLOC_ACCESSIBLE, nid);
> >>>> +                     if (!addr)
> >>>> +                             break;
> >>>> +                     m = addr;
> >>>> +                     BUG_ON(!IS_ALIGNED(virt_to_phys(m), huge_page_size(h)));
> >>>
> >>> We try very hard to avoid adding BUG calls.  Is there any way in which
> >>> this code can emit a WARNing then permit the kernel to keep operating?
> >>>
> >> Maybe we can rewrite it as below:
> >>                         if (WARN(!IS_ALIGNED(virt_to_phys(m),
> >> huge_page_size(h)),
> >>                                 "HugeTLB: page addr:%p is not aligned\n", m))
> >>                                 break;
> >> @Mike,  Do you think it's OK?
> > 
> > Sorry, I have not yet reviewed the latest version of this patch.
> > Quick thought on this question.
> > 
> > The required alignment passed to memblock_alloc_try_nid_raw() is
> > huge_page_size(h).  Therefore, we know the virtual address m is
> > huge_page_size(h) aligned.  The BUG is just checking to make sure
> > the physical address associated with the virtual address is aligned
> > the same.  I really do not see how this could not be the case.
> > In fact, the memblock allocator finds a physical address with the
> > required alignment and then returns phys_to_virt(alloc).
> > Someone please correct me if I am wrong.  Otherwise, we can drop
> > the BUG.

I agree with your analysis and I also think the BUG() can be dropped
entirely as well as the BUG() in __alloc_bootmem_huge_page().

> > Adding Mike Rapport on Cc:
> > 
> > This allocation code and the associated BUG was copied from
> > __alloc_bootmem_huge_page().  The BUG was added 12 years ago before
> > the memblock allocator existed and we were using the bootmem allocator.
> > If there is no need for a BUG in hugetlb_hstate_alloc_pages_onenode,
> > there is no need for one in __alloc_bootmem_huge_page.

Hmm, even bootmem had alignment guaranties so it seems to me that the BUG()
was over-protective even then.

-- 
Sincerely yours,
Mike.