Re: Fwd: [PATCH 1/1] RDMA/umem: add back hugepage sg list

[Jason Gunthorpe <jgg@nvidia.com>]

On Sun, Mar 21, 2021 at 08:54:31PM +0800, Zhu Yanjun wrote:
> On Sun, Mar 21, 2021 at 8:07 PM Jason Gunthorpe <jgg@nvidia.com> wrote:
> >
> > On Sun, Mar 21, 2021 at 04:06:07PM +0800, Zhu Yanjun wrote:
> > > > > > You are reporting 4k pages, if max_segment is larger than 4k there is
> > > > > > no such thing as "too big"
> > > > > >
> > > > > > I assume it is "too small" because of some maths overflow.
> > > > >
> > > > >  459                 while (n_pages && page_to_pfn(pages[0]) == paddr) {
> > > > >  460                         if (prv->length + PAGE_SIZE >
> > > > > max_segment)  <--it max_segment is big, n_pages is zero.
> > > >
> > > > What does n_pages have to do with max_segment?
> > >
> > > With the following snippet
> > > "
> > >         struct ib_umem *region;
> > >         region = ib_umem_get(pd->device, start, len, access);
> > >
> > >         page_size = ib_umem_find_best_pgsz(region,
> > >                                            SZ_4K | SZ_2M | SZ_1G,
> > >                                            virt);
> > > "
> >
> > Can you please stop posting random bits of code that do nothing to
> > answer the question?
> >
> > > IMHO, you can reproduce this problem in your local host.
> >
> > Uh, no, you need to communicate effectively or stop posting please.
> 
> can you read the following link again
> 
> https://patchwork.kernel.org/project/linux-rdma/patch/CAD=hENeqTTmpS5V+G646V0QvJFLVSd3Sq11ffQFcDXU-OSsQEg@mail.gmail.com/#24031643
> 
> In this link, I explained it in details.

No, this is all nonsense

> Since the max_segment is very big, so normally the prv->length +
> PAGE_SIZE is less than max_segment, it will loop over and over again,
> until n_pages is zero,

And each iteration increases

			prv->length += PAGE_SIZE;
			n_pages--;

Where prv is the last sgl entry, so decreasing n_pages cannot also
yield a 4k sgl entry like you are reporting.

You get 4k sgl entries if max_segment is *too small* not too big!

Jason