This is a (small) part of larger work of replacing page allocator calls
with kmalloc.

My initial intention a few month ago was to remove ugly casts [1], but then
willy pointed out that Linus objected to something like this [2] and it
looks like more than a decade old technical debt.

Largely, anything that doesn't need struct page (or a memdesc in the
future) should just use kmalloc() or kvmalloc() to allocate memory.
kmalloc() guarantees alignment, physical contiguity and working
virt_to_phys() and beside nicer API that returns void * on alloc and
doesn't require to know the allocation size on free, kmalloc() provides
better debugging capabilities than page allocator.

Another thing is that touching these allocation sites gives the reviewers
opportunity to see if a PAGE_SIZE buffer is actually needed or maybe
another size is appropriate.

For larger allocations that don't need physically contiguous memory
kvmalloc() can be a better option that __get_free_pages() because under
memory pressure it's is easier to allocate several order-0 pages than a
physically contiguous chunk with the same number of pages.

And last, but not least, removing needless calls to page allocator should
help with memdesc (aka project folio) conversion. There will be way less
places to audit to see if the user was actually using struct page.

Also in git:
https://git.kernel.org/pub/scm/linux/kernel/git/rppt/linux.git gfp-to-kmalloc/rdma

[1] https://lore.kernel.org/all/20251018093002.3660549-1-rppt@kernel.org/
[2] https://lore.kernel.org/all/CA+55aFwp4iy4rtX2gE2WjBGFL=NxMVnoFeHqYa2j1dYOMMGqxg@mail.gmail.com/ 

---
v2 changes:
* add comment to keep markers for sites that need "fast and as large as
  possible" allocation helper

v1: https://patch.msgid.link/20260630-b4-rdma-v1-0-ab42bcf0de92@kernel.org

---
Mike Rapoport (Microsoft) (5):
      RDMA/umem: ib_umem_get(): use kmalloc() to allocate page array
      RDMA/mlx5: replace __get_free_page() with kmalloc()
      IB/mthca: mthca_reg_user_mr(): use kmalloc() to allocate addresses array
      IB/mthca: allocate mthca_array memory with kzalloc()
      IB/rdmavt: use kzalloc() to allocate QPN-map pages

 drivers/infiniband/core/umem.c                | 5 +++--
 drivers/infiniband/hw/mlx5/odp.c              | 6 ++++--
 drivers/infiniband/hw/mthca/mthca_allocator.c | 6 +++---
 drivers/infiniband/hw/mthca/mthca_provider.c  | 5 +++--
 drivers/infiniband/sw/rdmavt/qp.c             | 8 ++++----
 5 files changed, 17 insertions(+), 13 deletions(-)
---
base-commit: dc59e4fea9d83f03bad6bddf3fa2e52491777482
change-id: 20260610-b4-rdma-44625922fe16

--
Sincerely yours,
Mike.

On Mon, 13 Jul 2026 10:17:21 +0300, Mike Rapoport (Microsoft) wrote:
> This is a (small) part of larger work of replacing page allocator calls
> with kmalloc.
> 
> My initial intention a few month ago was to remove ugly casts [1], but then
> willy pointed out that Linus objected to something like this [2] and it
> looks like more than a decade old technical debt.
> 
> [...]

Applied, thanks!

[1/5] RDMA/umem: ib_umem_get(): use kmalloc() to allocate page array
      https://git.kernel.org/rdma/rdma/c/66073100a7b857
[2/5] RDMA/mlx5: replace __get_free_page() with kmalloc()
      https://git.kernel.org/rdma/rdma/c/e3d8c413e2e9e8
[3/5] IB/mthca: mthca_reg_user_mr(): use kmalloc() to allocate addresses array
      https://git.kernel.org/rdma/rdma/c/5036553f0e39e2
[4/5] IB/mthca: allocate mthca_array memory with kzalloc()
      https://git.kernel.org/rdma/rdma/c/45bf0b3da47d75
[5/5] IB/rdmavt: use kzalloc() to allocate QPN-map pages
      https://git.kernel.org/rdma/rdma/c/f8d04b0c74e989

Best regards,
-- 
Leon Romanovsky <leon@kernel.org>

On Mon, Jul 13, 2026 at 10:17:21AM +0300, Mike Rapoport (Microsoft) wrote:
> This is a (small) part of larger work of replacing page allocator calls
> with kmalloc.
> 
> My initial intention a few month ago was to remove ugly casts [1], but then
> willy pointed out that Linus objected to something like this [2] and it
> looks like more than a decade old technical debt.
> 
> Largely, anything that doesn't need struct page (or a memdesc in the
> future) should just use kmalloc() or kvmalloc() to allocate memory.
> kmalloc() guarantees alignment, physical contiguity and working
> virt_to_phys() and beside nicer API that returns void * on alloc and
> doesn't require to know the allocation size on free, kmalloc() provides
> better debugging capabilities than page allocator.
> 
> Another thing is that touching these allocation sites gives the reviewers
> opportunity to see if a PAGE_SIZE buffer is actually needed or maybe
> another size is appropriate.
> 
> For larger allocations that don't need physically contiguous memory
> kvmalloc() can be a better option that __get_free_pages() because under
> memory pressure it's is easier to allocate several order-0 pages than a
> physically contiguous chunk with the same number of pages.
> 
> And last, but not least, removing needless calls to page allocator should
> help with memdesc (aka project folio) conversion. There will be way less
> places to audit to see if the user was actually using struct page.
> 
> Also in git:
> https://git.kernel.org/pub/scm/linux/kernel/git/rppt/linux.git gfp-to-kmalloc/rdma

Will you also change get_zeroed_page() in RDMA, or should I
prepare a patch?

Thanks

> 
> [1] https://lore.kernel.org/all/20251018093002.3660549-1-rppt@kernel.org/
> [2] https://lore.kernel.org/all/CA+55aFwp4iy4rtX2gE2WjBGFL=NxMVnoFeHqYa2j1dYOMMGqxg@mail.gmail.com/ 
> 
> ---
> v2 changes:
> * add comment to keep markers for sites that need "fast and as large as
>   possible" allocation helper
> 
> v1: https://patch.msgid.link/20260630-b4-rdma-v1-0-ab42bcf0de92@kernel.org
> 
> ---
> Mike Rapoport (Microsoft) (5):
>       RDMA/umem: ib_umem_get(): use kmalloc() to allocate page array
>       RDMA/mlx5: replace __get_free_page() with kmalloc()
>       IB/mthca: mthca_reg_user_mr(): use kmalloc() to allocate addresses array
>       IB/mthca: allocate mthca_array memory with kzalloc()
>       IB/rdmavt: use kzalloc() to allocate QPN-map pages
> 
>  drivers/infiniband/core/umem.c                | 5 +++--
>  drivers/infiniband/hw/mlx5/odp.c              | 6 ++++--
>  drivers/infiniband/hw/mthca/mthca_allocator.c | 6 +++---
>  drivers/infiniband/hw/mthca/mthca_provider.c  | 5 +++--
>  drivers/infiniband/sw/rdmavt/qp.c             | 8 ++++----
>  5 files changed, 17 insertions(+), 13 deletions(-)
> ---
> base-commit: dc59e4fea9d83f03bad6bddf3fa2e52491777482
> change-id: 20260610-b4-rdma-44625922fe16
> 
> --
> Sincerely yours,
> Mike.
>

On Tue, Jul 14, 2026 at 03:24:56PM +0300, Leon Romanovsky wrote:
> On Mon, Jul 13, 2026 at 10:17:21AM +0300, Mike Rapoport (Microsoft) wrote:
> > This is a (small) part of larger work of replacing page allocator calls
> > with kmalloc.
> > 
> > My initial intention a few month ago was to remove ugly casts [1], but then
> > willy pointed out that Linus objected to something like this [2] and it
> > looks like more than a decade old technical debt.
> > 
> > Largely, anything that doesn't need struct page (or a memdesc in the
> > future) should just use kmalloc() or kvmalloc() to allocate memory.
> > kmalloc() guarantees alignment, physical contiguity and working
> > virt_to_phys() and beside nicer API that returns void * on alloc and
> > doesn't require to know the allocation size on free, kmalloc() provides
> > better debugging capabilities than page allocator.
> > 
> > Another thing is that touching these allocation sites gives the reviewers
> > opportunity to see if a PAGE_SIZE buffer is actually needed or maybe
> > another size is appropriate.
> > 
> > For larger allocations that don't need physically contiguous memory
> > kvmalloc() can be a better option that __get_free_pages() because under
> > memory pressure it's is easier to allocate several order-0 pages than a
> > physically contiguous chunk with the same number of pages.
> > 
> > And last, but not least, removing needless calls to page allocator should
> > help with memdesc (aka project folio) conversion. There will be way less
> > places to audit to see if the user was actually using struct page.
> > 
> > Also in git:
> > https://git.kernel.org/pub/scm/linux/kernel/git/rppt/linux.git gfp-to-kmalloc/rdma
> 
> Will you also change get_zeroed_page() in RDMA, or should I
> prepare a patch?

➜  kernel git:(wip/leon-for-next) git grep get_zeroed_page drivers/infiniband/
drivers/infiniband/hw/bnxt_re/ib_verbs.c:                       srq->uctx_srq_page = (void *)get_zeroed_page(GFP_KERNEL);
drivers/infiniband/hw/bnxt_re/ib_verbs.c:               cq->uctx_cq_page = (void *)get_zeroed_page(GFP_KERNEL);
drivers/infiniband/hw/bnxt_re/ib_verbs.c:       uctx->shpg = (void *)get_zeroed_page(GFP_KERNEL);
drivers/infiniband/hw/mlx4/mr.c:        mr->pages = (__be64 *)get_zeroed_page(GFP_KERNEL);
drivers/infiniband/hw/qedr/verbs.c:     q->db_rec_data = (void *)get_zeroed_page(GFP_USER);

Thanks

> 
> Thanks
> 
> > 
> > [1] https://lore.kernel.org/all/20251018093002.3660549-1-rppt@kernel.org/
> > [2] https://lore.kernel.org/all/CA+55aFwp4iy4rtX2gE2WjBGFL=NxMVnoFeHqYa2j1dYOMMGqxg@mail.gmail.com/ 
> > 
> > ---
> > v2 changes:
> > * add comment to keep markers for sites that need "fast and as large as
> >   possible" allocation helper
> > 
> > v1: https://patch.msgid.link/20260630-b4-rdma-v1-0-ab42bcf0de92@kernel.org
> > 
> > ---
> > Mike Rapoport (Microsoft) (5):
> >       RDMA/umem: ib_umem_get(): use kmalloc() to allocate page array
> >       RDMA/mlx5: replace __get_free_page() with kmalloc()
> >       IB/mthca: mthca_reg_user_mr(): use kmalloc() to allocate addresses array
> >       IB/mthca: allocate mthca_array memory with kzalloc()
> >       IB/rdmavt: use kzalloc() to allocate QPN-map pages
> > 
> >  drivers/infiniband/core/umem.c                | 5 +++--
> >  drivers/infiniband/hw/mlx5/odp.c              | 6 ++++--
> >  drivers/infiniband/hw/mthca/mthca_allocator.c | 6 +++---
> >  drivers/infiniband/hw/mthca/mthca_provider.c  | 5 +++--
> >  drivers/infiniband/sw/rdmavt/qp.c             | 8 ++++----
> >  5 files changed, 17 insertions(+), 13 deletions(-)
> > ---
> > base-commit: dc59e4fea9d83f03bad6bddf3fa2e52491777482
> > change-id: 20260610-b4-rdma-44625922fe16
> > 
> > --
> > Sincerely yours,
> > Mike.
> > 
> 

On Tue, Jul 14, 2026 at 03:28:18PM +0300, Leon Romanovsky wrote:
> On Tue, Jul 14, 2026 at 03:24:56PM +0300, Leon Romanovsky wrote:
> > On Mon, Jul 13, 2026 at 10:17:21AM +0300, Mike Rapoport (Microsoft) wrote:
> > > This is a (small) part of larger work of replacing page allocator calls
> > > with kmalloc.
> > 
> > Will you also change get_zeroed_page() in RDMA, or should I
> > prepare a patch?

A patch would be great!

I do plan to audit all __get_free_pages()/get_zeroed_page() etc call sites,
but obviously it will take some time.

For now I only sent patches where the replacement was obvious and didn't
require me to dig deep and I surely could miss some.
 
> ➜  kernel git:(wip/leon-for-next) git grep get_zeroed_page drivers/infiniband/
> drivers/infiniband/hw/bnxt_re/ib_verbs.c:                       srq->uctx_srq_page = (void *)get_zeroed_page(GFP_KERNEL);
> drivers/infiniband/hw/bnxt_re/ib_verbs.c:               cq->uctx_cq_page = (void *)get_zeroed_page(GFP_KERNEL);
> drivers/infiniband/hw/bnxt_re/ib_verbs.c:       uctx->shpg = (void *)get_zeroed_page(GFP_KERNEL);
> drivers/infiniband/hw/mlx4/mr.c:        mr->pages = (__be64 *)get_zeroed_page(GFP_KERNEL);
> drivers/infiniband/hw/qedr/verbs.c:     q->db_rec_data = (void *)get_zeroed_page(GFP_USER);
> 
> Thanks

-- 
Sincerely yours,
Mike.

On Wed, Jul 15, 2026 at 09:32:36AM +0300, Mike Rapoport wrote:
> On Tue, Jul 14, 2026 at 03:28:18PM +0300, Leon Romanovsky wrote:
> > On Tue, Jul 14, 2026 at 03:24:56PM +0300, Leon Romanovsky wrote:
> > > On Mon, Jul 13, 2026 at 10:17:21AM +0300, Mike Rapoport (Microsoft) wrote:
> > > > This is a (small) part of larger work of replacing page allocator calls
> > > > with kmalloc.
> > > 
> > > Will you also change get_zeroed_page() in RDMA, or should I
> > > prepare a patch?
> 
> A patch would be great!
> 
> I do plan to audit all __get_free_pages()/get_zeroed_page() etc call sites,
> but obviously it will take some time.
> 
> For now I only sent patches where the replacement was obvious and didn't
> require me to dig deep and I surely could miss some.

No problem, I'll handle it.

Thanks

>  
> > ➜  kernel git:(wip/leon-for-next) git grep get_zeroed_page drivers/infiniband/
> > drivers/infiniband/hw/bnxt_re/ib_verbs.c:                       srq->uctx_srq_page = (void *)get_zeroed_page(GFP_KERNEL);
> > drivers/infiniband/hw/bnxt_re/ib_verbs.c:               cq->uctx_cq_page = (void *)get_zeroed_page(GFP_KERNEL);
> > drivers/infiniband/hw/bnxt_re/ib_verbs.c:       uctx->shpg = (void *)get_zeroed_page(GFP_KERNEL);
> > drivers/infiniband/hw/mlx4/mr.c:        mr->pages = (__be64 *)get_zeroed_page(GFP_KERNEL);
> > drivers/infiniband/hw/qedr/verbs.c:     q->db_rec_data = (void *)get_zeroed_page(GFP_USER);
> > 
> > Thanks
> 
> -- 
> Sincerely yours,
> Mike.