Re: [PATCH 5/5] s390/hmcdrv: Replace page allocator calls with k[mz]alloc()

From: Mike Rapoport

Date: Tue Nov 17 2026 - 07:45:03 EST


On Mon, Nov 16, 2026 at 02:46:56PM +0100, Heiko Carstens wrote:
> On Sun, Nov 08, 2026 at 10:49:02AM +0200, Mike Rapoport (Microsoft) wrote:
> > diag_ftp_cmd() allocates the DIAGNOSE 0x2C4 parameter list used for HMC
> > file transfers under z/VM, hmcdrv_ftp_probe() and hmcdrv_ftp_cmd()
> > allocate the data buffers used for the transfers and
> > hmcdrv_cache_startup() allocates the buffer that caches the contents of
> > the last file read from the HMC drive.
> >
> > These buffers can be allocated with kmalloc() as there's nothing special
> > about them to go directly to the page allocator.
> >
> > kmalloc() provides a better API that does not require ugly casts and
> > kfree() does not need to know the size of the freed object.
> >
> > Performance difference between kmalloc() and __get_free_pages() is not
> > measurable as both allocators take an object/page from a per-CPU list for
> > fast path allocations.
> >
> > For the slow path the performance is anyway determined by the amount of
> > reclaim involved rather than by what allocator is used.
> >
> > Replace use of get_zeroed_page() and __get_free_pages() with kzalloc()
> > and kmalloc() and free_page[s]() with kfree().
> >
> > Round transfer and cache buffers up to a page for the required 4K
> > alignment, without rounding to a power-of-two number of pages. Reject
> > unsupported sizes before rounding.
> >
> > Link: https://lore.kernel.org/all/635405e4-9423-4a25-a6e7-e03c8ea0bcbe@xxxxxxxxxx
> > Signed-off-by: Mike Rapoport (Microsoft) <rppt@xxxxxxxxxx>
> > ---
> > drivers/s390/char/diag_ftp.c | 5 +++--
> > drivers/s390/char/hmcdrv_cache.c | 18 +++++++++---------
> > drivers/s390/char/hmcdrv_ftp.c | 14 +++++++-------
> > 3 files changed, 19 insertions(+), 18 deletions(-)
>
> ...
>
> > int hmcdrv_cache_startup(size_t cachesize)
> > {
> > + if (cachesize > KMALLOC_MAX_SIZE)
> > + return -ENOMEM;
>
> What is the reason for this check? I can see that this seems to be a
> common pattern, however the buddy allocator would warn as well for
> allocations larger than MAX_PAGE_ORDER if I'm not mistaken.

Since cachesize is a module parameter there's a very corner case when it
could have value within a page of ULONG_MAX. Then round_up() wraps to 0 and
kmalloc(0) returns ZERO_SIZE_PTR rather than NULL and does not warn.

The module would load with a bogus cache instead of failing as it does now.

While it's really unlikely to happen in practice I'm sure Sashiko will be
unhappy without this check :)

> > +
> > if (cachesize > 0) { /* perform caching ? */
> > - hmcdrv_cache_order = get_order(cachesize);
> > + /* the buffer must be 4k aligned, hence round up to page size */
> > hmcdrv_cache_file.content =
> > - (void *) __get_free_pages(GFP_KERNEL | GFP_DMA,
> > - hmcdrv_cache_order);
> > + kmalloc(round_up(cachesize, PAGE_SIZE),
> > + GFP_KERNEL | GFP_DMA);
>
> ...and that check wasn't here before.
>
> > diff --git a/drivers/s390/char/hmcdrv_ftp.c b/drivers/s390/char/hmcdrv_ftp.c
> > index 3312b2ac00a9d..e22789b532de2 100644
> > --- a/drivers/s390/char/hmcdrv_ftp.c
> > +++ b/drivers/s390/char/hmcdrv_ftp.c
> > @@ -238,16 +238,16 @@ EXPORT_SYMBOL(hmcdrv_ftp_probe);
> > ssize_t hmcdrv_ftp_cmd(char __kernel *cmd, loff_t offset,
> > char __user *buf, size_t len)
> > {
> > - int order;
> > -
> > struct hmcdrv_ftp_cmdspec ftp = {.len = len, .ofs = offset};
> > ssize_t retlen = hmcdrv_ftp_parse(cmd, &ftp);
> >
> > if (retlen)
> > return retlen;
> >
> > - order = get_order(ftp.len);
> > - ftp.buf = (void *) __get_free_pages(GFP_KERNEL | GFP_DMA, order);
> > + if (!ftp.len || ftp.len > KMALLOC_MAX_SIZE)
> > + return -ENOMEM;
>
> Same here, and as far as I can tell the Sashiko finding seems to be
> valid as well, even though it looks to me as if this worked only by
> accident. Also, as far as I can see, the reported "delete" command is
> currently not used at all from the only existing user space
> application that I'm aware of. But let's not change this, please.

Sashiko is wrong about get_order(0) returning 0, it actually returns 52, so
__get_free_pages() returns NULL and we end up with -ENOMEM anyway :)

So it's either keep if (!ftp.len) and let it fail for "delete" like it does
today or have

size = round_up(ftp.len, PAGE_SIZE) ? : PAGE_SIZE;

and allocate at least a page for ftp.len == 0.

What's your preference?

The ftp.len > KMALLOC_MAX_SIZE check seems bogus indeed.

--
Sincerely yours,
Mike.