Re: [PATCH v4 2/3] mm: Add support for kmem caches in DMA32 zone

From: Nicolas Boichat
Date: Wed Dec 05 2018 - 22:50:17 EST


On Wed, Dec 5, 2018 at 10:02 PM Vlastimil Babka <vbabka@xxxxxxx> wrote:
>
> On 12/5/18 6:48 AM, Nicolas Boichat wrote:
> > In some cases (e.g. IOMMU ARMv7s page allocator), we need to allocate
> > data structures smaller than a page with GFP_DMA32 flag.
> >
> > This change makes it possible to create a custom cache in DMA32 zone
> > using kmem_cache_create, then allocate memory using kmem_cache_alloc.
> >
> > We do not create a DMA32 kmalloc cache array, as there are currently
> > no users of kmalloc(..., GFP_DMA32). The new test in check_slab_flags
> > ensures that such calls still fail (as they do before this change).
> >
> > Fixes: ad67f5a6545f ("arm64: replace ZONE_DMA with ZONE_DMA32")
>
> Same as my comment for 1/3.

I'll drop.

> > Signed-off-by: Nicolas Boichat <drinkcat@xxxxxxxxxxxx>
>
> In general,
> Acked-by: Vlastimil Babka <vbabka@xxxxxxx>
>
> Some comments below:
>
> > ---
> >
> > Changes since v2:
> > - Clarified commit message
> > - Add entry in sysfs-kernel-slab to document the new sysfs file
> >
> > (v3 used the page_frag approach)
> >
> > Documentation/ABI/testing/sysfs-kernel-slab | 9 +++++++++
> > include/linux/slab.h | 2 ++
> > mm/internal.h | 8 ++++++--
> > mm/slab.c | 4 +++-
> > mm/slab.h | 3 ++-
> > mm/slab_common.c | 2 +-
> > mm/slub.c | 18 +++++++++++++++++-
> > 7 files changed, 40 insertions(+), 6 deletions(-)
> >
> > diff --git a/Documentation/ABI/testing/sysfs-kernel-slab b/Documentation/ABI/testing/sysfs-kernel-slab
> > index 29601d93a1c2ea..d742c6cfdffbe9 100644
> > --- a/Documentation/ABI/testing/sysfs-kernel-slab
> > +++ b/Documentation/ABI/testing/sysfs-kernel-slab
> > @@ -106,6 +106,15 @@ Description:
> > are from ZONE_DMA.
> > Available when CONFIG_ZONE_DMA is enabled.
> >
> > +What: /sys/kernel/slab/cache/cache_dma32
> > +Date: December 2018
> > +KernelVersion: 4.21
> > +Contact: Nicolas Boichat <drinkcat@xxxxxxxxxxxx>
> > +Description:
> > + The cache_dma32 file is read-only and specifies whether objects
> > + are from ZONE_DMA32.
> > + Available when CONFIG_ZONE_DMA32 is enabled.
>
> I don't have a strong opinion. It's a new file, yeah, but consistent
> with already existing ones. I'd leave the decision with SL*B maintainers.
>
> > What: /sys/kernel/slab/cache/cpu_slabs
> > Date: May 2007
> > KernelVersion: 2.6.22
> > diff --git a/include/linux/slab.h b/include/linux/slab.h
> > index 11b45f7ae4057c..9449b19c5f107a 100644
> > --- a/include/linux/slab.h
> > +++ b/include/linux/slab.h
> > @@ -32,6 +32,8 @@
> > #define SLAB_HWCACHE_ALIGN ((slab_flags_t __force)0x00002000U)
> > /* Use GFP_DMA memory */
> > #define SLAB_CACHE_DMA ((slab_flags_t __force)0x00004000U)
> > +/* Use GFP_DMA32 memory */
> > +#define SLAB_CACHE_DMA32 ((slab_flags_t __force)0x00008000U)
> > /* DEBUG: Store the last owner for bug hunting */
> > #define SLAB_STORE_USER ((slab_flags_t __force)0x00010000U)
> > /* Panic if kmem_cache_create() fails */
> > diff --git a/mm/internal.h b/mm/internal.h
> > index a2ee82a0cd44ae..fd244ad716eaf8 100644
> > --- a/mm/internal.h
> > +++ b/mm/internal.h
> > @@ -14,6 +14,7 @@
> > #include <linux/fs.h>
> > #include <linux/mm.h>
> > #include <linux/pagemap.h>
> > +#include <linux/slab.h>
> > #include <linux/tracepoint-defs.h>
> >
> > /*
> > @@ -34,9 +35,12 @@
> > #define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
> >
> > /* Check for flags that must not be used with a slab allocator */
> > -static inline gfp_t check_slab_flags(gfp_t flags)
> > +static inline gfp_t check_slab_flags(gfp_t flags, slab_flags_t slab_flags)
> > {
> > - gfp_t bug_mask = __GFP_DMA32 | __GFP_HIGHMEM | ~__GFP_BITS_MASK;
> > + gfp_t bug_mask = __GFP_HIGHMEM | ~__GFP_BITS_MASK;
> > +
> > + if (!IS_ENABLED(CONFIG_ZONE_DMA32) || !(slab_flags & SLAB_CACHE_DMA32))
> > + bug_mask |= __GFP_DMA32;
>
> I'll point out that this is not even strictly needed AFAICS, as only
> flags passed to kmem_cache_alloc() are checked - the cache->allocflags
> derived from SLAB_CACHE_DMA32 are appended only after check_slab_flags()
> (in both SLAB and SLUB AFAICS). And for a cache created with
> SLAB_CACHE_DMA32, the caller of kmem_cache_alloc() doesn't need to also
> include __GFP_DMA32, the allocation will be from ZONE_DMA32 regardless.

Yes, you're right. I also looked at existing users of SLAB_CACHE_DMA,
and there is one case in drivers/scsi/scsi_lib.c where GFP_DMA is not
be passed (all the other users pass it).

I can drop GFP_DMA32 from my call in io-pgtable-arm-v7s.c.

> So it would be fine even unchanged. The check would anyway need some
> more love to catch the same with __GFP_DMA to be consistent and cover
> all corner cases.

Yes, the test is not complete. If we really wanted this to be
accurate, we'd need to check that GFP_* exactly matches SLAB_CACHE_*.

The only problem with dropping this is test that we should restore
GFP_DMA32 warning/errors somewhere else (as Christopher pointed out
here: https://lkml.org/lkml/2018/11/22/430), especially for kmalloc
case.

Maybe this can be done in kmalloc_slab.

> >
> > if (unlikely(flags & bug_mask)) {
> > gfp_t invalid_mask = flags & bug_mask;