Re: [PATCH v10 13/69] mm/mmap: use maple tree for unmapped_area{_topdown}

Re: [PATCH v10 13/69] mm/mmap: use maple tree for unmapped_area{_topdown}

[Sven Schnelle]

Hi Liam,

Liam Howlett <liam.howlett@oracle.com> writes:

> From: "Liam R. Howlett" <Liam.Howlett@Oracle.com>
>
> The maple tree code was added to find the unmapped area in a previous
> commit and was checked against what the rbtree returned, but the actual
> result was never used.  Start using the maple tree implementation and
> remove the rbtree code.
>
> Add kernel documentation comment for these functions.
>
> Link: https://lkml.kernel.org/r/20220504010716.661115-15-Liam.Howlett@oracle.com
> Signed-off-by: Liam R. Howlett <Liam.Howlett@Oracle.com>
> Cc: Catalin Marinas <catalin.marinas@arm.com>
> Cc: David Howells <dhowells@redhat.com>
> Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
> Cc: SeongJae Park <sj@kernel.org>
> Cc: Vlastimil Babka <vbabka@suse.cz>
> Cc: Will Deacon <will@kernel.org>
> Cc: Davidlohr Bueso <dave@stgolabs.net>
> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
> ---
>  mm/mmap.c | 253 +++++++-----------------------------------------------
>  1 file changed, 32 insertions(+), 221 deletions(-)

With next-20220718 git bisect pointed me to this commit because the
following code fails in compat mode (31 bit addressing) on s390:

#include <unistd.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <stdio.h>

#define __2GB__         0x80000000
#define VMA_SIZE        0x10000000
#define VMA_COUNT       (__2GB__ / VMA_SIZE)

/*
 * Repeated mapping below 2GB until all memory is exhausted
 */
int main(void)
{
        void *p;
        unsigned int i;

        for (i = 0; i < VMA_COUNT; i++) {
            p = mmap(NULL, VMA_SIZE, PROT_READ,MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
            printf("%d: %08x - %08x\n", i, (int)p, (int)p + VMA_SIZE);
            if (p == MAP_FAILED)
                  break;
            if (p > (void *)__2GB__) {
                  return 1;
            }
        }
        if ((i == 0) || (i >= VMA_COUNT))
               return 2;
        return 0;
}

With linux/master:
# ./mmap_compat_anon
0: 67a00000 - 77a00000
1: 57a00000 - 67a00000
2: 47a00000 - 57a00000
3: 37a00000 - 47a00000
4: 27a00000 - 37a00000
5: 17a00000 - 27a00000
6: 07a00000 - 17a00000
7: ffffffff - 0fffffff

With next-20220718 (and older, we're seeing this for a while, but i
didn't notice due to vacation and other priorities):

# ./mmap_compat_anon
0: 67600000 - 77600000
1: 57600000 - 67600000
2: 47600000 - 57600000
3: 37600000 - 47600000
4: 27600000 - 37600000
5: 17600000 - 27600000
6: 07600000 - 17600000
7: 40000000 - 50000000

So the first output is correct - mmap succeeds until there's no space
left, where it returns MAP_FAILED. In the lower (failing) one, mmap
instead returns an address that is already in use.

Regards
Sven

Re: [PATCH v10 13/69] mm/mmap: use maple tree for unmapped_area{_topdown}

[Liam Howlett]

* Sven Schnelle <svens@linux.ibm.com> [220719 10:21]:
> Hi Liam,
> 
> Liam Howlett <liam.howlett@oracle.com> writes:
> 
> > From: "Liam R. Howlett" <Liam.Howlett@Oracle.com>
> >
> > The maple tree code was added to find the unmapped area in a previous
> > commit and was checked against what the rbtree returned, but the actual
> > result was never used.  Start using the maple tree implementation and
> > remove the rbtree code.
> >
> > Add kernel documentation comment for these functions.
> >
> > Link: https://lkml.kernel.org/r/20220504010716.661115-15-Liam.Howlett@oracle.com
> > Signed-off-by: Liam R. Howlett <Liam.Howlett@Oracle.com>
> > Cc: Catalin Marinas <catalin.marinas@arm.com>
> > Cc: David Howells <dhowells@redhat.com>
> > Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
> > Cc: SeongJae Park <sj@kernel.org>
> > Cc: Vlastimil Babka <vbabka@suse.cz>
> > Cc: Will Deacon <will@kernel.org>
> > Cc: Davidlohr Bueso <dave@stgolabs.net>
> > Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
> > ---
> >  mm/mmap.c | 253 +++++++-----------------------------------------------
> >  1 file changed, 32 insertions(+), 221 deletions(-)
> 
> With next-20220718 git bisect pointed me to this commit because the
> following code fails in compat mode (31 bit addressing) on s390:
> 
> #include <unistd.h>
> #include <stdlib.h>
> #include <sys/mman.h>
> #include <stdio.h>
> 
> #define __2GB__         0x80000000
> #define VMA_SIZE        0x10000000
> #define VMA_COUNT       (__2GB__ / VMA_SIZE)
> 
> /*
>  * Repeated mapping below 2GB until all memory is exhausted
>  */
> int main(void)
> {
>         void *p;
>         unsigned int i;
> 
>         for (i = 0; i < VMA_COUNT; i++) {
>             p = mmap(NULL, VMA_SIZE, PROT_READ,MAP_ANONYMOUS | MAP_PRIVATE, -1, 0);
>             printf("%d: %08x - %08x\n", i, (int)p, (int)p + VMA_SIZE);
>             if (p == MAP_FAILED)
>                   break;
>             if (p > (void *)__2GB__) {
>                   return 1;
>             }
>         }
>         if ((i == 0) || (i >= VMA_COUNT))
>                return 2;
>         return 0;
> }
> 
> With linux/master:
> # ./mmap_compat_anon
> 0: 67a00000 - 77a00000
> 1: 57a00000 - 67a00000
> 2: 47a00000 - 57a00000
> 3: 37a00000 - 47a00000
> 4: 27a00000 - 37a00000
> 5: 17a00000 - 27a00000
> 6: 07a00000 - 17a00000
> 7: ffffffff - 0fffffff
> 
> With next-20220718 (and older, we're seeing this for a while, but i
> didn't notice due to vacation and other priorities):
> 
> # ./mmap_compat_anon
> 0: 67600000 - 77600000
> 1: 57600000 - 67600000
> 2: 47600000 - 57600000
> 3: 37600000 - 47600000
> 4: 27600000 - 37600000
> 5: 17600000 - 27600000
> 6: 07600000 - 17600000
> 7: 40000000 - 50000000
> 
> So the first output is correct - mmap succeeds until there's no space
> left, where it returns MAP_FAILED. In the lower (failing) one, mmap
> instead returns an address that is already in use.
> 

I'll have a look.

Thanks,
Liam