2016-04-21 11:07:59

by Sowmini Varadhan

Subject: [PATCH] lib/mpi: Fix kernel unaligned acces in mpi_write_to_sgl

Commit 2d4d1eea540b ("lib/mpi: Add mpi sgl helpers") added
mpi_write_to_sgl() which generates traps due to unaligned
access on some platforms like sparc. Fix this by using
the get_unaligned* and put_unaligned* functions.

Fixes: 2d4d1eea540b ("lib/mpi: Add mpi sgl helpers")
Signed-off-by: Sowmini Varadhan <[email protected]>
---
lib/mpi/mpicoder.c | 12 ++++++++----
1 files changed, 8 insertions(+), 4 deletions(-)

diff --git a/lib/mpi/mpicoder.c b/lib/mpi/mpicoder.c
index eb15e7d..6771378 100644
--- a/lib/mpi/mpicoder.c
+++ b/lib/mpi/mpicoder.c
@@ -21,6 +21,7 @@
#include <linux/bitops.h>
#include <linux/count_zeros.h>
#include "mpi-internal.h"
+#include <asm/unaligned.h>

#define MAX_EXTERN_MPI_BITS 16384

@@ -405,10 +406,13 @@ int mpi_write_to_sgl(MPI a, struct scatterlist *sgl, unsigned *nbytes,
p -= sizeof(alimb);
continue;
} else {
- mpi_limb_t *limb1 = (void *)p - sizeof(alimb);
- mpi_limb_t *limb2 = (void *)p - sizeof(alimb)
- + lzeros;
- *limb1 = *limb2;
+ mpi_limb_t tmp;
+
+ tmp = get_unaligned_be32((void *)p -
+ sizeof(alimb) +
+ lzeros);
+ put_unaligned_be32(tmp,
+ (void *)p - sizeof(alimb));
p -= lzeros;
y = lzeros;
}
--
1.7.1

2016-04-21 17:28:46

by Tadeusz Struk

Subject: Re: [PATCH] lib/mpi: Fix kernel unaligned acces in mpi_write_to_sgl

On 04/21/2016 04:07 AM, Sowmini Varadhan wrote:
>
> Commit 2d4d1eea540b ("lib/mpi: Add mpi sgl helpers") added
> mpi_write_to_sgl() which generates traps due to unaligned
> access on some platforms like sparc. Fix this by using
> the get_unaligned* and put_unaligned* functions.
>
> Fixes: 2d4d1eea540b ("lib/mpi: Add mpi sgl helpers")
> Signed-off-by: Sowmini Varadhan <[email protected]>
> ---
> lib/mpi/mpicoder.c | 12 ++++++++----
> 1 files changed, 8 insertions(+), 4 deletions(-)
>
> diff --git a/lib/mpi/mpicoder.c b/lib/mpi/mpicoder.c
> index eb15e7d..6771378 100644
> --- a/lib/mpi/mpicoder.c
> +++ b/lib/mpi/mpicoder.c
> @@ -21,6 +21,7 @@
> #include <linux/bitops.h>
> #include <linux/count_zeros.h>
> #include "mpi-internal.h"
> +#include <asm/unaligned.h>
>
> #define MAX_EXTERN_MPI_BITS 16384
>
> @@ -405,10 +406,13 @@ int mpi_write_to_sgl(MPI a, struct scatterlist *sgl, unsigned *nbytes,
> p -= sizeof(alimb);
> continue;
> } else {
> - mpi_limb_t *limb1 = (void *)p - sizeof(alimb);
> - mpi_limb_t *limb2 = (void *)p - sizeof(alimb)
> - + lzeros;
> - *limb1 = *limb2;
> + mpi_limb_t tmp;
> +
> + tmp = get_unaligned_be32((void *)p -
> + sizeof(alimb) +
> + lzeros);
> + put_unaligned_be32(tmp,
> + (void *)p - sizeof(alimb));
> p -= lzeros;
> y = lzeros;
> }

What if the mpi_limb_t will happen to be 64 bit?
Thanks,
--
TS

2016-04-21 17:45:22

by Sowmini Varadhan

Subject: Re: [PATCH] lib/mpi: Fix kernel unaligned acces in mpi_write_to_sgl

On (04/21/16 10:23), Tadeusz Struk wrote:
>
> What if the mpi_limb_t will happen to be 64 bit?
> Thanks,

When I checked this with cscope, I found

typedef unsigned long int mpi_limb_t;

thus I used the *32 functions.

But you obviously know better, since you wrote this code (and bug).
If you anticipate that mpi_limb_t in some environment today,
I can check for sizeof(mpi_limb_t), and predicate it to
use the *32 or *64 functions based on the result. Do you think that
is necessary?

Regards,
--Sowmini