On Thu, Aug 29, 2019 at 05:32:46PM +0200, Christophe Leroy wrote:
> Hi Eric,
>
>
> Le 01/02/2019 à 08:51, Eric Biggers a écrit :
> > From: Eric Biggers <ebiggers@xxxxxxxxxx>
> >
> > Convert alg_test_hash() to use the new test framework, adding a list of
> > testvec_configs to test by default. When the extra self-tests are
> > enabled, randomly generated testvec_configs are tested as well.
> >
> > This improves hash test coverage mainly because now all algorithms have
> > a variety of data layouts tested, whereas before each algorithm was
> > responsible for declaring its own chunked test cases which were often
> > missing or provided poor test coverage. The new code also tests both
> > the MAY_SLEEP and !MAY_SLEEP cases and buffers that cross pages.
> >
> > This already found bugs in the hash walk code and in the arm32 and arm64
> > implementations of crct10dif.
> >
> > I removed the hash chunked test vectors that were the same as
> > non-chunked ones, but left the ones that were unique.
> >
> > Signed-off-by: Eric Biggers <ebiggers@xxxxxxxxxx>
> > ---
> > crypto/testmgr.c | 795 ++++++++++++++++++++---------------------------
> > crypto/testmgr.h | 107 +------
> > 2 files changed, 352 insertions(+), 550 deletions(-)
> >
> > diff --git a/crypto/testmgr.c b/crypto/testmgr.c
> > index 7638090ff1b0a..96aa268ff4184 100644
> > --- a/crypto/testmgr.c
> > +++ b/crypto/testmgr.c
>
> [...]
>
>
> > -static int __test_hash(struct crypto_ahash *tfm,
> > - const struct hash_testvec *template, unsigned int tcount,
> > - enum hash_test test_type, const int align_offset)
> > +static int test_hash_vec_cfg(const char *driver,
> > + const struct hash_testvec *vec,
> > + unsigned int vec_num,
> > + const struct testvec_config *cfg,
> > + struct ahash_request *req,
> > + struct test_sglist *tsgl,
> > + u8 *hashstate)
> > {
> > - const char *algo = crypto_tfm_alg_driver_name(crypto_ahash_tfm(tfm));
> > - size_t digest_size = crypto_ahash_digestsize(tfm);
> > - unsigned int i, j, k, temp;
> > - struct scatterlist sg[8];
> > - char *result;
> > - char *key;
> > - struct ahash_request *req;
> > - struct crypto_wait wait;
> > - void *hash_buff;
> > - char *xbuf[XBUFSIZE];
> > - int ret = -ENOMEM;
> > -
> > - result = kmalloc(digest_size, GFP_KERNEL);
> > - if (!result)
> > - return ret;
> > - key = kmalloc(MAX_KEYLEN, GFP_KERNEL);
> > - if (!key)
> > - goto out_nobuf;
> > - if (testmgr_alloc_buf(xbuf))
> > - goto out_nobuf;
> > + struct crypto_ahash *tfm = crypto_ahash_reqtfm(req);
> > + const unsigned int alignmask = crypto_ahash_alignmask(tfm);
> > + const unsigned int digestsize = crypto_ahash_digestsize(tfm);
> > + const unsigned int statesize = crypto_ahash_statesize(tfm);
> > + const u32 req_flags = CRYPTO_TFM_REQ_MAY_BACKLOG | cfg->req_flags;
> > + const struct test_sg_division *divs[XBUFSIZE];
> > + DECLARE_CRYPTO_WAIT(wait);
> > + struct kvec _input;
> > + struct iov_iter input;
> > + unsigned int i;
> > + struct scatterlist *pending_sgl;
> > + unsigned int pending_len;
> > + u8 result[HASH_MAX_DIGESTSIZE + TESTMGR_POISON_LEN];
>
> Before this patch, result was allocated with kmalloc().
> Now, result is in the stack. Is there a reason for this change ?
>
> Due to this change, the talitos driver fails when using CONFIG_VMAP_STACK,
> because result is not dma-able anymore.
>
Yes, moving 'result' to the stack simplified the code slightly and also helps
detect buggy drivers that assume that 'result' is DMA-able. Note that some
"real" API users also put 'result' on the stack, as the API allows it.
This is not new; the only new thing is that the tests now test it.
>
> How should this be fixed ?
>
Take a look at how other drivers have fixed this issue in the past, e.g.
commit c19650d6ea9 ("crypto: caam - fix DMA mapping of stack memory").
Seems that you need to DMA into a temporary buffer.
- Eric
