[PATCH] crypto: testmgr: Use linear alias for test input

[PATCH] crypto: testmgr: Use linear alias for test input

[Laura Abbott]

Christopher Covington reported a crash on aarch64 on recent Fedora
kernels:

kernel BUG at ./include/linux/scatterlist.h:140!
Internal error: Oops - BUG: 0 [#1] PREEMPT SMP
Modules linked in:
CPU: 2 PID: 752 Comm: cryptomgr_test Not tainted 4.9.0-11815-ge93b1cc #162
Hardware name: linux,dummy-virt (DT)
task: ffff80007c650080 task.stack: ffff800008910000
PC is at sg_init_one+0xa0/0xb8
LR is at sg_init_one+0x24/0xb8
...
[<ffff000008398db8>] sg_init_one+0xa0/0xb8
[<ffff000008350a44>] test_acomp+0x10c/0x438
[<ffff000008350e20>] alg_test_comp+0xb0/0x118
[<ffff00000834f28c>] alg_test+0x17c/0x2f0
[<ffff00000834c6a4>] cryptomgr_test+0x44/0x50
[<ffff0000080dac70>] kthread+0xf8/0x128
[<ffff000008082ec0>] ret_from_fork+0x10/0x50

The test vectors used for input are part of the kernel image. These
inputs are passed as a buffer to sg_init_one which eventually blows up
with BUG_ON(!virt_addr_valid(buf)). On arm64, virt_addr_valid returns
false for the kernel image since virt_to_page will not return the
correct page. The kernel image is also aliased to the linear map so get
the linear alias and pass that to the scatterlist instead.

Reported-by: Christopher Covington <cov@codeaurora.org>
Fixes: d7db7a882deb ("crypto: acomp - update testmgr with support for acomp")
Signed-off-by: Laura Abbott <labbott@redhat.com>
---
x86 supports virt_addr_valid working on kernel image addresses but arm64 is
more strict. This is the direction things have been moving with my
CONFIG_DEBUG_VIRTUAL series for arm64 which is tightening the definition of
__pa/__pa_symbol.
---
 crypto/testmgr.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/crypto/testmgr.c b/crypto/testmgr.c
index f616ad7..f5bac10 100644
--- a/crypto/testmgr.c
+++ b/crypto/testmgr.c
@@ -1464,7 +1464,7 @@ static int test_acomp(struct crypto_acomp *tfm, struct comp_testvec *ctemplate,
 
 		memset(output, 0, dlen);
 		init_completion(&result.completion);
-		sg_init_one(&src, ctemplate[i].input, ilen);
+		sg_init_one(&src, __va(__pa_symbol(ctemplate[i].input)), ilen);
 		sg_init_one(&dst, output, dlen);
 
 		req = acomp_request_alloc(tfm);
@@ -1513,7 +1513,7 @@ static int test_acomp(struct crypto_acomp *tfm, struct comp_testvec *ctemplate,
 
 		memset(output, 0, dlen);
 		init_completion(&result.completion);
-		sg_init_one(&src, dtemplate[i].input, ilen);
+		sg_init_one(&src, __va(__pa_symbol(dtemplate[i].input)), ilen);
 		sg_init_one(&dst, output, dlen);
 
 		req = acomp_request_alloc(tfm);
-- 
2.7.4

Re: [PATCH] crypto: testmgr: Use linear alias for test input

[Ard Biesheuvel]

On 19 December 2016 at 23:37, Laura Abbott <labbott@redhat.com> wrote:
> Christopher Covington reported a crash on aarch64 on recent Fedora
> kernels:
>
> kernel BUG at ./include/linux/scatterlist.h:140!
> Internal error: Oops - BUG: 0 [#1] PREEMPT SMP
> Modules linked in:
> CPU: 2 PID: 752 Comm: cryptomgr_test Not tainted 4.9.0-11815-ge93b1cc #162
> Hardware name: linux,dummy-virt (DT)
> task: ffff80007c650080 task.stack: ffff800008910000
> PC is at sg_init_one+0xa0/0xb8
> LR is at sg_init_one+0x24/0xb8
> ...
> [<ffff000008398db8>] sg_init_one+0xa0/0xb8
> [<ffff000008350a44>] test_acomp+0x10c/0x438
> [<ffff000008350e20>] alg_test_comp+0xb0/0x118
> [<ffff00000834f28c>] alg_test+0x17c/0x2f0
> [<ffff00000834c6a4>] cryptomgr_test+0x44/0x50
> [<ffff0000080dac70>] kthread+0xf8/0x128
> [<ffff000008082ec0>] ret_from_fork+0x10/0x50
>
> The test vectors used for input are part of the kernel image. These
> inputs are passed as a buffer to sg_init_one which eventually blows up
> with BUG_ON(!virt_addr_valid(buf)). On arm64, virt_addr_valid returns
> false for the kernel image since virt_to_page will not return the
> correct page. The kernel image is also aliased to the linear map so get
> the linear alias and pass that to the scatterlist instead.
>
> Reported-by: Christopher Covington <cov@codeaurora.org>
> Fixes: d7db7a882deb ("crypto: acomp - update testmgr with support for acomp")
> Signed-off-by: Laura Abbott <labbott@redhat.com>
> ---
> x86 supports virt_addr_valid working on kernel image addresses but arm64 is
> more strict. This is the direction things have been moving with my
> CONFIG_DEBUG_VIRTUAL series for arm64 which is tightening the definition of
> __pa/__pa_symbol.

A helper function would be nice, so that we can call
sg_init_table/sg_set_page directly, and avoid the redundant
virt_to_phys(__va()) translation (and add a comment *why* we should
not use sg_init_one() with the address of a kernel symbol).

But I will leave it up to Herbert to decide whether he prefers that or not.

In any case,
Acked-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>

> ---
>  crypto/testmgr.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/crypto/testmgr.c b/crypto/testmgr.c
> index f616ad7..f5bac10 100644
> --- a/crypto/testmgr.c
> +++ b/crypto/testmgr.c
> @@ -1464,7 +1464,7 @@ static int test_acomp(struct crypto_acomp *tfm, struct comp_testvec *ctemplate,
>
>                 memset(output, 0, dlen);
>                 init_completion(&result.completion);
> -               sg_init_one(&src, ctemplate[i].input, ilen);
> +               sg_init_one(&src, __va(__pa_symbol(ctemplate[i].input)), ilen);
>                 sg_init_one(&dst, output, dlen);
>
>                 req = acomp_request_alloc(tfm);
> @@ -1513,7 +1513,7 @@ static int test_acomp(struct crypto_acomp *tfm, struct comp_testvec *ctemplate,
>
>                 memset(output, 0, dlen);
>                 init_completion(&result.completion);
> -               sg_init_one(&src, dtemplate[i].input, ilen);
> +               sg_init_one(&src, __va(__pa_symbol(dtemplate[i].input)), ilen);
>                 sg_init_one(&dst, output, dlen);
>
>                 req = acomp_request_alloc(tfm);
> --
> 2.7.4
>

Re: [PATCH] crypto: testmgr: Use linear alias for test input

[Herbert Xu]

On Mon, Dec 19, 2016 at 03:37:26PM -0800, Laura Abbott wrote:
> Christopher Covington reported a crash on aarch64 on recent Fedora
> kernels:
> 
> kernel BUG at ./include/linux/scatterlist.h:140!
> Internal error: Oops - BUG: 0 [#1] PREEMPT SMP
> Modules linked in:
> CPU: 2 PID: 752 Comm: cryptomgr_test Not tainted 4.9.0-11815-ge93b1cc #162
> Hardware name: linux,dummy-virt (DT)
> task: ffff80007c650080 task.stack: ffff800008910000
> PC is at sg_init_one+0xa0/0xb8
> LR is at sg_init_one+0x24/0xb8
> ...
> [<ffff000008398db8>] sg_init_one+0xa0/0xb8
> [<ffff000008350a44>] test_acomp+0x10c/0x438
> [<ffff000008350e20>] alg_test_comp+0xb0/0x118
> [<ffff00000834f28c>] alg_test+0x17c/0x2f0
> [<ffff00000834c6a4>] cryptomgr_test+0x44/0x50
> [<ffff0000080dac70>] kthread+0xf8/0x128
> [<ffff000008082ec0>] ret_from_fork+0x10/0x50
> 
> The test vectors used for input are part of the kernel image. These
> inputs are passed as a buffer to sg_init_one which eventually blows up
> with BUG_ON(!virt_addr_valid(buf)). On arm64, virt_addr_valid returns
> false for the kernel image since virt_to_page will not return the
> correct page. The kernel image is also aliased to the linear map so get
> the linear alias and pass that to the scatterlist instead.
> 
> Reported-by: Christopher Covington <cov@codeaurora.org>
> Fixes: d7db7a882deb ("crypto: acomp - update testmgr with support for acomp")
> Signed-off-by: Laura Abbott <labbott@redhat.com>
> ---
> x86 supports virt_addr_valid working on kernel image addresses but arm64 is
> more strict. This is the direction things have been moving with my
> CONFIG_DEBUG_VIRTUAL series for arm64 which is tightening the definition of
> __pa/__pa_symbol.

Please fix this by copying the templates to kmalloced memory like
the other test_* functions.

Thanks,
-- 
Email: Herbert Xu <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt