Port CC 310 sha 512 driver #10913

RonEld · 2019-06-27T15:31:12Z

Description

Port the cc310 SHA512 driver, even though it is sw implementation.
Because the linker could not remove the cc310 sha512 implementation,
there was duplicate implementation of SHA512, without enabling
the sha512 alternative implementation.

There is a minor performance degradation (284 KB/s instead of 322 KB/s), however improvement in memory usage.
According to the generated map file, using GCC_ARM:
ROM: 81880 bytes smaller
RAM: 48 bytes smaller
Downside s the SHA384 is not supported with this driver introduction anymore.

Pull request type

[ ] Fix
[ ] Refactor
[ x ] Target update
[ ] Functionality change
[ ] Docs update
[ ] Test update
[ ] Breaking change

Reviewers

@ARMmbed/mbed-os-crypto

Release Notes

SHA384 is not supported, returning MBEDTLS_ERR_PLATFORM_FEATURE_UNSUPPORTED

Port the cc310 SHA512 driver, even though it is sw implementation. Because the linker could not remove the cc310 sha512 implementation, there was duplicate implementation of SHA512, without enabling the sha512 alternative implementation.

0Grit · 2019-06-27T15:35:42Z

81880 bytes less in flash isn't a mistype?

RonEld · 2019-06-27T15:42:54Z

It looked to me strange as well, however, looking at the generated map file with GCC)ARM toolchain I see ROM value:
With Mbed TLS sha512: 251084 bytes ( Mbed TLS alone is 113320)
With CC 310 SHA 512 impl.: 169204 ( Mbed TLS alone 63324)

ciarmcom · 2019-06-27T17:00:30Z

@RonEld, thank you for your changes.
@ARMmbed/mbed-os-crypto @ARMmbed/mbed-os-maintainers please review.

0Grit · 2019-06-27T17:12:51Z

If this works out, you have made our summer more interesting. Thanks, keep the optimizations coming.

0xc0170 · 2019-06-28T13:11:08Z

If this works out, you have made our summer more interesting. Thanks, keep the optimizations coming.

🌞 🍹

RonEld · 2019-07-01T12:25:58Z

looking at the generated map file, I think the numbers I found are incorrect and not realted to this change, unfortunately. There was probably some environment issues causeing the differnce.
I made a new map file and in this file, ROM size of sha512.o when using Mbed TLS sha 512 is 4440 bytes, while when using cc310 it is 62 bytes, so the decrease is not as big as 82 KB.
The size of ROM in latest measurement is 172428 bytes ( and not 251084 as I got in previous measurement :( ), so the ROM improvement is probably only 3224 bytes, and no change in RAM

0xc0170 · 2019-07-01T14:04:01Z

@ARMmbed/mbed-os-crypto Please review

gilles-peskine-arm · 2019-07-01T15:34:28Z

the SHA384 is not supported with this driver introduction anymore

Is this acceptable? Note that SHA-384 is used in the real world, especially for HTTPS on the web. I think most clients and servers also accept SHA-256 but not SHA-512. The reason they use SHA-384 in preference to SHA-256 is that it's faster on 64-bit machines, but they normally also support SHA-256 to interoperate with 32-bit endpoints or to save on the transmitted data size. Even so, losing SHA-384 can affect applications and should at least be clearly documented.

Patater · 2019-07-01T17:42:04Z

the SHA384 is not supported with this driver introduction anymore

Is this acceptable? Note that SHA-384 is used in the real world, especially for HTTPS on the web. I think most clients and servers also accept SHA-256 but not SHA-512. The reason they use SHA-384 in preference to SHA-256 is that it's faster on 64-bit machines, but they normally also support SHA-256 to interoperate with 32-bit endpoints or to save on the transmitted data size. Even so, losing SHA-384 can affect applications and should at least be clearly documented.

It's a limitation of opting in to hardware acceleration on this platform, which the system integrator would need to accept as valid for their expected use cases. We are making it opt-in, right? FEATURE_CRYPTOCELL310 is an opt-in thing?

RonEld · 2019-07-02T08:49:22Z

We are making it opt-in, right? FEATURE_CRYPTOCELL310 is an opt-in thing?

It is defined by default in targets.json for MCU_NRF52840 target.
I guess it can be opted out in mbed_app.json, but not sure on that, but this will remove all hw accelerated algorithms.
However, one can just undefine MBEDTLS_SHA512_ALT in the user config file if they must have SHA384

bulislaw · 2019-07-02T11:29:51Z

I'd rather avoid braking changes and do it as an option. We can break at next major version.

RonEld · 2019-07-02T12:12:04Z

The idea was to make it for the next major release, unless it is required for current minor release, in which case we should make it optional, by removing MBEDTLS_SHA512_ALT from mbedtls_device.h and users that want this feature will need to add this definition in mbed_app.json

Have the alternative sha512 undefined by default, in order not to break backwards compatability.

RonEld · 2019-07-08T15:33:50Z

I made the alternative sha512 optional. I will make a separate PR, updating the readme file

artokin · 2019-07-15T12:52:54Z

@ARMmbed/mbed-os-crypto , please add your review comments!

mbed-ci · 2019-07-19T11:00:17Z

Test run: SUCCESS

Summary: 11 of 11 test jobs passed
Build number : 1
Build artifacts

Patater · 2019-07-19T15:30:38Z

features/cryptocell/FEATURE_CRYPTOCELL310/sha512_alt.c

+int mbedtls_internal_sha512_process( mbedtls_sha512_context *ctx,
+                                     const unsigned char data[128] )
+{
+    return( MBEDTLS_ERR_PLATFORM_FEATURE_UNSUPPORTED );


Will this break applications that were using software SHA-512 previously? Should we return non-error here?

no.
This function is called only by our internal test.
Please look at #8728 and discuss @yanesca for more information

Patater · 2019-07-19T15:31:05Z

features/cryptocell/FEATURE_CRYPTOCELL310/sha512_alt.c

+void mbedtls_sha512_init( mbedtls_sha512_context *ctx )
+{
+    memset( ctx, 0, sizeof( mbedtls_sha512_context ) );
+


Remove extra newline here.

Patater · 2019-07-19T15:31:30Z

features/cryptocell/FEATURE_CRYPTOCELL310/sha512_alt.c

+        return ( MBEDTLS_ERR_PLATFORM_HW_ACCEL_FAILED );
+}
+#endif //MBEDTLS_SHA512_ALT
+


Remove extra newline here

Remove redundant extra lines.

RonEld · 2019-07-23T10:55:16Z

@Patater Thank you for your comments! I have addressed them.

Patater

LGTM

RonEld · 2019-07-24T08:44:45Z

@ARMmbed/mbed-os-maintainers Can we start the CI?

evedon · 2019-07-27T17:02:47Z

CI started

mbed-ci · 2019-07-27T17:38:39Z

Test run: SUCCESS

Summary: 11 of 11 test jobs passed
Build number : 2
Build artifacts

Port CC 310 sha512 driver

a060c7b

Port the cc310 SHA512 driver, even though it is sw implementation. Because the linker could not remove the cc310 sha512 implementation, there was duplicate implementation of SHA512, without enabling the sha512 alternative implementation.

ciarmcom requested review from a team June 27, 2019 17:00

ciarmcom added the needs: review label Jun 27, 2019

0xc0170 approved these changes Jun 28, 2019

View reviewed changes

Make the alternative sha512 optional

6b17df7

Have the alternative sha512 undefined by default, in order not to break backwards compatability.

RonEld mentioned this pull request Jul 8, 2019

Explaining how to opt in modules #10991

Merged

SeppoTakalo added the needs: CI label Jul 17, 2019

Patater suggested changes Jul 19, 2019

View reviewed changes

SeppoTakalo added needs: work and removed needs: CI needs: review labels Jul 22, 2019

Style fixes

4cf3e1c

Remove redundant extra lines.

Patater approved these changes Jul 23, 2019

View reviewed changes

SeppoTakalo added ready for merge release-version: 5.14.0-rc1 and removed needs: work labels Aug 1, 2019

SeppoTakalo merged commit 54d7d7e into ARMmbed:master Aug 1, 2019

SeppoTakalo removed the ready for merge label Aug 1, 2019

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Port CC 310 sha 512 driver #10913

Port CC 310 sha 512 driver #10913

RonEld commented Jun 27, 2019

0Grit commented Jun 27, 2019

RonEld commented Jun 27, 2019

ciarmcom commented Jun 27, 2019

0Grit commented Jun 27, 2019 •

edited

Loading

0xc0170 commented Jun 28, 2019

RonEld commented Jul 1, 2019 •

edited

Loading

0xc0170 commented Jul 1, 2019

gilles-peskine-arm commented Jul 1, 2019

Patater commented Jul 1, 2019

RonEld commented Jul 2, 2019

bulislaw commented Jul 2, 2019

RonEld commented Jul 2, 2019

RonEld commented Jul 8, 2019

artokin commented Jul 15, 2019

mbed-ci commented Jul 19, 2019

Patater Jul 19, 2019

RonEld Jul 21, 2019

Patater Jul 19, 2019

Patater Jul 19, 2019

RonEld commented Jul 23, 2019

Patater left a comment

RonEld commented Jul 24, 2019

evedon commented Jul 27, 2019

mbed-ci commented Jul 27, 2019

Port CC 310 sha 512 driver #10913

Port CC 310 sha 512 driver #10913

Conversation

RonEld commented Jun 27, 2019

Description

Pull request type

Reviewers

Release Notes

0Grit commented Jun 27, 2019

RonEld commented Jun 27, 2019

ciarmcom commented Jun 27, 2019

0Grit commented Jun 27, 2019 • edited Loading

0xc0170 commented Jun 28, 2019

RonEld commented Jul 1, 2019 • edited Loading

0xc0170 commented Jul 1, 2019

gilles-peskine-arm commented Jul 1, 2019

Patater commented Jul 1, 2019

RonEld commented Jul 2, 2019

bulislaw commented Jul 2, 2019

RonEld commented Jul 2, 2019

RonEld commented Jul 8, 2019

artokin commented Jul 15, 2019

mbed-ci commented Jul 19, 2019

Test run: SUCCESS

Patater Jul 19, 2019

Choose a reason for hiding this comment

RonEld Jul 21, 2019

Choose a reason for hiding this comment

Patater Jul 19, 2019

Choose a reason for hiding this comment

Patater Jul 19, 2019

Choose a reason for hiding this comment

RonEld commented Jul 23, 2019

Patater left a comment

Choose a reason for hiding this comment

RonEld commented Jul 24, 2019

evedon commented Jul 27, 2019

mbed-ci commented Jul 27, 2019

Test run: SUCCESS

0Grit commented Jun 27, 2019 •

edited

Loading

RonEld commented Jul 1, 2019 •

edited

Loading