Hi Tom,
On Wed, 24 Jul 2024 at 18:42, Tom Rini trini@konsulko.com wrote:
On Wed, Jul 24, 2024 at 10:34:50AM -0400, Raymond Mao wrote:
Hi Tom,
On Tue, 2 Jul 2024 at 21:26, Tom Rini trini@konsulko.com wrote:
On Tue, Jul 02, 2024 at 11:22:36AM -0700, Raymond Mao wrote:
Integrate MbedTLS v3.6 LTS (currently v3.6.0-RC1) with U-Boot.
Motivations:
- MbedTLS is well maintained with LTS versions.
- LWIP is integrated with MbedTLS and easily to enable HTTPS.
- MbedTLS recently switched license back to GPLv2.
Prerequisite:
This patch series requires mbedtls git repo to be added as a subtree to the main U-Boot repo via: $ git subtree add --prefix lib/mbedtls/external/mbedtls \ https://github.com/Mbed-TLS/mbedtls.git \ v3.6.0 --squash Moreover, due to the Windows-style files from mbedtls git repo, we need to convert the CRLF endings to LF and do a commit manually: $ git add --renormalize . $ git commit
New Kconfig options:
`MBEDTLS_LIB` is for MbedTLS general switch. `MBEDTLS_LIB_CRYPTO` is for replacing original digest and crypto libs
with
MbedTLS. `MBEDTLS_LIB_X509` is for replacing original X509, PKCS7, MSCode,
ASN1,
and Pubkey parser with MbedTLS. `MBEDTLS_LIB_TLS` is for SSL/TLS (Disabled until LWIP port for
MbedTLS is
ready). `LEGACY_CRYPTO` is introduced as a main switch for legacy crypto
library.
`LEGACY_CRYPTO_BASIC` is for the basic crypto functionalities and `LEGACY_CRYPTO_CERT` is for the certificate related functionalities. For each of the algorithm, a pair of `<alg>_LEGACY` and
`<alg>_MBEDTLS`
Kconfig options are introduced. Meanwhile, `SPL_` Kconfig options are introduced.
In this patch set, MBEDTLS_LIB, MBEDTLS_LIB_CRYPTO and
MBEDTLS_LIB_X509
are by default enabled in qemu_arm64_defconfig for testing purpose.
Patches for external MbedTLS project:
Since U-Boot uses Microsoft Authentication Code to verify PE/COFFs executables which is not supported by MbedTLS at the moment, addtional patches for MbedTLS are created to adapt with the EFI
loader:
- Decoding of Microsoft Authentication Code.
- Decoding of PKCS#9 Authenticate Attributes.
- Extending MbedTLS PKCS#7 lib to support multiple signer's
certificates.
- MbedTLS native test suites for PKCS#7 signer's info.
All above 4 patches (tagged with `mbedtls/external`) are submitted to MbedTLS project and being reviewed, eventually they should be part of MbedTLS LTS release. But before that, please merge them into U-Boot, otherwise the
building
will be broken when MBEDTLS_LIB_X509 is enabled.
See below PR link for the reference: https://github.com/Mbed-TLS/mbedtls/pull/9001
Miscellaneous:
Optimized MbedTLS library size by tailoring the config file and disabling all unnecessary features for EFI loader. From v2, original libs (rsa, asn1_decoder, rsa_helper, md5, sha1,
sha256,
sha512) are completely replaced when MbedTLS is enabled. From v3, the size-growth is slightly reduced by refactoring Hash
functions.
Target(QEMU arm64) size-growth when enabling MbedTLS: v1: 6.03% v2: 4.66% v3 & v4: 4.55%
Please see the latest output of bloat-o-meter for the reference of
the
size-growth on QEMU arm64 target [1].
Tests done:
EFI Secure Boot test (EFI variables loading and verifying, EFI signed
image
verifying and booting) via U-Boot console. EFI Secure Boot and Capsule sandbox test passed.
Known issues:
None.
[1]: bloat-o-meter output between disabling/enabling MbedTLS (QEMU
arm64)
add/remove: 206/81 grow/shrink: 19/17 up/down: 55548/-17495 (38053)bloat-o-meter is a bit off then, since buildman shows: u-boot: add: 243/-17, grow: 18/-17 bytes: 65723/-8480 (57243)
(Please use buildman for the size comparisons in the future).
The reason that buildman is showing more growth is because I enable
"CONFIG_EFI_SECURE_BOOT=y" in my patch, which is off by default for qemu_arm64. Since the buildman is always comparing one local branch with the 'upstream' (I didn't find a way to let it compare two local branches or maybe I am wrong), I guess I have to first merge one commit with just enabling CONFIG_EFI_SECURE_BOOT to solve this. But I get this makes less value...
I think it is better to use sandbox for comparison from v5, and I will add one more platform (e.g. imx8mp) for reference.
Please note that I check the world before/after, not just single platforms, for size growth. I'll note some examples of issues when I find them, typically. And with the wrapper I posted, sometimes I will "--all" a platform to see which commit increases things.
For your reference, below size-growth is for v5 (qemu_arm64, nanopi_a64,
sandbox):
aarch64: (for 2/2 boards) all +582.0 bss +40.0 data -64.0 rodata +206.0 text +400.0 qemu_arm64 : all +7040 bss +80 data -64 rodata +212 text +6812 u-boot: add: 28/-17, grow: 12/-16 bytes: 15492/-8304 (7188) nanopi_a64 : all -5876 data -64 rodata +200 text -6012 u-boot: add: 21/-8, grow: 4/-8 bytes: 12312/-4364 (7948) sandbox: (for 1/1 boards) all +22416.0 data +1440.0 rodata -4160.0 text +25136.0 sandbox : all +22416 data +1440 rodata -4160 text +25136 u-boot: add: 253/-203, grow: 115/-61 bytes: 93168/-76647 (16521)
I think the size-growth should be reasonable. I will attach the details in v5 cover-letter.
Regards, Raymond