[PATCH v8 03/27] lib: Adapt digest header files to MbedTLS
Raymond Mao
raymond.mao at linaro.org
Fri Oct 11 20:25:20 CEST 2024
Hi Tom,
On Wed, 9 Oct 2024 at 13:52, Tom Rini <trini at konsulko.com> wrote:
> On Wed, Oct 09, 2024 at 08:32:29PM +0300, Ilias Apalodimas wrote:
> > On Wed, 9 Oct 2024 at 18:32, Raymond Mao <raymond.mao at linaro.org> wrote:
> > >
> > > Hi Tom,
> > >
> > > On Wed, 9 Oct 2024 at 10:38, Tom Rini <trini at konsulko.com> wrote:
> > >>
> > >> On Thu, Oct 03, 2024 at 02:50:16PM -0700, Raymond Mao wrote:
> > >>
> > >> > Adapt digest header files to support both original libs and MbedTLS
> > >> > by switching on/off MBEDTLS_LIB_CRYPTO.
> > >> > Introduce <alg>_LEGACY kconfig for legacy hash implementations.
> > >> > sha256.o should depend on SHA256 kconfig only but not
> SUPPORT_EMMC_RPMB,
> > >> > SHA256 should be selected when SUPPORT_EMMC_RPMB is enabled instead.
> > >> >
> > >> > `IS_ENABLED` or `CONFIG_IS_ENABLED` is not applicable here, since
> > >> > including <linux/kconfig.h> causes undefined reference on schedule()
> > >> > with sandbox build, as <linux/kconfig.h> includes
> <generated/autoconf.h>
> > >> > which enables `CONFIG_HW_WATCHDOG` and `CONFIG_WATCHDOG` but no
> schedule()
> > >> > are defined in sandbox build,
> > >> > Thus we use `#if defined(CONFIG_MBEDTLS_LIB_CRYPTO)` instead.
> > >> >
> > >> > Signed-off-by: Raymond Mao <raymond.mao at linaro.org>
> > >> > Reviewed-by: Ilias Apalodimas <ilias.apalodimas at linaro.org>
> > >>
> > >> There's three platforms where we see something like:
> > >> arm: (for 1/1 boards) all +5651.0 data +112.0 rodata +139.0
> text +5400.0
> > >> o4-imx6ull-nano: all +5651 data +112 rodata +139 text
> +5400
> > >> u-boot: add: 23/0, grow: 1/0 bytes: 1172/0 (1172)
> > >> function old
> new delta
> > >> hash_command 108
> 296 +188
> > >> sha1_finish -
> 156 +156
> > >> static.sha1_update -
> 114 +114
> > >> hash_algo -
> 112 +112
> > >> sha1_padding -
> 64 +64
> > >> hash_lookup_algo -
> 60 +60
> > >> sha1_starts -
> 52 +52
> > >> crc16_ccitt_wd_buf -
> 36 +36
> > >> sha256_csum_wd -
> 34 +34
> > >> sha1_csum_wd -
> 34 +34
> > >> hash_finish_sha256 -
> 34 +34
> > >> hash_finish_sha1 -
> 34 +34
> > >> crc32_wd_buf -
> 34 +34
> > >> hash_finish_crc32 -
> 28 +28
> > >> hash_finish_crc16_ccitt -
> 28 +28
> > >> hash_init_sha256 -
> 22 +22
> > >> hash_init_sha1 -
> 22 +22
> > >> hash_update_crc32 -
> 20 +20
> > >> hash_update_crc16_ccitt -
> 20 +20
> > >> hash_init_crc32 -
> 20 +20
> > >> hash_init_crc16_ccitt -
> 20 +20
> > >> hash_update_sha256 -
> 16 +16
> > >> hash_update_sha1 -
> 16 +16
> > >> sha1_update -
> 8 +8
> > >>
> > >> This is because:
> > >> > diff --git a/drivers/mmc/Kconfig b/drivers/mmc/Kconfig
> > >> > index 982e84dc3bc..5d7fd904950 100644
> > >> > --- a/drivers/mmc/Kconfig
> > >> > +++ b/drivers/mmc/Kconfig
> > >> > @@ -119,6 +119,7 @@ config MMC_HW_PARTITIONING
> > >> > config SUPPORT_EMMC_RPMB
> > >> > bool "Support eMMC replay protected memory block (RPMB)"
> > >> > imply CMD_MMC_RPMB
> > >> > + select SHA256
> > >> > help
> > >> > Enable support for reading, writing and programming the
> > >> > key for the Replay Protection Memory Block partition in
> eMMC.
> > >>
> > >> Wasn't true / required before now, no hashing algorithms were enabled.
> > >> This was fine because:
> > >> [snip]
> > >> > diff --git a/lib/Makefile b/lib/Makefile
> > >> > index c4950b78a29..33755778283 100644
> > >> > --- a/lib/Makefile
> > >> > +++ b/lib/Makefile
> > >> > @@ -50,7 +50,6 @@ obj-$(CONFIG_XXHASH) += xxhash.o
> > >> > obj-y += net_utils.o
> > >> > obj-$(CONFIG_PHYSMEM) += physmem.o
> > >> > obj-y += rc4.o
> > >> > -obj-$(CONFIG_SUPPORT_EMMC_RPMB) += sha256.o
> > >> > obj-$(CONFIG_RBTREE) += rbtree.o
> > >> > obj-$(CONFIG_BITREVERSE) += bitrev.o
> > >> > obj-y += list_sort.o
> > >>
> > >> Got us the library access without bringing in everything else. And
> since
> > >> two of the platforms that are hitting this now are "nano" this is an
> > >> important thing to figure out how to continue to support. If there's
> > >> just no way around it, we can likely live with the size increase, but
> > >> I'd like to see this looked in to specifically first, thanks!
> > >>
> > > To address this, I think there are two options.
> > > 1. Introduce SUPPORT_EMMC_RPMB into the MbedTLS sub makefile.
> > > 2. Make MBEDTLS_LIB_CRYPTO depends on !SUPPORT_EMMC_RPMB.
> > > 1) looks to be ugly, I prefer 2) if you agree.
> >
> > 2 is not a good idea either. We can't just drop RPMB support when
> > mbedTLS is enabled
>
> To be clear, while I hope we can do something about this growth, I would
> rather live with it (as it's not an unreasonable amount) than do 2, and
> if 1 is too ugly, probably skip that as well. If it's not a matter of
> loosening some select statements, or maybe introducing a library type
> symbol we can see if anyone else more motivated has a better idea as
> it's literally 3 platforms (ev-imx280-nano-x-mb is the other nano, and
> then uniphier_v8 where it's arguably a missing feature anyhow) rather
> than a large number of them. And not even other "mini" or "nano"
> configs.
>
> Yes. I figured out what the problem is.
The "select SHA256" change doesn't matter, but size growth was introduced by
the inline function in sha1_alt.h I added in patch #2 of v8.
I already fixed this in my working branch:
aarch64: (for 1/1 boards) all +8502.0 data +224.0 rodata +218.0 text
+8060.0
uniphier_v8 : all +8502 data +224 rodata +218 text +8060
arm: (for 2/2 boards) all +5649.0 data +112.0 rodata +137.0 text
+5400.0
o4-imx6ull-nano: all +5651 data +112 rodata +139 text +5400
ev-imx280-nano-x-mb: all +5647 data +112 rodata +135 text +5400
Now there is no "u-boot: add" any more on these three boards.
I will update v9 with this fix, and we don't need either 1) or 2) options
mentioned
in my previous reply.
Regards,
Raymond
More information about the U-Boot
mailing list