Re: [edk2-devel] [PATCH v4 0/7] CryptoPkg: Upgrade OpenSSL to 1.1.1b

["Laszlo Ersek" <lersek@redhat.com>]

On 05/17/19 12:12, Lu, XiaoyuX wrote:
> Hi, Lerszlo:

well... I agree that my first name may not be trivial to spell, but you
can always use the clipboard, whenever in doubt.

For the record, it's "Laszlo".

> 
> (1):
> 
>> Unfortunately, I've found another build issue with this series. (My apologies that I didn't discover it earlier.) It is reported in the 32-bit (ARM) build of the ArmVirtQemu platform:
>>
>>   CryptoPkg/Library/OpensslLib/openssl/crypto/rand/drbg_lib.c:1028:
>>   undefined reference to `__aeabi_ui2d'
>>
> 
> OpensslLib[Crypto].inf contains ArmSoftFloatLib as dependent library.
> 
> In ArmSoftFloatLib:
> 
>  softfloat-for-gcc.h|98| #define uint32_to_float64       __floatunsidf
>  softfloat-for-gcc.h|222| #define __floatunsidf       __aeabi_ui2d
> 
>  softfloat-for-gcc.h|128| #define float64_to_uint32_round_to_zero     __fixunsdfsi
>  softfloat-for-gcc.h|234| #define __fixunsdfsi        __aeabi_d2uiz
> 
> But *uint32_to_float64* and *float64_to_uint32_round_to_zero* aren't implemented in softfloat.c
> 
> If these two functions implement, the build will pass. (I use dummy functions and try)

See my response to Jian on this.

> (2):
> 
>> thus, preferably, a CryptoPkg patch series should be at least build tested (if not boot tested) for all arches, before being posted to the mailing list.
> 
> I should test ARM, since IA32 arch has Intrinsic problem(_ftol2). It is very likely that ARM arch does not support it either. 
> 
>> (Yes, CI would help a lot with such issues.)
> 
> Now I don't have a CI environment here. 
> I will setup one for building OvmfPkg, ArmVirtPkg, EmulatorPkg.

Sorry, I was unclear: I meant a community-level, central CI. Not a
personal one. And, the central CI is undergoing design discussions right
now.

Thanks
Laszlo



> -----Original Message-----
> From: devel@edk2.groups.io [mailto:devel@edk2.groups.io] On Behalf Of Laszlo Ersek
> Sent: Friday, May 17, 2019 2:26 AM
> To: devel@edk2.groups.io; Lu, XiaoyuX <xiaoyux.lu@intel.com>
> Cc: Wang, Jian J <jian.j.wang@intel.com>; Ye, Ting <ting.ye@intel.com>; Ard Biesheuvel <ard.biesheuvel@linaro.org>; Leif Lindholm <leif.lindholm@linaro.org>
> Subject: Re: [edk2-devel] [PATCH v4 0/7] CryptoPkg: Upgrade OpenSSL to 1.1.1b
> 
> Hi,
> 
> (+ Ard and Leif)
> 
> On 05/16/19 09:54, Xiaoyu lu wrote:
>> This series is also available at:
>> https://github.com/xiaoyuxlu/edk2/tree/bz_1089_upgrade_to_openssl_1_1_
>> 1b_v4
>>
>> Changes:
>>
>> (1) CryptoPkgOpensslLib: Modify process_files.pl for  upgrading 
>> OpenSSL
>>
>> (2) CryptoPkg/OpensslLib: Exclude unnecessary files in process_files.pl
>>     crypto/store/* are excluded.
>>     crypto/rand/randfile.c is excluded.
>>
>> (3) CryptoPkg/IntrinsicLib: Fix possible unresolved external symbol 
>> issue
>>
>> (4) CryptoPkg/OpensslLib: Prepare for upgrading OpenSSL
>>     Disable warnings for buiding OpenSSL_1_1_1b
>>
>> (5) CryptoPkg/OpensslLib: Fix cross-build problem for AARCH64
>>
>> (6) CryptoPkg: Upgrade OpenSSL to 1.1.1b
>>     The biggest change is use TSC as entropy source
>>     If TSC isn't avaiable, fallback to TimerLib(PerformanceCounter).
>>
>> (7) CryptoPkg/BaseCryptLib: Make HMAC_CTX size backward compatible
>>
>>
>> Verification done for this series:
>> * Https boot in OvmfPkg.
>> * BaseCrypt Library test. (Ovmf, EmulatorPkg)
>>
>> Important notice:
>> Nt32Pkg doesn't support TimerLib
>>> TimerLib|MdePkg/Library/BaseTimerLibNullTemplate/BaseTimerLibNullTemp
>>> TimerLib|late.inf
>> So it will failed in Nt32Pkg.
>>
>> Cc: Jian J Wang <jian.j.wang@intel.com>
>> Cc: Ting Ye <ting.ye@intel.com>
>>
>> Laszlo Ersek (1):
>>   CryptoPkg/OpensslLib: Fix cross-build problem for AARCH64
>>
>> Xiaoyu Lu (6):
>>   CryptoPkg/OpensslLib: Modify process_files.pl for upgrading OpenSSL
>>   CryptoPkg/OpensslLib: Exclude unnecessary files in process_files.pl
>>   CryptoPkg/IntrinsicLib: Fix possible unresolved external symbol issue
>>   CryptoPkg/OpensslLib: Prepare for upgrading OpenSSL
>>   CryptoPkg: Upgrade OpenSSL to 1.1.1b
>>   CryptoPkg/BaseCryptLib: Make HMAC_CTX size backward compatible
>>
>>  CryptoPkg/Library/IntrinsicLib/IntrinsicLib.inf    |   4 +-
>>  CryptoPkg/Library/OpensslLib/OpensslLib.inf        |  76 ++++-
>>  CryptoPkg/Library/OpensslLib/OpensslLibCrypto.inf  |  67 ++++-
>>  CryptoPkg/Library/Include/CrtLibSupport.h          |  13 +-
>>  CryptoPkg/Library/Include/openssl/opensslconf.h    |  54 +++-
>>  CryptoPkg/Library/Include/sys/syscall.h            |  11 +
>>  CryptoPkg/Library/OpensslLib/buildinf.h            |   2 +
>>  CryptoPkg/Library/OpensslLib/rand_pool_noise.h     |  29 ++
>>  CryptoPkg/Library/BaseCryptLib/Hmac/CryptHmacMd5.c |   8 +-
>>  .../Library/BaseCryptLib/Hmac/CryptHmacSha1.c      |   9 +-
>>  .../Library/BaseCryptLib/Hmac/CryptHmacSha256.c    |   8 +-
>>  CryptoPkg/Library/IntrinsicLib/Ia32/MathFtol.c     |  22 ++
>>  CryptoPkg/Library/OpensslLib/ossl_store.c          |  17 ++
>>  CryptoPkg/Library/OpensslLib/rand_pool.c           | 316 +++++++++++++++++++++
>>  CryptoPkg/Library/OpensslLib/rand_pool_noise.c     |  29 ++
>>  CryptoPkg/Library/OpensslLib/rand_pool_noise_tsc.c |  43 +++
>>  CryptoPkg/Library/OpensslLib/openssl               |   2 +-
>>  CryptoPkg/Library/OpensslLib/process_files.pl      |  11 +-
>>  18 files changed, 669 insertions(+), 52 deletions(-)  create mode 
>> 100644 CryptoPkg/Library/Include/sys/syscall.h
>>  create mode 100644 CryptoPkg/Library/OpensslLib/rand_pool_noise.h
>>  create mode 100644 CryptoPkg/Library/IntrinsicLib/Ia32/MathFtol.c
>>  create mode 100644 CryptoPkg/Library/OpensslLib/ossl_store.c
>>  create mode 100644 CryptoPkg/Library/OpensslLib/rand_pool.c
>>  create mode 100644 CryptoPkg/Library/OpensslLib/rand_pool_noise.c
>>  create mode 100644 CryptoPkg/Library/OpensslLib/rand_pool_noise_tsc.c
>>
> 
> Unfortunately, I've found another build issue with this series. (My apologies that I didn't discover it earlier.) It is reported in the 32-bit (ARM) build of the ArmVirtQemu platform:
> 
>   CryptoPkg/Library/OpensslLib/openssl/crypto/rand/drbg_lib.c:1028:
>   undefined reference to `__aeabi_ui2d'
> 
> The referenced line is from the drbg_add() function:
> 
>     if (buflen < seedlen || randomness < (double) seedlen) {
> 
> Beyond the failure to resolve the "__aeabi_ui2d" symbol, the edk2 coding style spec says, "Floating point operations are not recommended in UEFI firmware." (Even though the UEFI spec describes the required floating point environment for all architectures.)
> 
> So, I'm not sure what we should do here. If we think that floating point is plain evil in edk2, then we cannot rebase edk2 to OpenSSL-1.1.1b.
> 
> ... Hmmm, this seems to be the 32-bit ARM variant of [PATCH v4 3/7]!
> 
> If we find floating point generally acceptable in edk2, then Ard and Leif could help us decide please whether this 32-bit ARM issue should be fixed during the feature freeze (when fixes are still allowed), or if it justifies postponing OpenSSL 1.1.1b to the next edk2 stable tag.
> 
> Again, I'm sorry that I found this only now -- but "CryptoPkg/CryptoPkg.dsc" is multi-arch:
> 
>   SUPPORTED_ARCHITECTURES        = IA32|X64|ARM|AARCH64
> 
> thus, preferably, a CryptoPkg patch series should be at least build tested (if not boot tested) for all arches, before being posted to the mailing list.
> 
> (Yes, CI would help a lot with such issues.)
> 
> Thanks
> Laszlo
> 
> 
>