[Buildroot] [PATCH] package/openssl: fix build on riscv32

[Buildroot] [PATCH] package/openssl: fix build on riscv32

[Yann E. MORIN]

riscv32 is (surprise!) a 32-bit architecture. But it has been 2038-safe
from its inception. As such, there is no legacy binaries that may use
the 32-bit time syscalls, and thus they are anot available on riscv32.

Code tht directly calls to the syscalls without using the C libraries
wrappers thus need to handle this case by themselves.

Backport a patch from the upstream development branch that will
eventually be openssl 3.0, but has not yet been backported to the 1.1.1
stable branch.

Fixes:
    http://autobuild.buildroot.org/results/eb9/eb9a64d4ffae8569b5225083f282cf87ffa7c681/
    ...
    http://autobuild.buildroot.org/results/07e/07e413b24ba8adc9558c80267ce16dda339bf032/

Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
Cc: Matt Weber <matthew.weber@rockwellcollins.com>
Cc: Mark Corbin <mark@dibsco.co.uk>
---
 ...ort-for-io_pgetevents_time64-syscall.patch | 63 +++++++++++++++++++
 1 file changed, 63 insertions(+)
 create mode 100644 package/libopenssl/0006-Add-support-for-io_pgetevents_time64-syscall.patch

diff --git a/package/libopenssl/0006-Add-support-for-io_pgetevents_time64-syscall.patch b/package/libopenssl/0006-Add-support-for-io_pgetevents_time64-syscall.patch
new file mode 100644
index 0000000000..ab552de91a
--- /dev/null
+++ b/package/libopenssl/0006-Add-support-for-io_pgetevents_time64-syscall.patch
@@ -0,0 +1,63 @@
+From 5b5e2985f355c8e99c196d9ce5d02c15bebadfbc Mon Sep 17 00:00:00 2001
+From: Alistair Francis <alistair.francis@wdc.com>
+Date: Thu, 29 Aug 2019 13:56:21 -0700
+Subject: [PATCH] Add support for io_pgetevents_time64 syscall
+
+32-bit architectures that are y2038 safe don't include syscalls that use
+32-bit time_t. Instead these architectures have suffixed syscalls that
+always use a 64-bit time_t. In the case of the io_getevents syscall the
+syscall has been replaced with the io_pgetevents_time64 syscall instead.
+
+This patch changes the io_getevents() function to use the correct
+syscall based on the avaliable syscalls and the time_t size. We will
+only use the new 64-bit time_t syscall if the architecture is using a
+64-bit time_t. This is to avoid having to deal with 32/64-bit
+conversions and relying on a 64-bit timespec struct on 32-bit time_t
+platforms. As of Linux 5.3 there are no 32-bit time_t architectures
+without __NR_io_getevents. In the future if a 32-bit time_t architecture
+wants to use the 64-bit syscalls we can handle the conversion.
+
+This fixes build failures on 32-bit RISC-V.
+
+Signed-off-by: Alistair Francis <alistair.francis@wdc.com>
+
+Reviewed-by: Richard Levitte <levitte@openssl.org>
+Reviewed-by: Paul Dale <paul.dale@oracle.com>
+(Merged from https://github.com/openssl/openssl/pull/9819)
+[yann.morin.1998 at free.fr: backport from upstream 3.0 branch]
+Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
+---
+ engines/e_afalg.c | 16 ++++++++++++++++
+ 1 file changed, 16 insertions(+)
+
+diff --git a/engines/e_afalg.c b/engines/e_afalg.c
+index dacbe358cb..99516cb1bb 100644
+--- a/engines/e_afalg.c
++++ b/engines/e_afalg.c
+@@ -125,7 +125,23 @@ static ossl_inline int io_getevents(aio_context_t ctx, long min, long max,
+                                struct io_event *events,
+                                struct timespec *timeout)
+ {
++#if defined(__NR_io_getevents)
+     return syscall(__NR_io_getevents, ctx, min, max, events, timeout);
++#elif defined(__NR_io_pgetevents_time64)
++    /* Let's only support the 64 suffix syscalls for 64-bit time_t.
++     * This simplifies the code for us as we don't need to use a 64-bit
++     * version of timespec with a 32-bit time_t and handle converting
++     * between 64-bit and 32-bit times and check for overflows.
++     */
++    if (sizeof(timeout->tv_sec) == 8)
++        return syscall(__NR_io_pgetevents_time64, ctx, min, max, events, timeout, NULL);
++    else {
++        errno = ENOSYS;
++        return -1;
++    }
++#else
++# error "We require either the io_getevents syscall or __NR_io_pgetevents_time64."
++#endif
+ }
+ 
+ static void afalg_waitfd_cleanup(ASYNC_WAIT_CTX *ctx, const void *key,
+-- 
+2.25.1
+
-- 
2.25.1

[Buildroot] [PATCH] package/openssl: fix build on riscv32

[Arnout Vandecappelle]

On 02/03/2021 22:51, Yann E. MORIN wrote:
> riscv32 is (surprise!) a 32-bit architecture. But it has been 2038-safe
> from its inception. As such, there is no legacy binaries that may use
> the 32-bit time syscalls, and thus they are anot available on riscv32.
> 
> Code tht directly calls to the syscalls without using the C libraries
> wrappers thus need to handle this case by themselves.
> 
> Backport a patch from the upstream development branch that will
> eventually be openssl 3.0, but has not yet been backported to the 1.1.1
> stable branch.
> 
> Fixes:
>     http://autobuild.buildroot.org/results/eb9/eb9a64d4ffae8569b5225083f282cf87ffa7c681/
>     ...
>     http://autobuild.buildroot.org/results/07e/07e413b24ba8adc9558c80267ce16dda339bf032/
> 
> Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
> Cc: Matt Weber <matthew.weber@rockwellcollins.com>
> Cc: Mark Corbin <mark@dibsco.co.uk>


 Applied to master, thanks.

 Regards,
 Arnout

[Buildroot] [PATCH] package/openssl: fix build on riscv32

[Peter Korsgaard]

>>>>> "Yann" == Yann E MORIN <yann.morin.1998@free.fr> writes:

 > riscv32 is (surprise!) a 32-bit architecture. But it has been 2038-safe
 > from its inception. As such, there is no legacy binaries that may use
 > the 32-bit time syscalls, and thus they are anot available on riscv32.

 > Code tht directly calls to the syscalls without using the C libraries
 > wrappers thus need to handle this case by themselves.

 > Backport a patch from the upstream development branch that will
 > eventually be openssl 3.0, but has not yet been backported to the 1.1.1
 > stable branch.

 > Fixes:
 >     http://autobuild.buildroot.org/results/eb9/eb9a64d4ffae8569b5225083f282cf87ffa7c681/
 >     ...
 >     http://autobuild.buildroot.org/results/07e/07e413b24ba8adc9558c80267ce16dda339bf032/

 > Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
 > Cc: Matt Weber <matthew.weber@rockwellcollins.com>
 > Cc: Mark Corbin <mark@dibsco.co.uk>

Did you see Arndbs comment on the upstream commit?

https://github.com/openssl/openssl/commit/5b5e2985f355c8e99c196d9ce5d02c15bebadfbc

I cannot quite figure out if this causes issues for other archs?

-- 
Bye, Peter Korsgaard

[Buildroot] [PATCH] package/openssl: fix build on riscv32

[Yann E. MORIN]

Peter, All,

On 2021-03-03 08:12 +0100, Peter Korsgaard spake thusly:
> >>>>> "Yann" == Yann E MORIN <yann.morin.1998@free.fr> writes:
> 
>  > riscv32 is (surprise!) a 32-bit architecture. But it has been 2038-safe
>  > from its inception. As such, there is no legacy binaries that may use
>  > the 32-bit time syscalls, and thus they are anot available on riscv32.
> 
>  > Code tht directly calls to the syscalls without using the C libraries
>  > wrappers thus need to handle this case by themselves.
> 
>  > Backport a patch from the upstream development branch that will
>  > eventually be openssl 3.0, but has not yet been backported to the 1.1.1
>  > stable branch.
> 
>  > Fixes:
>  >     http://autobuild.buildroot.org/results/eb9/eb9a64d4ffae8569b5225083f282cf87ffa7c681/
>  >     ...
>  >     http://autobuild.buildroot.org/results/07e/07e413b24ba8adc9558c80267ce16dda339bf032/
> 
>  > Signed-off-by: Yann E. MORIN <yann.morin.1998@free.fr>
>  > Cc: Matt Weber <matthew.weber@rockwellcollins.com>
>  > Cc: Mark Corbin <mark@dibsco.co.uk>
> 
> Did you see Arndbs comment on the upstream commit?
> 
> https://github.com/openssl/openssl/commit/5b5e2985f355c8e99c196d9ce5d02c15bebadfbc
> 
> I cannot quite figure out if this causes issues for other archs?

Indeed, in light of the comment by Arnd, this patch looks a bit fishy.

Thus, I've now reverted it.

However, what should we do now?

The failing engine is not optional; it is always built (i.e. we can't
make BR2_PACKAGE_LIBOPENSSL_ENGINES depends on !riscv32).

A quick solution is to make libopenssl depends on !riscv32. This is not
very satisfying, especially because of BR2_PACKAGE_OPENSSL_FORCE_LIBOPENSSL
to which we should propagate the dependency; that's 5 packages: hostapd,
ibm-sw-tpm2, softether, tpm2-tss, wpa_suplicant (for mesh networking and
for wpa3), so not totally unfesasible either, but still... :-/

Thoughts?

Regards,
Yann E. MORIN.


-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 561 099 427 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'