gnu: libaio: Fix building on riscv64-linux.

* gnu/packages/linux.scm (libaio)[arguments]: When building for
riscv64-linux add a patch to account for undeclared behavior.
[native-inputs]: Add patch file, patch.
* gnu/packages/patches/libaio-riscv-test5.patch: New file.
* gnu/local.mk (dist_patch_DATA): Register it.
This commit is contained in:
Efraim Flashner 2023-05-02 12:27:14 +03:00
parent 44d7044094
commit b62c9fbdd3
No known key found for this signature in database
GPG key ID: 41AAE7DCCA3D8351
3 changed files with 62 additions and 1 deletions

View file

@ -1438,6 +1438,7 @@ dist_patch_DATA = \
%D%/packages/patches/liba52-set-soname.patch \
%D%/packages/patches/liba52-use-mtune-not-mcpu.patch \
%D%/packages/patches/libaio-32bit-test.patch \
%D%/packages/patches/libaio-riscv-test5.patch \
%D%/packages/patches/libbase-fix-includes.patch \
%D%/packages/patches/libbase-use-own-logging.patch \
%D%/packages/patches/libbonobo-activation-test-race.patch \

View file

@ -5621,7 +5621,20 @@ (define-public libaio
#:test-target "partcheck" ; need root for a full 'check'
#:phases
#~(modify-phases %standard-phases
(delete 'configure)))) ; no configure script
(delete 'configure) ; no configure script
#$@(if (target-riscv64?)
#~((add-after 'unpack 'patch-test
(lambda* (#:key build-inputs #:allow-other-keys)
(invoke "patch" "-p1" "-i"
#$(local-file
(search-patch
"libaio-riscv-test5.patch"))))))
#~()))))
(native-inputs
(if (target-riscv64?)
(list (search-patch "libaio-riscv-test5.patch")
patch)
'()))
(home-page "https://pagure.io/libaio")
(synopsis "Linux-native asynchronous I/O access library")
(description

View file

@ -0,0 +1,47 @@
https://pagure.io/libaio/pull-request/23
From f68c69b6cbc1260a6034997d5f146e3d0a197ed8 Mon Sep 17 00:00:00 2001
From: Xiongchuan Tan <xc-tan@outlook.com>
Date: Jun 28 2022 15:53:38 +0000
Subject: As of June 28th, 2022, the RISC-V spec[1] reserves the PTE permission bit
combination of "write+!read", and the kernel would have incoherent behavior in
the last test case of "harness/cases/5.t". Since it leads to undefined behavior,
until further spec update, this test case should be disabled for RISC-V.
A patch to disallow such permission in mmap() can be found here[2].
[1]: https://github.com/riscv/riscv-isa-manual/releases/download/Priv-v1.12/riscv-privileged-20211203.pdf
[2]: https://www.spinics.net/lists/kernel/msg4412421.html
---
diff --git a/harness/cases/5.t b/harness/cases/5.t
index b0a7c56..8d6c959 100644
--- a/harness/cases/5.t
+++ b/harness/cases/5.t
@@ -37,6 +37,14 @@ int test_main(void)
status |= attempt_rw(rwfd, buf, SIZE, 0, READ, -EFAULT);
res = munmap(buf, page_size); assert(res == 0);
+
+ /* As of June 28th, 2022, the RISC-V spec Volume 2 Section 4.3
+ * version "20211203 Privileged Architecture v1.12, Ratified"
+ * reserves the usage of the PTE permission bit combination of
+ * "write+!read", so the next test leads to undefined behavior
+ * and should be disabled. */
+#ifndef __riscv
+
buf = mmap(0, page_size, PROT_WRITE, MAP_SHARED, rwfd, 0);
assert(buf != (char *)-1);
@@ -48,6 +56,8 @@ int test_main(void)
status |= attempt_rw(rwfd, buf, SIZE, 0, READ, SIZE);
status |= attempt_rw(rwfd, buf, SIZE, 0, WRITE, res);
+#endif
+
return status;
}