[v2,2/2] build: fix crash by disabling AVX512 with binutils 2.31

Message ID 20190416153932.21788-2-ferruh.yigit@intel.com
State Superseded, archived
Delegated to: Thomas Monjalon
Headers show
Series
  • [v2,1/2] build: fix meson binutils workaround
Related show

Checks

Context Check Description
ci/Intel-compilation success Compilation OK
ci/checkpatch success coding style OK

Commit Message

Ferruh Yigit April 16, 2019, 3:39 p.m.
On Skylake platform, with native build, KNI kernel module crashes
because of the corrupted values passed to kernel module.

The corruption occurs because the userspace kni library works
unexpectedly. Compiler [1] is using AVX512 instructions and generated
binary is wrong [2].

It turned around gcc does its job correct, but gas is generating binary
wrong. And expected binutils 2.30, 2.31 & 2.31.1 are affected. Issue has
been fixed in binutils 2.32 with:
Commit x86: don't mistakenly scale non-8-bit displacements

AVX512 was already disabled with bintuils 2.30 [3], extending it to
2.31 & 2.31.1 too.

[1] gcc (GCC) 8.3.1 20190223 (Red Hat 8.3.1-2)

[2] gcc bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90028

[3] Bugzilla ID 97 has the details.

Bugzilla ID: 249
Cc: stable@dpdk.org

Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
---
v2:
* Release notes "Known Issues" section updated
---
 config/x86/meson.build                   | 6 ++++++
 doc/guides/rel_notes/release_19_05.rst   | 6 ++++++
 mk/toolchain/gcc/rte.toolchain-compat.mk | 9 ++++++++-
 3 files changed, 20 insertions(+), 1 deletion(-)

Comments

Bruce Richardson April 16, 2019, 3:47 p.m. | #1
On Tue, Apr 16, 2019 at 04:39:32PM +0100, Ferruh Yigit wrote:
> On Skylake platform, with native build, KNI kernel module crashes
> because of the corrupted values passed to kernel module.
> 
> The corruption occurs because the userspace kni library works
> unexpectedly. Compiler [1] is using AVX512 instructions and generated
> binary is wrong [2].
> 
> It turned around gcc does its job correct, but gas is generating binary
> wrong. And expected binutils 2.30, 2.31 & 2.31.1 are affected. Issue has
> been fixed in binutils 2.32 with:
> Commit x86: don't mistakenly scale non-8-bit displacements
> 
> AVX512 was already disabled with bintuils 2.30 [3], extending it to
> 2.31 & 2.31.1 too.
> 
> [1] gcc (GCC) 8.3.1 20190223 (Red Hat 8.3.1-2)
> 
> [2] gcc bug: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90028
> 
> [3] Bugzilla ID 97 has the details.
> 
> Bugzilla ID: 249
> Cc: stable@dpdk.org
> 
> Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
> ---
> v2:
> * Release notes "Known Issues" section updated
> ---
>  config/x86/meson.build                   | 6 ++++++
>  doc/guides/rel_notes/release_19_05.rst   | 6 ++++++
>  mk/toolchain/gcc/rte.toolchain-compat.mk | 9 ++++++++-
>  3 files changed, 20 insertions(+), 1 deletion(-)
> 
> diff --git a/config/x86/meson.build b/config/x86/meson.build
> index 692aebe7a..9e9d5dc8c 100644
> --- a/config/x86/meson.build
> +++ b/config/x86/meson.build
> @@ -10,6 +10,12 @@ if host_machine.system() != 'windows'
>  			message('Binutils 2.30 detected, disabling AVX512 support as workaround for bug #97')
>  		endif
>  	endif
> +	if ldver.contains('2.31')
> +		if cc.has_argument('-mno-avx512f')
> +			machine_args += '-mno-avx512f'
> +			message('Binutils 2.31 detected, disabling AVX512 support as workaround for bug #249')
> +		endif
> +	endif

Is this not the same as the previous block just with a slightly different
error message? Should we merge the two, and print out both bug numbers?

If not merging, we can reduce the indentation by putting the second
cc.has_argument() condition on the same line as the previous check, i.e.
	"if ldver.contains(...) and cc.has_argument(...)"

Apart from this nit:

Acked-by: Bruce Richardson <bruce.richardson@intel.com>

Patch

diff --git a/config/x86/meson.build b/config/x86/meson.build
index 692aebe7a..9e9d5dc8c 100644
--- a/config/x86/meson.build
+++ b/config/x86/meson.build
@@ -10,6 +10,12 @@  if host_machine.system() != 'windows'
 			message('Binutils 2.30 detected, disabling AVX512 support as workaround for bug #97')
 		endif
 	endif
+	if ldver.contains('2.31')
+		if cc.has_argument('-mno-avx512f')
+			machine_args += '-mno-avx512f'
+			message('Binutils 2.31 detected, disabling AVX512 support as workaround for bug #249')
+		endif
+	endif
 endif
 
 # we require SSE4.2 for DPDK
diff --git a/doc/guides/rel_notes/release_19_05.rst b/doc/guides/rel_notes/release_19_05.rst
index 25cd61ca2..cf04b9b26 100644
--- a/doc/guides/rel_notes/release_19_05.rst
+++ b/doc/guides/rel_notes/release_19_05.rst
@@ -338,6 +338,12 @@  Known Issues
    Also, make sure to start the actual text at the margin.
    =========================================================
 
+* **On x86 platforms, AVX512 support is disabled with binutils 2.31**
+
+  Because a defect in binutils 2.31 AVX512 support is disabled.
+  DPDK defect: https://bugs.dpdk.org/show_bug.cgi?id=249
+  GCC defect: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90028
+
 * **No software AES-XTS implementation.**
 
   There are currently no cryptodev software PMDs available which implement
diff --git a/mk/toolchain/gcc/rte.toolchain-compat.mk b/mk/toolchain/gcc/rte.toolchain-compat.mk
index df71e4a8b..ea40a11c0 100644
--- a/mk/toolchain/gcc/rte.toolchain-compat.mk
+++ b/mk/toolchain/gcc/rte.toolchain-compat.mk
@@ -27,7 +27,14 @@  ifneq ($(filter 2.30%,$(LD_VERSION)),)
 FORCE_DISABLE_AVX512 := y
 # print warning only once for librte_eal
 ifneq ($(filter %librte_eal,$(CURDIR)),)
-$(warning AVX512 support disabled because of ld 2.30. See Bug 97)
+$(warning AVX512 support disabled because of binutils 2.30. See Bug 97)
+endif
+endif
+ifneq ($(filter 2.31%,$(LD_VERSION)),)
+FORCE_DISABLE_AVX512 := y
+# print warning only once for librte_eal
+ifneq ($(filter %librte_eal,$(CURDIR)),)
+$(warning AVX512 support disabled because of binutils 2.31. See Bug 249)
 endif
 endif
 endif