| Message ID | 1515499487-29937-1-git-send-email-matthew.weber@rockwellcollins.com |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | [v2] kvm-unit-tests: test for rdseed/rdrand | expand |
Hello, Thanks for your work on this! On Tue, 9 Jan 2018 06:04:47 -0600, Matt Weber wrote: > The build fails when the host binutils isn't at least 2.23 > (2.22.x introduced RDSEED). > > The host toolchain is used for x86_64 target builds where > we need to do a 32bit build. Most other buildroot builds > are using a much newer binutils unless it's a external > older toolchain. > > Fixes: > http://autobuild.buildroot.net/results/c39/c3987a3cbd2960b0ff50f872636bdfd8d1a9c820/ > > Upstream: > https://www.spinics.net/lists/kvm/msg161542.html > > Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com> > --- > > Thomas/Peter could you test this on one of your builders? > I've requested a gcc farm account but still waiting at this > point. Here's the builders which can reproduce the failure > (ie have old enough host binutils). > http://autobuild.buildroot.net/?reason=kvm-unit-tests-kvm-unit-tests-20171020 Sadly, it doesn't work. It correctly detects that rdseed isn't available: >>> kvm-unit-tests kvm-unit-tests-20171020 Configuring cd /home/test/buildroot/output/build/kvm-unit-tests-kvm-unit-tests-20171020 && ./configure --arch="x86_64" --processor="" --endian="little" Checking for rdseed/rdrand... No. but the CFLAGS you assign have no effect. And indeed you assign CFLAGS, but there's nothing that actually uses the CFLAGS set by the configure script. In fact, you can perfectly test this on your side: just introduce a stupid typo in the rdseed test program that makes it fail to build unconditionally, and verify that the NO_RDSEEDRAND define really exists when the C file is built. Thanks! Thomas
Thomas, On Fri, Jan 12, 2018 at 4:17 PM, Thomas Petazzoni <thomas.petazzoni@free-electrons.com> wrote: > Hello, > >> >>>> kvm-unit-tests kvm-unit-tests-20171020 Configuring > cd /home/test/buildroot/output/build/kvm-unit-tests-kvm-unit-tests-20171020 && ./configure --arch="x86_64" --processor="" --endian="little" > Checking for rdseed/rdrand... No. > > but the CFLAGS you assign have no effect. And indeed you assign CFLAGS, > but there's nothing that actually uses the CFLAGS set by the configure > script. > > In fact, you can perfectly test this on your side: just introduce a > stupid typo in the rdseed test program that makes it fail to build > unconditionally, and verify that the NO_RDSEEDRAND define really exists > when the C file is built. > Ah shoot, correct they force CFLAGS to empty. Fixed in v2 and I typo tested it successfully. https://patchwork.ozlabs.org/patch/860274/ Matt
diff --git a/package/kvm-unit-tests/0002-x86-vmx_tests-test-for-rdseed-rdrand.patch b/package/kvm-unit-tests/0002-x86-vmx_tests-test-for-rdseed-rdrand.patch new file mode 100644 index 0000000..5fc15b4 --- /dev/null +++ b/package/kvm-unit-tests/0002-x86-vmx_tests-test-for-rdseed-rdrand.patch @@ -0,0 +1,82 @@ +From 9a8ddd7fdde14f4f1085fe89a2e2dfba6b291aa3 Mon Sep 17 00:00:00 2001 +From: Matt Weber <matthew.weber@rockwellcollins.com> +Date: Mon, 8 Jan 2018 22:39:29 -0600 +Subject: [PATCH] x86/vmx_tests: test for rdseed/rdrand + +A binutils of at least 2.23 is required +(2.22.x introduced RDSEED). + +Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com> +--- + configure | 17 +++++++++++++++++ + x86/vmx_tests.c | 6 ++++++ + 2 files changed, 23 insertions(+) + +diff --git a/configure b/configure +index dd9d361..4912582 100755 +--- a/configure ++++ b/configure +@@ -171,6 +171,23 @@ mkdir -p lib + ln -sf "$asm" lib/asm + + ++cat > rd_test.c <<EOF ++#include <stdint.h> ++int main() { ++ uint16_t seed=0; ++ unsigned char ok; ++ asm volatile ("rdseed %0; setc %1" ++ : "=r" (seed), "=qm" (ok)); ++ return ok; ++} ++EOF ++if $cross_prefix$cc -o /dev/null rd_test.c &> /dev/null; then ++ echo "Checking for rdseed/rdrand... Yes." ++else ++ CFLAGS="${CFLAGS} -DNO_RDSEEDRAND" ++ echo "Checking for rdseed/rdrand... No." ++fi ++ + # create the config + cat <<EOF > config.mak + SRCDIR=$srcdir +diff --git a/x86/vmx_tests.c b/x86/vmx_tests.c +index 4a3e94b..2cbe3eb 100644 +--- a/x86/vmx_tests.c ++++ b/x86/vmx_tests.c +@@ -770,8 +770,10 @@ asm( + "insn_sldt: sldt %ax;ret\n\t" + "insn_lldt: xor %eax, %eax; lldt %ax;ret\n\t" + "insn_str: str %ax;ret\n\t" ++#ifndef NO_RDSEEDRAND + "insn_rdrand: rdrand %rax;ret\n\t" + "insn_rdseed: rdseed %rax;ret\n\t" ++#endif + ); + extern void insn_hlt(); + extern void insn_invlpg(); +@@ -796,8 +798,10 @@ extern void insn_lldt(); + extern void insn_str(); + extern void insn_cpuid(); + extern void insn_invd(); ++#ifndef NO_RDSEEDRAND + extern void insn_rdrand(); + extern void insn_rdseed(); ++#endif + + u32 cur_insn; + u64 cr3; +@@ -853,8 +857,10 @@ static struct insn_table insn_table[] = { + {"DESC_TABLE (LLDT)", CPU_DESC_TABLE, insn_lldt, INSN_CPU1, 47, 0, 0, 0}, + {"DESC_TABLE (STR)", CPU_DESC_TABLE, insn_str, INSN_CPU1, 47, 0, 0, 0}, + /* LTR causes a #GP if done with a busy selector, so it is not tested. */ ++#ifndef NO_RDSEEDRAND + {"RDRAND", CPU_RDRAND, insn_rdrand, INSN_CPU1, VMX_RDRAND, 0, 0, 0}, + {"RDSEED", CPU_RDSEED, insn_rdseed, INSN_CPU1, VMX_RDSEED, 0, 0, 0}, ++#endif + // Instructions always trap + {"CPUID", 0, insn_cpuid, INSN_ALWAYS_TRAP, 10, 0, 0, 0}, + {"INVD", 0, insn_invd, INSN_ALWAYS_TRAP, 13, 0, 0, 0}, +-- +1.9.1 +
The build fails when the host binutils isn't at least 2.23 (2.22.x introduced RDSEED). The host toolchain is used for x86_64 target builds where we need to do a 32bit build. Most other buildroot builds are using a much newer binutils unless it's a external older toolchain. Fixes: http://autobuild.buildroot.net/results/c39/c3987a3cbd2960b0ff50f872636bdfd8d1a9c820/ Upstream: https://www.spinics.net/lists/kvm/msg161542.html Signed-off-by: Matthew Weber <matthew.weber@rockwellcollins.com> --- Thomas/Peter could you test this on one of your builders? I've requested a gcc farm account but still waiting at this point. Here's the builders which can reproduce the failure (ie have old enough host binutils). http://autobuild.buildroot.net/?reason=kvm-unit-tests-kvm-unit-tests-20171020 Changes v1 -> v2 [Thomas - Suggested a configure check vs the original revert of the upstream commit which added these instructions to the package. --- ...0002-x86-vmx_tests-test-for-rdseed-rdrand.patch | 82 ++++++++++++++++++++++ 1 file changed, 82 insertions(+) create mode 100644 package/kvm-unit-tests/0002-x86-vmx_tests-test-for-rdseed-rdrand.patch