On 10/09/2025 06:50, Thomas Huth wrote:
On 12/05/2025 21.05, Paolo Bonzini wrote:
Now that we can do so after the error code has been pushed, raising
the #DB exception for task-switch traps is trivial.
Signed-off-by: Paolo Bonzini <[email protected]>
---
target/i386/tcg/seg_helper.c | 9 +++++----
1 file changed, 5 insertions(+), 4 deletions(-)
diff --git a/target/i386/tcg/seg_helper.c b/target/i386/tcg/seg_helper.c
index cb90ccd2adc..071f3fbd83d 100644
--- a/target/i386/tcg/seg_helper.c
+++ b/target/i386/tcg/seg_helper.c
@@ -473,10 +473,6 @@ static void switch_tss_ra(CPUX86State *env, int
tss_selector,
new_segs[R_GS] = 0;
new_trap = 0;
}
- /* XXX: avoid a compiler warning, see
- https://urldefense.proofpoint.com/v2/url?
u=http-3A__support.amd.com_us_Processor-5FTechDocs_24593.pdf&d=DwICaQ&c=s883GpUCOChKOHiocYtGcg&r=c23RpsaH4D2MKyD3EPJTDa0BAxz6tV8aUJqVSoytEiY&m=ne1e2OYZDArdmBhjIXmv-d6hN8DFQV2i9elKEJJ2rgw_Rjs4bbFAj9BI-B8Y8SUa&s=Lx1uoh3Mv7iduPgFcshKa1nly0lcsCF6Z1G0neDOxQ4&e= - chapters 12.2.5 and 13.2.4 on how to implement TSS Trap bit */
- (void)new_trap;
/* clear busy bit (it is restartable) */
if (source == SWITCH_TSS_JMP || source == SWITCH_TSS_IRET) {
@@ -622,6 +618,11 @@ static void switch_tss_ra(CPUX86State *env, int
tss_selector,
}
SET_ESP(sa.sp, sa.sp_mask);
}
+
+ if (new_trap) {
+ env->dr[6] |= DR6_BT;
+ raise_exception_ra(env, EXCP01_DB, retaddr);
+ }
}
Hi Paolo,
as already quickly mentioned in IRC, the kvm-unit-tests CI had a
regression after homebrew updated its QEMU from 10.0 to 10.1, the
"eventinj" and the "taskwitch" test started failing:
10.0: https://urldefense.proofpoint.com/v2/url?
u=https-3A__gitlab.com_kvm-2Dunit-2Dtests_kvm-2Dunit-2Dtests_-2D_jobs_10871048973&d=DwICaQ&c=s883GpUCOChKOHiocYtGcg&r=c23RpsaH4D2MKyD3EPJTDa0BAxz6tV8aUJqVSoytEiY&m=ne1e2OYZDArdmBhjIXmv-d6hN8DFQV2i9elKEJJ2rgw_Rjs4bbFAj9BI-B8Y8SUa&s=2-qwz0tlJsb4ChkM4ZqunWv6Bmf3Zri5BDO1guawTyc&e= 10.1: https://urldefense.proofpoint.com/v2/url?u=https-3A__gitlab.com_kvm-2Dunit-2Dtests_kvm-2Dunit-2Dtests_-2D_jobs_11282832498&d=DwICaQ&c=s883GpUCOChKOHiocYtGcg&r=c23RpsaH4D2MKyD3EPJTDa0BAxz6tV8aUJqVSoytEiY&m=ne1e2OYZDArdmBhjIXmv-d6hN8DFQV2i9elKEJJ2rgw_Rjs4bbFAj9BI-B8Y8SUa&s=U7Y_mIKPuJtmnCc5eDadnSt1qVAWtKHPBqnXRrPcG1s&e=
I've now bisected the problem (painfully in the terminal window of the
cirrus-ci), and it seems to be this commit here that is causing the issue:
ad441b8b7913a26b18edbc076c74ca0cdbfa4ee5 is the first bad commit
commit ad441b8b7913a26b18edbc076c74ca0cdbfa4ee5
Author: Paolo Bonzini <[email protected]>
Date: Wed Aug 14 12:33:02 2024 +0200
target/i386: implement TSS trap bit
Could you please have a look?
Thanks,
Thomas
Possibly related: https://gitlab.com/qemu-project/qemu/-/issues/3101?
The submitter has also provided a suggested patch.
ATB,
Mark.
