linux-kernel - [PATCH] kvm,x86: load guest FPU context more eagerly

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <20150423115237.10598d5f@cuia.bos.redhat.com>
Date:	Thu, 23 Apr 2015 11:52:37 -0400
From:	Rik van Riel <riel@...hat.com>
To:	kvm@...r.kernel.org
Cc:	pbonzini@...hat.com, mtosatti@...hat.com,
	linux-kernel@...r.kernel.org, bp@...en8.de, tglx@...utronix.de,
	hpa@...or.com
Subject: [PATCH] kvm,x86: load guest FPU context more eagerly

Currently KVM will clear the FPU bits in CR0.TS in the VMCS, and trap to 
re-load them every time the guest accesses the FPU after a switch back into
the guest from the host.

This patch copies the x86 task switch semantics for FPU loading, with the
FPU loaded eagerly after first use if the system uses eager fpu mode,
or if the guest uses the FPU frequently.

In the latter case, after loading the FPU for 255 times, the fpu_counter
will roll over, and we will revert to loading the FPU on demand, until
it has been established that the guest is still actively using the FPU.

This mirrors the x86 task switch policy, which seems to work.

Signed-off-by: Rik van Riel <riel@...hat.com>
---
I still hope to put the larger FPU changes in at some point, but with
all the current changes to the FPU code I am somewhat uncomfortable
causing even more churn. After 4.1 I may send in the changes to defer
loading of user space FPU context to do_notify_resume() - unless people
want them sooner.

 arch/x86/kvm/x86.c       | 15 +++++++++++++--
 include/linux/kvm_host.h |  1 +
 2 files changed, 14 insertions(+), 2 deletions(-)

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index e1a81267f3f6..2cdb2472a633 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -7031,14 +7031,25 @@ void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
 {
 	kvm_put_guest_xcr0(vcpu);
 
-	if (!vcpu->guest_fpu_loaded)
+	if (!vcpu->guest_fpu_loaded) {
+		vcpu->fpu_counter = 0;
 		return;
+	}
 
 	vcpu->guest_fpu_loaded = 0;
 	fpu_save_init(&vcpu->arch.guest_fpu);
 	__kernel_fpu_end();
 	++vcpu->stat.fpu_reload;
-	kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
+	/*
+	 * If using eager FPU mode, or if the guest is a frequent user
+	 * of the FPU, just leave the FPU active for next time.
+	 * Every 255 times fpu_counter rolls over to 0; a guest that uses
+	 * the FPU in bursts will revert to loading it on demand.
+	 */
+	if (!use_eager_fpu()) {
+		if (++vcpu->fpu_counter < 5)
+			kvm_make_request(KVM_REQ_DEACTIVATE_FPU, vcpu);
+	}
 	trace_kvm_fpu(0);
 }
 
diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
index ad45054309a0..f197ad3f6316 100644
--- a/include/linux/kvm_host.h
+++ b/include/linux/kvm_host.h
@@ -230,6 +230,7 @@ struct kvm_vcpu {
 
 	int fpu_active;
 	int guest_fpu_loaded, guest_xcr0_loaded;
+	unsigned char fpu_counter;
 	wait_queue_head_t wq;
 	struct pid *pid;
 	int sigset_active;

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/