From ec6449a9c2296b1c04f6219f7473e0c2fedecfed Mon Sep 17 00:00:00 2001 From: Christoffer Dall Date: Mon, 20 Nov 2017 12:10:15 +0100 Subject: KVM: arm/arm64: Don't enable/disable physical timer access on VHE After the timer optimization rework we accidentally end up calling physical timer enable/disable functions on VHE systems, which is neither needed nor correct, since the CNTHCTL_EL2 register format is different when HCR_EL2.E2H is set. The CNTHCTL_EL2 is initialized when CPUs become online in kvm_timer_init_vhe() and we don't have to call these functions on VHE systems, which also allows us to inline the non-VHE functionality. Reported-by: Jintack Lim Signed-off-by: Christoffer Dall --- include/kvm/arm_arch_timer.h | 3 --- 1 file changed, 3 deletions(-) (limited to 'include') diff --git a/include/kvm/arm_arch_timer.h b/include/kvm/arm_arch_timer.h index 01ee473517e2..6e45608b2399 100644 --- a/include/kvm/arm_arch_timer.h +++ b/include/kvm/arm_arch_timer.h @@ -93,7 +93,4 @@ void kvm_timer_init_vhe(void); #define vcpu_vtimer(v) (&(v)->arch.timer_cpu.vtimer) #define vcpu_ptimer(v) (&(v)->arch.timer_cpu.ptimer) -void enable_el1_phys_timer_access(void); -void disable_el1_phys_timer_access(void); - #endif -- cgit v1.2.3 From f775b13eedee2f7f3c6fdd4e90fb79090ce5d339 Mon Sep 17 00:00:00 2001 From: Rik van Riel Date: Tue, 14 Nov 2017 16:54:23 -0500 Subject: x86,kvm: move qemu/guest FPU switching out to vcpu_run MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Currently, every time a VCPU is scheduled out, the host kernel will first save the guest FPU/xstate context, then load the qemu userspace FPU context, only to then immediately save the qemu userspace FPU context back to memory. When scheduling in a VCPU, the same extraneous FPU loads and saves are done. This could be avoided by moving from a model where the guest FPU is loaded and stored with preemption disabled, to a model where the qemu userspace FPU is swapped out for the guest FPU context for the duration of the KVM_RUN ioctl. This is done under the VCPU mutex, which is also taken when other tasks inspect the VCPU FPU context, so the code should already be safe for this change. That should come as no surprise, given that s390 already has this optimization. This can fix a bug where KVM calls get_user_pages while owning the FPU, and the file system ends up requesting the FPU again: [258270.527947] __warn+0xcb/0xf0 [258270.527948] warn_slowpath_null+0x1d/0x20 [258270.527951] kernel_fpu_disable+0x3f/0x50 [258270.527953] __kernel_fpu_begin+0x49/0x100 [258270.527955] kernel_fpu_begin+0xe/0x10 [258270.527958] crc32c_pcl_intel_update+0x84/0xb0 [258270.527961] crypto_shash_update+0x3f/0x110 [258270.527968] crc32c+0x63/0x8a [libcrc32c] [258270.527975] dm_bm_checksum+0x1b/0x20 [dm_persistent_data] [258270.527978] node_prepare_for_write+0x44/0x70 [dm_persistent_data] [258270.527985] dm_block_manager_write_callback+0x41/0x50 [dm_persistent_data] [258270.527988] submit_io+0x170/0x1b0 [dm_bufio] [258270.527992] __write_dirty_buffer+0x89/0x90 [dm_bufio] [258270.527994] __make_buffer_clean+0x4f/0x80 [dm_bufio] [258270.527996] __try_evict_buffer+0x42/0x60 [dm_bufio] [258270.527998] dm_bufio_shrink_scan+0xc0/0x130 [dm_bufio] [258270.528002] shrink_slab.part.40+0x1f5/0x420 [258270.528004] shrink_node+0x22c/0x320 [258270.528006] do_try_to_free_pages+0xf5/0x330 [258270.528008] try_to_free_pages+0xe9/0x190 [258270.528009] __alloc_pages_slowpath+0x40f/0xba0 [258270.528011] __alloc_pages_nodemask+0x209/0x260 [258270.528014] alloc_pages_vma+0x1f1/0x250 [258270.528017] do_huge_pmd_anonymous_page+0x123/0x660 [258270.528021] handle_mm_fault+0xfd3/0x1330 [258270.528025] __get_user_pages+0x113/0x640 [258270.528027] get_user_pages+0x4f/0x60 [258270.528063] __gfn_to_pfn_memslot+0x120/0x3f0 [kvm] [258270.528108] try_async_pf+0x66/0x230 [kvm] [258270.528135] tdp_page_fault+0x130/0x280 [kvm] [258270.528149] kvm_mmu_page_fault+0x60/0x120 [kvm] [258270.528158] handle_ept_violation+0x91/0x170 [kvm_intel] [258270.528162] vmx_handle_exit+0x1ca/0x1400 [kvm_intel] No performance changes were detected in quick ping-pong tests on my 4 socket system, which is expected since an FPU+xstate load is on the order of 0.1us, while ping-ponging between CPUs is on the order of 20us, and somewhat noisy. Cc: stable@vger.kernel.org Signed-off-by: Rik van Riel Suggested-by: Christian Borntraeger Signed-off-by: Paolo Bonzini [Fixed a bug where reset_vcpu called put_fpu without preceding load_fpu, which happened inside from KVM_CREATE_VCPU ioctl. - Radim] Signed-off-by: Radim Krčmář --- include/linux/kvm_host.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'include') diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h index 893d6d606cd0..6bdd4b9f6611 100644 --- a/include/linux/kvm_host.h +++ b/include/linux/kvm_host.h @@ -232,7 +232,7 @@ struct kvm_vcpu { struct mutex mutex; struct kvm_run *run; - int guest_fpu_loaded, guest_xcr0_loaded; + int guest_xcr0_loaded; struct swait_queue_head wq; struct pid __rcu *pid; int sigset_active; -- cgit v1.2.3 From bb64da9aba89765fee74b395967b18a7d6c364e9 Mon Sep 17 00:00:00 2001 From: Christian Borntraeger Date: Tue, 21 Nov 2017 16:02:52 +0100 Subject: KVM: s390: mark irq_state.flags as non-usable Old kernels did not check for zero in the irq_state.flags field and old QEMUs did not zero the flag/reserved fields when calling KVM_S390_*_IRQ_STATE. Let's add comments to prevent future uses of these fields. Signed-off-by: Christian Borntraeger Reviewed-by: Thomas Huth Reviewed-by: Cornelia Huck Reviewed-by: David Hildenbrand Signed-off-by: Christian Borntraeger --- include/uapi/linux/kvm.h | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'include') diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h index 282d7613fce8..496e59a2738b 100644 --- a/include/uapi/linux/kvm.h +++ b/include/uapi/linux/kvm.h @@ -630,9 +630,9 @@ struct kvm_s390_irq { struct kvm_s390_irq_state { __u64 buf; - __u32 flags; + __u32 flags; /* will stay unused for compatibility reasons */ __u32 len; - __u32 reserved[4]; + __u32 reserved[4]; /* will stay unused for compatibility reasons */ }; /* for KVM_SET_GUEST_DEBUG */ -- cgit v1.2.3