KVM: s390: make use of ultravisor AIV support
v4 cover KVM: s390: make use of ultravisor AIV support
This patch enables the ultravisor adapter interruption vitualization support.
Changes in v4:
- All vcpus are pulled out of SIE and disbled before the control blocks are changed to disable/enable the gisa and re-enabled afterwards instead of taking the vcpu specific lock.
Changes in v3:
- cache and test GISA descriptor only once in kvm_s390_gisa_enable()
- modified some comments
- removed some whitespace issues
Changes in v2:
- moved GISA disable into "put CPUs in PV mode" routine
- moved GISA enable into "pull CPUs out of PV mode" routine
[1] https://lore.kernel.org/lkml/[email protected]/T/#mcb67699bf458ba7482f6b7529afe589d1dbb5930
[2] https://lore.kernel.org/all/[email protected]/
[3] https://lore.kernel.org/all/[email protected]/
Michael Mueller (2):
KVM: s390: pv: pull all vcpus out of SIE before converting to/from pv
vcpu
KVM: s390: pv: make use of ultravisor AIV support
arch/s390/include/asm/uv.h | 1 + arch/s390/kvm/interrupt.c | 54 +++++++++++++++++++++++++++++++++----- arch/s390/kvm/kvm-s390.c | 19 +++++++++----- arch/s390/kvm/kvm-s390.h | 11 ++++++++ 4 files changed, 72 insertions(+), 13 deletions(-)
Signature · 1 line
2.32.0
PATCH v4 1/2 KVM: s390: pv: pull all vcpus out of SIE before converting to/from pv vcpu
That will allow to modify the control blocks without locking the vcpus. A kvm-unit-test surfaced a possible DEADLOCK situation when guest cpus are executing intercepted instructions at this time.
[ 319.799638] ======================================================
[ 319.799639] WARNING: possible circular locking dependency detected
[ 319.799641] 5.17.0-rc5-08427-gfd14b6309198 #4661 Not tainted
[ 319.799643] ------------------------------------------------------
[ 319.799644] qemu-system-s39/14220 is trying to acquire lock:
[ 319.799646] 00000000b30c0b50 (&kvm->lock){+.+.}-{3:3}, at: kvm_s390_set_tod_clock+0x36/0x250
[ 319.799659]
but task is already holding lock:
[ 319.799660] 00000000b5beda60 (&vcpu->mutex){+.+.}-{3:3}, at: kvm_vcpu_ioctl+0x9a/0x958
[ 319.799665]
which lock already depends on the new lock.[ 319.799667]
the existing dependency chain (in reverse order) is:
[ 319.799668]
-> #1 (&vcpu->mutex){+.+.}-{3:3}:
[ 319.799671] __mutex_lock+0x8a/0x798
[ 319.799677] mutex_lock_nested+0x32/0x40
[ 319.799679] kvm_arch_vm_ioctl+0x1902/0x2c58
[ 319.799682] kvm_vm_ioctl+0x5b0/0xa80
[ 319.799685] __s390x_sys_ioctl+0xbe/0x100
[ 319.799688] __do_syscall+0x1da/0x208
[ 319.799689] system_call+0x82/0xb0
[ 319.799692]
-> #0 (&kvm->lock){+.+.}-{3:3}:
[ 319.799694] __lock_acquire+0x1916/0x2e70
[ 319.799699] lock_acquire+0x164/0x388
[ 319.799702] __mutex_lock+0x8a/0x798
[ 319.799757] mutex_lock_nested+0x32/0x40
[ 319.799759] kvm_s390_set_tod_clock+0x36/0x250
[ 319.799761] kvm_s390_handle_b2+0x6cc/0x26f0
[ 319.799764] kvm_handle_sie_intercept+0x1fe/0xe98
[ 319.799765] kvm_arch_vcpu_ioctl_run+0xec8/0x1880
[ 319.799768] kvm_vcpu_ioctl+0x29e/0x958
[ 319.799769] __s390x_sys_ioctl+0xbe/0x100
[ 319.799771] __do_syscall+0x1da/0x208
[ 319.799773] system_call+0x82/0xb0
[ 319.799774]
other info that might help us debug this:[ 319.799776] Possible unsafe locking scenario:
[ 319.799777] CPU0 CPU1
[ 319.799778] ---- ----
[ 319.799779] lock(&vcpu->mutex);
[ 319.799780] lock(&kvm->lock);
[ 319.799782] lock(&vcpu->mutex);
[ 319.799783] lock(&kvm->lock);
[ 319.799784]
*** DEADLOCK ***[ 319.799785] 2 locks held by qemu-system-s39/14220:
[ 319.799787] #0: 00000000b5beda60 (&vcpu->mutex){+.+.}-{3:3}, at: kvm_vcpu_ioctl+0x9a/0x958
[ 319.799791] #1: 00000000b30c4588 (&kvm->srcu){....}-{0:0}, at: kvm_arch_vcpu_ioctl_run+0x6f2/0x1880
[ 319.799796]
stack backtrace:
[ 319.799798] CPU: 5 PID: 14220 Comm: qemu-system-s39 Not tainted 5.17.0-rc5-08427-gfd14b6309198 #4661
[ 319.799801] Hardware name: IBM 8561 T01 701 (LPAR)
[ 319.799802] Call Trace:
[ 319.799803] [<000000020d7410de>] dump_stack_lvl+0x76/0x98
[ 319.799808] [<000000020cbbd268>] check_noncircular+0x140/0x160
[ 319.799811] [<000000020cbc0efe>] __lock_acquire+0x1916/0x2e70
[ 319.799813] [<000000020cbc2dbc>] lock_acquire+0x164/0x388
[ 319.799816] [<000000020d75013a>] __mutex_lock+0x8a/0x798
[ 319.799818] [<000000020d75087a>] mutex_lock_nested+0x32/0x40
[ 319.799820] [<000000020cb029a6>] kvm_s390_set_tod_clock+0x36/0x250
[ 319.799823] [<000000020cb14d14>] kvm_s390_handle_b2+0x6cc/0x26f0
[ 319.799825] [<000000020cb09b6e>] kvm_handle_sie_intercept+0x1fe/0xe98
[ 319.799827] [<000000020cb06c28>] kvm_arch_vcpu_ioctl_run+0xec8/0x1880
[ 319.799829] [<000000020caeddc6>] kvm_vcpu_ioctl+0x29e/0x958
[ 319.799831] [<000000020ce4e82e>] __s390x_sys_ioctl+0xbe/0x100
[ 319.799833] [<000000020d744a72>] __do_syscall+0x1da/0x208
[ 319.799835] [<000000020d757322>] system_call+0x82/0xb0
[ 319.799836] INFO: lockdep is turned off.--- arch/s390/kvm/kvm-s390.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
arch/s390/kvm/kvm-s390.c +4 -4
@@ -2185,15 +2185,15 @@ static int kvm_s390_cpus_from_pv(struct kvm *kvm, u16 *rcp, u16 *rrcp)* behind. * We want to return the first failure rc and rrc, though. */ +kvm_s390_vcpu_block_all (kvm);kvm_for_each_vcpu (i, vcpu, kvm) { -mutex_lock (&vcpu->mutex);if (kvm_s390_pv_destroy_cpu (vcpu, &rc, &rrc) && !ret) { *rcp = rc; *rrcp = rrc; ret = -EIO ; } -mutex_unlock (&vcpu->mutex); } +kvm_s390_vcpu_unblock_all (kvm);return ret; }
@@ -2205,13 +2205,13 @@ static int kvm_s390_cpus_to_pv(struct kvm *kvm, u16 *rc, u16 *rrc)struct kvm_vcpu *vcpu; +kvm_s390_vcpu_block_all (kvm);kvm_for_each_vcpu (i, vcpu, kvm) { -mutex_lock (&vcpu->mutex); r =kvm_s390_pv_create_cpu (vcpu, rc, rrc); -mutex_unlock (&vcpu->mutex);if (r)break ; } +kvm_s390_vcpu_unblock_all (kvm);if (r)kvm_s390_cpus_from_pv (kvm, &dummy, &dummy);return r;
Signature · 1 line
2.32.0
PATCH v4 2/2 KVM: s390: pv: make use of ultravisor AIV support
In reply to Michael MuellerThis patch enables the ultravisor adapter interruption vitualization support indicated by UV feature BIT_UV_FEAT_AIV. This allows ISC interruption injection directly into the GISA IPM for PV kvm guests.
Hardware that does not support this feature will continue to use the UV interruption interception method to deliver ISC interruptions to PV kvm guests. For this purpose, the ECA_AIV bit for all guest cpus will be cleared and the GISA will be disabled during PV CPU setup.
In addition a check in __inject_io() has been removed. That reduces the required instructions for interruption handling for PV and traditional kvm guests.
--- arch/s390/include/asm/uv.h | 1 + arch/s390/kvm/interrupt.c | 54 +++++++++++++++++++++++++++++++++----- arch/s390/kvm/kvm-s390.c | 11 +++++--- arch/s390/kvm/kvm-s390.h | 11 ++++++++ 4 files changed, 68 insertions(+), 9 deletions(-)
arch/s390/include/asm/uv.h +1 -0
@@ -80,6 +80,7 @@ enum uv_cmds_inst {enum uv_feat_ind {BIT_UV_FEAT_MISC =0 , +BIT_UV_FEAT_AIV =1 , };struct uv_cb_header {
arch/s390/kvm/interrupt.c +48 -6
76 lines not on this page · raw message
arch/s390/kvm/kvm-s390.c +8 -3
@@ -2194,6 +2194,9 @@ static int kvm_s390_cpus_from_pv(struct kvm *kvm, u16 *rcp, u16 *rrcp) } }kvm_s390_vcpu_unblock_all (kvm); +/* Ensure that we re-enable gisa if the non-PV guest used it but the PV guest did not. */ +if (use_gisa) +kvm_s390_gisa_enable (kvm);return ret; }
@@ -2205,6 +2208,10 @@ static int kvm_s390_cpus_to_pv(struct kvm *kvm, u16 *rc, u16 *rrc)struct kvm_vcpu *vcpu; +/* Disable the GISA if the ultravisor does not support AIV. */ +if (!test_bit_inv (BIT_UV_FEAT_AIV , &uv_info.uv_feature_indications)) +kvm_s390_gisa_disable (kvm); +kvm_s390_vcpu_block_all (kvm);kvm_for_each_vcpu (i, vcpu, kvm) { r =kvm_s390_pv_create_cpu (vcpu, rc, rrc);
@@ -3263,9 +3270,7 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) vcpu->arch.sie_block->icpua = vcpu->vcpu_id;spin_lock_init (&vcpu->arch.local_int.lock); - vcpu->arch.sie_block->gd = (u32 )(u64 )vcpu->kvm->arch.gisa_int.origin; -if (vcpu->arch.sie_block->gd && sclp.has_gisaf) - vcpu->arch.sie_block->gd |=GISA_FORMAT1 ; + vcpu->arch.sie_block->gd =kvm_s390_get_gisa_desc (vcpu->kvm);seqcount_init (&vcpu->arch.cputm_seqcount); vcpu->arch.pfault_token =KVM_S390_PFAULT_TOKEN_INVALID ;
arch/s390/kvm/kvm-s390.h +11 -0
@@ -231,6 +231,15 @@ static inline unsigned long kvm_s390_get_gfn_end(struct kvm_memslots *slots)return ms->base_gfn + ms->npages; } +static inline u32 kvm_s390_get_gisa_desc (struct kvm *kvm) +{ +u32 gd = (u32 )(u64 )kvm->arch.gisa_int.origin; + +if (gd && sclp.has_gisaf) + gd |=GISA_FORMAT1 ; +return gd; +} +/* implemented in pv.c */ int kvm_s390_pv_destroy_cpu (struct kvm_vcpu *vcpu,u16 *rc,u16 *rrc);int kvm_s390_pv_create_cpu (struct kvm_vcpu *vcpu,u16 *rc,u16 *rrc);
@@ -450,6 +459,8 @@ int kvm_s390_get_irq_state(struct kvm_vcpu *vcpu,void kvm_s390_gisa_init (struct kvm *kvm);void kvm_s390_gisa_clear (struct kvm *kvm);void kvm_s390_gisa_destroy (struct kvm *kvm); +void kvm_s390_gisa_disable (struct kvm *kvm); +void kvm_s390_gisa_enable (struct kvm *kvm);int kvm_s390_gib_init (u8 nisc);void kvm_s390_gib_destroy (void );
Signature · 1 line
2.32.0
Quoted 7 lines · Michael Mueller
This patch enables the ultravisor adapter interruption vitualization support.
Changes in v4:
- All vcpus are pulled out of SIE and disbled before the control blocks are changed to disable/enable the gisa and re-enabled afterwards instead of taking the vcpu specific lock.
Christian,
I'm pulling back this v4 and ask for v3 to be re-applied as it turned out that the described issue is related to kvm_s390_set_tod_clock() being called in SIE intercepted context.
Quoted 25 lines
Changes in v3:
- cache and test GISA descriptor only once inkvm_s390_gisa_enable()
- modified some comments
- removed some whitespace issuesChanges in v2:
- moved GISA disable into "put CPUs in PV mode" routine
- moved GISA enable into "pull CPUs out of PV mode" routine[1] https://lore.kernel.org/lkml/[email protected]/T/#mcb67699bf458ba7482f6b7529afe589d1dbb5930
[2] https://lore.kernel.org/all/[email protected]/
[3] https://lore.kernel.org/all/[email protected]/Michael Mueller (2):
KVM: s390: pv: pull all vcpus out of SIE before converting to/from pv
vcpu
KVM: s390: pv: make use of ultravisor AIV supportarch/s390/include/asm/uv.h | 1 + arch/s390/kvm/interrupt.c | 54 +++++++++++++++++++++++++++++++++----- arch/s390/kvm/kvm-s390.c | 19 +++++++++----- arch/s390/kvm/kvm-s390.h | 11 ++++++++ 4 files changed, 72 insertions(+), 13 deletions(-)