| Message ID | 1474425470-3629-7-git-send-email-feng.wu@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
>>> On 21.09.16 at 04:37, <feng.wu@intel.com> wrote: > +void vmx_pi_desc_fixup(int cpu) > +{ > + unsigned int new_cpu, dest; > + unsigned long flags; > + struct arch_vmx_struct *vmx, *tmp; > + spinlock_t *new_lock, *old_lock = &per_cpu(vmx_pi_blocking, cpu).lock; > + struct list_head *blocked_vcpus = &per_cpu(vmx_pi_blocking, cpu).list; > + > + if ( !iommu_intpost ) > + return; > + > + spin_lock_irqsave(old_lock, flags); > + > + list_for_each_entry_safe(vmx, tmp, blocked_vcpus, pi_blocking.list) > + { > + /* > + * We need to find an online cpu as the NDST of the PI descriptor, it > + * doesn't matter whether it is within the cpupool of the domain or > + * not. As long as it is online, the vCPU will be woken up once the > + * notification event arrives. > + */ > + new_cpu = cpumask_any(&cpu_online_map); > + new_lock = &per_cpu(vmx_pi_blocking, new_cpu).lock; > + > + spin_lock(new_lock); Without extra consideration this is a classical ABBA deadlock scenario. Please add a comment (perhaps at the first lock location above) at least briefly explaining why there can't be any deadlock here. Apart from that the patch looks fine to me now. Jan
> -----Original Message----- > From: Jan Beulich [mailto:JBeulich@suse.com] > Sent: Monday, September 26, 2016 9:03 PM > To: Wu, Feng <feng.wu@intel.com> > Cc: andrew.cooper3@citrix.com; dario.faggioli@citrix.com; > george.dunlap@eu.citrix.com; Tian, Kevin <kevin.tian@intel.com>; xen- > devel@lists.xen.org > Subject: Re: [PATCH v4 6/6] VMX: Fixup PI descritpor when cpu is offline > > >>> On 21.09.16 at 04:37, <feng.wu@intel.com> wrote: > > +void vmx_pi_desc_fixup(int cpu) > > +{ > > + unsigned int new_cpu, dest; > > + unsigned long flags; > > + struct arch_vmx_struct *vmx, *tmp; > > + spinlock_t *new_lock, *old_lock = &per_cpu(vmx_pi_blocking, cpu).lock; > > + struct list_head *blocked_vcpus = &per_cpu(vmx_pi_blocking, cpu).list; > > + > > + if ( !iommu_intpost ) > > + return; > > + > > + spin_lock_irqsave(old_lock, flags); > > + > > + list_for_each_entry_safe(vmx, tmp, blocked_vcpus, pi_blocking.list) > > + { > > + /* > > + * We need to find an online cpu as the NDST of the PI descriptor, it > > + * doesn't matter whether it is within the cpupool of the domain or > > + * not. As long as it is online, the vCPU will be woken up once the > > + * notification event arrives. > > + */ > > + new_cpu = cpumask_any(&cpu_online_map); > > + new_lock = &per_cpu(vmx_pi_blocking, new_cpu).lock; > > + > > + spin_lock(new_lock); > > Without extra consideration this is a classical ABBA deadlock > scenario. Please add a comment (perhaps at the first lock location > above) at least briefly explaining why there can't be any deadlock > here. Yes, I also noted this. I will add a comment here, which is really needed. Thanks a lot! Thanks, Feg > > Apart from that the patch looks fine to me now. > > Jan
diff --git a/xen/arch/x86/hvm/vmx/vmcs.c b/xen/arch/x86/hvm/vmx/vmcs.c index e733753..9f56c7c 100644 --- a/xen/arch/x86/hvm/vmx/vmcs.c +++ b/xen/arch/x86/hvm/vmx/vmcs.c @@ -578,6 +578,7 @@ void vmx_cpu_dead(unsigned int cpu) vmx_free_vmcs(per_cpu(vmxon_region, cpu)); per_cpu(vmxon_region, cpu) = 0; nvmx_cpu_dead(cpu); + vmx_pi_desc_fixup(cpu); } int vmx_cpu_up(void) diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c index 09262d5..ff87444 100644 --- a/xen/arch/x86/hvm/vmx/vmx.c +++ b/xen/arch/x86/hvm/vmx/vmx.c @@ -208,6 +208,48 @@ static void vmx_pi_list_cleanup(struct vcpu *v) vmx_pi_list_remove(v); } +void vmx_pi_desc_fixup(int cpu) +{ + unsigned int new_cpu, dest; + unsigned long flags; + struct arch_vmx_struct *vmx, *tmp; + spinlock_t *new_lock, *old_lock = &per_cpu(vmx_pi_blocking, cpu).lock; + struct list_head *blocked_vcpus = &per_cpu(vmx_pi_blocking, cpu).list; + + if ( !iommu_intpost ) + return; + + spin_lock_irqsave(old_lock, flags); + + list_for_each_entry_safe(vmx, tmp, blocked_vcpus, pi_blocking.list) + { + /* + * We need to find an online cpu as the NDST of the PI descriptor, it + * doesn't matter whether it is within the cpupool of the domain or + * not. As long as it is online, the vCPU will be woken up once the + * notification event arrives. + */ + new_cpu = cpumask_any(&cpu_online_map); + new_lock = &per_cpu(vmx_pi_blocking, new_cpu).lock; + + spin_lock(new_lock); + + ASSERT(vmx->pi_blocking.lock == old_lock); + + dest = cpu_physical_id(new_cpu); + write_atomic(&vmx->pi_desc.ndst, + x2apic_enabled ? dest : MASK_INSR(dest, PI_xAPIC_NDST_MASK)); + + list_move(&vmx->pi_blocking.list, + &per_cpu(vmx_pi_blocking, new_cpu).list); + vmx->pi_blocking.lock = new_lock; + + spin_unlock(new_lock); + } + + spin_unlock_irqrestore(old_lock, flags); +} + /* This function is called when pcidevs_lock is held */ void vmx_pi_hooks_assign(struct domain *d) { diff --git a/xen/include/asm-x86/hvm/vmx/vmx.h b/xen/include/asm-x86/hvm/vmx/vmx.h index 4cdd9b1..9783c70 100644 --- a/xen/include/asm-x86/hvm/vmx/vmx.h +++ b/xen/include/asm-x86/hvm/vmx/vmx.h @@ -569,6 +569,7 @@ void free_p2m_hap_data(struct p2m_domain *p2m); void p2m_init_hap_data(struct p2m_domain *p2m); void vmx_pi_per_cpu_init(unsigned int cpu); +void vmx_pi_desc_fixup(int cpu); void vmx_pi_hooks_assign(struct domain *d); void vmx_pi_hooks_deassign(struct domain *d);
When cpu is offline, we need to move all the vcpus in its blocking list to another online cpu, this patch handles it. Signed-off-by: Feng Wu <feng.wu@intel.com> --- v4: - Remove the pointless check since we are in machine stop context and no other cpus go down in parallel. xen/arch/x86/hvm/vmx/vmcs.c | 1 + xen/arch/x86/hvm/vmx/vmx.c | 42 +++++++++++++++++++++++++++++++++++++++ xen/include/asm-x86/hvm/vmx/vmx.h | 1 + 3 files changed, 44 insertions(+)