On 2025/11/24 下午5:03, Huacai Chen wrote:
On Mon, Nov 24, 2025 at 4:35 PM Bibo Mao <[email protected]> wrote:
On 2025/11/24 下午4:03, Huacai Chen wrote:
On Mon, Nov 24, 2025 at 3:50 PM Bibo Mao <[email protected]> wrote:
On 2025/11/24 下午3:13, Huacai Chen wrote:
On Mon, Nov 24, 2025 at 3:03 PM Bibo Mao <[email protected]> wrote:
On 2025/11/24 下午2:33, Huacai Chen wrote:
Hi, Bibo,
On Mon, Nov 24, 2025 at 11:54 AM Bibo Mao <[email protected]> wrote:
Function vcpu_is_preempted() is used to check whether vCPU is preempted
or not. Here add implementation with vcpu_is_preempted() when option
CONFIG_PARAVIRT is enabled.
Signed-off-by: Bibo Mao <[email protected]>steal_time
---
arch/loongarch/include/asm/qspinlock.h | 5 +++++
arch/loongarch/kernel/paravirt.c | 16 ++++++++++++++++
2 files changed, 21 insertions(+)
diff --git a/arch/loongarch/include/asm/qspinlock.h
b/arch/loongarch/include/asm/qspinlock.h
index e76d3aa1e1eb..9a5b7ba1f4cb 100644
--- a/arch/loongarch/include/asm/qspinlock.h
+++ b/arch/loongarch/include/asm/qspinlock.h
@@ -34,6 +34,11 @@ static inline bool virt_spin_lock(struct qspinlock *lock)
return true;
}
+#ifdef CONFIG_SMP
+#define vcpu_is_preempted vcpu_is_preempted
+bool vcpu_is_preempted(int cpu);
In V1 there is a build error because you reference mp_ops, so in V2
you needn't put it in CONFIG_SMP.
The compile failure problem is that vcpu_is_preempted() is redefined in
both arch/loongarch/kernel/paravirt.c and include/linux/sched.h
But other archs don't define vcpu_is_preempted() under CONFIG_SMP, and
so what is advantage to implement this function if CONFIG_SMP is disabled?
1. Keep consistency with other architectures.
2. Keep it simple to reduce #ifdefs (and !SMP is just for build, not
very useful in practice).
It seems that CONFIG_SMP can be removed in header file
include/asm/qspinlock.h, since asm/spinlock.h and asm/qspinlock.h is
only included when CONFIG_SMP is set, otherwise only linux/spinlock_up.h
is included.
you can consider to inline the whole vcpu_is_preempted() here.
Defining the function vcpu_is_preempted() as inlined is not so easy for
me, it beyond my ability now :(
With static key method, the static key need be exported, all modules
need apply the jump label, that is dangerous and I doubt whether it is
deserved.
No, you have already done similar things in virt_spin_lock(), it is an
inline function and uses virt_spin_lock_key.
virt_spin_lock is only called qspinlock in function
queued_spin_lock_slowpath(). Function vcpu_is_preempted() is defined
header file linux/sched.h, kernel module may use it.
Yes, if modules want to use it we need to EXPORT_SYMBOL. But don't
worry, static key infrastructure can handle this. Please see
cpu_feature_keys defined and used in
arch/powerpc/include/asm/cpu_has_feature.h, which is exported in
arch/powerpc/kernel/cputable.c.
No, I do not want to do so. export static key and percpu steal_time
structure, just in order to implement one inline function.
Huacai
Huacai
Regards
Bibo Mao
The problem is that <asm/spinlock.h> is not included by sched.h, if
CONFIG_SMP is disabled. Here is part of file include/linux/spinlock.h
#ifdef CONFIG_SMP
# include <asm/spinlock.h>
#else
# include <linux/spinlock_up.h>
#endif
On the other hand, even if you really build a UP guest kernel, when
multiple guests run together, you probably need vcpu_is_preemtped.
It is not relative with multiple VMs. When vcpu_is_preempted() is
called, it is to detect whether dest CPU is preempted or not, the cpu
from smp_processor_id() should not be preempted. So in generic
vcpu_is_preempted() works on multiple vCPUs.
OK, I'm wrong here.
Huacai
Regards
Bibo Mao
Huacai
+#endif
+
#endif /* CONFIG_PARAVIRT */
#include <asm-generic/qspinlock.h>
diff --git a/arch/loongarch/kernel/paravirt.c b/arch/loongarch/kernel/paravirt.c
index b1b51f920b23..d4163679adc4 100644
--- a/arch/loongarch/kernel/paravirt.c
+++ b/arch/loongarch/kernel/paravirt.c
@@ -246,6 +246,7 @@ static void pv_disable_steal_time(void)
}
#ifdef CONFIG_SMP
+DEFINE_STATIC_KEY_FALSE(virt_preempt_key);
static int pv_time_cpu_online(unsigned int cpu)
{
unsigned long flags;
@@ -267,6 +268,18 @@ static int pv_time_cpu_down_prepare(unsigned int cpu)
return 0;
}
+
+bool notrace vcpu_is_preempted(int cpu)
+{
+ struct kvm_steal_time *src;
+
+ if (!static_branch_unlikely(&virt_preempt_key))
+ return false;
+
+ src = &per_cpu(steal_time, cpu);
+ return !!(src->preempted & KVM_VCPU_PREEMPTED);
+}
+EXPORT_SYMBOL(vcpu_is_preempted);
#endif
static void pv_cpu_reboot(void *unused)
@@ -308,6 +321,9 @@ int __init pv_time_init(void)
pr_err("Failed to install cpu hotplug callbacks\n");
return r;
}
+
+ if (kvm_para_has_feature(KVM_FEATURE_PREEMPT))
+ static_branch_enable(&virt_preempt_key);
#endif
static_call_update(pv_steal_clock, paravt_steal_clock);
--
2.39.3
