Re: [PATCH 1/2] asm-generic/hyperv: provide cpumask_to_vpset_noself

On Thu, Sep 09, 2021 at 07:38:28AM +0200, Vitaly Kuznetsov wrote:

Wei Liu [off-list ref] writes:

quoted

This is a new variant which removes `self' cpu from the vpset. It will
be used in Hyper-V enlightened IPI code.

Signed-off-by: Wei Liu <wei.liu@kernel.org>
---
Provide a new variant instead of adding a new parameter because it makes
it easier to backport -- we don't need to fix the users of
cpumask_to_vpset.
---
 include/asm-generic/mshyperv.h | 20 ++++++++++++++++++--
 1 file changed, 18 insertions(+), 2 deletions(-)

diff --git a/include/asm-generic/mshyperv.h b/include/asm-generic/mshyperv.h
index 9a000ba2bb75..d89690ee95aa 100644
--- a/include/asm-generic/mshyperv.h
+++ b/include/asm-generic/mshyperv.h

@@ -184,10 +184,12 @@ static inline int hv_cpu_number_to_vp_number(int cpu_number)
 	return hv_vp_index[cpu_number];
 }
 
-static inline int cpumask_to_vpset(struct hv_vpset *vpset,
-				    const struct cpumask *cpus)
+static inline int cpumask_to_vpset_ex(struct hv_vpset *vpset,

I'd suggest to avoid '_ex' suffix as we use it for 'extended hypercalls'
(e.g. __send_ipi_mask_ex). Assuming nobody needs to call
cpumask_to_vpset_ex() directly, should we just go with
__cpumask_to_vpset() instead?

Sure. I'm not too fussed about the name.

I will wait a bit for other people to express their opinions.

quoted

+				    const struct cpumask *cpus,
+				    bool exclude_self)
 {
 	int cpu, vcpu, vcpu_bank, vcpu_offset, nr_bank = 1;
+	int this_cpu = smp_processor_id();
 
 	/* valid_bank_mask can represent up to 64 banks */
 	if (hv_max_vp_index / 64 >= 64)

@@ -205,6 +207,8 @@ static inline int cpumask_to_vpset(struct hv_vpset *vpset,
 	 * Some banks may end up being empty but this is acceptable.
 	 */
 	for_each_cpu(cpu, cpus) {
+		if (exclude_self && cpu == this_cpu)
+			continue;
 		vcpu = hv_cpu_number_to_vp_number(cpu);
 		if (vcpu == VP_INVAL)
 			return -1;

@@ -219,6 +223,18 @@ static inline int cpumask_to_vpset(struct hv_vpset *vpset,
 	return nr_bank;
 }
 
+static inline int cpumask_to_vpset(struct hv_vpset *vpset,
+				    const struct cpumask *cpus)
+{
+	return cpumask_to_vpset_ex(vpset, cpus, false);
+}
+
+static inline int cpumask_to_vpset_noself(struct hv_vpset *vpset,
+				    const struct cpumask *cpus)
+{
+	return cpumask_to_vpset_ex(vpset, cpus, true);

We need to make sure this is called with preemption disabled. We
could've just swapped smp_processor_id() for get_cpu() in
cpumask_to_vpset_ex() but this is hardly a solution: we can get
preempted right after put_cpu() so it's really the caller of this
function which needs to be protected.

TL;DR: I suggest we add 'WARN_ON_ONCE(preemptible());' or something like
this here to catch wrong usage.

Good suggestion. I can add this check to the noself variant. Or if
people prefer, this check can also be moved into the leaf helper.

Wei.

quoted

+}
+
 void hyperv_report_panic(struct pt_regs *regs, long err, bool in_die);
 bool hv_is_hyperv_initialized(void);
 bool hv_is_hibernation_supported(void);

-- 
Vitaly