[PATCH v2] Drivers: hv: vmbus: Optimize boot time by concurrent execution of hv_synic_init()

[Saurabh Sengar]

Currently on a very large system with 1780 CPUs, hv_acpi_init() takes
around 3 seconds to complete. This is because of sequential synic
initialization for each CPU performed by hv_synic_init().

Schedule these tasks parallelly so that each CPU executes hv_synic_init()
in parallel to take full advantage of multiple CPUs.

This solution saves around 2 seconds of boot time on a 1780 CPU system,
which is around 66% improvement in the existing logic.

Signed-off-by: Saurabh Sengar <sse...@linux.microsoft.com>
---
[V2]
- used cpuhp_setup_state_nocalls_cpuslocked instead of internal function
- improve commit message and subject
- Added a comment for cpu hotplug callback


drivers/hv/vmbus_drv.c | 34 +++++++++++++++++++++++++++++++---
1 file changed, 31 insertions(+), 3 deletions(-)

diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
index c857dc3975be..f769b34445b8 100644
--- a/drivers/hv/vmbus_drv.c
+++ b/drivers/hv/vmbus_drv.c
@@ -1306,6 +1306,13 @@ static irqreturn_t vmbus_percpu_isr(int irq, void *dev_id)
return IRQ_HANDLED;
}

+static void vmbus_percpu_work(struct work_struct *work)
+{
+ unsigned int cpu = smp_processor_id();
+
+ hv_synic_init(cpu);
+}
+
/*
* vmbus_bus_init -Main vmbus driver initialization routine.
*
@@ -1316,7 +1323,8 @@ static irqreturn_t vmbus_percpu_isr(int irq, void *dev_id)
*/
static int vmbus_bus_init(void)
{
- int ret;
+ int ret, cpu;
+ struct work_struct __percpu *works;

ret = hv_init();
if (ret != 0) {
@@ -1355,12 +1363,32 @@ static int vmbus_bus_init(void)
if (ret)
goto err_alloc;

+ works = alloc_percpu(struct work_struct);
+ if (!works) {
+ ret = -ENOMEM;
+ goto err_alloc;
+ }
+
/*
* Initialize the per-cpu interrupt state and stimer state.
* Then connect to the host.
*/
- ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
- hv_synic_init, hv_synic_cleanup);
+ cpus_read_lock();
+ for_each_online_cpu(cpu) {
+ struct work_struct *work = per_cpu_ptr(works, cpu);
+
+ INIT_WORK(work, vmbus_percpu_work);
+ schedule_work_on(cpu, work);
+ }
+
+ for_each_online_cpu(cpu)
+ flush_work(per_cpu_ptr(works, cpu));
+
+ /* register the callback for hotplug CPUs */
+ ret = cpuhp_setup_state_nocalls_cpuslocked(CPUHP_AP_ONLINE_DYN, "hyperv/vmbus:online",
+ hv_synic_init, hv_synic_cleanup);
+ cpus_read_unlock();
+ free_percpu(works);
if (ret < 0)
goto err_alloc;
hyperv_cpuhp_online = ret;
--
2.43.0

[Srivatsa S. Bhat]

On Mon, Jul 29, 2024 at 12:57:18AM -0700, Saurabh Sengar wrote:
> Currently on a very large system with 1780 CPUs, hv_acpi_init() takes
> around 3 seconds to complete. This is because of sequential synic
> initialization for each CPU performed by hv_synic_init().
>
> Schedule these tasks parallelly so that each CPU executes hv_synic_init()
> in parallel to take full advantage of multiple CPUs.
>
> This solution saves around 2 seconds of boot time on a 1780 CPU system,
> which is around 66% improvement in the existing logic.
>
> Signed-off-by: Saurabh Sengar <sse...@linux.microsoft.com>
> ---

Reviewed-by: Srivatsa S. Bhat (Microsoft) <sriv...@csail.mit.edu>

Regards,
Srivatsa
Microsoft Linux Systems Group