[PATCH v5 1/2] drivers/hv: introduce vmbus_channel_set_cpu()

[PATCH v5 1/2] drivers/hv: introduce vmbus_channel_set_cpu()

[Hamza Mahfooz]

The core functionality in target_cpu_store() is also needed in a
subsequent patch for automatically changing the CPU when taking
a CPU offline. As such, factor out the body of target_cpu_store()
into new function vmbus_channel_set_cpu() that can also be used
elsewhere.

No functional change is intended.

Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Michael Kelley <mhklinux@outlook.com>
Cc: Wei Liu <wei.liu@kernel.org>
Signed-off-by: Hamza Mahfooz <hamzamahfooz@linux.microsoft.com>
---
v2: separate vmbus_channel_set_cpu() changes from
    cpu offlining changes.

v3: address comments from Michael.
---
 drivers/hv/vmbus_drv.c | 50 +++++++++++++++++++++++++-----------------
 include/linux/hyperv.h |  1 +
 2 files changed, 31 insertions(+), 20 deletions(-)

diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
index 2892b8da20a5..0ca0e85e6edd 100644
--- a/drivers/hv/vmbus_drv.c
+++ b/drivers/hv/vmbus_drv.c
@@ -1611,16 +1611,16 @@ static ssize_t target_cpu_show(struct vmbus_channel *channel, char *buf)
 {
 	return sprintf(buf, "%u\n", channel->target_cpu);
 }
-static ssize_t target_cpu_store(struct vmbus_channel *channel,
-				const char *buf, size_t count)
+
+int vmbus_channel_set_cpu(struct vmbus_channel *channel, u32 target_cpu)
 {
-	u32 target_cpu, origin_cpu;
-	ssize_t ret = count;
+	u32 origin_cpu;
+	int ret = 0;
 
-	if (vmbus_proto_version < VERSION_WIN10_V4_1)
-		return -EIO;
+	lockdep_assert_cpus_held();
+	lockdep_assert_held(&vmbus_connection.channel_mutex);
 
-	if (sscanf(buf, "%uu", &target_cpu) != 1)
+	if (vmbus_proto_version < VERSION_WIN10_V4_1)
 		return -EIO;
 
 	/* Validate target_cpu for the cpumask_test_cpu() operation below. */
@@ -1630,22 +1630,17 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,
 	if (!cpumask_test_cpu(target_cpu, housekeeping_cpumask(HK_TYPE_MANAGED_IRQ)))
 		return -EINVAL;
 
-	/* No CPUs should come up or down during this. */
-	cpus_read_lock();
-
-	if (!cpu_online(target_cpu)) {
-		cpus_read_unlock();
+	if (!cpu_online(target_cpu))
 		return -EINVAL;
-	}
 
 	/*
-	 * Synchronizes target_cpu_store() and channel closure:
+	 * Synchronizes vmbus_channel_set_cpu() and channel closure:
 	 *
 	 * { Initially: state = CHANNEL_OPENED }
 	 *
 	 * CPU1				CPU2
 	 *
-	 * [target_cpu_store()]		[vmbus_disconnect_ring()]
+	 * [vmbus_channel_set_cpu()]	[vmbus_disconnect_ring()]
 	 *
 	 * LOCK channel_mutex		LOCK channel_mutex
 	 * LOAD r1 = state		LOAD r2 = state
@@ -1660,7 +1655,6 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,
 	 * Note.  The host processes the channel messages "sequentially", in
 	 * the order in which they are received on a per-partition basis.
 	 */
-	mutex_lock(&vmbus_connection.channel_mutex);
 
 	/*
 	 * Hyper-V will ignore MODIFYCHANNEL messages for "non-open" channels;
@@ -1668,17 +1662,17 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,
 	 */
 	if (channel->state != CHANNEL_OPENED_STATE) {
 		ret = -EIO;
-		goto cpu_store_unlock;
+		goto end;
 	}
 
 	origin_cpu = channel->target_cpu;
 	if (target_cpu == origin_cpu)
-		goto cpu_store_unlock;
+		goto end;
 
 	if (vmbus_send_modifychannel(channel,
 				     hv_cpu_number_to_vp_number(target_cpu))) {
 		ret = -EIO;
-		goto cpu_store_unlock;
+		goto end;
 	}
 
 	/*
@@ -1708,9 +1702,25 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,
 				origin_cpu, target_cpu);
 	}
 
-cpu_store_unlock:
+end:
+	return ret;
+}
+
+static ssize_t target_cpu_store(struct vmbus_channel *channel,
+				const char *buf, size_t count)
+{
+	ssize_t ret = count;
+	u32 target_cpu;
+
+	if (sscanf(buf, "%uu", &target_cpu) != 1)
+		return -EIO;
+
+	cpus_read_lock();
+	mutex_lock(&vmbus_connection.channel_mutex);
+	ret = vmbus_channel_set_cpu(channel, target_cpu);
 	mutex_unlock(&vmbus_connection.channel_mutex);
 	cpus_read_unlock();
+
 	return ret;
 }
 static VMBUS_CHAN_ATTR(cpu, 0644, target_cpu_show, target_cpu_store);
diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
index 02a226bcf0ed..25e9e982f1b0 100644
--- a/include/linux/hyperv.h
+++ b/include/linux/hyperv.h
@@ -1670,6 +1670,7 @@ int vmbus_send_tl_connect_request(const guid_t *shv_guest_servie_id,
 				  const guid_t *shv_host_servie_id);
 int vmbus_send_modifychannel(struct vmbus_channel *channel, u32 target_vp);
 void vmbus_set_event(struct vmbus_channel *channel);
+int vmbus_channel_set_cpu(struct vmbus_channel *channel, u32 target_cpu);
 
 /* Get the start of the ring buffer. */
 static inline void *
-- 
2.47.1

[PATCH v5 2/2] drivers/hv: add CPU offlining support

[Hamza Mahfooz]

Currently, it is tedious to offline CPUs in a Hyper-V VM since CPUs may
have VMBus channels attached to them that a user would have to manually
rebind elsewhere. So, as made mention of in
commit d570aec0f2154 ("Drivers: hv: vmbus: Synchronize init_vp_index()
vs. CPU hotplug"), rebind channels associated with CPUs that a user is
trying to offline to a new "randomly" selected CPU.

Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Michael Kelley <mhklinux@outlook.com>
Cc: Wei Liu <wei.liu@kernel.org>
Signed-off-by: Hamza Mahfooz <hamzamahfooz@linux.microsoft.com>
---
v2: remove cpus_read_{un,}lock() from hv_pick_new_cpu() and add
    lockdep_assert_cpus_held().

v3: use for_each_cpu_wrap() in hv_pick_new_cpu().

v4: store get_random_u32_below() in start.

v5: style fixes, use target_cpu and set ret to -EBUSY by default.
---
 drivers/hv/hv.c | 72 ++++++++++++++++++++++++++++++++++---------------
 1 file changed, 51 insertions(+), 21 deletions(-)

diff --git a/drivers/hv/hv.c b/drivers/hv/hv.c
index 36d9ba097ff5..fab0690b5c41 100644
--- a/drivers/hv/hv.c
+++ b/drivers/hv/hv.c
@@ -433,13 +433,47 @@ static bool hv_synic_event_pending(void)
 	return pending;
 }
 
+static int hv_pick_new_cpu(struct vmbus_channel *channel)
+{
+	int ret = -EBUSY;
+	int start;
+	int cpu;
+
+	lockdep_assert_cpus_held();
+	lockdep_assert_held(&vmbus_connection.channel_mutex);
+
+	/*
+	 * We can't assume that the relevant interrupts will be sent before
+	 * the cpu is offlined on older versions of hyperv.
+	 */
+	if (vmbus_proto_version < VERSION_WIN10_V5_3)
+		return -EBUSY;
+
+	start = get_random_u32_below(nr_cpu_ids);
+
+	for_each_cpu_wrap(cpu, cpu_online_mask, start) {
+		if (channel->target_cpu == cpu ||
+		    channel->target_cpu == VMBUS_CONNECT_CPU)
+			continue;
+
+		ret = vmbus_channel_set_cpu(channel, cpu);
+		if (!ret)
+			break;
+	}
+
+	if (ret)
+		ret = vmbus_channel_set_cpu(channel, VMBUS_CONNECT_CPU);
+
+	return ret;
+}
+
 /*
  * hv_synic_cleanup - Cleanup routine for hv_synic_init().
  */
 int hv_synic_cleanup(unsigned int cpu)
 {
 	struct vmbus_channel *channel, *sc;
-	bool channel_found = false;
+	int ret = 0;
 
 	if (vmbus_connection.conn_state != CONNECTED)
 		goto always_cleanup;
@@ -456,38 +490,34 @@ int hv_synic_cleanup(unsigned int cpu)
 
 	/*
 	 * Search for channels which are bound to the CPU we're about to
-	 * cleanup.  In case we find one and vmbus is still connected, we
-	 * fail; this will effectively prevent CPU offlining.
-	 *
-	 * TODO: Re-bind the channels to different CPUs.
+	 * cleanup.
 	 */
 	mutex_lock(&vmbus_connection.channel_mutex);
 	list_for_each_entry(channel, &vmbus_connection.chn_list, listentry) {
 		if (channel->target_cpu == cpu) {
-			channel_found = true;
-			break;
+			ret = hv_pick_new_cpu(channel);
+			if (ret) {
+				mutex_unlock(&vmbus_connection.channel_mutex);
+				return ret;
+			}
 		}
 		list_for_each_entry(sc, &channel->sc_list, sc_list) {
 			if (sc->target_cpu == cpu) {
-				channel_found = true;
-				break;
+				ret = hv_pick_new_cpu(sc);
+				if (ret) {
+					mutex_unlock(&vmbus_connection.channel_mutex);
+					return ret;
+				}
 			}
 		}
-		if (channel_found)
-			break;
 	}
 	mutex_unlock(&vmbus_connection.channel_mutex);
 
-	if (channel_found)
-		return -EBUSY;
-
 	/*
-	 * channel_found == false means that any channels that were previously
-	 * assigned to the CPU have been reassigned elsewhere with a call of
-	 * vmbus_send_modifychannel().  Scan the event flags page looking for
-	 * bits that are set and waiting with a timeout for vmbus_chan_sched()
-	 * to process such bits.  If bits are still set after this operation
-	 * and VMBus is connected, fail the CPU offlining operation.
+	 * Scan the event flags page looking for bits that are set and waiting
+	 * with a timeout for vmbus_chan_sched() to process such bits. If bits
+	 * are still set after this operation and VMBus is connected, fail the
+	 * CPU offlining operation.
 	 */
 	if (vmbus_proto_version >= VERSION_WIN10_V4_1 && hv_synic_event_pending())
 		return -EBUSY;
@@ -497,5 +527,5 @@ int hv_synic_cleanup(unsigned int cpu)
 
 	hv_synic_disable_regs(cpu);
 
-	return 0;
+	return ret;
 }
-- 
2.47.1

RE: [PATCH v5 1/2] drivers/hv: introduce vmbus_channel_set_cpu()

[Michael Kelley]

From: Hamza Mahfooz <hamzamahfooz@linux.microsoft.com> Sent: Thursday, January 16, 2025 3:07 PM
> 
> The core functionality in target_cpu_store() is also needed in a
> subsequent patch for automatically changing the CPU when taking
> a CPU offline. As such, factor out the body of target_cpu_store()
> into new function vmbus_channel_set_cpu() that can also be used
> elsewhere.
> 
> No functional change is intended.
> 
> Cc: Boqun Feng <boqun.feng@gmail.com>
> Cc: Michael Kelley <mhklinux@outlook.com>
> Cc: Wei Liu <wei.liu@kernel.org>
> Signed-off-by: Hamza Mahfooz <hamzamahfooz@linux.microsoft.com>
> ---
> v2: separate vmbus_channel_set_cpu() changes from
>     cpu offlining changes.
> 
> v3: address comments from Michael.
> ---
>  drivers/hv/vmbus_drv.c | 50 +++++++++++++++++++++++++-----------------
>  include/linux/hyperv.h |  1 +
>  2 files changed, 31 insertions(+), 20 deletions(-)
> 
> diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
> index 2892b8da20a5..0ca0e85e6edd 100644
> --- a/drivers/hv/vmbus_drv.c
> +++ b/drivers/hv/vmbus_drv.c
> @@ -1611,16 +1611,16 @@ static ssize_t target_cpu_show(struct vmbus_channel *channel, char *buf)
>  {
>  	return sprintf(buf, "%u\n", channel->target_cpu);
>  }
> -static ssize_t target_cpu_store(struct vmbus_channel *channel,
> -				const char *buf, size_t count)
> +
> +int vmbus_channel_set_cpu(struct vmbus_channel *channel, u32 target_cpu)
>  {
> -	u32 target_cpu, origin_cpu;
> -	ssize_t ret = count;
> +	u32 origin_cpu;
> +	int ret = 0;
> 
> -	if (vmbus_proto_version < VERSION_WIN10_V4_1)
> -		return -EIO;
> +	lockdep_assert_cpus_held();

With the previous bugs fixed, I applied both patches in this series
and did a test build. Unfortunately, lockdep_assert_cpus_held() is
not exported, and the build fails if CONFIG_HYPERV=m. 

cpus_read_lock(), cpus_read_unlock(), and cpus_read_trylock() are
exported, but lockdep_assert_cpus_held() is not. I don't know why
they are inconsistent in that regard. You'll need to drop the use of
lockdep_assert_cpus_held() here and in hv_pick_new_cpu(), or
add a patch to export lockdep_assert_cpus_held(), assuming it's
valid to do so.

With any Hyper-V related changes, it's necessary to build and
test with CONFIG_HYPERV=m to catch problems like this.

> +	lockdep_assert_held(&vmbus_connection.channel_mutex);
> 
> -	if (sscanf(buf, "%uu", &target_cpu) != 1)
> +	if (vmbus_proto_version < VERSION_WIN10_V4_1)
>  		return -EIO;
> 
>  	/* Validate target_cpu for the cpumask_test_cpu() operation below. */
> @@ -1630,22 +1630,17 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,
>  	if (!cpumask_test_cpu(target_cpu, housekeeping_cpumask(HK_TYPE_MANAGED_IRQ)))
>  		return -EINVAL;
> 
> -	/* No CPUs should come up or down during this. */
> -	cpus_read_lock();
> -
> -	if (!cpu_online(target_cpu)) {
> -		cpus_read_unlock();
> +	if (!cpu_online(target_cpu))
>  		return -EINVAL;
> -	}
> 
>  	/*
> -	 * Synchronizes target_cpu_store() and channel closure:
> +	 * Synchronizes vmbus_channel_set_cpu() and channel closure:
>  	 *
>  	 * { Initially: state = CHANNEL_OPENED }
>  	 *
>  	 * CPU1				CPU2
>  	 *
> -	 * [target_cpu_store()]		[vmbus_disconnect_ring()]
> +	 * [vmbus_channel_set_cpu()]	[vmbus_disconnect_ring()]
>  	 *
>  	 * LOCK channel_mutex		LOCK channel_mutex
>  	 * LOAD r1 = state		LOAD r2 = state
> @@ -1660,7 +1655,6 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,
>  	 * Note.  The host processes the channel messages "sequentially", in
>  	 * the order in which they are received on a per-partition basis.
>  	 */
> -	mutex_lock(&vmbus_connection.channel_mutex);
> 
>  	/*
>  	 * Hyper-V will ignore MODIFYCHANNEL messages for "non-open" channels;
> @@ -1668,17 +1662,17 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,
>  	 */
>  	if (channel->state != CHANNEL_OPENED_STATE) {
>  		ret = -EIO;
> -		goto cpu_store_unlock;
> +		goto end;
>  	}
> 
>  	origin_cpu = channel->target_cpu;
>  	if (target_cpu == origin_cpu)
> -		goto cpu_store_unlock;
> +		goto end;
> 
>  	if (vmbus_send_modifychannel(channel,
>  				     hv_cpu_number_to_vp_number(target_cpu))) {
>  		ret = -EIO;
> -		goto cpu_store_unlock;
> +		goto end;
>  	}
> 
>  	/*
> @@ -1708,9 +1702,25 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,
>  				origin_cpu, target_cpu);
>  	}
> 
> -cpu_store_unlock:
> +end:
> +	return ret;
> +}
> +
> +static ssize_t target_cpu_store(struct vmbus_channel *channel,
> +				const char *buf, size_t count)
> +{
> +	ssize_t ret = count;
> +	u32 target_cpu;
> +
> +	if (sscanf(buf, "%uu", &target_cpu) != 1)
> +		return -EIO;
> +
> +	cpus_read_lock();
> +	mutex_lock(&vmbus_connection.channel_mutex);
> +	ret = vmbus_channel_set_cpu(channel, target_cpu);

There's a problem here that I didn't previously see. Like all other
functions for setting a sysfs attribute, this function should return
the number of bytes processed, or a negative errno value. That's
why "ret" is initialized to "count". But vmbus_channel_set_cpu()
sets "ret" to zero if it succeeds, and the wrong value will be
returned. The original code correctly returned "count".

This problem would have been evident with a simple test of
writing to /sys/bus/vmbus/devices/<guid>/channels/<nn>/cpu.

Michael

>  	mutex_unlock(&vmbus_connection.channel_mutex);
>  	cpus_read_unlock();
> +
>  	return ret;
>  }
>  static VMBUS_CHAN_ATTR(cpu, 0644, target_cpu_show, target_cpu_store);
> diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
> index 02a226bcf0ed..25e9e982f1b0 100644
> --- a/include/linux/hyperv.h
> +++ b/include/linux/hyperv.h
> @@ -1670,6 +1670,7 @@ int vmbus_send_tl_connect_request(const guid_t
> *shv_guest_servie_id,
>  				  const guid_t *shv_host_servie_id);
>  int vmbus_send_modifychannel(struct vmbus_channel *channel, u32 target_vp);
>  void vmbus_set_event(struct vmbus_channel *channel);
> +int vmbus_channel_set_cpu(struct vmbus_channel *channel, u32 target_cpu);
> 
>  /* Get the start of the ring buffer. */
>  static inline void *
> --
> 2.47.1