[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <ADA89F4D-5789-44BF-A4C6-3003F8B95A20@linux.microsoft.com>
Date: Thu, 11 Apr 2024 16:58:56 -0700
From: Allen Pais <apais@...ux.microsoft.com>
To: Michael Kelley <mhklinux@...look.com>
Cc: "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"tj@...nel.org" <tj@...nel.org>,
"keescook@...omium.org" <keescook@...omium.org>,
"kys@...rosoft.com" <kys@...rosoft.com>,
"haiyangz@...rosoft.com" <haiyangz@...rosoft.com>,
"wei.liu@...nel.org" <wei.liu@...nel.org>,
"decui@...rosoft.com" <decui@...rosoft.com>,
"linux-hyperv@...r.kernel.org" <linux-hyperv@...r.kernel.org>
Subject: Re: [PATCH v2 1/2] hyperv: Convert from tasklet to BH workqueue
> On Apr 10, 2024, at 11:08 AM, Michael Kelley <mhklinux@...look.com> wrote:
>
> From: Allen Pais <apais@...ux.microsoft.com> Sent: Wednesday, April 3, 2024 9:56 AM
>>
>> The only generic interface to execute asynchronously in the BH context is
>> tasklet; however, it's marked deprecated and has some design flaws. To
>> replace tasklets, BH workqueue support was recently added. A BH workqueue
>> behaves similarly to regular workqueues except that the queued work items
>> are executed in the BH context.
>>
>> This patch converts drivers/hv/* from tasklet to BH workqueue.
>>
>> Based on the work done by Tejun Heo <tj@...nel.org>
>> Branch: https://git.kernel.org/pub/scm/linux/kernel/git/tj/wq.git for-6.10
>>
>> Signed-off-by: Allen Pais <allen.lkml@...il.com>
>> ---
>> drivers/hv/channel.c | 8 ++++----
>> drivers/hv/channel_mgmt.c | 5 ++---
>> drivers/hv/connection.c | 9 +++++----
>> drivers/hv/hv.c | 3 +--
>> drivers/hv/hv_balloon.c | 4 ++--
>> drivers/hv/hv_fcopy.c | 8 ++++----
>> drivers/hv/hv_kvp.c | 8 ++++----
>> drivers/hv/hv_snapshot.c | 8 ++++----
>> drivers/hv/hyperv_vmbus.h | 9 +++++----
>> drivers/hv/vmbus_drv.c | 20 +++++++++++---------
>> include/linux/hyperv.h | 2 +-
>> 11 files changed, 43 insertions(+), 41 deletions(-)
>>
>> diff --git a/drivers/hv/channel.c b/drivers/hv/channel.c
>> index adbf674355b2..876d78eb4dce 100644
>> --- a/drivers/hv/channel.c
>> +++ b/drivers/hv/channel.c
>> @@ -859,7 +859,7 @@ void vmbus_reset_channel_cb(struct vmbus_channel
>> *channel)
>> unsigned long flags;
>>
>> /*
>> - * vmbus_on_event(), running in the per-channel tasklet, can race
>> + * vmbus_on_event(), running in the per-channel work, can race
>> * with vmbus_close_internal() in the case of SMP guest, e.g., when
>> * the former is accessing channel->inbound.ring_buffer, the latter
>> * could be freeing the ring_buffer pages, so here we must stop it
>> @@ -871,7 +871,7 @@ void vmbus_reset_channel_cb(struct vmbus_channel *channel)
>> * and that the channel ring buffer is no longer being accessed, cf.
>> * the calls to napi_disable() in netvsc_device_remove().
>> */
>> - tasklet_disable(&channel->callback_event);
>> + disable_work_sync(&channel->callback_event);
>>
>> /* See the inline comments in vmbus_chan_sched(). */
>> spin_lock_irqsave(&channel->sched_lock, flags);
>> @@ -880,8 +880,8 @@ void vmbus_reset_channel_cb(struct vmbus_channel *channel)
>>
>> channel->sc_creation_callback = NULL;
>>
>> - /* Re-enable tasklet for use on re-open */
>> - tasklet_enable(&channel->callback_event);
>> + /* Re-enable work for use on re-open */
>> + enable_and_queue_work(system_bh_wq, &channel->callback_event);
>
> In this case and in several other cases in the Hyper-V related code, you've
> used enable_and_queue_work() as the replacement for tasklet_enable().
> I would have expected just enable_work() as the equivalent. tasklet_enable()
> just re-enables the tasklet; it does not do tasklet_schedule().
Thank you. I see your point. Let me update the call accordingly and send out
A new version.
>
> Doing the additional queue_work() shouldn't break anything; the work
> function will run and find nothing to do, which is benign. But it seems
> conceptually wrong to have these places in the code queueing the work
> to run.
Okay.
>
> Other than that, the code looks good to me. I can see that there's
> considerably more overhead in using a workqueue instead of a
> tasklet. Tasklets access with only per-CPU data and have no spin locks,
> whereas the workqueue code reads some global data and does
> a spin lock obtain/release on per-CPU data. I haven't done any
> perf testing, and won't be able to at least over the next week. But
> the key scenario will be to test VMs with high CPU counts and lots
> of synthetic and/or storage interrupts. I suspect the additional
> overhead won't be noticeable/measurable, but I agree with your
> initial statement that this should be checked.
I will try and grab hold of a vm with high CPU count and run some tests.
Thanks for the quick review.
- Allen
>
> Michael
>
>> }
>>
>> static int vmbus_close_internal(struct vmbus_channel *channel)
>> diff --git a/drivers/hv/channel_mgmt.c b/drivers/hv/channel_mgmt.c
>> index 2f4d09ce027a..58397071a0de 100644
>> --- a/drivers/hv/channel_mgmt.c
>> +++ b/drivers/hv/channel_mgmt.c
>> @@ -353,8 +353,7 @@ static struct vmbus_channel *alloc_channel(void)
>>
>> INIT_LIST_HEAD(&channel->sc_list);
>>
>> - tasklet_init(&channel->callback_event,
>> - vmbus_on_event, (unsigned long)channel);
>> + INIT_WORK(&channel->callback_event, vmbus_on_event);
>>
>> hv_ringbuffer_pre_init(channel);
>>
>> @@ -366,7 +365,7 @@ static struct vmbus_channel *alloc_channel(void)
>> */
>> static void free_channel(struct vmbus_channel *channel)
>> {
>> - tasklet_kill(&channel->callback_event);
>> + cancel_work_sync(&channel->callback_event);
>> vmbus_remove_channel_attr_group(channel);
>>
>> kobject_put(&channel->kobj);
>> diff --git a/drivers/hv/connection.c b/drivers/hv/connection.c
>> index 3cabeeabb1ca..f2a3394a8303 100644
>> --- a/drivers/hv/connection.c
>> +++ b/drivers/hv/connection.c
>> @@ -372,12 +372,13 @@ struct vmbus_channel *relid2channel(u32 relid)
>> * 3. Once we return, enable signaling from the host. Once this
>> * state is set we check to see if additional packets are
>> * available to read. In this case we repeat the process.
>> - * If this tasklet has been running for a long time
>> + * If this work has been running for a long time
>> * then reschedule ourselves.
>> */
>> -void vmbus_on_event(unsigned long data)
>> +void vmbus_on_event(struct work_struct *t)
>> {
>> - struct vmbus_channel *channel = (void *) data;
>> + struct vmbus_channel *channel = from_work(channel, t,
>> + callback_event);
>> void (*callback_fn)(void *context);
>>
>> trace_vmbus_on_event(channel);
>> @@ -401,7 +402,7 @@ void vmbus_on_event(unsigned long data)
>> return;
>>
>> hv_begin_read(&channel->inbound);
>> - tasklet_schedule(&channel->callback_event);
>> + queue_work(system_bh_wq, &channel->callback_event);
>> }
>>
>> /*
>> diff --git a/drivers/hv/hv.c b/drivers/hv/hv.c
>> index a8ad728354cb..2af92f08f9ce 100644
>> --- a/drivers/hv/hv.c
>> +++ b/drivers/hv/hv.c
>> @@ -119,8 +119,7 @@ int hv_synic_alloc(void)
>> for_each_present_cpu(cpu) {
>> hv_cpu = per_cpu_ptr(hv_context.cpu_context, cpu);
>>
>> - tasklet_init(&hv_cpu->msg_dpc,
>> - vmbus_on_msg_dpc, (unsigned long) hv_cpu);
>> + INIT_WORK(&hv_cpu->msg_dpc, vmbus_on_msg_dpc);
>>
>> if (ms_hyperv.paravisor_present && hv_isolation_type_tdx())
>> {
>> hv_cpu->post_msg_page = (void *)get_zeroed_page(GFP_ATOMIC);
>> diff --git a/drivers/hv/hv_balloon.c b/drivers/hv/hv_balloon.c
>> index e000fa3b9f97..c7efa2ff4cdf 100644
>> --- a/drivers/hv/hv_balloon.c
>> +++ b/drivers/hv/hv_balloon.c
>> @@ -2083,7 +2083,7 @@ static int balloon_suspend(struct hv_device *hv_dev)
>> {
>> struct hv_dynmem_device *dm = hv_get_drvdata(hv_dev);
>>
>> - tasklet_disable(&hv_dev->channel->callback_event);
>> + disable_work_sync(&hv_dev->channel->callback_event);
>>
>> cancel_work_sync(&dm->balloon_wrk.wrk);
>> cancel_work_sync(&dm->ha_wrk.wrk);
>> @@ -2094,7 +2094,7 @@ static int balloon_suspend(struct hv_device *hv_dev)
>> vmbus_close(hv_dev->channel);
>> }
>>
>> - tasklet_enable(&hv_dev->channel->callback_event);
>> + enable_and_queue_work(system_bh_wq, &hv_dev->channel->callback_event);
>>
>> return 0;
>>
>> diff --git a/drivers/hv/hv_fcopy.c b/drivers/hv/hv_fcopy.c
>> index 922d83eb7ddf..fd6799293c17 100644
>> --- a/drivers/hv/hv_fcopy.c
>> +++ b/drivers/hv/hv_fcopy.c
>> @@ -71,7 +71,7 @@ static void fcopy_poll_wrapper(void *channel)
>> {
>> /* Transaction is finished, reset the state here to avoid races. */
>> fcopy_transaction.state = HVUTIL_READY;
>> - tasklet_schedule(&((struct vmbus_channel *)channel)->callback_event);
>> + queue_work(system_bh_wq, &((struct vmbus_channel *)channel)->callback_event);
>> }
>>
>> static void fcopy_timeout_func(struct work_struct *dummy)
>> @@ -391,7 +391,7 @@ int hv_fcopy_pre_suspend(void)
>> if (!fcopy_msg)
>> return -ENOMEM;
>>
>> - tasklet_disable(&channel->callback_event);
>> + disable_work_sync(&channel->callback_event);
>>
>> fcopy_msg->operation = CANCEL_FCOPY;
>>
>> @@ -404,7 +404,7 @@ int hv_fcopy_pre_suspend(void)
>>
>> fcopy_transaction.state = HVUTIL_READY;
>>
>> - /* tasklet_enable() will be called in hv_fcopy_pre_resume(). */
>> + /* enable_and_queue_work(system_bh_wq, ) will be called in hv_fcopy_pre_resume(). */
>> return 0;
>> }
>>
>> @@ -412,7 +412,7 @@ int hv_fcopy_pre_resume(void)
>> {
>> struct vmbus_channel *channel = fcopy_transaction.recv_channel;
>>
>> - tasklet_enable(&channel->callback_event);
>> + enable_and_queue_work(system_bh_wq, &channel->callback_event);
>>
>> return 0;
>> }
>> diff --git a/drivers/hv/hv_kvp.c b/drivers/hv/hv_kvp.c
>> index d35b60c06114..85b8fb4a3d2e 100644
>> --- a/drivers/hv/hv_kvp.c
>> +++ b/drivers/hv/hv_kvp.c
>> @@ -113,7 +113,7 @@ static void kvp_poll_wrapper(void *channel)
>> {
>> /* Transaction is finished, reset the state here to avoid races. */
>> kvp_transaction.state = HVUTIL_READY;
>> - tasklet_schedule(&((struct vmbus_channel *)channel)->callback_event);
>> + queue_work(system_bh_wq, &((struct vmbus_channel *)channel)->callback_event);
>> }
>>
>> static void kvp_register_done(void)
>> @@ -160,7 +160,7 @@ static void kvp_timeout_func(struct work_struct *dummy)
>>
>> static void kvp_host_handshake_func(struct work_struct *dummy)
>> {
>> - tasklet_schedule(&kvp_transaction.recv_channel->callback_event);
>> + queue_work(system_bh_wq, &kvp_transaction.recv_channel->callback_event);
>> }
>>
>> static int kvp_handle_handshake(struct hv_kvp_msg *msg)
>> @@ -786,7 +786,7 @@ int hv_kvp_pre_suspend(void)
>> {
>> struct vmbus_channel *channel = kvp_transaction.recv_channel;
>>
>> - tasklet_disable(&channel->callback_event);
>> + disable_work_sync(&channel->callback_event);
>>
>> /*
>> * If there is a pending transtion, it's unnecessary to tell the host
>> @@ -809,7 +809,7 @@ int hv_kvp_pre_resume(void)
>> {
>> struct vmbus_channel *channel = kvp_transaction.recv_channel;
>>
>> - tasklet_enable(&channel->callback_event);
>> + enable_and_queue_work(system_bh_wq, &channel->callback_event);
>>
>> return 0;
>> }
>> diff --git a/drivers/hv/hv_snapshot.c b/drivers/hv/hv_snapshot.c
>> index 0d2184be1691..46c2263d2591 100644
>> --- a/drivers/hv/hv_snapshot.c
>> +++ b/drivers/hv/hv_snapshot.c
>> @@ -83,7 +83,7 @@ static void vss_poll_wrapper(void *channel)
>> {
>> /* Transaction is finished, reset the state here to avoid races. */
>> vss_transaction.state = HVUTIL_READY;
>> - tasklet_schedule(&((struct vmbus_channel *)channel)->callback_event);
>> + queue_work(system_bh_wq, &((struct vmbus_channel *)channel)->callback_event);
>> }
>>
>> /*
>> @@ -421,7 +421,7 @@ int hv_vss_pre_suspend(void)
>> if (!vss_msg)
>> return -ENOMEM;
>>
>> - tasklet_disable(&channel->callback_event);
>> + disable_work_sync(&channel->callback_event);
>>
>> vss_msg->vss_hdr.operation = VSS_OP_THAW;
>>
>> @@ -435,7 +435,7 @@ int hv_vss_pre_suspend(void)
>>
>> vss_transaction.state = HVUTIL_READY;
>>
>> - /* tasklet_enable() will be called in hv_vss_pre_resume(). */
>> + /* enable_and_queue_work() will be called in hv_vss_pre_resume(). */
>> return 0;
>> }
>>
>> @@ -443,7 +443,7 @@ int hv_vss_pre_resume(void)
>> {
>> struct vmbus_channel *channel = vss_transaction.recv_channel;
>>
>> - tasklet_enable(&channel->callback_event);
>> + enable_and_queue_work(system_bh_wq, &channel->callback_event);
>>
>> return 0;
>> }
>> diff --git a/drivers/hv/hyperv_vmbus.h b/drivers/hv/hyperv_vmbus.h
>> index f6b1e710f805..95ca570ac7af 100644
>> --- a/drivers/hv/hyperv_vmbus.h
>> +++ b/drivers/hv/hyperv_vmbus.h
>> @@ -19,6 +19,7 @@
>> #include <linux/atomic.h>
>> #include <linux/hyperv.h>
>> #include <linux/interrupt.h>
>> +#include <linux/workqueue.h>
>>
>> #include "hv_trace.h"
>>
>> @@ -136,10 +137,10 @@ struct hv_per_cpu_context {
>>
>> /*
>> * Starting with win8, we can take channel interrupts on any CPU;
>> - * we will manage the tasklet that handles events messages on a per CPU
>> + * we will manage the work that handles events messages on a per CPU
>> * basis.
>> */
>> - struct tasklet_struct msg_dpc;
>> + struct work_struct msg_dpc;
>> };
>>
>> struct hv_context {
>> @@ -366,8 +367,8 @@ void vmbus_disconnect(void);
>>
>> int vmbus_post_msg(void *buffer, size_t buflen, bool can_sleep);
>>
>> -void vmbus_on_event(unsigned long data);
>> -void vmbus_on_msg_dpc(unsigned long data);
>> +void vmbus_on_event(struct work_struct *t);
>> +void vmbus_on_msg_dpc(struct work_struct *t);
>>
>> int hv_kvp_init(struct hv_util_service *srv);
>> void hv_kvp_deinit(void);
>> diff --git a/drivers/hv/vmbus_drv.c b/drivers/hv/vmbus_drv.c
>> index 4cb17603a828..28490068cacc 100644
>> --- a/drivers/hv/vmbus_drv.c
>> +++ b/drivers/hv/vmbus_drv.c
>> @@ -1025,9 +1025,9 @@ static void vmbus_onmessage_work(struct work_struct *work)
>> kfree(ctx);
>> }
>>
>> -void vmbus_on_msg_dpc(unsigned long data)
>> +void vmbus_on_msg_dpc(struct work_struct *t)
>> {
>> - struct hv_per_cpu_context *hv_cpu = (void *)data;
>> + struct hv_per_cpu_context *hv_cpu = from_work(hv_cpu, t, msg_dpc);
>> void *page_addr = hv_cpu->synic_message_page;
>> struct hv_message msg_copy, *msg = (struct hv_message *)page_addr +
>> VMBUS_MESSAGE_SINT;
>> @@ -1131,7 +1131,7 @@ void vmbus_on_msg_dpc(unsigned long data)
>> * before sending the rescind message of the same
>> * channel. These messages are sent to the guest's
>> * connect CPU; the guest then starts processing them
>> - * in the tasklet handler on this CPU:
>> + * in the work handler on this CPU:
>> *
>> * VMBUS_CONNECT_CPU
>> *
>> @@ -1276,7 +1276,7 @@ static void vmbus_chan_sched(struct hv_per_cpu_context *hv_cpu)
>> hv_begin_read(&channel->inbound);
>> fallthrough;
>> case HV_CALL_DIRECT:
>> - tasklet_schedule(&channel->callback_event);
>> + queue_work(system_bh_wq, &channel->callback_event);
>> }
>>
>> sched_unlock:
>> @@ -1304,7 +1304,7 @@ static void vmbus_isr(void)
>> hv_stimer0_isr();
>> vmbus_signal_eom(msg, HVMSG_TIMER_EXPIRED);
>> } else
>> - tasklet_schedule(&hv_cpu->msg_dpc);
>> + queue_work(system_bh_wq, &hv_cpu->msg_dpc);
>> }
>>
>> add_interrupt_randomness(vmbus_interrupt);
>> @@ -2371,10 +2371,12 @@ static int vmbus_bus_suspend(struct device *dev)
>> hv_context.cpu_context, VMBUS_CONNECT_CPU);
>> struct vmbus_channel *channel, *sc;
>>
>> - tasklet_disable(&hv_cpu->msg_dpc);
>> + disable_work_sync(&hv_cpu->msg_dpc);
>> vmbus_connection.ignore_any_offer_msg = true;
>> - /* The tasklet_enable() takes care of providing a memory barrier */
>> - tasklet_enable(&hv_cpu->msg_dpc);
>> + /* The enable_and_queue_work() takes care of
>> + * providing a memory barrier
>> + */
>> + enable_and_queue_work(system_bh_wq, &hv_cpu->msg_dpc);
>>
>> /* Drain all the workqueues as we are in suspend */
>> drain_workqueue(vmbus_connection.rescind_work_queue);
>> @@ -2692,7 +2694,7 @@ static void __exit vmbus_exit(void)
>> struct hv_per_cpu_context *hv_cpu
>> = per_cpu_ptr(hv_context.cpu_context, cpu);
>>
>> - tasklet_kill(&hv_cpu->msg_dpc);
>> + cancel_work_sync(&hv_cpu->msg_dpc);
>> }
>> hv_debug_rm_all_dir();
>>
>> diff --git a/include/linux/hyperv.h b/include/linux/hyperv.h
>> index 6ef0557b4bff..db3d85ea5ce6 100644
>> --- a/include/linux/hyperv.h
>> +++ b/include/linux/hyperv.h
>> @@ -882,7 +882,7 @@ struct vmbus_channel {
>> bool out_full_flag;
>>
>> /* Channel callback's invoked in softirq context */
>> - struct tasklet_struct callback_event;
>> + struct work_struct callback_event;
>> void (*onchannel_callback)(void *context);
>> void *channel_callback_context;
>>
>> --
>> 2.17.1
>>
Powered by blists - more mailing lists