Drivers: hv: vmbus: Miscellaneous cleanups
Subsystems (3)
cover Drivers: hv: vmbus: Miscellaneous cleanups
Hi all,
I went back to my "cleanup list" recently and I wrote these patches:
here you can find, among other things,
1) the removal of the fields 'target_vp' and 'numa_node' from the
channel data structure, as suggested by Michael back in May;
2) various cleanups for channel->lock, which is actually *removed by the end of this series! ;-)
I'm sure there is room for further "cleanups", ;-) but let me check if these (relatively small) changes make sense first...
Thanks,
Andrea
Andrea Parri (Microsoft) (8):
Drivers: hv: vmbus: Remove the target_vp field from the vmbus_channel
struct
Drivers: hv: vmbus: Remove the numa_node field from the vmbus_channel
struct
Drivers: hv: vmbus: Replace cpumask_test_cpu(, cpu_online_mask) with
cpu_online()
Drivers: hv: vmbus: Remove unnecessary channel->lock critical sections
(sc_list readers)
Drivers: hv: vmbus: Use channel_mutex in channel_vp_mapping_show()
Drivers: hv: vmbus: Remove unnecessary channel->lock critical sections
(sc_list updaters)
scsi: storvsc: Introduce the per-storvsc_device spinlock
Drivers: hv: vmbus: Remove the lock field from the vmbus_channel
struct
drivers/hv/channel.c | 9 +++------ drivers/hv/channel_mgmt.c | 31 ++++++------------------------- drivers/hv/hv.c | 3 --- drivers/hv/vmbus_drv.c | 17 +++++------------ drivers/scsi/storvsc_drv.c | 16 +++++++++++----- include/linux/hyperv.h | 22 +++++++--------------- 6 files changed, 32 insertions(+), 66 deletions(-)
Signature · 1 line
2.25.1
PATCH 1/8 Drivers: hv: vmbus: Remove the target_vp field from the vmbus_channel struct
The field is read only in __vmbus_open() and it is already stored twice (after a call to hv_cpu_number_to_vp_number()) in target_cpu_store() and init_vp_index(); there is no need to "cache" its value in the channel data structure.
--- drivers/hv/channel.c | 3 ++- drivers/hv/channel_mgmt.c | 3 --- drivers/hv/vmbus_drv.c | 2 -- include/linux/hyperv.h | 15 +++++++-------- 4 files changed, 9 insertions(+), 14 deletions(-)
drivers/hv/channel.c +2 -1
@@ -18,6 +18,7 @@#include <linux/uio.h> #include <linux/interrupt.h> #include <asm/page.h> +#include <asm/mshyperv.h> #include "hyperv_vmbus.h"
@@ -176,7 +177,7 @@ static int __vmbus_open(struct vmbus_channel *newchannel, open_msg->child_relid = newchannel->offermsg.child_relid; open_msg->ringbuffer_gpadlhandle = newchannel->ringbuffer_gpadlhandle; open_msg->downstream_ringbuffer_pageoffset = newchannel->ringbuffer_send_offset; - open_msg->target_vp = newchannel->target_vp; + open_msg->target_vp =hv_cpu_number_to_vp_number (newchannel->target_cpu);if (userdatalen)memcpy (open_msg->userdata, userdata, userdatalen);
drivers/hv/channel_mgmt.c +0 -3
@@ -704,8 +704,6 @@ static void init_vp_index(struct vmbus_channel *channel)*/ channel->numa_node =cpu_to_node (VMBUS_CONNECT_CPU ); channel->target_cpu =VMBUS_CONNECT_CPU ; - channel->target_vp = -hv_cpu_number_to_vp_number (VMBUS_CONNECT_CPU );if (perf_chn)hv_set_alloced_cpu (VMBUS_CONNECT_CPU );return ;
@@ -739,7 +737,6 @@ static void init_vp_index(struct vmbus_channel *channel)cpumask_set_cpu (target_cpu, alloced_mask); channel->target_cpu = target_cpu; - channel->target_vp =hv_cpu_number_to_vp_number (target_cpu);free_cpumask_var (available_mask); }
drivers/hv/vmbus_drv.c +0 -2
@@ -23,7 +23,6 @@#include <linux/cpu.h> #include <linux/sched/task_stack.h> -#include <asm/mshyperv.h> #include <linux/delay.h> #include <linux/notifier.h> #include <linux/ptrace.h>
@@ -1765,7 +1764,6 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,*/ channel->target_cpu = target_cpu; - channel->target_vp =hv_cpu_number_to_vp_number (target_cpu); channel->numa_node =cpu_to_node (target_cpu);/* See init_vp_index(). */
include/linux/hyperv.h +7 -8
@@ -803,15 +803,14 @@ struct vmbus_channel {u64 sig_event;/* -* Starting with win8, this field will be used to specify -* the target virtual processor on which to deliver the interrupt for -* the host to guest communication. -* Prior to win8, incoming channel interrupts would only -* be delivered on cpu 0. Setting this value to 0 would -* preserve the earlier behavior. +* Starting with win8, this field will be used to specify the +* target CPU on which to deliver the interrupt for the host +* to guest communication. +* +* Prior to win8, incoming channel interrupts would only be +* delivered on CPU 0. Setting this value to 0 would preserve +* the earlier behavior. */ -u32 target_vp; -/* The corresponding CPUID in the guest */ u32 target_cpu;int numa_node;/*
Signature · 1 line
2.25.1
RE: [PATCH 1/8] Drivers: hv: vmbus: Remove the target_vp field from the vmbus_channel struct
From: Andrea Parri (Microsoft) <[email protected]> Sent: Wednesday, June 17, 2020 9:47 AM
Quoted 6 lines
The field is read only in
__vmbus_open()and it is already stored twice (after a call tohv_cpu_number_to_vp_number()) intarget_cpu_store()andinit_vp_index(); there is no need to "cache" its value in the channel data structure.
Quoted 8 lines
Suggested-by Michael Kelley <[email protected]>Signed-off-by Andrea Parri (Microsoft) <[email protected]>--- drivers/hv/channel.c | 3 ++- drivers/hv/channel_mgmt.c | 3 --- drivers/hv/vmbus_drv.c | 2 -- include/linux/hyperv.h | 15 +++++++-------- 4 files changed, 9 insertions(+), 14 deletions(-)
PATCH 2/8 Drivers: hv: vmbus: Remove the numa_node field from the vmbus_channel struct
In reply to Andrea Parri (Microsoft)The field is read only in numa_node_show() and it is already stored twice (after a call to cpu_to_node()) in target_cpu_store() and init_vp_index(); there is no need to "cache" its value in the channel data structure.
--- drivers/hv/channel_mgmt.c | 2 -- drivers/hv/vmbus_drv.c | 3 +-- include/linux/hyperv.h | 1 - 3 files changed, 1 insertion(+), 5 deletions(-)
drivers/hv/channel_mgmt.c +0 -2
@@ -702,7 +702,6 @@ static void init_vp_index(struct vmbus_channel *channel)* In case alloc_cpumask_var() fails, bind it to * VMBUS_CONNECT_CPU. */ - channel->numa_node =cpu_to_node (VMBUS_CONNECT_CPU ); channel->target_cpu =VMBUS_CONNECT_CPU ;if (perf_chn)hv_set_alloced_cpu (VMBUS_CONNECT_CPU );
@@ -719,7 +718,6 @@ static void init_vp_index(struct vmbus_channel *channel)continue ;break ; } - channel->numa_node = numa_node; alloced_mask = &hv_context.hv_numa_map[numa_node]; if (cpumask_weight(alloced_mask) ==
drivers/hv/vmbus_drv.c +1 -2
@@ -226,7 +226,7 @@ static ssize_t numa_node_show(struct device *dev,if (!hv_dev->channel)return -ENODEV ; - return sprintf(buf, "%d\n", hv_dev->channel->numa_node); + return sprintf(buf, "%d\n", cpu_to_node(hv_dev->channel->target_cpu)); }static DEVICE_ATTR_RO (numa_node);#endif
@@ -1764,7 +1764,6 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,*/ channel->target_cpu = target_cpu; - channel->numa_node =cpu_to_node (target_cpu);/* See init_vp_index(). */ if (hv_is_perf_channel (channel))
include/linux/hyperv.h +0 -1
@@ -812,7 +812,6 @@ struct vmbus_channel {* the earlier behavior. */ u32 target_cpu; -int numa_node;/* * Support for sub-channels. For high performance devices, * it will be useful to have multiple sub-channels to support
Signature · 1 line
2.25.1
RE: [PATCH 2/8] Drivers: hv: vmbus: Remove the numa_node field from the vmbus_channel struct
From: Andrea Parri (Microsoft) <[email protected]> Sent: Wednesday, June 17, 2020 9:47 AM
Quoted 11 lines
The field is read only in
numa_node_show()and it is already stored twice (after a call tocpu_to_node()) intarget_cpu_store()andinit_vp_index(); there is no need to "cache" its value in the channel data structure.Signed-off-by Andrea Parri (Microsoft) <[email protected]>--- drivers/hv/channel_mgmt.c | 2 -- drivers/hv/vmbus_drv.c | 3 +-- include/linux/hyperv.h | 1 - 3 files changed, 1 insertion(+), 5 deletions(-)
PATCH 3/8 Drivers: hv: vmbus: Replace cpumask_test_cpu(, cpu_online_mask) with cpu_online()
In reply to Andrea Parri (Microsoft)A slight improvement in readability, and this does also remove one memory access when NR_CPUS == 1! ;-)
--- drivers/hv/vmbus_drv.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/hv/vmbus_drv.c +1 -1
@@ -1702,7 +1702,7 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,/* No CPUs should come up or down during this. */ cpus_read_lock (); -if (!cpumask_test_cpu (target_cpu, cpu_online_mask)) { +if (!cpu_online (target_cpu)) {cpus_read_unlock ();return -EINVAL ; }
Signature · 1 line
2.25.1
RE: [PATCH 3/8] Drivers: hv: vmbus: Replace cpumask_test_cpu(, cpu_online_mask) with cpu_online()
From: Andrea Parri (Microsoft) <[email protected]> Sent: Wednesday, June 17, 2020 9:47 AM
Quoted 23 lines
A slight improvement in readability, and this does also remove one memory access when
NR_CPUS== 1! ;-)Signed-off-by Andrea Parri (Microsoft) <[email protected]>--- drivers/hv/vmbus_drv.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)drivers/hv/vmbus_drv.c +1
-2@@ -1702,7 +1702,7 @@ static ssize_t target_cpu_store(struct vmbus_channel *channel,/* No CPUs should come up or down during this. */ cpus_read_lock (); -if (!cpumask_test_cpu (target_cpu, cpu_online_mask)) { +if (!cpu_online (target_cpu)) {cpus_read_unlock ();return -EINVAL ; } --
Quoted 1 line
2.25.1
PATCH 4/8 Drivers: hv: vmbus: Remove unnecessary channel->lock critical sections (sc_list readers)
In reply to Andrea Parri (Microsoft)Additions/deletions to/from sc_list (as well as modifications of target_cpu(s)) are protected by channel_mutex, which hv_synic_cleanup() and vmbus_bus_suspend() own for the duration of the channel->lock critical section in question.
--- drivers/hv/hv.c | 3 --- drivers/hv/vmbus_drv.c | 3 --- 2 files changed, 6 deletions(-)
drivers/hv/hv.c +0 -3
@@ -245,7 +245,6 @@ int hv_synic_cleanup(unsigned int cpu) {struct vmbus_channel *channel, *sc;bool channel_found = false; -unsigned long flags;/* * Hyper-V does not provide a way to change the connect CPU once
@@ -267,14 +266,12 @@ int hv_synic_cleanup(unsigned int cpu) channel_found = true;break ; } -spin_lock_irqsave (&channel->lock, flags);list_for_each_entry (sc, &channel->sc_list, sc_list) {if (sc->target_cpu == cpu) { channel_found = true;break ; } } -spin_unlock_irqrestore (&channel->lock, flags);if (channel_found)break ; }
drivers/hv/vmbus_drv.c +0 -3
@@ -2330,7 +2330,6 @@ static int vmbus_acpi_add(struct acpi_device *device)static int vmbus_bus_suspend (struct device *dev) {struct vmbus_channel *channel, *sc; -unsigned long flags;while (atomic_read (&vmbus_connection.offer_in_progress) !=0 ) {/*
@@ -2388,12 +2387,10 @@ static int vmbus_bus_suspend(struct device *dev)continue ; } -spin_lock_irqsave (&channel->lock, flags);list_for_each_entry (sc, &channel->sc_list, sc_list) {pr_err ("Sub-channel not deleted!\n" );WARN_ON_ONCE (1 ); } -spin_unlock_irqrestore (&channel->lock, flags);atomic_inc (&vmbus_connection.nr_chan_fixup_on_resume); }
Signature · 1 line
2.25.1
RE: [PATCH 4/8] Drivers: hv: vmbus: Remove unnecessary channel->lock critical sections (sc_list readers)
From: Andrea Parri (Microsoft) <[email protected]> Sent: Wednesday, June 17, 2020 9:47 AM
Quoted 6 lines
Additions/deletions to/from
sc_list(as well as modifications oftarget_cpu(s)) are protected bychannel_mutex, whichhv_synic_cleanup()andvmbus_bus_suspend()own for the duration of thechannel->lockcritical section in question.
Quoted 6 lines
Signed-off-by Andrea Parri (Microsoft) <[email protected]>--- drivers/hv/hv.c | 3 --- drivers/hv/vmbus_drv.c | 3 --- 2 files changed, 6 deletions(-)
PATCH 5/8 Drivers: hv: vmbus: Use channel_mutex in channel_vp_mapping_show()
In reply to Andrea Parri (Microsoft)The primitive currently uses channel->lock to protect the loop over sc_list w.r.t. list additions/deletions but it doesn't protect the target_cpu(s) loads w.r.t. a concurrent target_cpu_store(): while the data races on target_cpu are hardly of any concern here, replace the channel->lock critical section with a channel_mutex critical section and extend the latter to include the loads of target_cpu; this same pattern is also used in hv_synic_cleanup().
--- drivers/hv/vmbus_drv.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
drivers/hv/vmbus_drv.c +3 -4
@@ -507,18 +507,17 @@ static ssize_t channel_vp_mapping_show(struct device *dev, {struct hv_device *hv_dev =device_to_hv_device (dev);struct vmbus_channel *channel = hv_dev->channel, *cur_sc; -unsigned long flags;int buf_size =PAGE_SIZE , n_written, tot_written;struct list_head *cur;if (!channel)return -ENODEV ; +mutex_lock (&vmbus_connection.channel_mutex); + tot_written =snprintf (buf, buf_size,"%u:%u\n" , channel->offermsg.child_relid, channel->target_cpu); -spin_lock_irqsave (&channel->lock, flags); -list_for_each (cur, &channel->sc_list) {if (tot_written >= buf_size -1 )break ;
@@ -532,7 +531,7 @@ static ssize_t channel_vp_mapping_show(struct device *dev, tot_written += n_written; } -spin_unlock_irqrestore (&channel->lock, flags); +mutex_unlock (&vmbus_connection.channel_mutex);return tot_written; }
Signature · 1 line
2.25.1
RE: [PATCH 5/8] Drivers: hv: vmbus: Use channel_mutex in channel_vp_mapping_show()
From: Andrea Parri (Microsoft) <[email protected]> Sent: Wednesday, June 17, 2020 9:47 AM
Quoted 9 lines
The primitive currently uses
channel->lockto protect the loop oversc_listw.r.t. list additions/deletions but it doesn't protect thetarget_cpu(s)loads w.r.t. a concurrenttarget_cpu_store(): while the data races ontarget_cpuare hardly of any concern here, replace thechannel->lockcritical section with achannel_mutexcritical section and extend the latter to include the loads oftarget_cpu; this same pattern is also used inhv_synic_cleanup().
Quoted 4 lines
Signed-off-by Andrea Parri (Microsoft) <[email protected]>--- drivers/hv/vmbus_drv.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)
PATCH 6/8 Drivers: hv: vmbus: Remove unnecessary channel->lock critical sections (sc_list updaters)
In reply to Andrea Parri (Microsoft)None of the readers/updaters of sc_list rely on channel->lock for synchronization.
--- drivers/hv/channel_mgmt.c | 25 ++++++------------------- 1 file changed, 6 insertions(+), 19 deletions(-)
drivers/hv/channel_mgmt.c +6 -19
@@ -400,8 +400,6 @@ static void vmbus_release_relid(u32 relid)void hv_process_channel_removal (struct vmbus_channel *channel) { -unsigned long flags; -lockdep_assert_held (&vmbus_connection.channel_mutex);BUG_ON (!channel->rescind);
@@ -422,14 +420,10 @@ void hv_process_channel_removal(struct vmbus_channel *channel)if (channel->offermsg.child_relid !=INVALID_RELID )vmbus_channel_unmap_relid (channel); -if (channel->primary_channel ==NULL ) { +if (channel->primary_channel ==NULL )list_del (&channel->listentry); - }else { -struct vmbus_channel *primary_channel = channel->primary_channel; -spin_lock_irqsave (&primary_channel->lock, flags); +else list_del (&channel->sc_list); -spin_unlock_irqrestore (&primary_channel->lock, flags); - }/* * If this is a "perf" channel, updates the hv_numa_map[] masks so that
@@ -470,7 +464,6 @@ static void vmbus_add_channel_work(struct work_struct *work)struct vmbus_channel *newchannel = container_of(work, struct vmbus_channel, add_channel_work);struct vmbus_channel *primary_channel = newchannel->primary_channel; -unsigned long flags;int ret;/*
@@ -531,13 +524,10 @@ static void vmbus_add_channel_work(struct work_struct *work)*/ newchannel->probe_done = true; -if (primary_channel ==NULL ) { +if (primary_channel ==NULL )list_del (&newchannel->listentry); - }else { -spin_lock_irqsave (&primary_channel->lock, flags); +else list_del (&newchannel->sc_list); -spin_unlock_irqrestore (&primary_channel->lock, flags); - }/* vmbus_process_offer() has mapped the channel. */ vmbus_channel_unmap_relid (newchannel);
@@ -557,7 +547,6 @@ static void vmbus_process_offer(struct vmbus_channel *newchannel) {struct vmbus_channel *channel;struct workqueue_struct *wq; -unsigned long flags;bool fnew = true;/*
@@ -609,10 +598,10 @@ static void vmbus_process_offer(struct vmbus_channel *newchannel) } } -if (fnew) +if (fnew) {list_add_tail (&newchannel->listentry, &vmbus_connection.chn_list); -else { + }else {/* * Check to see if this is a valid sub-channel. */
@@ -630,9 +619,7 @@ static void vmbus_process_offer(struct vmbus_channel *newchannel)* Process the sub-channel. */ newchannel->primary_channel = channel; -spin_lock_irqsave (&channel->lock, flags);list_add_tail (&newchannel->sc_list, &channel->sc_list); -spin_unlock_irqrestore (&channel->lock, flags); }vmbus_channel_map_relid (newchannel);
Signature · 1 line
2.25.1
RE: [PATCH 6/8] Drivers: hv: vmbus: Remove unnecessary channel->lock critical sections (sc_list updaters)
From: Andrea Parri (Microsoft) <[email protected]> Sent: Wednesday, June 17, 2020 9:47 AM
Quoted 8 lines
None of the readers/updaters of
sc_listrely onchannel->lockfor synchronization.Signed-off-by Andrea Parri (Microsoft) <[email protected]>--- drivers/hv/channel_mgmt.c | 25 ++++++------------------- 1 file changed, 6 insertions(+), 19 deletions(-)
PATCH 7/8 scsi: storvsc: Introduce the per-storvsc_device spinlock
In reply to Andrea Parri (Microsoft)storvsc uses the spinlock of the hv_device's primary channel to serialize modifications of stor_chns[] performed by storvsc_do_io() and storvsc_change_target_cpu(), when it could/should use a (per-) storvsc_device spinlock: this change untangles the synchronization mechanism for the (storvsc-specific) stor_chns[] array from the "generic" VMBus code and data structures, clarifying the scope of this synchronization mechanism.
--- drivers/scsi/storvsc_drv.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)
drivers/scsi/storvsc_drv.c +11 -5
@@ -462,6 +462,11 @@ struct storvsc_device {* Mask of CPUs bound to subchannels. */ struct cpumask alloced_cpus; +/* +* Serializes modifications of stor_chns[] from storvsc_do_io() +* and storvsc_change_target_cpu(). +*/ +spinlock_t lock;/* Used for vsc/vsp channel reset process */ struct storvsc_cmd_request init_request;struct storvsc_cmd_request reset_request;
@@ -639,7 +644,7 @@ static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,return ;/* See storvsc_do_io() -> get_og_chn(). */ -spin_lock_irqsave (&device->channel->lock, flags); +spin_lock_irqsave (&stor_device->lock, flags);/* * Determines if the storvsc device has other channels assigned to
@@ -676,7 +681,7 @@ static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,WRITE_ONCE (stor_device->stor_chns[new], channel);cpumask_set_cpu (new, &stor_device->alloced_cpus); -spin_unlock_irqrestore (&device->channel->lock, flags); +spin_unlock_irqrestore (&stor_device->lock, flags); }static void handle_sc_creation (struct vmbus_channel *new_sc)
@@ -1433,14 +1438,14 @@ static int storvsc_do_io(struct hv_device *device, } } }else { -spin_lock_irqsave (&device->channel->lock, flags); +spin_lock_irqsave (&stor_device->lock, flags); outgoing_channel = stor_device->stor_chns[q_num];if (outgoing_channel !=NULL ) { -spin_unlock_irqrestore (&device->channel->lock, flags); +spin_unlock_irqrestore (&stor_device->lock, flags); goto found_channel; } outgoing_channel =get_og_chn (stor_device, q_num); -spin_unlock_irqrestore (&device->channel->lock, flags); +spin_unlock_irqrestore (&stor_device->lock, flags); } found_channel:
@@ -1881,6 +1886,7 @@ static int storvsc_probe(struct hv_device *device,init_waitqueue_head (&stor_device->waiting_to_drain); stor_device->device = device; stor_device->host = host; +spin_lock_init (&stor_device->lock);hv_set_drvdata (device, stor_device); stor_device->port_number = host->host_no;
Signature · 1 line
2.25.1
RE: [PATCH 7/8] scsi: storvsc: Introduce the per-storvsc_device spinlock
From: Andrea Parri (Microsoft) <[email protected]> Sent: Wednesday, June 17, 2020 9:47 AM
Quoted 9 lines
storvsc uses the spinlock of the hv_device's primary channel to serialize modifications of stor_chns[] performed by
storvsc_do_io()andstorvsc_change_target_cpu(), when it could/should use a (per-)storvsc_devicespinlock: this change untangles the synchronization mechanism for the (storvsc-specific) stor_chns[] array from the "generic" VMBus code and data structures, clarifying the scope of this synchronization mechanism.
Quoted 5 lines
Signed-off-by Andrea Parri (Microsoft) <[email protected]>--- drivers/scsi/storvsc_drv.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)
FWIW, this patch will need to get routed to the SCSI maintainers and go through the SCSI tree.
Wei Liu
maintainerRe: [PATCH 7/8] scsi: storvsc: Introduce the per-storvsc_device spinlock
In reply to Andrea Parri (Microsoft)Cc SCSI maintainers
This patch should go via the hyperv tree because a later patch is dependent on it. It requires and ack from SCSI maintainers though.
Thanks,
Wei.
Quoted 76 lines · Andrea Parri (Microsoft)
storvsc uses the spinlock of the hv_device's primary channel to serialize modifications of stor_chns[] performed by
storvsc_do_io()andstorvsc_change_target_cpu(), when it could/should use a (per-)storvsc_devicespinlock: this change untangles the synchronization mechanism for the (storvsc-specific) stor_chns[] array from the "generic" VMBus code and data structures, clarifying the scope of this synchronization mechanism.Signed-off-by Andrea Parri (Microsoft) <[email protected]>--- drivers/scsi/storvsc_drv.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-)drivers/scsi/storvsc_drv.c +11
-5@@ -462,6 +462,11 @@ struct storvsc_device {* Mask of CPUs bound to subchannels. */ struct cpumask alloced_cpus; +/* +* Serializes modifications of stor_chns[] from storvsc_do_io() +* and storvsc_change_target_cpu(). +*/ +spinlock_t lock;/* Used for vsc/vsp channel reset process */ struct storvsc_cmd_request init_request;struct storvsc_cmd_request reset_request;@@ -639,7 +644,7 @@ static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,return ;/* See storvsc_do_io() -> get_og_chn(). */ -spin_lock_irqsave (&device->channel->lock, flags); +spin_lock_irqsave (&stor_device->lock, flags);/* * Determines if the storvsc device has other channels assigned to @@ -676,7 +681,7 @@ static void storvsc_change_target_cpu(struct vmbus_channel *channel, u32 old,WRITE_ONCE (stor_device->stor_chns[new], channel);cpumask_set_cpu (new, &stor_device->alloced_cpus); -spin_unlock_irqrestore (&device->channel->lock, flags); +spin_unlock_irqrestore (&stor_device->lock, flags); }static void handle_sc_creation (struct vmbus_channel *new_sc)@@ -1433,14 +1438,14 @@ static int storvsc_do_io(struct hv_device *device, } } }else { -spin_lock_irqsave (&device->channel->lock, flags); +spin_lock_irqsave (&stor_device->lock, flags); outgoing_channel = stor_device->stor_chns[q_num];if (outgoing_channel !=NULL ) { -spin_unlock_irqrestore (&device->channel->lock, flags); +spin_unlock_irqrestore (&stor_device->lock, flags); goto found_channel; } outgoing_channel =get_og_chn (stor_device, q_num); -spin_unlock_irqrestore (&device->channel->lock, flags); +spin_unlock_irqrestore (&stor_device->lock, flags); } found_channel:@@ -1881,6 +1886,7 @@ static int storvsc_probe(struct hv_device *device,init_waitqueue_head (&stor_device->waiting_to_drain); stor_device->device = device; stor_device->host = host; +spin_lock_init (&stor_device->lock);hv_set_drvdata (device, stor_device); stor_device->port_number = host->host_no;Signature · 1 line
2.25.1
Re: [PATCH 7/8] scsi: storvsc: Introduce the per-storvsc_device spinlock
Quoted 4 lines · Wei Liu
Cc SCSI maintainers
This patch should go via the hyperv tree because a later patch is dependent on it. It requires and ack from SCSI maintainers though.
Right. Sorry for the Cc: omission... ;-/
SCSI maintainers, please let me know if you prefer me to send you a new series with this patch (7/8) and the later/dependent hyperv patch (8/8).
(1-6/8 of this series are hyperv-specific only and have been applied to the hyperv tree, so this would only 7-8/8 of this series out.)
Thanks,
Andrea
Martin K. Petersen
Acked-byRe: [PATCH 7/8] scsi: storvsc: Introduce the per-storvsc_device spinlock
Andrea,
Quoted 2 lines
Quoted 2 lines
This patch should go via the hyperv tree because a later patch is dependent on it. It requires and ack from SCSI maintainers though.
Looks OK to me.
Signature · 1 line
Martin K. Petersen Oracle Linux Engineering
Wei Liu
maintainerRe: [PATCH 7/8] scsi: storvsc: Introduce the per-storvsc_device spinlock
Quoted 9 lines · Martin K. Petersen
Andrea,
Quoted 2 lines
Quoted 2 lines
This patch should go via the hyperv tree because a later patch is dependent on it. It requires and ack from SCSI maintainers though.
Looks OK to me.
Thanks Martin.
Quoted 3 lines
Signature · 2 lines
-- Martin K. Petersen Oracle Linux Engineering
PATCH 8/8 Drivers: hv: vmbus: Remove the lock field from the vmbus_channel struct
In reply to Andrea Parri (Microsoft)The spinlock is (now) *not used to protect test-and-set accesses to attributes of the structure or sc_list operations.
There is, AFAICT, a distinct lack of {WRITE,READ}_ONCE()s in the handling of channel->state, but the changes below do not seem to make things "worse". ;-)
--- drivers/hv/channel.c | 6 +----- drivers/hv/channel_mgmt.c | 1 - include/linux/hyperv.h | 6 ------ 3 files changed, 1 insertion(+), 12 deletions(-)
drivers/hv/channel.c +1 -5
@@ -129,12 +129,8 @@ static int __vmbus_open(struct vmbus_channel *newchannel, send_pages = newchannel->ringbuffer_send_offset; recv_pages = newchannel->ringbuffer_pagecount - send_pages; -spin_lock_irqsave (&newchannel->lock, flags); -if (newchannel->state !=CHANNEL_OPEN_STATE ) { -spin_unlock_irqrestore (&newchannel->lock, flags); +if (newchannel->state !=CHANNEL_OPEN_STATE )return -EINVAL ; - } -spin_unlock_irqrestore (&newchannel->lock, flags); newchannel->state =CHANNEL_OPENING_STATE ; newchannel->onchannel_callback = onchannelcallback;
drivers/hv/channel_mgmt.c +0 -1
@@ -317,7 +317,6 @@ static struct vmbus_channel *alloc_channel(void)return NULL ;spin_lock_init (&channel->sched_lock); -spin_lock_init (&channel->lock);init_completion (&channel->rescind_event);INIT_LIST_HEAD (&channel->sc_list);
include/linux/hyperv.h +0 -6
@@ -840,12 +840,6 @@ struct vmbus_channel {*/ void (*chn_rescind_callback)(struct vmbus_channel *channel); -/* -* The spinlock to protect the structure. It is being used to protect -* test-and-set access to various attributes of the structure as well -* as all sc_list operations. -*/ -spinlock_t lock;/* * All Sub-channels of a primary channel are linked here. */
Signature · 1 line
2.25.1
RE: [PATCH 8/8] Drivers: hv: vmbus: Remove the lock field from the vmbus_channel struct
From: Andrea Parri (Microsoft) <[email protected]> Sent: Wednesday, June 17, 2020 9:47 AM
Quoted 8 lines
The spinlock is (now) *not used to protect test-and-set accesses to attributes of the structure or
sc_listoperations.There is, AFAICT, a distinct lack of {WRITE,READ}_ONCE()s in the handling of
channel->state, but the changes below do not seem to make things "worse". ;-)
Quoted 7 lines
Signed-off-by Andrea Parri (Microsoft) <[email protected]>--- drivers/hv/channel.c | 6 +----- drivers/hv/channel_mgmt.c | 1 - include/linux/hyperv.h | 6 ------ 3 files changed, 1 insertion(+), 12 deletions(-)
Wei Liu
maintainerQuoted 17 lines · Andrea Parri (Microsoft)
Hi all,
I went back to my "cleanup list" recently and I wrote these patches:
here you can find, among other things,1) the removal of the fields '
target_vp' and 'numa_node' from thechannel data structure, as suggested by Michael back in May;2) various cleanups for
channel->lock, which is actually *removed by the end of this series! ;-)I'm sure there is room for further "cleanups", ;-) but let me check if these (relatively small) changes make sense first...
Thanks,
Andrea
Quoted 15 lines
Andrea Parri (Microsoft) (8):
Drivers: hv: vmbus: Remove thetarget_vpfield from thevmbus_channel
struct
Drivers: hv: vmbus: Remove thenuma_nodefield from thevmbus_channel
struct
Drivers: hv: vmbus: Replacecpumask_test_cpu(, cpu_online_mask)with
cpu_online()
Drivers: hv: vmbus: Remove unnecessarychannel->lockcritical sections
(sc_listreaders)
Drivers: hv: vmbus: Usechannel_mutexinchannel_vp_mapping_show()
Drivers: hv: vmbus: Remove unnecessarychannel->lockcritical sections
(sc_listupdaters)
scsi: storvsc: Introduce the per-storvsc_device spinlock
Drivers: hv: vmbus: Remove the lock field from thevmbus_channel
struct
I've queued these up to hyperv-next except for the scsi patch.
Thanks,
Wei.
Wei Liu
maintainerQuoted 34 lines · Wei Liu
Quoted 17 lines · Andrea Parri (Microsoft)
Hi all,
I went back to my "cleanup list" recently and I wrote these patches:
here you can find, among other things,1) the removal of the fields '
target_vp' and 'numa_node' from thechannel data structure, as suggested by Michael back in May;2) various cleanups for
channel->lock, which is actually *removed by the end of this series! ;-)I'm sure there is room for further "cleanups", ;-) but let me check if these (relatively small) changes make sense first...
Thanks,
AndreaQuoted 15 lines
Andrea Parri (Microsoft) (8):
Drivers: hv: vmbus: Remove thetarget_vpfield from thevmbus_channel
struct
Drivers: hv: vmbus: Remove thenuma_nodefield from thevmbus_channel
struct
Drivers: hv: vmbus: Replacecpumask_test_cpu(, cpu_online_mask)with
cpu_online()
Drivers: hv: vmbus: Remove unnecessarychannel->lockcritical sections
(sc_listreaders)
Drivers: hv: vmbus: Usechannel_mutexinchannel_vp_mapping_show()
Drivers: hv: vmbus: Remove unnecessarychannel->lockcritical sections
(sc_listupdaters)
scsi: storvsc: Introduce the per-storvsc_device spinlock
Drivers: hv: vmbus: Remove the lock field from thevmbus_channel
struct
Quoted 2 lines
I've queued these up to hyperv-next except for the scsi patch.
Andrea just pointed out that the last patch can't build without the scsi patch, so I've only applied only the first 6 patches for now.
Wei.
Quoted 2 lines
Thanks,
Wei.