Now the libsas hotplug has some issues, Dan Williams report
a similar bug here before
https://www.mail-archive.com/[email protected]/msg39187.html
The issues we have found
1. if LLDD burst reports lots of phy-up/phy-down sas events, some events
may lost because a same sas events is pending now, finally libsas topo
may different the hardware.
2. receive a phy down sas event, libsas call sas_deform_port to remove
devices, it would first delete the sas port, then put a destruction
discovery event in a new work, and queue it at the tail of workqueue,
once the sas port be deleted, its children device will be deleted too,
when the destruction work start, it will found the target device has
been removed, and report a sysfs warnning.
3. since a hotplug process will be devided into several works, if a phy up
sas event insert into phydown works, like
destruction work ---> PORTE_BYTES_DMAED (sas_form_port) ---->PHYE_LOSS_OF_SIGNAL
the hot remove flow would broken by PORTE_BYTES_DMAED event, it's not
we expected, and issues would occur.
The first patch fix the sas events lost, and the second one introudce wait-complete
to fix the hotplug order issues.
v1->v2: some code improvements suggested by John Garry
Yijing Wang (2):
libsas: Don't process sas events in static works
libsas: Enhance libsas hotplug
drivers/scsi/libsas/sas_discover.c | 58 +++++++++++++++++-------
drivers/scsi/libsas/sas_event.c | 90 ++++++++++++++++++++++++++------------
drivers/scsi/libsas/sas_expander.c | 9 +++-
drivers/scsi/libsas/sas_init.c | 29 +++++++++---
drivers/scsi/libsas/sas_internal.h | 64 +++++++++++++++++++++++++++
drivers/scsi/libsas/sas_phy.c | 45 ++++---------------
drivers/scsi/libsas/sas_port.c | 22 +++++-----
include/scsi/libsas.h | 19 ++++----
8 files changed, 232 insertions(+), 104 deletions(-)
--
2.5.0
Now libsas hotplug work is static, LLDD driver queue
the hotplug work into shost->work_q. If LLDD driver
burst post lots hotplug events to libsas, the hotplug
events may pending in the workqueue like
shost->work_q
new work[PORTE_BYTES_DMAED] --> |[PHYE_LOSS_OF_SIGNAL][PORTE_BYTES_DMAED] -> processing
|<-------wait worker to process-------->|
In this case, a new PORTE_BYTES_DMAED event coming, libsas try to queue it
to shost->work_q, but this work is already pending, so it would be lost.
Finally, libsas delete the related sas port and sas devices, but LLDD driver
expect libsas add the sas port and devices(last sas event).
This patch remove the static defined hotplug work, and use dynamic work to
avoid missing hotplug events.
Signed-off-by: Yijing Wang <[email protected]>
Signed-off-by: Yousong He <[email protected]>
Signed-off-by: Qilin Chen <[email protected]>
---
drivers/scsi/libsas/sas_event.c | 88 +++++++++++++++++++++++++++-----------
drivers/scsi/libsas/sas_init.c | 6 ---
drivers/scsi/libsas/sas_internal.h | 3 ++
drivers/scsi/libsas/sas_phy.c | 45 ++++---------------
drivers/scsi/libsas/sas_port.c | 18 ++++----
include/scsi/libsas.h | 10 +----
6 files changed, 84 insertions(+), 86 deletions(-)
diff --git a/drivers/scsi/libsas/sas_event.c b/drivers/scsi/libsas/sas_event.c
index aadbd53..06c5c4b 100644
--- a/drivers/scsi/libsas/sas_event.c
+++ b/drivers/scsi/libsas/sas_event.c
@@ -27,6 +27,10 @@
#include "sas_internal.h"
#include "sas_dump.h"
+static const work_func_t sas_ha_event_fns[HA_NUM_EVENTS] = {
+ [HAE_RESET] = sas_hae_reset,
+};
+
void sas_queue_work(struct sas_ha_struct *ha, struct sas_work *sw)
{
if (!test_bit(SAS_HA_REGISTERED, &ha->state))
@@ -40,17 +44,14 @@ void sas_queue_work(struct sas_ha_struct *ha, struct sas_work *sw)
scsi_queue_work(ha->core.shost, &sw->work);
}
-static void sas_queue_event(int event, unsigned long *pending,
- struct sas_work *work,
+static void sas_queue_event(int event, struct sas_work *work,
struct sas_ha_struct *ha)
{
- if (!test_and_set_bit(event, pending)) {
- unsigned long flags;
+ unsigned long flags;
- spin_lock_irqsave(&ha->lock, flags);
- sas_queue_work(ha, work);
- spin_unlock_irqrestore(&ha->lock, flags);
- }
+ spin_lock_irqsave(&ha->lock, flags);
+ sas_queue_work(ha, work);
+ spin_unlock_irqrestore(&ha->lock, flags);
}
@@ -111,52 +112,87 @@ void sas_enable_revalidation(struct sas_ha_struct *ha)
if (!test_and_clear_bit(ev, &d->pending))
continue;
- sas_queue_event(ev, &d->pending, &d->disc_work[ev].work, ha);
+ sas_queue_event(ev, &d->disc_work[ev].work, ha);
}
mutex_unlock(&ha->disco_mutex);
}
+static void sas_ha_event_worker(struct work_struct *work)
+{
+ struct sas_ha_event *ev = to_sas_ha_event(work);
+
+ sas_ha_event_fns[ev->type](work);
+ kfree(ev);
+}
+
+static void sas_port_event_worker(struct work_struct *work)
+{
+ struct asd_sas_event *ev = to_asd_sas_event(work);
+
+ sas_port_event_fns[ev->type](work);
+ kfree(ev);
+}
+
+static void sas_phy_event_worker(struct work_struct *work)
+{
+ struct asd_sas_event *ev = to_asd_sas_event(work);
+
+ sas_phy_event_fns[ev->type](work);
+ kfree(ev);
+}
+
static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event event)
{
+ struct sas_ha_event *ev;
+
BUG_ON(event >= HA_NUM_EVENTS);
- sas_queue_event(event, &sas_ha->pending,
- &sas_ha->ha_events[event].work, sas_ha);
+ ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
+ if (!ev)
+ return;
+
+ INIT_SAS_WORK(&ev->work, sas_ha_event_worker);
+ ev->ha = sas_ha;
+ ev->type = event;
+ sas_queue_event(event, &ev->work, sas_ha);
}
static void notify_port_event(struct asd_sas_phy *phy, enum port_event event)
{
+ struct asd_sas_event *ev;
struct sas_ha_struct *ha = phy->ha;
BUG_ON(event >= PORT_NUM_EVENTS);
- sas_queue_event(event, &phy->port_events_pending,
- &phy->port_events[event].work, ha);
+ ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
+ if (!ev)
+ return;
+
+ INIT_SAS_WORK(&ev->work, sas_port_event_worker);
+ ev->phy = phy;
+ ev->type = event;
+ sas_queue_event(event, &ev->work, ha);
}
void sas_notify_phy_event(struct asd_sas_phy *phy, enum phy_event event)
{
+ struct asd_sas_event *ev;
struct sas_ha_struct *ha = phy->ha;
BUG_ON(event >= PHY_NUM_EVENTS);
- sas_queue_event(event, &phy->phy_events_pending,
- &phy->phy_events[event].work, ha);
+ ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
+ if (!ev)
+ return;
+
+ INIT_SAS_WORK(&ev->work, sas_phy_event_worker);
+ ev->phy = phy;
+ ev->type = event;
+ sas_queue_event(event, &ev->work, ha);
}
int sas_init_events(struct sas_ha_struct *sas_ha)
{
- static const work_func_t sas_ha_event_fns[HA_NUM_EVENTS] = {
- [HAE_RESET] = sas_hae_reset,
- };
-
- int i;
-
- for (i = 0; i < HA_NUM_EVENTS; i++) {
- INIT_SAS_WORK(&sas_ha->ha_events[i].work, sas_ha_event_fns[i]);
- sas_ha->ha_events[i].ha = sas_ha;
- }
-
sas_ha->notify_ha_event = notify_ha_event;
sas_ha->notify_port_event = notify_port_event;
sas_ha->notify_phy_event = sas_notify_phy_event;
diff --git a/drivers/scsi/libsas/sas_init.c b/drivers/scsi/libsas/sas_init.c
index 64e9cdd..c227a8b 100644
--- a/drivers/scsi/libsas/sas_init.c
+++ b/drivers/scsi/libsas/sas_init.c
@@ -111,10 +111,6 @@ void sas_hash_addr(u8 *hashed, const u8 *sas_addr)
void sas_hae_reset(struct work_struct *work)
{
- struct sas_ha_event *ev = to_sas_ha_event(work);
- struct sas_ha_struct *ha = ev->ha;
-
- clear_bit(HAE_RESET, &ha->pending);
}
int sas_register_ha(struct sas_ha_struct *sas_ha)
@@ -375,8 +371,6 @@ void sas_prep_resume_ha(struct sas_ha_struct *ha)
struct asd_sas_phy *phy = ha->sas_phy[i];
memset(phy->attached_sas_addr, 0, SAS_ADDR_SIZE);
- phy->port_events_pending = 0;
- phy->phy_events_pending = 0;
phy->frame_rcvd_size = 0;
}
}
diff --git a/drivers/scsi/libsas/sas_internal.h b/drivers/scsi/libsas/sas_internal.h
index b306b78..33ce7e5 100644
--- a/drivers/scsi/libsas/sas_internal.h
+++ b/drivers/scsi/libsas/sas_internal.h
@@ -97,6 +97,9 @@ void sas_hae_reset(struct work_struct *work);
void sas_free_device(struct kref *kref);
+extern const work_func_t sas_phy_event_fns[PHY_NUM_EVENTS];
+extern const work_func_t sas_port_event_fns[PORT_NUM_EVENTS];
+
#ifdef CONFIG_SCSI_SAS_HOST_SMP
extern int sas_smp_host_handler(struct Scsi_Host *shost, struct request *req,
struct request *rsp);
diff --git a/drivers/scsi/libsas/sas_phy.c b/drivers/scsi/libsas/sas_phy.c
index cdee446c..7c4576d 100644
--- a/drivers/scsi/libsas/sas_phy.c
+++ b/drivers/scsi/libsas/sas_phy.c
@@ -35,7 +35,6 @@ static void sas_phye_loss_of_signal(struct work_struct *work)
struct asd_sas_event *ev = to_asd_sas_event(work);
struct asd_sas_phy *phy = ev->phy;
- clear_bit(PHYE_LOSS_OF_SIGNAL, &phy->phy_events_pending);
phy->error = 0;
sas_deform_port(phy, 1);
}
@@ -45,7 +44,6 @@ static void sas_phye_oob_done(struct work_struct *work)
struct asd_sas_event *ev = to_asd_sas_event(work);
struct asd_sas_phy *phy = ev->phy;
- clear_bit(PHYE_OOB_DONE, &phy->phy_events_pending);
phy->error = 0;
}
@@ -58,8 +56,6 @@ static void sas_phye_oob_error(struct work_struct *work)
struct sas_internal *i =
to_sas_internal(sas_ha->core.shost->transportt);
- clear_bit(PHYE_OOB_ERROR, &phy->phy_events_pending);
-
sas_deform_port(phy, 1);
if (!port && phy->enabled && i->dft->lldd_control_phy) {
@@ -88,8 +84,6 @@ static void sas_phye_spinup_hold(struct work_struct *work)
struct sas_internal *i =
to_sas_internal(sas_ha->core.shost->transportt);
- clear_bit(PHYE_SPINUP_HOLD, &phy->phy_events_pending);
-
phy->error = 0;
i->dft->lldd_control_phy(phy, PHY_FUNC_RELEASE_SPINUP_HOLD, NULL);
}
@@ -99,8 +93,6 @@ static void sas_phye_resume_timeout(struct work_struct *work)
struct asd_sas_event *ev = to_asd_sas_event(work);
struct asd_sas_phy *phy = ev->phy;
- clear_bit(PHYE_RESUME_TIMEOUT, &phy->phy_events_pending);
-
/* phew, lldd got the phy back in the nick of time */
if (!phy->suspended) {
dev_info(&phy->phy->dev, "resume timeout cancelled\n");
@@ -112,46 +104,18 @@ static void sas_phye_resume_timeout(struct work_struct *work)
sas_deform_port(phy, 1);
}
-
/* ---------- Phy class registration ---------- */
int sas_register_phys(struct sas_ha_struct *sas_ha)
{
int i;
- static const work_func_t sas_phy_event_fns[PHY_NUM_EVENTS] = {
- [PHYE_LOSS_OF_SIGNAL] = sas_phye_loss_of_signal,
- [PHYE_OOB_DONE] = sas_phye_oob_done,
- [PHYE_OOB_ERROR] = sas_phye_oob_error,
- [PHYE_SPINUP_HOLD] = sas_phye_spinup_hold,
- [PHYE_RESUME_TIMEOUT] = sas_phye_resume_timeout,
-
- };
-
- static const work_func_t sas_port_event_fns[PORT_NUM_EVENTS] = {
- [PORTE_BYTES_DMAED] = sas_porte_bytes_dmaed,
- [PORTE_BROADCAST_RCVD] = sas_porte_broadcast_rcvd,
- [PORTE_LINK_RESET_ERR] = sas_porte_link_reset_err,
- [PORTE_TIMER_EVENT] = sas_porte_timer_event,
- [PORTE_HARD_RESET] = sas_porte_hard_reset,
- };
-
/* Now register the phys. */
for (i = 0; i < sas_ha->num_phys; i++) {
- int k;
struct asd_sas_phy *phy = sas_ha->sas_phy[i];
phy->error = 0;
INIT_LIST_HEAD(&phy->port_phy_el);
- for (k = 0; k < PORT_NUM_EVENTS; k++) {
- INIT_SAS_WORK(&phy->port_events[k].work, sas_port_event_fns[k]);
- phy->port_events[k].phy = phy;
- }
-
- for (k = 0; k < PHY_NUM_EVENTS; k++) {
- INIT_SAS_WORK(&phy->phy_events[k].work, sas_phy_event_fns[k]);
- phy->phy_events[k].phy = phy;
- }
phy->port = NULL;
phy->ha = sas_ha;
@@ -179,3 +143,12 @@ int sas_register_phys(struct sas_ha_struct *sas_ha)
return 0;
}
+
+const work_func_t sas_phy_event_fns[PHY_NUM_EVENTS] = {
+ [PHYE_LOSS_OF_SIGNAL] = sas_phye_loss_of_signal,
+ [PHYE_OOB_DONE] = sas_phye_oob_done,
+ [PHYE_OOB_ERROR] = sas_phye_oob_error,
+ [PHYE_SPINUP_HOLD] = sas_phye_spinup_hold,
+ [PHYE_RESUME_TIMEOUT] = sas_phye_resume_timeout,
+
+};
diff --git a/drivers/scsi/libsas/sas_port.c b/drivers/scsi/libsas/sas_port.c
index d3c5297..9326628 100644
--- a/drivers/scsi/libsas/sas_port.c
+++ b/drivers/scsi/libsas/sas_port.c
@@ -261,8 +261,6 @@ void sas_porte_bytes_dmaed(struct work_struct *work)
struct asd_sas_event *ev = to_asd_sas_event(work);
struct asd_sas_phy *phy = ev->phy;
- clear_bit(PORTE_BYTES_DMAED, &phy->port_events_pending);
-
sas_form_port(phy);
}
@@ -273,8 +271,6 @@ void sas_porte_broadcast_rcvd(struct work_struct *work)
unsigned long flags;
u32 prim;
- clear_bit(PORTE_BROADCAST_RCVD, &phy->port_events_pending);
-
spin_lock_irqsave(&phy->sas_prim_lock, flags);
prim = phy->sas_prim;
spin_unlock_irqrestore(&phy->sas_prim_lock, flags);
@@ -288,8 +284,6 @@ void sas_porte_link_reset_err(struct work_struct *work)
struct asd_sas_event *ev = to_asd_sas_event(work);
struct asd_sas_phy *phy = ev->phy;
- clear_bit(PORTE_LINK_RESET_ERR, &phy->port_events_pending);
-
sas_deform_port(phy, 1);
}
@@ -298,8 +292,6 @@ void sas_porte_timer_event(struct work_struct *work)
struct asd_sas_event *ev = to_asd_sas_event(work);
struct asd_sas_phy *phy = ev->phy;
- clear_bit(PORTE_TIMER_EVENT, &phy->port_events_pending);
-
sas_deform_port(phy, 1);
}
@@ -308,8 +300,6 @@ void sas_porte_hard_reset(struct work_struct *work)
struct asd_sas_event *ev = to_asd_sas_event(work);
struct asd_sas_phy *phy = ev->phy;
- clear_bit(PORTE_HARD_RESET, &phy->port_events_pending);
-
sas_deform_port(phy, 1);
}
@@ -353,3 +343,11 @@ void sas_unregister_ports(struct sas_ha_struct *sas_ha)
sas_deform_port(sas_ha->sas_phy[i], 0);
}
+
+const work_func_t sas_port_event_fns[PORT_NUM_EVENTS] = {
+ [PORTE_BYTES_DMAED] = sas_porte_bytes_dmaed,
+ [PORTE_BROADCAST_RCVD] = sas_porte_broadcast_rcvd,
+ [PORTE_LINK_RESET_ERR] = sas_porte_link_reset_err,
+ [PORTE_TIMER_EVENT] = sas_porte_timer_event,
+ [PORTE_HARD_RESET] = sas_porte_hard_reset,
+};
diff --git a/include/scsi/libsas.h b/include/scsi/libsas.h
index dd0f72c..e6ebad0 100644
--- a/include/scsi/libsas.h
+++ b/include/scsi/libsas.h
@@ -300,6 +300,7 @@ struct asd_sas_port {
struct asd_sas_event {
struct sas_work work;
struct asd_sas_phy *phy;
+ int type;
};
static inline struct asd_sas_event *to_asd_sas_event(struct work_struct *work)
@@ -314,11 +315,6 @@ static inline struct asd_sas_event *to_asd_sas_event(struct work_struct *work)
*/
struct asd_sas_phy {
/* private: */
- struct asd_sas_event port_events[PORT_NUM_EVENTS];
- struct asd_sas_event phy_events[PHY_NUM_EVENTS];
-
- unsigned long port_events_pending;
- unsigned long phy_events_pending;
int error;
int suspended;
@@ -365,6 +361,7 @@ struct scsi_core {
struct sas_ha_event {
struct sas_work work;
struct sas_ha_struct *ha;
+ int type;
};
static inline struct sas_ha_event *to_sas_ha_event(struct work_struct *work)
@@ -383,9 +380,6 @@ enum sas_ha_state {
struct sas_ha_struct {
/* private: */
- struct sas_ha_event ha_events[HA_NUM_EVENTS];
- unsigned long pending;
-
struct list_head defer_q; /* work queued while draining */
struct mutex drain_mutex;
unsigned long state;
--
2.5.0
Libsas complete a hotplug event notified by LLDD in several works,
for example, if libsas receive a PHYE_LOSS_OF_SIGNAL, we process it
in following steps:
notify_phy_event [interrupt context]
sas_queue_event [queue work on shost->work_q]
sas_phye_loss_of_signal [running in shost->work_q]
sas_deform_port [remove sas port]
sas_unregister_dev
sas_discover_event [queue destruct work on shost->work_q tail]
In above case, complete whole hotplug in two works, remove sas port first, then
put the destruction of device in another work and queue it on in the tail of
workqueue, since sas port is the parent of the children rphy device, so if remove
sas port first, the children rphy device would also be deleted, when the destruction
work coming, it would find the target has been removed already, and report a
sysfs warning calltrace.
queue tail queue head
DISCE_DESTRUCT----> PORTE_BYTES_DMAED event ----->PHYE_LOSS_OF_SIGNAL[running]
There are other hotplug issues in current framework, in above case, if there is
hotadd sas event queued between hotremove works, the hotplug order would be broken
and unexpected issues would happen.
In this patch, we try to solve these issues in following steps:
1. create a new workqueue used to run sas event work, instead of scsi host workqueue,
because we may block sas event work, we cannot block the normal scsi works.
When libsas receive a phy down event, sas_deform_port would be called, and now we
block sas_deform_port and wait for destruction work finish, in sas_destruct_devices,
we may wait ata error handler, it would take a long time, so if do all stuff in scsi
host workq, libsas may block other scsi works too long.
2. create a new workqueue used to run sas discovery events work, instead of scsi host
workqueue, because in some cases, eg. in revalidate domain event, we may unregister
a sas device and discover new one, we must sync the execution, wait the remove process
finish, then start a new discovery. So we must put the probe and destruct discovery
events in a new workqueue to avoid deadlock.
3. introudce a asd_sas_port level wait-complete and a sas_discovery level wait-complete
we use former wait-complete to achieve a sas event atomic process and use latter to
make a sas discovery sync.
4. remove disco_mutex in sas_revalidate_domain, since now sas_revalidate_domain sync
the destruct discovery event execution, it's no need to lock disco mutex there.
Signed-off-by: Yijing Wang <[email protected]>
---
drivers/scsi/libsas/sas_discover.c | 58 ++++++++++++++++++++++++++----------
drivers/scsi/libsas/sas_event.c | 2 +-
drivers/scsi/libsas/sas_expander.c | 9 +++++-
drivers/scsi/libsas/sas_init.c | 23 +++++++++++++-
drivers/scsi/libsas/sas_internal.h | 61 ++++++++++++++++++++++++++++++++++++++
drivers/scsi/libsas/sas_port.c | 4 +++
include/scsi/libsas.h | 9 ++++++
7 files changed, 148 insertions(+), 18 deletions(-)
diff --git a/drivers/scsi/libsas/sas_discover.c b/drivers/scsi/libsas/sas_discover.c
index 60de662..43e8a1e 100644
--- a/drivers/scsi/libsas/sas_discover.c
+++ b/drivers/scsi/libsas/sas_discover.c
@@ -503,11 +503,10 @@ static void sas_revalidate_domain(struct work_struct *work)
struct domain_device *ddev = port->port_dev;
/* prevent revalidation from finding sata links in recovery */
- mutex_lock(&ha->disco_mutex);
if (test_bit(SAS_HA_ATA_EH_ACTIVE, &ha->state)) {
SAS_DPRINTK("REVALIDATION DEFERRED on port %d, pid:%d\n",
port->id, task_pid_nr(current));
- goto out;
+ return;
}
clear_bit(DISCE_REVALIDATE_DOMAIN, &port->disc.pending);
@@ -521,20 +520,57 @@ static void sas_revalidate_domain(struct work_struct *work)
SAS_DPRINTK("done REVALIDATING DOMAIN on port %d, pid:%d, res 0x%x\n",
port->id, task_pid_nr(current), res);
- out:
- mutex_unlock(&ha->disco_mutex);
+}
+
+static const work_func_t sas_event_fns[DISC_NUM_EVENTS] = {
+ [DISCE_DISCOVER_DOMAIN] = sas_discover_domain,
+ [DISCE_REVALIDATE_DOMAIN] = sas_revalidate_domain,
+ [DISCE_PROBE] = sas_probe_devices,
+ [DISCE_SUSPEND] = sas_suspend_devices,
+ [DISCE_RESUME] = sas_resume_devices,
+ [DISCE_DESTRUCT] = sas_destruct_devices,
+};
+
+/* a simple wrapper for sas discover event funtions */
+static void sas_discover_common_fn(struct work_struct *work)
+{
+ struct sas_discovery_event *ev = to_sas_discovery_event(work);
+ struct asd_sas_port *port = ev->port;
+
+ sas_event_fns[ev->type](work);
+ sas_unbusy_port(port);
}
/* ---------- Events ---------- */
static void sas_chain_work(struct sas_ha_struct *ha, struct sas_work *sw)
{
+ int ret;
+ struct sas_discovery_event *ev = to_sas_discovery_event(&sw->work);
+ struct asd_sas_port *port = ev->port;
+
/* chained work is not subject to SA_HA_DRAINING or
* SAS_HA_REGISTERED, because it is either submitted in the
* workqueue, or known to be submitted from a context that is
* not racing against draining
*/
- scsi_queue_work(ha->core.shost, &sw->work);
+ sas_busy_port(port);
+
+ /*
+ * discovery event probe and destruct would be called in other
+ * discovery event like discover domain and revalidate domain
+ * events, in some cases, we need to sync execute probe and destruct
+ * events, so run discover events except probe/destruct in a new
+ * workqueue.
+ */
+ if (ev->type == DISCE_PROBE || ev->type == DISCE_DESTRUCT)
+ ret = scsi_queue_work(ha->core.shost, &sw->work);
+ else
+ ret = queue_work(ha->disc_q, &sw->work);
+
+ if (ret != 1)
+ /* queue a work fail, unbusy the ha before return */
+ sas_unbusy_port(port);
}
static void sas_chain_event(int event, unsigned long *pending,
@@ -575,18 +611,10 @@ void sas_init_disc(struct sas_discovery *disc, struct asd_sas_port *port)
{
int i;
- static const work_func_t sas_event_fns[DISC_NUM_EVENTS] = {
- [DISCE_DISCOVER_DOMAIN] = sas_discover_domain,
- [DISCE_REVALIDATE_DOMAIN] = sas_revalidate_domain,
- [DISCE_PROBE] = sas_probe_devices,
- [DISCE_SUSPEND] = sas_suspend_devices,
- [DISCE_RESUME] = sas_resume_devices,
- [DISCE_DESTRUCT] = sas_destruct_devices,
- };
-
disc->pending = 0;
for (i = 0; i < DISC_NUM_EVENTS; i++) {
- INIT_SAS_WORK(&disc->disc_work[i].work, sas_event_fns[i]);
+ INIT_SAS_WORK(&disc->disc_work[i].work, sas_discover_common_fn);
disc->disc_work[i].port = port;
+ disc->disc_work[i].type = i;
}
}
diff --git a/drivers/scsi/libsas/sas_event.c b/drivers/scsi/libsas/sas_event.c
index 06c5c4b..c0fc07d 100644
--- a/drivers/scsi/libsas/sas_event.c
+++ b/drivers/scsi/libsas/sas_event.c
@@ -41,7 +41,7 @@ void sas_queue_work(struct sas_ha_struct *ha, struct sas_work *sw)
if (list_empty(&sw->drain_node))
list_add(&sw->drain_node, &ha->defer_q);
} else
- scsi_queue_work(ha->core.shost, &sw->work);
+ queue_work(ha->event_q, &sw->work);
}
static void sas_queue_event(int event, struct sas_work *work,
diff --git a/drivers/scsi/libsas/sas_expander.c b/drivers/scsi/libsas/sas_expander.c
index 570b2cb..7db4a28 100644
--- a/drivers/scsi/libsas/sas_expander.c
+++ b/drivers/scsi/libsas/sas_expander.c
@@ -822,7 +822,9 @@ static struct domain_device *sas_ex_discover_end_dev(
list_add_tail(&child->disco_list_node, &parent->port->disco_list);
+ sas_wait_discover_event_init(child->port);
res = sas_discover_sata(child);
+ sas_wait_for_discover_event_finish(child->port);
if (res) {
SAS_DPRINTK("sas_discover_sata() for device %16llx at "
"%016llx:0x%x returned 0x%x\n",
@@ -847,7 +849,9 @@ static struct domain_device *sas_ex_discover_end_dev(
list_add_tail(&child->disco_list_node, &parent->port->disco_list);
+ sas_wait_discover_event_init(child->port);
res = sas_discover_end_dev(child);
+ sas_wait_for_discover_event_finish(child->port);
if (res) {
SAS_DPRINTK("sas_discover_end_dev() for device %16llx "
"at %016llx:0x%x returned 0x%x\n",
@@ -1890,8 +1894,11 @@ static void sas_unregister_devs_sas_addr(struct domain_device *parent,
if (child->dev_type == SAS_EDGE_EXPANDER_DEVICE ||
child->dev_type == SAS_FANOUT_EXPANDER_DEVICE)
sas_unregister_ex_tree(parent->port, child);
- else
+ else {
+ sas_wait_discover_event_init(parent->port);
sas_unregister_dev(parent->port, child);
+ sas_wait_for_discover_event_finish(parent->port);
+ }
found = child;
break;
}
diff --git a/drivers/scsi/libsas/sas_init.c b/drivers/scsi/libsas/sas_init.c
index c227a8b..c9907ec1 100644
--- a/drivers/scsi/libsas/sas_init.c
+++ b/drivers/scsi/libsas/sas_init.c
@@ -38,6 +38,8 @@
#include "../scsi_sas_internal.h"
+static DEFINE_IDA(sas_ida);
+
static struct kmem_cache *sas_task_cache;
struct sas_task *sas_alloc_task(gfp_t flags)
@@ -116,6 +118,7 @@ void sas_hae_reset(struct work_struct *work)
int sas_register_ha(struct sas_ha_struct *sas_ha)
{
int error = 0;
+ char name[64];
mutex_init(&sas_ha->disco_mutex);
spin_lock_init(&sas_ha->phy_port_lock);
@@ -146,6 +149,22 @@ int sas_register_ha(struct sas_ha_struct *sas_ha)
goto Undo_ports;
}
+ snprintf(name, 64, "%s_event_q", dev_name(sas_ha->dev));
+ sas_ha->event_q = create_singlethread_workqueue(name);
+ if (!sas_ha->event_q)
+ goto Undo_ports;
+
+ /*
+ * sas-disc-xx workqueue run the discover work except
+ * probe and destruct.
+ */
+ snprintf(name, 64, "%s_disc_q", dev_name(sas_ha->dev));
+ sas_ha->disc_q = create_singlethread_workqueue(name);
+ if(!sas_ha->disc_q) {
+ destroy_workqueue(sas_ha->event_q);
+ goto Undo_ports;
+ }
+
INIT_LIST_HEAD(&sas_ha->eh_done_q);
INIT_LIST_HEAD(&sas_ha->eh_ata_q);
@@ -181,6 +200,9 @@ int sas_unregister_ha(struct sas_ha_struct *sas_ha)
__sas_drain_work(sas_ha);
mutex_unlock(&sas_ha->drain_mutex);
+ destroy_workqueue(sas_ha->event_q);
+ destroy_workqueue(sas_ha->disc_q);
+ ida_simple_remove(&sas_ida, sas_ha->id);
return 0;
}
@@ -561,7 +583,6 @@ sas_domain_attach_transport(struct sas_domain_function_template *dft)
EXPORT_SYMBOL_GPL(sas_domain_attach_transport);
/* ---------- SAS Class register/unregister ---------- */
-
static int __init sas_class_init(void)
{
sas_task_cache = KMEM_CACHE(sas_task, SLAB_HWCACHE_ALIGN);
diff --git a/drivers/scsi/libsas/sas_internal.h b/drivers/scsi/libsas/sas_internal.h
index 33ce7e5..d5fcae8 100644
--- a/drivers/scsi/libsas/sas_internal.h
+++ b/drivers/scsi/libsas/sas_internal.h
@@ -100,6 +100,67 @@ void sas_free_device(struct kref *kref);
extern const work_func_t sas_phy_event_fns[PHY_NUM_EVENTS];
extern const work_func_t sas_port_event_fns[PORT_NUM_EVENTS];
+static inline void sas_wait_discover_event_init(struct asd_sas_port *port)
+{
+ if (port) {
+ init_completion(&port->disc.completion);
+ port->disc.wait = 1;
+ port->disc.busy = 0;
+ }
+}
+
+static inline void sas_wait_for_discover_event_finish(
+ struct asd_sas_port *port)
+{
+ if (port && port->disc.busy > 0)
+ wait_for_completion(&port->disc.completion);
+}
+
+static inline void sas_wait_sas_event_init(struct asd_sas_port *port)
+{
+ if (port) {
+ init_completion(&port->completion);
+ kref_init(&port->ref);
+ }
+}
+
+static inline void sas_wait_for_sas_event_finish(
+ struct asd_sas_port *port)
+{
+ if (port && (kref_read(&port->ref) > 1))
+ wait_for_completion(&port->completion);
+}
+
+static inline void sas_busy_port(struct asd_sas_port *port)
+{
+ if (port)
+ kref_get(&port->ref);
+
+ if (port->disc.wait == 1)
+ port->disc.busy++;
+}
+
+static void sas_complete_event(struct kref *kref)
+{
+ struct asd_sas_port *port = container_of(kref, struct asd_sas_port, ref);
+
+ complete(&port->completion);
+}
+
+static inline void sas_unbusy_port(struct asd_sas_port *port)
+{
+ if (!port)
+ return;
+
+ kref_put(&port->ref, sas_complete_event);
+
+ if (port->disc.wait == 1) {
+ complete(&port->disc.completion);
+ port->disc.wait = 0;
+ port->disc.busy = 0;
+ }
+}
+
#ifdef CONFIG_SCSI_SAS_HOST_SMP
extern int sas_smp_host_handler(struct Scsi_Host *shost, struct request *req,
struct request *rsp);
diff --git a/drivers/scsi/libsas/sas_port.c b/drivers/scsi/libsas/sas_port.c
index 9326628..459f852 100644
--- a/drivers/scsi/libsas/sas_port.c
+++ b/drivers/scsi/libsas/sas_port.c
@@ -191,7 +191,9 @@ static void sas_form_port(struct asd_sas_phy *phy)
if (si->dft->lldd_port_formed)
si->dft->lldd_port_formed(phy);
+ sas_wait_sas_event_init(port);
sas_discover_event(phy->port, DISCE_DISCOVER_DOMAIN);
+ sas_wait_for_sas_event_finish(port);
}
/**
@@ -218,7 +220,9 @@ void sas_deform_port(struct asd_sas_phy *phy, int gone)
dev->pathways--;
if (port->num_phys == 1) {
+ sas_wait_sas_event_init(port);
sas_unregister_domain_devices(port, gone);
+ sas_wait_for_sas_event_finish(port);
sas_port_delete(port->port);
port->port = NULL;
} else {
diff --git a/include/scsi/libsas.h b/include/scsi/libsas.h
index e6ebad0..be09491 100644
--- a/include/scsi/libsas.h
+++ b/include/scsi/libsas.h
@@ -240,6 +240,7 @@ static inline void INIT_SAS_WORK(struct sas_work *sw, void (*fn)(struct work_str
struct sas_discovery_event {
struct sas_work work;
struct asd_sas_port *port;
+ enum discover_event type;
};
static inline struct sas_discovery_event *to_sas_discovery_event(struct work_struct *work)
@@ -256,6 +257,9 @@ struct sas_discovery {
u8 eeds_a[8];
u8 eeds_b[8];
int max_level;
+ int wait;
+ int busy;
+ struct completion completion;
};
/* The port struct is Class:RW, driver:RO */
@@ -273,6 +277,8 @@ struct asd_sas_port {
struct sas_work work;
int suspended;
+ struct kref ref;
+ struct completion completion;
/* public: */
int id;
@@ -387,6 +393,7 @@ struct sas_ha_struct {
int eh_active;
wait_queue_head_t eh_wait_q;
struct list_head eh_dev_q;
+ int id; /* for create workqueue */
struct mutex disco_mutex;
@@ -396,6 +403,8 @@ struct sas_ha_struct {
char *sas_ha_name;
struct device *dev; /* should be set */
struct module *lldd_module; /* should be set */
+ struct workqueue_struct *event_q;
+ struct workqueue_struct *disc_q;
u8 *sas_addr; /* must be set */
u8 hashed_sas_addr[HASHED_SAS_ADDR_SIZE];
--
2.5.0
On 06/14/2017 09:33 AM, Yijing Wang wrote:
> Now libsas hotplug work is static, LLDD driver queue
> the hotplug work into shost->work_q. If LLDD driver
> burst post lots hotplug events to libsas, the hotplug
> events may pending in the workqueue like
>
> shost->work_q
> new work[PORTE_BYTES_DMAED] --> |[PHYE_LOSS_OF_SIGNAL][PORTE_BYTES_DMAED] -> processing
> |<-------wait worker to process-------->|
> In this case, a new PORTE_BYTES_DMAED event coming, libsas try to queue it
> to shost->work_q, but this work is already pending, so it would be lost.
> Finally, libsas delete the related sas port and sas devices, but LLDD driver
> expect libsas add the sas port and devices(last sas event).
>
> This patch remove the static defined hotplug work, and use dynamic work to
> avoid missing hotplug events.
>
> Signed-off-by: Yijing Wang <[email protected]>
> Signed-off-by: Yousong He <[email protected]>
> Signed-off-by: Qilin Chen <[email protected]>
> ---
[...]
> static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event event)
> {
> + struct sas_ha_event *ev;
> +
> BUG_ON(event >= HA_NUM_EVENTS);
>
> - sas_queue_event(event, &sas_ha->pending,
> - &sas_ha->ha_events[event].work, sas_ha);
> + ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
> + if (!ev)
> + return;
GFP_ATOMIC allocations can fail and then no events will be queued *and* we
don't report the error back to the caller.
> index 64e9cdd..c227a8b 100644
> --- a/drivers/scsi/libsas/sas_init.c
> +++ b/drivers/scsi/libsas/sas_init.c
> @@ -111,10 +111,6 @@ void sas_hash_addr(u8 *hashed, const u8 *sas_addr)
>
> void sas_hae_reset(struct work_struct *work)
> {
> - struct sas_ha_event *ev = to_sas_ha_event(work);
> - struct sas_ha_struct *ha = ev->ha;
> -
> - clear_bit(HAE_RESET, &ha->pending);
> }
I don't really get why you need a stubbed out sas_hae_reset(). Can't we
just kill it if it doesn't have anything left to do?
--
Johannes Thumshirn Storage
[email protected] +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850
On 06/14/2017 09:33 AM, Yijing Wang wrote:
> Libsas complete a hotplug event notified by LLDD in several works,
> for example, if libsas receive a PHYE_LOSS_OF_SIGNAL, we process it
> in following steps:
>
> notify_phy_event [interrupt context]
> sas_queue_event [queue work on shost->work_q]
> sas_phye_loss_of_signal [running in shost->work_q]
> sas_deform_port [remove sas port]
> sas_unregister_dev
> sas_discover_event [queue destruct work on shost->work_q tail]
>
> In above case, complete whole hotplug in two works, remove sas port first, then
> put the destruction of device in another work and queue it on in the tail of
> workqueue, since sas port is the parent of the children rphy device, so if remove
> sas port first, the children rphy device would also be deleted, when the destruction
> work coming, it would find the target has been removed already, and report a
> sysfs warning calltrace.
>
> queue tail queue head
> DISCE_DESTRUCT----> PORTE_BYTES_DMAED event ----->PHYE_LOSS_OF_SIGNAL[running]
>
> There are other hotplug issues in current framework, in above case, if there is
> hotadd sas event queued between hotremove works, the hotplug order would be broken
> and unexpected issues would happen.
>
> In this patch, we try to solve these issues in following steps:
> 1. create a new workqueue used to run sas event work, instead of scsi host workqueue,
> because we may block sas event work, we cannot block the normal scsi works.
> When libsas receive a phy down event, sas_deform_port would be called, and now we
> block sas_deform_port and wait for destruction work finish, in sas_destruct_devices,
> we may wait ata error handler, it would take a long time, so if do all stuff in scsi
> host workq, libsas may block other scsi works too long.
> 2. create a new workqueue used to run sas discovery events work, instead of scsi host
> workqueue, because in some cases, eg. in revalidate domain event, we may unregister
> a sas device and discover new one, we must sync the execution, wait the remove process
> finish, then start a new discovery. So we must put the probe and destruct discovery
> events in a new workqueue to avoid deadlock.
> 3. introudce a asd_sas_port level wait-complete and a sas_discovery level wait-complete
> we use former wait-complete to achieve a sas event atomic process and use latter to
> make a sas discovery sync.
> 4. remove disco_mutex in sas_revalidate_domain, since now sas_revalidate_domain sync
> the destruct discovery event execution, it's no need to lock disco mutex there.
The way you've written the changelog suggests this patch should be split
into 4 patches, each one taking care of one of your change items.
--
Johannes Thumshirn Storage
[email protected] +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850
>> static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event event)
>> {
>> + struct sas_ha_event *ev;
>> +
>> BUG_ON(event >= HA_NUM_EVENTS);
>>
>> - sas_queue_event(event, &sas_ha->pending,
>> - &sas_ha->ha_events[event].work, sas_ha);
>> + ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
>> + if (!ev)
>> + return;
> GFP_ATOMIC allocations can fail and then no events will be queued *and* we
> don't report the error back to the caller.
>
Yes, it's really a problem, but I don't find a better solution, do you have some suggestion ?
>
>
>> index 64e9cdd..c227a8b 100644
>> --- a/drivers/scsi/libsas/sas_init.c
>> +++ b/drivers/scsi/libsas/sas_init.c
>> @@ -111,10 +111,6 @@ void sas_hash_addr(u8 *hashed, const u8 *sas_addr)
>>
>> void sas_hae_reset(struct work_struct *work)
>> {
>> - struct sas_ha_event *ev = to_sas_ha_event(work);
>> - struct sas_ha_struct *ha = ev->ha;
>> -
>> - clear_bit(HAE_RESET, &ha->pending);
>> }
>
> I don't really get why you need a stubbed out sas_hae_reset(). Can't we
> just kill it if it doesn't have anything left to do?
I have no idea about this function history, I agree clean it out.
>
>
>> In this patch, we try to solve these issues in following steps:
>> 1. create a new workqueue used to run sas event work, instead of scsi host workqueue,
>> because we may block sas event work, we cannot block the normal scsi works.
>> When libsas receive a phy down event, sas_deform_port would be called, and now we
>> block sas_deform_port and wait for destruction work finish, in sas_destruct_devices,
>> we may wait ata error handler, it would take a long time, so if do all stuff in scsi
>> host workq, libsas may block other scsi works too long.
>> 2. create a new workqueue used to run sas discovery events work, instead of scsi host
>> workqueue, because in some cases, eg. in revalidate domain event, we may unregister
>> a sas device and discover new one, we must sync the execution, wait the remove process
>> finish, then start a new discovery. So we must put the probe and destruct discovery
>> events in a new workqueue to avoid deadlock.
>> 3. introudce a asd_sas_port level wait-complete and a sas_discovery level wait-complete
>> we use former wait-complete to achieve a sas event atomic process and use latter to
>> make a sas discovery sync.
>> 4. remove disco_mutex in sas_revalidate_domain, since now sas_revalidate_domain sync
>> the destruct discovery event execution, it's no need to lock disco mutex there.
>
> The way you've written the changelog suggests this patch should be split
> into 4 patches, each one taking care of one of your change items.
I will split it in next version.
Thanks!
Yijing.
>
On 06/14/2017 11:04 AM, wangyijing wrote:
>>> static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event event)
>>> {
>>> + struct sas_ha_event *ev;
>>> +
>>> BUG_ON(event >= HA_NUM_EVENTS);
>>>
>>> - sas_queue_event(event, &sas_ha->pending,
>>> - &sas_ha->ha_events[event].work, sas_ha);
>>> + ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
>>> + if (!ev)
>>> + return;
>> GFP_ATOMIC allocations can fail and then no events will be queued *and* we
>> don't report the error back to the caller.
>>
>
> Yes, it's really a problem, but I don't find a better solution, do you have some suggestion ?
Something like this?
diff --git a/drivers/scsi/hisi_sas/hisi_sas_main.c
b/drivers/scsi/hisi_sas/hisi_sas_main.c
index d622db502ec9..27cd7307d308 100644
--- a/drivers/scsi/hisi_sas/hisi_sas_main.c
+++ b/drivers/scsi/hisi_sas/hisi_sas_main.c
@@ -894,7 +894,11 @@ static int hisi_sas_controller_reset(struct
hisi_hba *hisi_hba)
hisi_sas_release_tasks(hisi_hba);
spin_unlock_irqrestore(&hisi_hba->lock, flags);
- sas_ha->notify_ha_event(sas_ha, HAE_RESET);
+ rc = sas_ha->notify_ha_event(sas_ha, HAE_RESET);
+ if (rc) {
+ dev_warn(dev, "failed to queue HA event (%d)\n",
rc);
+ goto out;
+ }
dev_dbg(dev, "controller reset successful!\n");
} else
return -1;
diff --git a/drivers/scsi/libsas/sas_event.c
b/drivers/scsi/libsas/sas_event.c
index aadbd5314c5c..6c91fb31f891 100644
--- a/drivers/scsi/libsas/sas_event.c
+++ b/drivers/scsi/libsas/sas_event.c
@@ -116,7 +116,7 @@ void sas_enable_revalidation(struct sas_ha_struct *ha)
mutex_unlock(&ha->disco_mutex);
}
-static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event
event)
+static int notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event
event)
{
BUG_ON(event >= HA_NUM_EVENTS);
diff --git a/include/scsi/libsas.h b/include/scsi/libsas.h
index dd0f72c95abe..5d9cd6d20855 100644
--- a/include/scsi/libsas.h
+++ b/include/scsi/libsas.h
@@ -415,7 +415,7 @@ struct sas_ha_struct {
* their siblings when forming wide ports */
/* LLDD calls these to notify the class of an event. */
- void (*notify_ha_event)(struct sas_ha_struct *, enum ha_event);
+ int (*notify_ha_event)(struct sas_ha_struct *, enum ha_event);
void (*notify_port_event)(struct asd_sas_phy *, enum port_event);
void (*notify_phy_event)(struct asd_sas_phy *, enum phy_event);
--
Johannes Thumshirn Storage
[email protected] +49 911 74053 689
SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg
GF: Felix Imendörffer, Jane Smithard, Graham Norton
HRB 21284 (AG Nürnberg)
Key fingerprint = EC38 9CAB C2C4 F25D 8600 D0D0 0393 969D 2D76 0850
在 2017/6/14 17:18, Johannes Thumshirn 写道:
> On 06/14/2017 11:04 AM, wangyijing wrote:
>>>> static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event event)
>>>> {
>>>> + struct sas_ha_event *ev;
>>>> +
>>>> BUG_ON(event >= HA_NUM_EVENTS);
>>>>
>>>> - sas_queue_event(event, &sas_ha->pending,
>>>> - &sas_ha->ha_events[event].work, sas_ha);
>>>> + ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
>>>> + if (!ev)
>>>> + return;
>>> GFP_ATOMIC allocations can fail and then no events will be queued *and* we
>>> don't report the error back to the caller.
>>>
>>
>> Yes, it's really a problem, but I don't find a better solution, do you have some suggestion ?
>
> Something like this?
>
> diff --git a/drivers/scsi/hisi_sas/hisi_sas_main.c
> b/drivers/scsi/hisi_sas/hisi_sas_main.c
> index d622db502ec9..27cd7307d308 100644
> --- a/drivers/scsi/hisi_sas/hisi_sas_main.c
> +++ b/drivers/scsi/hisi_sas/hisi_sas_main.c
> @@ -894,7 +894,11 @@ static int hisi_sas_controller_reset(struct
> hisi_hba *hisi_hba)
> hisi_sas_release_tasks(hisi_hba);
> spin_unlock_irqrestore(&hisi_hba->lock, flags);
>
> - sas_ha->notify_ha_event(sas_ha, HAE_RESET);
> + rc = sas_ha->notify_ha_event(sas_ha, HAE_RESET);
> + if (rc) {
> + dev_warn(dev, "failed to queue HA event (%d)\n",
> rc);
> + goto out;
> + }
> dev_dbg(dev, "controller reset successful!\n");
> } else
> return -1;
> diff --git a/drivers/scsi/libsas/sas_event.c
> b/drivers/scsi/libsas/sas_event.c
> index aadbd5314c5c..6c91fb31f891 100644
> --- a/drivers/scsi/libsas/sas_event.c
> +++ b/drivers/scsi/libsas/sas_event.c
> @@ -116,7 +116,7 @@ void sas_enable_revalidation(struct sas_ha_struct *ha)
> mutex_unlock(&ha->disco_mutex);
> }
>
> -static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event
> event)
> +static int notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event
> event)
> {
> BUG_ON(event >= HA_NUM_EVENTS);
>
> diff --git a/include/scsi/libsas.h b/include/scsi/libsas.h
> index dd0f72c95abe..5d9cd6d20855 100644
> --- a/include/scsi/libsas.h
> +++ b/include/scsi/libsas.h
> @@ -415,7 +415,7 @@ struct sas_ha_struct {
> * their siblings when forming wide ports */
>
> /* LLDD calls these to notify the class of an event. */
> - void (*notify_ha_event)(struct sas_ha_struct *, enum ha_event);
> + int (*notify_ha_event)(struct sas_ha_struct *, enum ha_event);
> void (*notify_port_event)(struct asd_sas_phy *, enum port_event);
> void (*notify_phy_event)(struct asd_sas_phy *, enum phy_event);
This make sense, return a fail status and report it to the callers.
>
>
On 14/06/2017 10:04, wangyijing wrote:
>>> static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event event)
>>> >> {
>>> >> + struct sas_ha_event *ev;
>>> >> +
>>> >> BUG_ON(event >= HA_NUM_EVENTS);
>>> >>
>>> >> - sas_queue_event(event, &sas_ha->pending,
>>> >> - &sas_ha->ha_events[event].work, sas_ha);
>>> >> + ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
>>> >> + if (!ev)
>>> >> + return;
>> > GFP_ATOMIC allocations can fail and then no events will be queued *and* we
>> > don't report the error back to the caller.
>> >
> Yes, it's really a problem, but I don't find a better solution, do you have some suggestion ?
>
Dan raised an issue with this approach, regarding a malfunctioning PHY
which spews out events. I still don't think we're handling it safely.
Here's the suggestion:
- each asd_sas_phy owns a finite-sized pool of events
- when the event pool becomes exhausted, libsas stops queuing events
(obviously) and disables the PHY in the LLDD
- upon attempting to re-enable the PHY from sysfs, libsas first checks
that the pool is still not exhausted
If you cannot find a good solution, then let us know and we can help.
John
在 2017/6/14 21:08, John Garry 写道:
> On 14/06/2017 10:04, wangyijing wrote:
>>>> static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event event)
>>>> >> {
>>>> >> + struct sas_ha_event *ev;
>>>> >> +
>>>> >> BUG_ON(event >= HA_NUM_EVENTS);
>>>> >>
>>>> >> - sas_queue_event(event, &sas_ha->pending,
>>>> >> - &sas_ha->ha_events[event].work, sas_ha);
>>>> >> + ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
>>>> >> + if (!ev)
>>>> >> + return;
>>> > GFP_ATOMIC allocations can fail and then no events will be queued *and* we
>>> > don't report the error back to the caller.
>>> >
>> Yes, it's really a problem, but I don't find a better solution, do you have some suggestion ?
>>
>
> Dan raised an issue with this approach, regarding a malfunctioning PHY which spews out events. I still don't think we're handling it safely. Here's the suggestion:
> - each asd_sas_phy owns a finite-sized pool of events
> - when the event pool becomes exhausted, libsas stops queuing events (obviously) and disables the PHY in the LLDD
> - upon attempting to re-enable the PHY from sysfs, libsas first checks that the pool is still not exhausted
>
> If you cannot find a good solution, then let us know and we can help.
Hi John and Dan, what's event you found on malfunctioning PHY, if the event is PORTE_BROADCAST_RCVD, since
every PORTE_BROADCAST_RCVD libsas always call sas_revalidate_domain(), what about keeping a broadcast waiting(not queued in workqueue)
and discard others. If the event is other types, things may become knotty.
>
> John
>
>
> .
>
On 15/06/2017 08:37, wangyijing wrote:
>
>
> 在 2017/6/14 21:08, John Garry 写道:
>> On 14/06/2017 10:04, wangyijing wrote:
>>>>> static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event event)
>>>>>>> {
>>>>>>> + struct sas_ha_event *ev;
>>>>>>> +
>>>>>>> BUG_ON(event >= HA_NUM_EVENTS);
>>>>>>>
>>>>>>> - sas_queue_event(event, &sas_ha->pending,
>>>>>>> - &sas_ha->ha_events[event].work, sas_ha);
>>>>>>> + ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
>>>>>>> + if (!ev)
>>>>>>> + return;
>>>>> GFP_ATOMIC allocations can fail and then no events will be queued *and* we
>>>>> don't report the error back to the caller.
>>>>>
>>> Yes, it's really a problem, but I don't find a better solution, do you have some suggestion ?
>>>
>>
>> Dan raised an issue with this approach, regarding a malfunctioning PHY which spews out events. I still don't think we're handling it safely. Here's the suggestion:
>> - each asd_sas_phy owns a finite-sized pool of events
>> - when the event pool becomes exhausted, libsas stops queuing events (obviously) and disables the PHY in the LLDD
>> - upon attempting to re-enable the PHY from sysfs, libsas first checks that the pool is still not exhausted
>>
>> If you cannot find a good solution, then let us know and we can help.
>
> Hi John and Dan, what's event you found on malfunctioning PHY, if the event is PORTE_BROADCAST_RCVD, since
> every PORTE_BROADCAST_RCVD libsas always call sas_revalidate_domain(), what about keeping a broadcast waiting(not queued in workqueue)
> and discard others. If the event is other types, things may become knotty.
>
As I mentioned in the v1 series discussion, I found a poorly connected
expander PHY was spewing out PHY up and loss of signal events
continuously. This is the sort of situation we should protect against.
Current solution is ok, as it uses a static event per port/PHY/HA.
The point is that we cannot allow a PHY to continuously send events to
libsas, which may lead to memory exhaustion.
John
>
>>
>> John
>>
>>
>> .
>>
>
>
> .
>
在 2017/6/15 16:00, John Garry 写道:
> On 15/06/2017 08:37, wangyijing wrote:
>>
>>
>> 在 2017/6/14 21:08, John Garry 写道:
>>> On 14/06/2017 10:04, wangyijing wrote:
>>>>>> static void notify_ha_event(struct sas_ha_struct *sas_ha, enum ha_event event)
>>>>>>>> {
>>>>>>>> + struct sas_ha_event *ev;
>>>>>>>> +
>>>>>>>> BUG_ON(event >= HA_NUM_EVENTS);
>>>>>>>>
>>>>>>>> - sas_queue_event(event, &sas_ha->pending,
>>>>>>>> - &sas_ha->ha_events[event].work, sas_ha);
>>>>>>>> + ev = kzalloc(sizeof(*ev), GFP_ATOMIC);
>>>>>>>> + if (!ev)
>>>>>>>> + return;
>>>>>> GFP_ATOMIC allocations can fail and then no events will be queued *and* we
>>>>>> don't report the error back to the caller.
>>>>>>
>>>> Yes, it's really a problem, but I don't find a better solution, do you have some suggestion ?
>>>>
>>>
>>> Dan raised an issue with this approach, regarding a malfunctioning PHY which spews out events. I still don't think we're handling it safely. Here's the suggestion:
>>> - each asd_sas_phy owns a finite-sized pool of events
>>> - when the event pool becomes exhausted, libsas stops queuing events (obviously) and disables the PHY in the LLDD
>>> - upon attempting to re-enable the PHY from sysfs, libsas first checks that the pool is still not exhausted
>>>
>>> If you cannot find a good solution, then let us know and we can help.
>>
>> Hi John and Dan, what's event you found on malfunctioning PHY, if the event is PORTE_BROADCAST_RCVD, since
>> every PORTE_BROADCAST_RCVD libsas always call sas_revalidate_domain(), what about keeping a broadcast waiting(not queued in workqueue)
>> and discard others. If the event is other types, things may become knotty.
>>
>
> As I mentioned in the v1 series discussion, I found a poorly connected expander PHY was spewing out PHY up and loss of signal events continuously. This is the sort of situation we should protect against. Current solution is ok, as it uses a static event per port/PHY/HA.
>
> The point is that we cannot allow a PHY to continuously send events to libsas, which may lead to memory exhaustion.
The current solution won't introduce memory exhaustion, but it's not ok, since the root of this issue is it may lost event which is normal.
If we cannot identify the abnormal PHY, I think your mem pool idea is a candidate solution.
>
> John
>
>>
>>>
>>> John
>>>
>>>
>>> .
>>>
>>
>>
>> .
>>
>
>
>
> .
>