All of lore.kernel.org
 help / color / mirror / Atom feed
From: Tejun Heo <htejun@gmail.com>
To: jgarzik@pobox.com, lkml@rtr.ca, axboe@suse.de,
	forrest.zhao@intel.com, alan@lxorguk.ukuu.org.uk,
	linux-ide@vger.kernel.org
Cc: Tejun Heo <htejun@gmail.com>
Subject: [PATCH 1/3] libata: implement hotplug by polling
Date: Tue, 4 Jul 2006 23:16:17 +0900	[thread overview]
Message-ID: <11520225773310-git-send-email-htejun@gmail.com> (raw)
In-Reply-To: <11520225771424-git-send-email-htejun@gmail.com>

This patch implements hotplug by polling - hp-poll.  It is used for

* hotplug event detection on controllers which don't provide PHY
  status changed interrupt.

* hotplug event detection on disabled ports after reset failure.
  libata used to leave such ports in frozen state to protect the
  system from malfunctioning controller/device.  With hp-poll, hotplug
  events no such ports are watched safely by polling, such that
  removing/replacing the malfunctioning device triggers EH retry on
  the port.

There are three port ops for hp-poll - hp_poll_activate, hp_poll,
hp_poll_deactivate.  Only hp_poll is mandatory for hp-poll to work.
This patch also implements SATA standard polling callbacks which poll
SError.N/X bits - sata_std_hp_poll_activate(), sata_std_hp_poll(),
sata_std_hp_poll_port_excl() and sata_std_hp_poll_hostset_excl().

By default, hp-poll is enabled only on disabled ports.  If a LLD
doesn't support hotplug interrupts but can poll for hotplug events, it
should indicate it by setting ATA_FLAG_HP_POLLING which tells libata
to turn on hp-poll by default.

Signed-off-by: Tejun Heo <htejun@gmail.com>

---

 drivers/scsi/libata-core.c |  134 ++++++++++++++++++++++++++++++++++++++++++++
 drivers/scsi/libata-eh.c   |  134 +++++++++++++++++++++++++++++++++++++++++++-
 drivers/scsi/libata.h      |    5 ++
 include/linux/libata.h     |   12 ++++
 4 files changed, 283 insertions(+), 2 deletions(-)

7af09db1d75b71f1e45ec9ea1163edab03af56c9
diff --git a/drivers/scsi/libata-core.c b/drivers/scsi/libata-core.c
index 46ae4bc..8ed0c37 100644
--- a/drivers/scsi/libata-core.c
+++ b/drivers/scsi/libata-core.c
@@ -2840,6 +2840,126 @@ void ata_std_postreset(struct ata_port *
 }
 
 /**
+ *	sata_std_hp_poll_activate - standard SATA hotplug polling activation
+ *	@ap: the target ata_port
+ *
+ *	Activate SATA std hotplug polling.
+ *
+ *	LOCKING:
+ *	Kernel thread context (may sleep).
+ */
+void sata_std_hp_poll_activate(struct ata_port *ap)
+{
+	u32 serror;
+
+	sata_scr_read(ap, SCR_ERROR, &serror);
+	serror &= SERR_PHYRDY_CHG | SERR_DEV_XCHG;
+	if (serror)
+		sata_scr_write(ap, SCR_ERROR, serror);
+
+	ap->hp_poll_data = 0;
+}
+
+/**
+ *	sata_std_hp_poll - standard SATA hotplug polling callback
+ *	@ap: the target ata_port
+ *
+ *	Poll SError.N/X for hotplug event.  Hotplug event is triggered
+ *	only after PHY stays stable for at least one full polling
+ *	interval after the first event detection.
+ *
+ *	LOCKING:
+ *	spin_lock_irqsave(host_set lock)
+ *
+ *	RETURNS:
+ *	1 if hotplug event has occurred, 0 otherwise.
+ */
+int sata_std_hp_poll(struct ata_port *ap)
+{
+	unsigned long state = (unsigned long)ap->hp_poll_data;
+	u32 serror;
+	int rc = 0;
+
+	sata_scr_read(ap, SCR_ERROR, &serror);
+	serror &= SERR_PHYRDY_CHG | SERR_DEV_XCHG;
+
+	switch (state) {
+	case 0:
+		if (serror) {
+			/* PHY status could be bouncing crazy due to
+			 * faulty controller or device.  Don't fire
+			 * hotplug event till hotplug event stays
+			 * quiescent for one full polling interval.
+			 */
+			sata_scr_write(ap, SCR_ERROR, serror);
+			state = 1;
+		}
+		break;
+
+	case 1:
+		if (!serror)
+			rc = 1;
+		else
+			sata_scr_write(ap, SCR_ERROR, serror);
+		break;
+	}
+
+	ap->hp_poll_data = (void *)(unsigned long)state;
+
+	return rc;
+}
+
+/**
+ *	sata_std_hp_poll_port_excl - sata_std_hp_poll with port exclusion
+ *	@ap: the target ata_port
+ *
+ *	SATA standard hp_poll callback for controllers which don't
+ *	allow access to SCR registers while the port is active.
+ *
+ *	LOCKING:
+ *	spin_lock_irqsave(host_set lock)
+ *
+ *	RETURNS:
+ *	1 if hotplug event has occurred, 0 otherwise.
+ */
+int sata_std_hp_poll_port_excl(struct ata_port *ap)
+{
+	if (ap->qc_active)
+		return 0;
+
+	return sata_std_hp_poll(ap);
+}
+
+/**
+ *	sata_std_hp_poll_hostset_excl - sata_std_hp_poll with host_set exclusion
+ *	@ap: the target ata_port
+ *
+ *	SATA standard hp_poll callback for controllers which don't
+ *	allow access to SCR registers while any port in the host_set
+ *	is active.  How dumb can you get?  Surprise, VIA did it.
+ *
+ *	LOCKING:
+ *	spin_lock_irqsave(host_set lock)
+ *
+ *	RETURNS:
+ *	1 if hotplug event has occurred, 0 otherwise.
+ */
+int sata_std_hp_poll_hostset_excl(struct ata_port *ap)
+{
+	struct ata_host_set *host_set = ap->host_set;
+	int i;
+
+	for (i = 0; i < host_set->n_ports; i++) {
+		struct ata_port *t_ap = host_set->ports[i];
+
+		if (t_ap->qc_active)
+			return 0;
+	}
+
+	return sata_std_hp_poll(ap);
+}
+
+/**
  *	ata_dev_same_device - Determine whether new ID matches configured device
  *	@dev: device to compare against
  *	@new_class: class of the new device
@@ -5290,6 +5410,7 @@ #endif
 	INIT_WORK(&ap->scsi_rescan_task, ata_scsi_dev_rescan, ap);
 	INIT_LIST_HEAD(&ap->eh_done_q);
 	init_waitqueue_head(&ap->eh_wait_q);
+	INIT_LIST_HEAD(&ap->hp_poll_entry);
 
 	/* set cable type */
 	ap->cbl = ATA_CBL_NONE;
@@ -5586,6 +5707,9 @@ void ata_port_detach(struct ata_port *ap
 	cancel_delayed_work(&ap->hotplug_task);
 	flush_workqueue(ata_aux_wq);
 
+	/* deactivate hotplug polling */
+	ata_hp_poll_deactivate(ap);
+
 	/* remove the associated SCSI host */
 	scsi_remove_host(ap->host);
 }
@@ -5832,6 +5956,12 @@ static int __init ata_init(void)
 
 static void __exit ata_exit(void)
 {
+	/* hp_poll_list gotta be empty by now and thus hp_poll_work
+	 * isn't rearming.
+	 */
+	WARN_ON(!list_empty(&hp_poll_list));
+	cancel_delayed_work(&hp_poll_work);
+
 	destroy_workqueue(ata_wq);
 	destroy_workqueue(ata_aux_wq);
 }
@@ -5966,6 +6096,10 @@ EXPORT_SYMBOL_GPL(ata_std_prereset);
 EXPORT_SYMBOL_GPL(ata_std_softreset);
 EXPORT_SYMBOL_GPL(sata_std_hardreset);
 EXPORT_SYMBOL_GPL(ata_std_postreset);
+EXPORT_SYMBOL_GPL(sata_std_hp_poll_activate);
+EXPORT_SYMBOL_GPL(sata_std_hp_poll);
+EXPORT_SYMBOL_GPL(sata_std_hp_poll_port_excl);
+EXPORT_SYMBOL_GPL(sata_std_hp_poll_hostset_excl);
 EXPORT_SYMBOL_GPL(ata_dev_revalidate);
 EXPORT_SYMBOL_GPL(ata_dev_classify);
 EXPORT_SYMBOL_GPL(ata_dev_pair);
diff --git a/drivers/scsi/libata-eh.c b/drivers/scsi/libata-eh.c
index 4b6aa30..bd89608 100644
--- a/drivers/scsi/libata-eh.c
+++ b/drivers/scsi/libata-eh.c
@@ -34,6 +34,8 @@
 
 #include <linux/config.h>
 #include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
 #include <scsi/scsi.h>
 #include <scsi/scsi_host.h>
 #include <scsi/scsi_eh.h>
@@ -45,6 +47,17 @@ #include <linux/libata.h>
 
 #include "libata.h"
 
+static unsigned long hotplug_polling_interval = 2000;
+module_param(hotplug_polling_interval, ulong, 0644);
+MODULE_PARM_DESC(hotplug_polling_interval,
+		 "Hotplug polling interval in milliseconds (default 2000)");
+
+static void ata_hp_poll_worker(void *data);
+
+static DEFINE_MUTEX(hp_poll_mutex);
+struct list_head hp_poll_list = LIST_HEAD_INIT(hp_poll_list);
+DECLARE_WORK(hp_poll_work, ata_hp_poll_worker, NULL);
+
 static void __ata_port_freeze(struct ata_port *ap);
 static void ata_eh_finish(struct ata_port *ap);
 static void ata_eh_handle_port_suspend(struct ata_port *ap);
@@ -617,7 +630,10 @@ int ata_port_freeze(struct ata_port *ap)
  *	ata_eh_freeze_port - EH helper to freeze port
  *	@ap: ATA port to freeze
  *
- *	Freeze @ap.
+ *	Freeze @ap.  As the 'freeze' operation means 'shutdown event
+ *	reporting', it is a perfect place to deactivate hp-poll.  Note
+ *	that ata_port_freeze() always invokes EH and eventually this
+ *	function, so deactivating hp-poll in this function is enough.
  *
  *	LOCKING:
  *	None.
@@ -629,6 +645,8 @@ void ata_eh_freeze_port(struct ata_port 
 	if (!ap->ops->error_handler)
 		return;
 
+	ata_hp_poll_deactivate(ap);
+
 	spin_lock_irqsave(ap->lock, flags);
 	__ata_port_freeze(ap);
 	spin_unlock_irqrestore(ap->lock, flags);
@@ -638,7 +656,7 @@ void ata_eh_freeze_port(struct ata_port 
  *	ata_port_thaw_port - EH helper to thaw port
  *	@ap: ATA port to thaw
  *
- *	Thaw frozen port @ap.
+ *	Thaw frozen port @ap and activate hp-poll if necessary.
  *
  *	LOCKING:
  *	None.
@@ -659,6 +677,9 @@ void ata_eh_thaw_port(struct ata_port *a
 
 	spin_unlock_irqrestore(ap->lock, flags);
 
+	if (ap->flags & ATA_FLAG_HP_POLLING)
+		ata_hp_poll_activate(ap);
+
 	DPRINTK("ata%u port thawed\n", ap->id);
 }
 
@@ -2037,6 +2058,9 @@ static int ata_eh_recover(struct ata_por
 
  out:
 	if (rc) {
+		/* recovery failed, activate hp-poll */
+		ata_hp_poll_activate(ap);
+
 		for (i = 0; i < ATA_MAX_DEVICES; i++)
 			ata_dev_disable(&ap->device[i]);
 	}
@@ -2225,3 +2249,109 @@ static void ata_eh_handle_port_resume(st
 	}
 	spin_unlock_irqrestore(ap->lock, flags);
 }
+
+static void ata_hp_poll_worker(void *data)
+{
+	unsigned long flags = 0;	/* shut up, gcc */
+	struct ata_port *ap, *next;
+	spinlock_t *held_lock = NULL;
+
+	mutex_lock(&hp_poll_mutex);
+
+	list_for_each_entry_safe(ap, next, &hp_poll_list, hp_poll_entry) {
+		int rc;
+
+		/* Ports belonging to sharing a lock are registered
+		 * sequentially.  Cache locking.
+		 */
+		if (ap->lock != held_lock) {
+			if (held_lock)
+				spin_unlock_irqrestore(held_lock, flags);
+			spin_lock_irqsave(ap->lock, flags);
+			held_lock = ap->lock;
+		}
+
+		/* Poll.  Positive return value indicates hotplug
+		 * event while negative indicates error condition.
+		 */
+		rc = ap->ops->hp_poll(ap);
+		if (rc) {
+			if (rc > 0) {
+				ata_ehi_hotplugged(&ap->eh_info);
+				ata_port_freeze(ap);
+			}
+			list_del_init(&ap->hp_poll_entry);
+		}
+	}
+
+	if (held_lock)
+		spin_unlock_irqrestore(held_lock, flags);
+
+	if (!list_empty(&hp_poll_list))
+		schedule_delayed_work(&hp_poll_work,
+				      hotplug_polling_interval * HZ / 1000);
+
+	mutex_unlock(&hp_poll_mutex);
+}
+
+/**
+ *	ata_hp_poll_activate - activate hotplug polling
+ *	@ap: host port to activate hotplug polling for
+ *
+ *	Activate hotplug probing for @ap.
+ *
+ *	LOCKING:
+ *	Kernel thread context (may sleep).
+ */
+void ata_hp_poll_activate(struct ata_port *ap)
+{
+	struct list_head *pos;
+
+	if (!ap->ops->hp_poll || !list_empty(&ap->hp_poll_entry))
+		return;
+
+	if (ap->ops->hp_poll_activate)
+		ap->ops->hp_poll_activate(ap);
+
+	mutex_lock(&hp_poll_mutex);
+
+	if (list_empty(&hp_poll_list))
+		schedule_delayed_work(&hp_poll_work,
+				      hotplug_polling_interval * HZ / 1000);
+
+	/* group poll entries by lock to cache locking when polling */
+	list_for_each(pos, &hp_poll_list) {
+		struct ata_port *pos_ap =
+			list_entry(pos, struct ata_port, hp_poll_entry);
+
+		if (ap->lock == pos_ap->lock)
+			break;
+	}
+
+	list_add(&ap->hp_poll_entry, pos);
+	ap->hp_poll_data = 0;
+
+	mutex_unlock(&hp_poll_mutex);
+}
+
+/**
+ *	ata_hp_poll_deactivate - deactivate hotplug polling
+ *	@ap: host port to deactivate hotplug polling for
+ *
+ *	Deactivate hotplug probing for @ap.
+ *
+ *	LOCKING:
+ *	Kernel thread context (may sleep).
+ */
+void ata_hp_poll_deactivate(struct ata_port *ap)
+{
+	if (list_empty(&ap->hp_poll_entry))
+		return;
+
+	mutex_lock(&hp_poll_mutex);
+	list_del_init(&ap->hp_poll_entry);
+	mutex_unlock(&hp_poll_mutex);
+
+	if (ap->ops->hp_poll_deactivate)
+		ap->ops->hp_poll_deactivate(ap);
+}
diff --git a/drivers/scsi/libata.h b/drivers/scsi/libata.h
index c325679..3171dde 100644
--- a/drivers/scsi/libata.h
+++ b/drivers/scsi/libata.h
@@ -109,9 +109,14 @@ extern void ata_schedule_scsi_eh(struct 
 extern void ata_scsi_dev_rescan(void *data);
 
 /* libata-eh.c */
+extern struct list_head hp_poll_list;
+extern struct work_struct hp_poll_work;
+
 extern enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd);
 extern void ata_scsi_error(struct Scsi_Host *host);
 extern void ata_port_wait_eh(struct ata_port *ap);
 extern void ata_qc_schedule_eh(struct ata_queued_cmd *qc);
+extern void ata_hp_poll_activate(struct ata_port *ap);
+extern void ata_hp_poll_deactivate(struct ata_port *ap);
 
 #endif /* __LIBATA_H__ */
diff --git a/include/linux/libata.h b/include/linux/libata.h
index 6cc497a..2ceca5f 100644
--- a/include/linux/libata.h
+++ b/include/linux/libata.h
@@ -162,6 +162,7 @@ enum {
 	ATA_FLAG_SKIP_D2H_BSY	= (1 << 12), /* can't wait for the first D2H
 					      * Register FIS clearing BSY */
 	ATA_FLAG_DEBUGMSG	= (1 << 13),
+	ATA_FLAG_HP_POLLING	= (1 << 14), /* hotplug by polling */
 
 	/* The following flag belongs to ap->pflags but is kept in
 	 * ap->flags because it's referenced in many LLDs and will be
@@ -553,6 +554,9 @@ struct ata_port {
 	pm_message_t		pm_mesg;
 	int			*pm_result;
 
+	struct list_head	hp_poll_entry;	/* hotplug poll list entry */
+	void			*hp_poll_data;
+
 	void			*private_data;
 
 	u8			sector_buf[ATA_SECT_SIZE]; /* owned by EH */
@@ -600,6 +604,10 @@ struct ata_port_operations {
 	void (*error_handler) (struct ata_port *ap);
 	void (*post_internal_cmd) (struct ata_queued_cmd *qc);
 
+	void (*hp_poll_activate) (struct ata_port *ap);
+	void (*hp_poll_deactivate) (struct ata_port *ap);
+	int (*hp_poll) (struct ata_port *ap);
+
 	irqreturn_t (*irq_handler)(int, void *, struct pt_regs *);
 	void (*irq_clear) (struct ata_port *);
 
@@ -667,6 +675,10 @@ extern int ata_std_prereset(struct ata_p
 extern int ata_std_softreset(struct ata_port *ap, unsigned int *classes);
 extern int sata_std_hardreset(struct ata_port *ap, unsigned int *class);
 extern void ata_std_postreset(struct ata_port *ap, unsigned int *classes);
+extern void sata_std_hp_poll_activate(struct ata_port *ap);
+extern int sata_std_hp_poll(struct ata_port *ap);
+extern int sata_std_hp_poll_port_excl(struct ata_port *ap);
+extern int sata_std_hp_poll_hostset_excl(struct ata_port *ap);
 extern int ata_dev_revalidate(struct ata_device *dev, int post_reset);
 extern void ata_port_disable(struct ata_port *);
 extern void ata_std_ports(struct ata_ioports *ioaddr);
-- 
1.3.2



  reply	other threads:[~2006-07-04 14:15 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2006-07-04 14:16 [RFT][PATCHSET] hotplug polling Tejun Heo
2006-07-04 14:16 ` Tejun Heo [this message]
2006-07-04 14:16 ` [PATCH 2/3] libata: add hp-poll support to controllers with hotplug interrutps Tejun Heo
2006-07-04 14:16 ` [PATCH 3/3] libata: add hp-poll support to controllers without hotplug interrupts Tejun Heo
2006-07-04 20:28 ` [RFT][PATCHSET] hotplug polling Tejun Heo
  -- strict thread matches above, loose matches on Subject: below --
2006-07-05  6:06 [RFT][PATCHSET] hotplug polling, take 2 Tejun Heo
2006-07-05  6:06 ` [PATCH 1/3] libata: implement hotplug by polling Tejun Heo
2006-07-17  7:00 [RFT][PATCHSET] hotplug polling, take 3 Tejun Heo
2006-07-17  7:00 ` [PATCH 1/3] libata: implement hotplug by polling Tejun Heo
2006-07-19 20:57   ` Jeff Garzik
2006-07-24  7:06     ` Tejun Heo
2006-10-10  5:36 [PATCHSET] hotplug polling, take 4 Tejun Heo
2006-10-10  5:36 ` [PATCH 1/3] libata: implement hotplug by polling Tejun Heo
2006-10-15 22:37 [PATCHSET] hotplug polling, take 5 Tejun Heo
2006-10-15 22:37 ` [PATCH 1/3] libata: implement hotplug by polling Tejun Heo
2007-04-25  6:25 [PATCH 0/3] hotplug polling, respin Robin H. Johnson
2007-04-25  6:28 ` [PATCH 1/3] libata: implement hotplug by polling Robin H. Johnson

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=11520225773310-git-send-email-htejun@gmail.com \
    --to=htejun@gmail.com \
    --cc=alan@lxorguk.ukuu.org.uk \
    --cc=axboe@suse.de \
    --cc=forrest.zhao@intel.com \
    --cc=jgarzik@pobox.com \
    --cc=linux-ide@vger.kernel.org \
    --cc=lkml@rtr.ca \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.