All of lore.kernel.org
 help / color / mirror / Atom feed
From: Joe Lawrence <joe.lawrence@stratus.com>
To: linux-scsi@vger.kernel.org
Cc: Chad Dupuis <chad.dupuis@qlogic.com>,
	Giridhar Malavali <giridhar.malavali@qlogic.com>,
	Saurav Kashyap <saurav.kashyap@qlogic.com>,
	Don Zickus <dzickus@redhat.com>,
	Prarit Bhargava <prarit@redhat.com>,
	Bill Kuzeja <william.kuzeja@stratus.com>,
	Dave Bulkow <david.bulkow@stratus.com>,
	Joe Lawrence <joe.lawrence@stratus.com>
Subject: [PATCH 5/6] qla2xxx: Prevent removal and board_disable race
Date: Wed, 18 Jun 2014 10:04:43 -0400	[thread overview]
Message-ID: <1403100284-1872-1-git-send-email-joe.lawrence@stratus.com> (raw)
In-Reply-To: <1403100139-1798-1-git-send-email-joe.lawrence@stratus.com>

Introduce mutual exclusion between the qla2xxx_remove_one PCI driver
callback and qla2x00_disable_board_on_pci_error, which is scheduled as
board_disable work by qla2x00_check_reg{32,16}_for_disconnect:

* Leave the driver-specific data attached to the underlying PCI device
intact in qla2x00_disable_board_on_pci_error, so that qla2x00_remove_one
has enough breadcrumbs to determine that any board_disable work has been
completed.

* In qla2xxx_remove_one, set a bit to prevent any subsequent
board_disable work from scheduling, then cancel and wait until pending
work has completed.

* Reuse the PCI device enable count check in qla2x00_remove_one to
determine if board_disable has occured.  The original purpose of this
check was unnecessary since the driver remove function wasn't called
when the probe fails.

Signed-off-by: Joe Lawrence <joe.lawrence@stratus.com>
---
 drivers/scsi/qla2xxx/qla_def.h |    1 +
 drivers/scsi/qla2xxx/qla_isr.c |    3 ++-
 drivers/scsi/qla2xxx/qla_os.c  |   31 +++++++++++++++++++------------
 3 files changed, 22 insertions(+), 13 deletions(-)

diff --git a/drivers/scsi/qla2xxx/qla_def.h b/drivers/scsi/qla2xxx/qla_def.h
index 1267b11..7c441c9 100644
--- a/drivers/scsi/qla2xxx/qla_def.h
+++ b/drivers/scsi/qla2xxx/qla_def.h
@@ -3404,6 +3404,7 @@ typedef struct scsi_qla_host {
 
 	unsigned long	pci_flags;
 #define PFLG_DISCONNECTED	0	/* PCI device removed */
+#define PFLG_DRIVER_REMOVING	1	/* PCI driver .remove */
 
 	uint32_t	device_flags;
 #define SWITCH_FOUND		BIT_0
diff --git a/drivers/scsi/qla2xxx/qla_isr.c b/drivers/scsi/qla2xxx/qla_isr.c
index 2741ad8..ee5eef4 100644
--- a/drivers/scsi/qla2xxx/qla_isr.c
+++ b/drivers/scsi/qla2xxx/qla_isr.c
@@ -117,7 +117,8 @@ qla2x00_check_reg32_for_disconnect(scsi_qla_host_t *vha, uint32_t reg)
 {
 	/* Check for PCI disconnection */
 	if (reg == 0xffffffff) {
-		if (!test_and_set_bit(PFLG_DISCONNECTED, &vha->pci_flags)) {
+		if (!test_and_set_bit(PFLG_DISCONNECTED, &vha->pci_flags) &&
+		    !test_bit(PFLG_DRIVER_REMOVING, &vha->pci_flags)) {
 			/*
 			 * Schedule this (only once) on the default system
 			 * workqueue so that all the adapter workqueues and the
diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
index 39c9953..51cba37 100644
--- a/drivers/scsi/qla2xxx/qla_os.c
+++ b/drivers/scsi/qla2xxx/qla_os.c
@@ -3123,15 +3123,25 @@ qla2x00_remove_one(struct pci_dev *pdev)
 	scsi_qla_host_t *base_vha;
 	struct qla_hw_data  *ha;
 
+	base_vha = pci_get_drvdata(pdev);
+	ha = base_vha->hw;
+
+	/* Indicate device removal to prevent future board_disable and wait
+	 * until any pending board_disable has completed. */
+	set_bit(PFLG_DRIVER_REMOVING, &base_vha->pci_flags);
+	cancel_work_sync(&ha->board_disable);
+
 	/*
-	 * If the PCI device is disabled that means that probe failed and any
-	 * resources should be have cleaned up on probe exit.
+	 * If the PCI device is disabled then there was a PCI-disconnect and
+	 * qla2x00_disable_board_on_pci_error has taken care of most of the
+	 * resources.
 	 */
-	if (!atomic_read(&pdev->enable_cnt))
+	if (!atomic_read(&pdev->enable_cnt)) {
+		scsi_host_put(base_vha->host);
+		kfree(ha);
+		pci_set_drvdata(pdev, NULL);
 		return;
-
-	base_vha = pci_get_drvdata(pdev);
-	ha = base_vha->hw;
+	}
 
 	qla2x00_wait_for_hba_ready(base_vha);
 
@@ -4791,18 +4801,15 @@ qla2x00_disable_board_on_pci_error(struct work_struct *work)
 	qla82xx_md_free(base_vha);
 	qla2x00_free_queues(ha);
 
-	scsi_host_put(base_vha->host);
-
 	qla2x00_unmap_iobases(ha);
 
 	pci_release_selected_regions(ha->pdev, ha->bars);
-	kfree(ha);
-	ha = NULL;
-
 	pci_disable_pcie_error_reporting(pdev);
 	pci_disable_device(pdev);
-	pci_set_drvdata(pdev, NULL);
 
+	/*
+	 * Let qla2x00_remove_one cleanup qla_hw_data on device removal.
+	 */
 }
 
 /**************************************************************************
-- 
1.7.10.4


  parent reply	other threads:[~2014-06-18 14:05 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-06-18 14:02 [PATCH 0/6] qla2xxx device removal fixups Joe Lawrence
2014-06-18 14:02 ` [PATCH 1/6] qla2xxx: Fix shost use-after-free on device removal Joe Lawrence
2014-06-18 14:02 ` [PATCH 2/6] qla2xxx: Use qla2x00_clear_drv_active on probe failure Joe Lawrence
2014-06-18 14:02 ` [PATCH 3/6] qla2xxx: Collect PCI register checks and board_disable scheduling Joe Lawrence
2014-06-18 14:02 ` [PATCH 4/6] qla2xxx: Schedule board_disable only once Joe Lawrence
2014-06-18 14:04 ` Joe Lawrence [this message]
2014-06-18 14:04   ` [PATCH 6/6] qla2xxx: Prevent probe and board_disable race Joe Lawrence
2014-07-25 15:23   ` [PATCH 5/6] qla2xxx: Prevent removal " Chad Dupuis
2014-07-25 19:00     ` Joe Lawrence
2014-07-25 19:45       ` Chad Dupuis
2014-06-18 15:35 ` [PATCH 0/6] qla2xxx device removal fixups Giridhar Malavali
2014-07-28 18:26 ` Chad Dupuis
2014-08-26 13:20 ` Joe Lawrence
2014-08-26 13:52   ` Christoph Hellwig

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=1403100284-1872-1-git-send-email-joe.lawrence@stratus.com \
    --to=joe.lawrence@stratus.com \
    --cc=chad.dupuis@qlogic.com \
    --cc=david.bulkow@stratus.com \
    --cc=dzickus@redhat.com \
    --cc=giridhar.malavali@qlogic.com \
    --cc=linux-scsi@vger.kernel.org \
    --cc=prarit@redhat.com \
    --cc=saurav.kashyap@qlogic.com \
    --cc=william.kuzeja@stratus.com \
    /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.