libata: PCI device should be powered up before being accessed

PCI device should be powered up or powered up before its PCI regsiters
are accessed.  Although PCI configuration register access is allowed
in D3hot, PCI device is free to reset its status when transiting from
D3hot to D0 causing configuration data to change.

Many libata SFF drivers which use ata_pci_init_one() read and update
configuration registers before calling ata_pci_init_one() which
enables the PCI device.  Also, in resume paths, some drivers access
registers without resuming the PCI device.

This patch adds a call to pcim_enable_device() in init path if
register is accessed before calling ata_pci_init_one() and make resume
paths first resume PCI devices, access PCI configuration regiters then
resume ATA host.

While at it...

* cmd640 was strange in that it set ->resume even when CONFIG_PM is
  not.  This is by-product of minimal build fix.  Updated.

* In cs5530, Don't BUG() on reinit failure.  Just whine and fail
  resume.

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

1 files changed, 13 insertions, 1 deletions

diff --git a/drivers/ata/pata_ali.c b/drivers/ata/pata_ali.c
index 511a830b6256..3814aebefb2d 100644
--- a/drivers/ata/pata_ali.c
+++ b/drivers/ata/pata_ali.c
@@ -637,6 +637,11 @@ static int ali_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
 	const struct ata_port_info *ppi[] = { NULL, NULL };
 	u8 tmp;
 	struct pci_dev *isa_bridge;
+	int rc;
+
+	rc = pcim_enable_device(pdev);
+	if (rc)
+		return rc;
 
 	/*
 	 * The chipset revision selects the driver operations and
@@ -672,8 +677,15 @@ static int ali_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
 #ifdef CONFIG_PM
 static int ali_reinit_one(struct pci_dev *pdev)
 {
+	struct ata_host *host = dev_get_drvdata(&pdev->dev);
+	int rc;
+
+	rc = ata_pci_device_do_resume(pdev);
+	if (rc)
+		return rc;
 	ali_init_chipset(pdev);
-	return ata_pci_device_resume(pdev);
+	ata_host_resume(host);
+	return 0;
 }
 #endif