Re: [RFC PATCH #upstream-fixes] libata: don't use 32bit PIO for small transfers
From: Ming Lei <tom.leiming@gmail.com>
Date: 2011-09-14 07:44:45
Subsystem:
libata subsystem (serial and parallel ata drivers), the rest · Maintainers:
Damien Le Moal, Niklas Cassel, Linus Torvalds
Hi Tejun and Alan, Thanks for your spending time on the bug. On Tue, Sep 6, 2011 at 5:49 PM, Alan Cox [off-list ref] wrote:
So NACK. Add a piix_sata_data_xfer32_maybe() function to the ata_piix driver specifically for this chipset.
Could you accept the quirk patch below to make these controllers working at least now?
diff --git a/drivers/ata/ata_piix.c b/drivers/ata/ata_piix.c
index 43107e9..eb7ea56 100644
--- a/drivers/ata/ata_piix.c
+++ b/drivers/ata/ata_piix.c@@ -341,11 +341,12 @@ static struct ata_port_operations piix_sata_ops = { }; static struct ata_port_operations piix_pata_ops = { - .inherits = &piix_sata_ops, + .inherits = &ata_bmdma32_port_ops, .cable_detect = ata_cable_40wire, .set_piomode = piix_set_piomode, .set_dmamode = piix_set_dmamode, .prereset = piix_pata_prereset, + .sff_irq_check = piix_irq_check, }; static struct ata_port_operations piix_vmw_ops = {
@@ -1585,6 +1586,15 @@ static int __devinit piix_init_one(struct pci_dev *pdev, "on poweroff and hibernation\n"); } + /* + * Sandybridge chipset H61/P67/H67 have broken 32 mode up to now + * see https://bugzilla.kernel.org/show_bug.cgi?id=40592 + */ + if (pdev->vendor == PCI_VENDOR_ID_INTEL && pdev->device == 0x1c00) + piix_sata_ops.inherits = &ata_bmdma_port_ops; + else + piix_sata_ops.inherits = &ata_bmdma32_port_ops; + port_info[0] = piix_port_info[ent->driver_data]; port_info[1] = piix_port_info[ent->driver_data];
thanks, -- Ming Lei