Patchwork enable AHCI mode on certain ich chipsets

login
register
mail settings
Submitter Joerg Dorchain
Date Feb. 16, 2011, 9:05 a.m.
Message ID <20110216090548.GQ5778@Redstar.dorchain.net>
Download mbox | patch
Permalink /patch/83346/
State Not Applicable
Delegated to: David Miller
Headers show

Comments

Joerg Dorchain - Feb. 16, 2011, 9:05 a.m.
Hello all,

this patch allows to force ICH7/8/9 into AHCI mode.  This is needed
because some BIOSes do not make AHCI-mode operation available to the
user.
As the Intel documentation states that the OS should not carry
out the operation - the user must force this on the kernel
commandline using quirk_ich_force_ahci

As this quirk gets called whilst the PCI subsystem is
walking the PCI bus, we declare this quirk against the LPC
(device 00:1f.0), so that we can frob 00:1f.2 before the PCI
code has scanned it.
Note: the pci id might change due to this (e.g. from 27c4 to 27c5)

For working suspend/resume, the next patch is required, too.

Bye,

Joerg


Signed-Off-By: joerg Dorchain <joerg@dorchain.net>
Jeff Garzik - Feb. 16, 2011, 7:45 p.m.
On 02/16/2011 04:05 AM, Joerg Dorchain wrote:
> Hello all,
>
> this patch allows to force ICH7/8/9 into AHCI mode.  This is needed
> because some BIOSes do not make AHCI-mode operation available to the
> user.
> As the Intel documentation states that the OS should not carry
> out the operation - the user must force this on the kernel
> commandline using quirk_ich_force_ahci
>
> As this quirk gets called whilst the PCI subsystem is
> walking the PCI bus, we declare this quirk against the LPC
> (device 00:1f.0), so that we can frob 00:1f.2 before the PCI
> code has scanned it.
> Note: the pci id might change due to this (e.g. from 27c4 to 27c5)
>
> For working suspend/resume, the next patch is required, too.
>
> Bye,
>
> Joerg
>
>
> Signed-Off-By: joerg Dorchain<joerg@dorchain.net>
>
> --- linux/drivers/pci/quirks.c.orig	2011-02-04 18:29:03.000000000 +0100
> +++ linux/drivers/pci/quirks.c	2011-02-12 07:07:57.000000000 +0100
> @@ -2684,6 +2684,74 @@
>   DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_HINT, 0x0020, quirk_hotplug_bridge);
>
>   /*
> + * Force ICH7/8/9 into AHCI mode.  This is needed because some
> + * BIOSes do not make AHCI-mode operation available to the user.
> + * As the Intel documentation states that the OS should not carry
> + * out the operation - the user must force this on the kernel
> + * commandline using quirk_ich_force_ahci
> + *
> + * As this quirk gets called whilst the PCI subsystem is
> + * walking the PCI bus, we declare this quirk against the LPC
> + * (device 00:1f.0), so that we can frob 00:1f.2 before the PCI
> + * code has scanned it.
> + * Note: the pci id might change due to this (e.g. from 27c4 to 27c5)
> + *
> + */
> +
> +static bool ich_force_ahci_mode; /* defaults to false */
> +
> +static int __init ich789_force_ahci_mode_setup(char *str)
> +{
> +	ich_force_ahci_mode = true;
> +	return 0;
> +}
> +early_param("quirk_ich_force_ahci", ich789_force_ahci_mode_setup);
> +
> +static void ich789_force_ahci_mode(struct pci_dev *pdev)
> +{
> +	u8 amrval;
> +	u8 sclkgc;
> +	const int ich89_address_map_reg = 0x90;
> +	const int ich89_sata_clock_gen_config_reg = 0x9c;
> +
> +	if (!ich_force_ahci_mode)
> +		return;
> +
> +	/* ICH8 datasheet section 12.1.33 */
> +	if (!pci_bus_read_config_byte(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
> +		ich89_address_map_reg,&amrval)) {
> +
> +		if (amrval&  (BIT(6) | BIT(7))) {
> +			dev_printk(KERN_DEBUG,&pdev->dev,
> +				"ICH7/8/9 SATA controller not in IDE mode.  Not modifying.\n");
> +			return;
> +		}
> +		if (amrval&  (BIT(0) | BIT(1)))
> +			dev_printk(KERN_DEBUG,&pdev->dev,
> +				"ICH7/8/9 in SATA/PATA combined mode.  Untested.\n");
> +		/* AHCI mode */
> +		amrval |= BIT(6);
> +		amrval&= ~BIT(7);
> +		pci_bus_read_config_byte(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
> +			ich89_sata_clock_gen_config_reg,&sclkgc);
> +		dev_printk(KERN_DEBUG,&pdev->dev, "sclkgc is %#0x\n", sclkgc);
> +		pci_bus_write_config_byte(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
> +			ich89_address_map_reg, amrval);
> +		dev_printk(KERN_DEBUG,&pdev->dev, "Forced ICH7/8/9 mode PIIX->AHCI\n");

How is the assignment of the AHCI BAR handled?  Does this happen 
automatically because it's an early fixup?

That is traditionally the stumbling block...

	Jeff


--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Tejun Heo - Feb. 17, 2011, 10:03 a.m.
On Wed, Feb 16, 2011 at 10:05:48AM +0100, Joerg Dorchain wrote:
> this patch allows to force ICH7/8/9 into AHCI mode.  This is needed
> because some BIOSes do not make AHCI-mode operation available to the
> user.
> As the Intel documentation states that the OS should not carry
> out the operation - the user must force this on the kernel
> commandline using quirk_ich_force_ahci
> 
> As this quirk gets called whilst the PCI subsystem is
> walking the PCI bus, we declare this quirk against the LPC
> (device 00:1f.0), so that we can frob 00:1f.2 before the PCI
> code has scanned it.
> Note: the pci id might change due to this (e.g. from 27c4 to 27c5)
> 
> For working suspend/resume, the next patch is required, too.

There have been multiple attempts at this and as Jeff pointed out ABAR
allocation usually is one of the technical roadblocks.

I'm not particularly inclined toward including this patch in upstream
as long as it can't be enabled by default.  It's just gonna be an
extra piece of code to carry around which only a handful of people use
and nobody really knows on which hardware it works or not and to what
extent.

Thanks.
Matthew Garrett - Feb. 17, 2011, 4:36 p.m.
On Wed, Feb 16, 2011 at 02:45:56PM -0500, Jeff Garzik wrote:

> How is the assignment of the AHCI BAR handled?  Does this happen  
> automatically because it's an early fixup?
>
> That is traditionally the stumbling block...

Yes, because it's an early fixup the kernel should allocate it itself. 
It's small enough that I'd hope we don't have trouble finding one...
Joerg Dorchain - Feb. 17, 2011, 5:42 p.m.
On Wed, Feb 16, 2011 at 02:45:56PM -0500, Jeff Garzik wrote:
> >this patch allows to force ICH7/8/9 into AHCI mode.  This is needed
> >because some BIOSes do not make AHCI-mode operation available to the
> >user.
> >As the Intel documentation states that the OS should not carry
> >out the operation - the user must force this on the kernel
> >commandline using quirk_ich_force_ahci
> >
> >As this quirk gets called whilst the PCI subsystem is
> >walking the PCI bus, we declare this quirk against the LPC
> >(device 00:1f.0), so that we can frob 00:1f.2 before the PCI
> >code has scanned it.
> >Note: the pci id might change due to this (e.g. from 27c4 to 27c5)
> >
> >For working suspend/resume, the next patch is required, too.
> >
> >Bye,
> >
> >Joerg
> >
> >
> >Signed-Off-By: joerg Dorchain<joerg@dorchain.net>
> >
> >--- linux/drivers/pci/quirks.c.orig	2011-02-04 18:29:03.000000000 +0100
> >+++ linux/drivers/pci/quirks.c	2011-02-12 07:07:57.000000000 +0100
> >@@ -2684,6 +2684,74 @@
> >  DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_HINT, 0x0020, quirk_hotplug_bridge);
> >
> >  /*
> >+ * Force ICH7/8/9 into AHCI mode.  This is needed because some
> >+ * BIOSes do not make AHCI-mode operation available to the user.
> >+ * As the Intel documentation states that the OS should not carry
> >+ * out the operation - the user must force this on the kernel
> >+ * commandline using quirk_ich_force_ahci
> >+ *
> >+ * As this quirk gets called whilst the PCI subsystem is
> >+ * walking the PCI bus, we declare this quirk against the LPC
> >+ * (device 00:1f.0), so that we can frob 00:1f.2 before the PCI
> >+ * code has scanned it.
> >+ * Note: the pci id might change due to this (e.g. from 27c4 to 27c5)
> >+ *
> >+ */
> >+
> >+static bool ich_force_ahci_mode; /* defaults to false */
> >+
> >+static int __init ich789_force_ahci_mode_setup(char *str)
> >+{
> >+	ich_force_ahci_mode = true;
> >+	return 0;
> >+}
> >+early_param("quirk_ich_force_ahci", ich789_force_ahci_mode_setup);
> >+
> >+static void ich789_force_ahci_mode(struct pci_dev *pdev)
> >+{
> >+	u8 amrval;
> >+	u8 sclkgc;
> >+	const int ich89_address_map_reg = 0x90;
> >+	const int ich89_sata_clock_gen_config_reg = 0x9c;
> >+
> >+	if (!ich_force_ahci_mode)
> >+		return;
> >+
> >+	/* ICH8 datasheet section 12.1.33 */
> >+	if (!pci_bus_read_config_byte(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
> >+		ich89_address_map_reg,&amrval)) {
> >+
> >+		if (amrval&  (BIT(6) | BIT(7))) {
> >+			dev_printk(KERN_DEBUG,&pdev->dev,
> >+				"ICH7/8/9 SATA controller not in IDE mode.  Not modifying.\n");
> >+			return;
> >+		}
> >+		if (amrval&  (BIT(0) | BIT(1)))
> >+			dev_printk(KERN_DEBUG,&pdev->dev,
> >+				"ICH7/8/9 in SATA/PATA combined mode.  Untested.\n");
> >+		/* AHCI mode */
> >+		amrval |= BIT(6);
> >+		amrval&= ~BIT(7);
> >+		pci_bus_read_config_byte(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
> >+			ich89_sata_clock_gen_config_reg,&sclkgc);
> >+		dev_printk(KERN_DEBUG,&pdev->dev, "sclkgc is %#0x\n", sclkgc);
> >+		pci_bus_write_config_byte(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
> >+			ich89_address_map_reg, amrval);
> >+		dev_printk(KERN_DEBUG,&pdev->dev, "Forced ICH7/8/9 mode PIIX->AHCI\n");
> 
> How is the assignment of the AHCI BAR handled?  Does this happen
> automatically because it's an early fixup?

For me, it does:
        Region 0: I/O ports at b400 [size=8]
        Region 1: I/O ports at b080 [size=4]
        Region 2: I/O ports at b000 [size=8]
        Region 3: I/O ports at ac80 [size=4]
        Region 4: I/O ports at ac00 [size=16]
        Region 5: Memory at fe837800 (32-bit, non-prefetchable) [size=1K]

> 
> That is traditionally the stumbling block...

This quirk relies on the fact the the device on 00:1f.0 is
scanned earlier than the device at 00:1f.2, so when the scan
reaches 00:1f.2, the device there is just ok for the ahci driver.

Bye,

Joerg
Joerg Dorchain - Feb. 17, 2011, 6:09 p.m.
On Thu, Feb 17, 2011 at 11:03:30AM +0100, Tejun Heo wrote:
> 
> I'm not particularly inclined toward including this patch in upstream
> as long as it can't be enabled by default.  It's just gonna be an
> extra piece of code to carry around which only a handful of people use
> and nobody really knows on which hardware it works or not and to what
> extent.

I have a feedback for the chipsets listed in the patch. The
protection via the extra parameter is owned to the fact that
intel recommends against the OS doing this. Quote: "SW shall not
manipulate SMS during runtime operation; that is, the OS will not
do this." I would not see early fixup as runtime, however, better
save than sorry.

Advantages are better performance and lower power consumption.

I don't wanna break anybody's data with that patch, hence the
precaution.

Bye,

Joerg
Mark Lord - Feb. 17, 2011, 10:52 p.m.
On 11-02-17 11:36 AM, Matthew Garrett wrote:
> On Wed, Feb 16, 2011 at 02:45:56PM -0500, Jeff Garzik wrote:
> 
>> How is the assignment of the AHCI BAR handled?  Does this happen  
>> automatically because it's an early fixup?
>>
>> That is traditionally the stumbling block...
> 
> Yes, because it's an early fixup the kernel should allocate it itself. 
> It's small enough that I'd hope we don't have trouble finding one...


It works here (the AHCI BAR setup) with this patch,
though the workaround itself doesn't work because my
Dell machine requires extra/secret fiddling to switch
things fully over to AHCI.

The patch does seem harmless enough to include in the kernel, though.
--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Bjørn Mork - Feb. 18, 2011, 2:03 p.m.
Tejun Heo <tj@kernel.org> writes:

> I'm not particularly inclined toward including this patch in upstream
> as long as it can't be enabled by default.  It's just gonna be an
> extra piece of code to carry around which only a handful of people use
> and nobody really knows on which hardware it works or not and to what
> extent.

Then enable it by default :-)



Bjørn

--
To unsubscribe from this list: send the line "unsubscribe linux-ide" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Tejun Heo - Feb. 18, 2011, 2:42 p.m.
(please always reply-to-all)

On Fri, Feb 18, 2011 at 03:03:20PM +0100, Bjørn Mork wrote:
> Tejun Heo <tj@kernel.org> writes:
> 
> > I'm not particularly inclined toward including this patch in upstream
> > as long as it can't be enabled by default.  It's just gonna be an
> > extra piece of code to carry around which only a handful of people use
> > and nobody really knows on which hardware it works or not and to what
> > extent.
> 
> Then enable it by default :-)

That's the point tho.  We can't enable it by default.  It's too
dangerous.  If there's a way we can safely, I'd be happy to do it.

Patch

--- linux/drivers/pci/quirks.c.orig	2011-02-04 18:29:03.000000000 +0100
+++ linux/drivers/pci/quirks.c	2011-02-12 07:07:57.000000000 +0100
@@ -2684,6 +2684,74 @@ 
 DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_HINT, 0x0020, quirk_hotplug_bridge);
 
 /*
+ * Force ICH7/8/9 into AHCI mode.  This is needed because some
+ * BIOSes do not make AHCI-mode operation available to the user.
+ * As the Intel documentation states that the OS should not carry
+ * out the operation - the user must force this on the kernel
+ * commandline using quirk_ich_force_ahci
+ *
+ * As this quirk gets called whilst the PCI subsystem is
+ * walking the PCI bus, we declare this quirk against the LPC
+ * (device 00:1f.0), so that we can frob 00:1f.2 before the PCI
+ * code has scanned it.
+ * Note: the pci id might change due to this (e.g. from 27c4 to 27c5)
+ *
+ */
+
+static bool ich_force_ahci_mode; /* defaults to false */
+
+static int __init ich789_force_ahci_mode_setup(char *str)
+{
+	ich_force_ahci_mode = true;
+	return 0;
+}
+early_param("quirk_ich_force_ahci", ich789_force_ahci_mode_setup);
+
+static void ich789_force_ahci_mode(struct pci_dev *pdev)
+{
+	u8 amrval;
+	u8 sclkgc;
+	const int ich89_address_map_reg = 0x90;
+	const int ich89_sata_clock_gen_config_reg = 0x9c;
+
+	if (!ich_force_ahci_mode)
+		return;
+
+	/* ICH8 datasheet section 12.1.33 */
+	if (!pci_bus_read_config_byte(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
+		ich89_address_map_reg, &amrval)) {
+
+		if (amrval & (BIT(6) | BIT(7))) {
+			dev_printk(KERN_DEBUG, &pdev->dev,
+				"ICH7/8/9 SATA controller not in IDE mode.  Not modifying.\n");
+			return;
+		}
+		if (amrval & (BIT(0) | BIT(1)))
+			dev_printk(KERN_DEBUG, &pdev->dev,
+				"ICH7/8/9 in SATA/PATA combined mode.  Untested.\n");
+		/* AHCI mode */
+		amrval |= BIT(6);
+		amrval &= ~BIT(7);
+		pci_bus_read_config_byte(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
+			ich89_sata_clock_gen_config_reg, &sclkgc);
+		dev_printk(KERN_DEBUG, &pdev->dev, "sclkgc is %#0x\n", sclkgc);
+		pci_bus_write_config_byte(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 2),
+			ich89_address_map_reg, amrval);
+		dev_printk(KERN_DEBUG, &pdev->dev, "Forced ICH7/8/9 mode PIIX->AHCI\n");
+	}
+}
+/* ICH7 */
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, 0x27b9, ich789_force_ahci_mode);
+/* ICH8 */
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_ICH8_0, ich789_force_ahci_mode);
+/* ICH9R LPC */
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, 0x2916, ich789_force_ahci_mode);
+/* ICH9M LPC */
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, 0x2917, ich789_force_ahci_mode);
+/* ICH9M-E LPC */
+DECLARE_PCI_FIXUP_EARLY(PCI_VENDOR_ID_INTEL, 0x2919, ich789_force_ahci_mode);
+
+/*
  * This is a quirk for the Ricoh MMC controller found as a part of
  * some mulifunction chips.