Re: [PATCH] iommu/amd: Add quirk for broken Marvell 88SE91xx SATA

[Manfred Schwarb <manfred99-OI3hZJvNYWs@public.gmane.org>]

Thanks, Joerg!

Am 30.07.2013 13:37, schrieb Joerg Roedel:
> Hi Manfred,
>
> On Wed, Jul 03, 2013 at 10:53:37AM +0200, Manfred Schwarb wrote:
>>> I will look into the best way to port this to the AMD IOMMU driver. Will
>>> send you a patch for testing when I have somthing.
>>>
>>
>> Any news on this? Had you time to look at this issue?
>
> Can you test the attached patch please? It applies to v3.11-rc3. I
> compiled and boot-tested it, but I don't have a broken Marvell
> controller myself, so not sure if the patch works.
>

I tried this patch with 3.11-rc3 on opensuse 12.1, doing an "make oldconfig"
and choosing defaults for the new items.
This linux version seems to have some issues with my somewhat ancient user
space, boot hangs quite early without any special messages (independent of
iommu settings). However, in the "failsafe" mode I was able to boot.

When turning on iommu support in BIOS, I got a timeout at fsck/mount time
for the second disk on the Marvell controller. The first one did work
straight on. Then I got to the rescue prompt, where I did a "mount -a", and
hey, also the second disk started working, I could read and write to this
disk also.

Things seem to work halfway therefore. The timeout for the second attached
disk is repeatable. When doing the "mount -a", first some kernel boot messages
appeared, as if this command unblocked some unfinished disk initialization
or so.
The boot logs (boot.msg) where nearly identical for the iommu and no-iommu
case. But I was not able to get your additional print into the log file,
I tried "loglevel=6" on the command line and setting /proc/sys/kernel/printk
in /etc/init.d/boot.local, to no avail.

Hope this helps somewhat,
cheers,

Manfred


> Thanks,
>
> 	Joerg
>
>>From 04c7d4ea2171780656c4c8ff3500a97ea45aabcb Mon Sep 17 00:00:00 2001
> From: Joerg Roedel <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
> Date: Tue, 30 Jul 2013 13:10:54 +0200
> Subject: [PATCH] iommu/amd: Add quirk for broken Marvell 88SE91xx SATA
>   controllers
>
> These controlers use the wrong requestor id when sending DMA
> requests. Add a quirk for this to the AMD IOMMU driver to
> make them work with IOMMU enabled.
>
> Signed-off-by: Joerg Roedel <joro-zLv9SwRftAIdnm+yROfE0A@public.gmane.org>
> ---
>   drivers/iommu/amd_iommu_init.c |   54 ++++++++++++++++++++++++++++++++++++++++
>   1 file changed, 54 insertions(+)
>
> diff --git a/drivers/iommu/amd_iommu_init.c b/drivers/iommu/amd_iommu_init.c
> index 7acbf35..f4e0c82 100644
> --- a/drivers/iommu/amd_iommu_init.c
> +++ b/drivers/iommu/amd_iommu_init.c
> @@ -1025,6 +1025,58 @@ static void __init free_iommu_all(void)
>   }
>
>   /*
> + * Update IOMMU data structures to handle broken devices
> + */
> +static const struct pci_dev_merge_functions {
> +	u16 vendor;
> +	u16 device;
> +} pci_dev_merge_fn_devs[] = {
> +	{ PCI_VENDOR_ID_MARVELL_EXT, 0x9123 },
> +	{ PCI_VENDOR_ID_MARVELL_EXT, 0x9125 },
> +	{ PCI_VENDOR_ID_MARVELL_EXT, 0x9128 },
> +	{ PCI_VENDOR_ID_MARVELL_EXT, 0x9130 },
> +	{ PCI_VENDOR_ID_MARVELL_EXT, 0x9172 },
> +	{ 0 }
> +};
> +
> +static void amd_iommu_merge_fn_quirk(struct pci_dev *pdev)
> +{
> +	const struct pci_dev_merge_functions *fn;
> +
> +	for (fn = pci_dev_merge_fn_devs; fn->vendor; fn++) {
> +		u16 base_devid, devid, i;
> +
> +		if ((pdev->vendor != fn->vendor && fn->vendor != PCI_ANY_ID) ||
> +		    (pdev->device != fn->device && fn->device != PCI_ANY_ID))
> +			continue;
> +
> +		pr_info("AMD-Vi: Applying merge-fn quirk for device %s\n",
> +			dev_name(&pdev->dev));
> +
> +		devid      = PCI_DEVID(pdev->bus->number, pdev->devfn);
> +		base_devid = PCI_DEVID(PCI_BUS_NUM(devid),
> +				       PCI_DEVFN(PCI_SLOT(devid), 0));
> +
> +		/* Update alias table */
> +		for (i = 1; i < 8; ++i)
> +			amd_iommu_alias_table[base_devid + i] = devid;
> +	}
> +}
> +
> +static void amd_iommu_pci_quirks(void)
> +{
> +        struct pci_dev *pdev = NULL;
> +
> +	for_each_pci_dev(pdev) {
> +		/*
> +		 * Check for known PCI devices that use
> +		 * the wrong requestor id
> +		 */
> +		amd_iommu_merge_fn_quirk(pdev);
> +	}
> +}
> +
> +/*
>    * Family15h Model 10h-1fh erratum 746 (IOMMU Logging May Stall Translations)
>    * Workaround:
>    *     BIOS should disable L2B micellaneous clock gating by setting
> @@ -1337,6 +1389,8 @@ static int __init amd_iommu_init_pci(void)
>   			break;
>   	}
>
> +	amd_iommu_pci_quirks();
> +
>   	ret = amd_iommu_init_devices();
>
>   	print_iommu_info();
>