nvme-pci: Add suspend quirk for HP mt645 thin client

[PATCH] nvme-pci: Add suspend quirk for HP mt645 thin client

Posted by Alexandru Gagniuc 2 years, 7 months ago

The HP Elite mt645 thin client exhibits resume times of over one
minute when using the normal nvme resume path. BIOS has tried to work
around this by setting the "StorageD3Enable" ACPI property, but only
if it detected the "Linux-HPI-Hybrid-Graphics" _OSI() flag. This flag
does not exist, so the BIOS workaround can't work.

Instead, just set NVME_QUIRK_SIMPLE_SUSPEND when running on an mt645.
The DMI_PRODUCT_NAME cannot be used because this string can be changed
in the field. Match against DMI_BOARD_NAME, which should be immutable.

Cc: stable@vger.kernel.org
Signed-off-by: Alexandru Gagniuc <alexandru.gagniuc@hp.com>
---
 drivers/nvme/host/pci.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
index 492f319ebdf3..25b59f5ce874 100644
--- a/drivers/nvme/host/pci.c
+++ b/drivers/nvme/host/pci.c
@@ -2897,6 +2897,15 @@ static unsigned long check_vendor_combination_bug(struct pci_dev *pdev)
 		if ((dmi_match(DMI_BOARD_VENDOR, "LENOVO")) &&
 		     dmi_match(DMI_BOARD_NAME, "LNVNB161216"))
 			return NVME_QUIRK_SIMPLE_SUSPEND;
+	} else if (dmi_match(DMI_SYS_VENDOR, "HP") &&
+		   (dmi_match(DMI_BOARD_NAME, "8B0F") ||
+		    dmi_match(DMI_BOARD_NAME, "8B59"))) {
+		/*
+		 * Force simple suspend to work around long resume latencies
+		 * (1 minute or longer).
+		 */
+		dev_info(&pdev->dev, "simple suspend quirk for HP mt645\n");
+		return NVME_QUIRK_SIMPLE_SUSPEND;
 	}
 
 	return 0;
-- 
2.39.1

Re: [PATCH] nvme-pci: Add suspend quirk for HP mt645 thin client

Posted by Limonciello, Mario 2 years, 7 months ago

On 7/6/2023 1:06 PM, Alexandru Gagniuc wrote:
> The HP Elite mt645 thin client exhibits resume times of over one
> minute when using the normal nvme resume path.

Is this the specific product you're referring to?

https://www.amd.com/system/files/documents/hp-elite-mt645-g7-mobile-thin-client.pdf

> BIOS has tried to work
> around this by setting the "StorageD3Enable" ACPI property, but only
> if it detected the "Linux-HPI-Hybrid-Graphics" _OSI() flag. This flag
> does not exist, so the BIOS workaround can't work.
>
> Instead, just set NVME_QUIRK_SIMPLE_SUSPEND when running on an mt645.
> The DMI_PRODUCT_NAME cannot be used because this string can be changed
> in the field. Match against DMI_BOARD_NAME, which should be immutable.
I'm quite confused why this is necessary.  If the product link
I sent above is correct, this is an AMD Barcelo APU.

Cezanne and Barcelo report as the exact same X86 model.

Cezanne will already always identify as simple
suspend due to bugs that were reported on some other machines.
It's been this way since kernel 6.3-rc1.

e2a56364485e ("ACPI: x86: utils: Add Cezanne to the list for forcing 
StorageD3Enable")

>
> Cc: stable@vger.kernel.org
> Signed-off-by: Alexandru Gagniuc <alexandru.gagniuc@hp.com>
> ---
>   drivers/nvme/host/pci.c | 9 +++++++++
>   1 file changed, 9 insertions(+)
>
> diff --git a/drivers/nvme/host/pci.c b/drivers/nvme/host/pci.c
> index 492f319ebdf3..25b59f5ce874 100644
> --- a/drivers/nvme/host/pci.c
> +++ b/drivers/nvme/host/pci.c
> @@ -2897,6 +2897,15 @@ static unsigned long check_vendor_combination_bug(struct pci_dev *pdev)
>   		if ((dmi_match(DMI_BOARD_VENDOR, "LENOVO")) &&
>   		     dmi_match(DMI_BOARD_NAME, "LNVNB161216"))
>   			return NVME_QUIRK_SIMPLE_SUSPEND;
> +	} else if (dmi_match(DMI_SYS_VENDOR, "HP") &&
> +		   (dmi_match(DMI_BOARD_NAME, "8B0F") ||
> +		    dmi_match(DMI_BOARD_NAME, "8B59"))) {
> +		/*
> +		 * Force simple suspend to work around long resume latencies
> +		 * (1 minute or longer).
> +		 */
> +		dev_info(&pdev->dev, "simple suspend quirk for HP mt645\n");
> +		return NVME_QUIRK_SIMPLE_SUSPEND;
>   	}
>   
>   	return 0;

Re: [PATCH] nvme-pci: Add suspend quirk for HP mt645 thin client

Posted by Gagniuc, Alexandru 2 years, 7 months ago


________________________________________
From: Limonciello, Mario <mario.limonciello@amd.com>
Sent: Friday, July 7, 2023 11:28 PM
To: Gagniuc, Alexandru; linux-nvme@lists.infradead.org
Cc: kbusch@kernel.org; axboe@fb.com; hch@lst.de; sagi@grimberg.me; linux-kernel@vger.kernel.org; Zhang, Eniac; stable@vger.kernel.org
Subject: Re: [PATCH] nvme-pci: Add suspend quirk for HP mt645 thin client

CAUTION: External Email

On 7/6/2023 1:06 PM, Alexandru Gagniuc wrote:
> The HP Elite mt645 thin client exhibits resume times of over one
> minute when using the normal nvme resume path.

Is this the specific product you're referring to?

https://www.amd.com/system/files/documents/hp-elite-mt645-g7-mobile-thin-client.pdf<https://www.amd.com/system/files/documents/hp-elite-mt645-g7-mobile-thin-client.pdf>

> BIOS has tried to work
> around this by setting the "StorageD3Enable" ACPI property, but only
> if it detected the "Linux-HPI-Hybrid-Graphics" _OSI() flag. This flag
> does not exist, so the BIOS workaround can't work.
>
> Instead, just set NVME_QUIRK_SIMPLE_SUSPEND when running on an mt645.
> The DMI_PRODUCT_NAME cannot be used because this string can be changed
> in the field. Match against DMI_BOARD_NAME, which should be immutable.
I'm quite confused why this is necessary.  If the product link
I sent above is correct, this is an AMD Barcelo APU.

Cezanne and Barcelo report as the exact same X86 model.

Cezanne will already always identify as simple
suspend due to bugs that were reported on some other machines.
It's been this way since kernel 6.3-rc1.

e2a56364485e ("ACPI: x86: utils: Add Cezanne to the list for forcing
StorageD3Enable")
[snip]

Hi Mario,

The product link you posted is correct. I was not aware of e2a56364485e
when I developed the fix herein before v6.3. After some quick testing, I can
confirm that e2a56364485e solves the original issue.

We do not need this patch then.

Thank you,
Alex

Re: [PATCH] nvme-pci: Add suspend quirk for HP mt645 thin client

Posted by Limonciello, Mario 2 years, 7 months ago

On 7/10/2023 3:06 PM, Gagniuc, Alexandru wrote:
> [snip]
> Hi Mario,
>
> The product link you posted is correct. I was not aware of e2a56364485e
> when I developed the fix herein before v6.3. After some quick testing, I can
> confirm that e2a56364485e solves the original issue.
>
> We do not need this patch then.
>
> Thank you,
> Alex
>
Great, thanks for confirming.  Just as an FYI to you we don't have any 
intention
to grow that list any further.  Those were added because the platforms 
came at
a time that StorageD3Enable was just being introduced and there were 
bugs with too
many missing it.

If you have future AMD based products that use newer hardware such as 
Rembrandt or
Phoenix that you need to fix this in the BIOS properly.