Setting storage password requires iommu=pt to wake from suspend

[deepcube]

Device Information

System Model or SKU

Framework Laptop 13 (AMD Ryzen™ 7040 Series)

BIOS VERSION

3.05

DIY Edition information

Memory: Crucial 2x16GB DDR5 5600MHz CT2K16G56C46S5
Storage: Samsung 990 PRO 2 TB MZ-V9P2T0B/AM

Describe the bug

When using storage password in the BIOS the system fails to resume from suspend to RAM (s2idle) correctly. Adding iommu=pt works around the issue.

Steps To Reproduce

Steps to reproduce the behavior:

1. Enable storage password for NVMe drive in BIOS.
2. Boot into Linux (confirm CONFIG_BLK_SED_OPAL is enabled in kernel).
3. Suspend to RAM echo mem > /sys/power/state.
4. Wake laptop with keypress.
5. Note laptop does not wake correctly (hangs or resumes but disk is unavailable).
6. Add iommu=pt to kernel cmdline, repeat, note resume works.

Expected behavior

Can resume from suspend to RAM successfully without addition of iommu=pt to kernel cmdline.

Operating System (please complete the following information):

• OS/Distribution: Void Linux
• Linux Kernel Version: 6.13.7

Additional context

Forum thread: https://community.frame.work/t/responded-amd-7040-setting-storage-password-breaks-resume-from-suspend/42554/24

[Eloston]

For those reproducing, I would like to note that the Storage password might be disabled in the BIOS settings unless a Master password is set first

I am able to reproduce this issue with the following setup:

• AMD Ryzen 5 7640U
• Storage: WD_BLACK SN850X 1TB
• OS: Debian 12 (bookworm)
• Linux Kernel Version: 6.12.12 (from Debian backports)
• Secure Boot enabled (I believe OP's system has it disabled)

[deepcube]

Secure Boot enabled (I believe OP's system has it disabled)

I have reproduced both with secure boot enabled and disabled.

[superm1]

Can you guys please try this patch?

[Eloston]

@superm1 I tried your v2 patch (link) on Linux 6.12.22 (technically on top of Debian's 6.12.22-1~bpo12+1). I tested suspend and resume using amd-debug-tools' amd_s2idle.py script

• When I remove iommu=pt from the kernel cmdline, resume from suspend crashes the system (screen is black, unresponsive to inputs, and I have no system logs nor s2idle results persisted when I hard reset the system).
• When I keep iommu=pt in the cmdline, resume from suspend works normally.

tl;dr there seems to be no change with your patch applied.

[superm1]

@superm1 I tried your v2 patch (link) on Linux 6.12.22 (technically on top of Debian's 6.12.22-1~bpo12+1). I tested suspend and resume using amd-debug-tools' amd_s2idle.py script

• When I remove iommu=pt from the kernel cmdline, resume from suspend crashes the system (screen is black, unresponsive to inputs, and I have no system logs nor s2idle results persisted when I hard reset the system).
• When I keep iommu=pt in the cmdline, resume from suspend works normally.

tl;dr there seems to be no change with your patch applied.

Thanks for checking. It was a bit of a long shot.

I feel that likely the problem here is a missing unity mapping in the ACPI IVRS table for the region that the BIOS uses for authentication.