Skip to content

UFS/PHY dev_info instrumentation needs proper cleanup before upstream #2

@ccrutchf

Description

@ccrutchf

What

On 2026-05-03 the worst UART offenders in drivers/ufs/core/ufshcd.c, drivers/ufs/host/ufs-exynos.c, and drivers/phy/samsung/phy-gs101-ufs.c got bulk-converted from dev_info(...) to dev_dbg(...) (commit dcefb14acacf). Without that change each boot dumps ~700 KB of UART output = ~70s of throttle alone at 115200 baud, masking real boot time and the kmscon login prompt.

The worst single offender was ufs-cmd-issue in ufshcd_send_command() — fires per SCSI submission at ~110 bytes/line.

Why this is interim, not final

The instrumentation was load-bearing while we didn't know what was breaking. Once root causes were nailed down (terminator-fix, GSA mailbox shim, post-PMC misplaced writes, three cal-if walk fixes, FMPSECURITY0.DESCTYPE=0), the prints became pure noise.

dev_info → dev_dbg is silent unless dyndbg= cmdline or sysfs control re-enables them — so the data stays available for debugging while shipping defaults are quiet. But this is a quick "shut it up" sweep, not the upstream-quality state.

What needs to happen before upstream

  • Keep dev_warn / dev_err for actual error paths
  • Drop dev_info for routine link-up state (delete, don't downgrade to dev_dbg)
  • Delete the dump helpers entirely, OR guard them behind a Kconfig debug option
  • Audit phy-gs101-ufs.c cdr-lock instrumentation specifically — added during dl_err 0x80000002 chase, can probably go entirely

Tracked patches affected

Affects upstream-patches/0012-ufs-phy-gs201-silence-debug-instrumentation-dev_info.patch in particular, but most of the 0001-0012 series carries some debug residue.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions