Hi Niklas, On Wed, 9 Jan 2019, Niklas Söderlund wrote:
When removing the driver make sure to undo the PM autosuspend configured when probing the host. Signed-off-by: Niklas Söderlund <niklas.soderlund+renesas@xxxxxxxxxxxx> Reviewed-by: Wolfram Sang <wsa+renesas@xxxxxxxxxxxxxxxxxxxx>
This is commit bb60023c6387842b ("mmc: tmio: undo PM autosuspend when removing the host"), and I've bisected a regression to it. Sorry for taking that long to discover this issue (by loading a DT overlay to enable SCIFA5 on Koelsch, which requires unbinding SHDI0 first).
--- a/drivers/mmc/host/tmio_mmc_core.c +++ b/drivers/mmc/host/tmio_mmc_core.c @@ -1287,6 +1287,7 @@ void tmio_mmc_host_remove(struct tmio_mmc_host *host) cancel_delayed_work_sync(&host->delayed_reset_work); tmio_mmc_release_dma(host); + pm_runtime_dont_use_autosuspend(&pdev->dev); pm_runtime_put_sync(&pdev->dev); pm_runtime_disable(&pdev->dev); }
When unbinding the sdhi driver on e.g. Koelsch or Salvator-XS: echo ee100000.sd > /sys/bus/platform/drivers/sh_mobile_sdhi/unbind echo ee100000.sd > /sys/bus/platform/drivers/renesas_sdhi_internal_dmac/unbind Two warnings are triggered: WARNING: CPU: 0 PID: 586 at drivers/clk/clk.c:954 clk_core_disable+0x40/0x214 sdhi0 already disabled CPU: 0 PID: 586 Comm: bash Not tainted 5.7.0-rc3-shmobile-00508-g3c2c87c467f1c34e-dirty #547 Hardware name: Generic R-Car Gen2 (Flattened Device Tree) [<c010de5c>] (unwind_backtrace) from [<c0109cbc>] (show_stack+0x10/0x14) [<c0109cbc>] (show_stack) from [<c072d5ac>] (dump_stack+0x88/0xa8) [<c072d5ac>] (dump_stack) from [<c011c190>] (__warn+0xd0/0xec) [<c011c190>] (__warn) from [<c011c21c>] (warn_slowpath_fmt+0x70/0x9c) [<c011c21c>] (warn_slowpath_fmt) from [<c03a09e4>] (clk_core_disable+0x40/0x214) [<c03a09e4>] (clk_core_disable) from [<c03a0bd0>] (clk_core_disable_lock+0x18/0x24) [<c03a0bd0>] (clk_core_disable_lock) from [<c059e910>] (renesas_sdhi_clk_disable+0x14/0x34) [<c059e910>] (renesas_sdhi_clk_disable) from [<c059f2fc>] (renesas_sdhi_remove+0x18/0x20) [<c059f2fc>] (renesas_sdhi_remove) from [<c043c054>] (platform_drv_remove+0x24/0x3c) [<c043c054>] (platform_drv_remove) from [<c043afb4>] (device_release_driver_internal+0xf0/0x198) [<c043afb4>] (device_release_driver_internal) from [<c0439140>] (unbind_store+0x44/0x68) [<c0439140>] (unbind_store) from [<c02941fc>] (kernfs_fop_write+0x128/0x1e4) [<c02941fc>] (kernfs_fop_write) from [<c0225598>] (__vfs_write+0x28/0xe0) [<c0225598>] (__vfs_write) from [<c0226664>] (vfs_write+0x98/0xbc) [<c0226664>] (vfs_write) from [<c02267ec>] (ksys_write+0x68/0xb4) [<c02267ec>] (ksys_write) from [<c0100060>] (ret_fast_syscall+0x0/0x54) ... WARNING: CPU: 0 PID: 586 at drivers/clk/clk.c:813 clk_core_unprepare+0x40/0x108 sdhi0 already unprepared CPU: 0 PID: 586 Comm: bash Tainted: G W 5.7.0-rc3-shmobile-00508-g3c2c87c467f1c34e-dirty #547 Hardware name: Generic R-Car Gen2 (Flattened Device Tree) [<c010de5c>] (unwind_backtrace) from [<c0109cbc>] (show_stack+0x10/0x14) [<c0109cbc>] (show_stack) from [<c072d5ac>] (dump_stack+0x88/0xa8) [<c072d5ac>] (dump_stack) from [<c011c190>] (__warn+0xd0/0xec) [<c011c190>] (__warn) from [<c011c21c>] (warn_slowpath_fmt+0x70/0x9c) [<c011c21c>] (warn_slowpath_fmt) from [<c03a1150>] (clk_core_unprepare+0x40/0x108) [<c03a1150>] (clk_core_unprepare) from [<c03a2d34>] (clk_unprepare+0x24/0x2c) [<c03a2d34>] (clk_unprepare) from [<c044e244>] (__pm_clk_remove+0x38/0x54) [<c044e244>] (__pm_clk_remove) from [<c044e6cc>] (pm_clk_destroy+0xc4/0xdc) [<c044e6cc>] (pm_clk_destroy) from [<c044c4ec>] (genpd_remove_device+0x10c/0x128) [<c044c4ec>] (genpd_remove_device) from [<c044c54c>] (genpd_dev_pm_detach+0x44/0xe4) [<c044c54c>] (genpd_dev_pm_detach) from [<c043c064>] (platform_drv_remove+0x34/0x3c) [<c043c064>] (platform_drv_remove) from [<c043afb4>] (device_release_driver_internal+0xf0/0x198) [<c043afb4>] (device_release_driver_internal) from [<c0439140>] (unbind_store+0x44/0x68) [<c0439140>] (unbind_store) from [<c02941fc>] (kernfs_fop_write+0x128/0x1e4) [<c02941fc>] (kernfs_fop_write) from [<c0225598>] (__vfs_write+0x28/0xe0) [<c0225598>] (__vfs_write) from [<c0226664>] (vfs_write+0x98/0xbc) [<c0226664>] (vfs_write) from [<c02267ec>] (ksys_write+0x68/0xb4) [<c02267ec>] (ksys_write) from [<c0100060>] (ret_fast_syscall+0x0/0x54) Removing the call to pm_runtime_dont_use_autosuspend() fixes that symptom. But that is definitely not the right fix, as doing so causes genpd_stop_dev() to no longer being called on unbind. What seems to happen is: - During driver probe, the clocks are enabled twice, by 1. renesas_sdhi_clk_enable(), called from renesas_sdhi_probe(), 2. pm_clk_resume(), called from tmio_mmc_host_probe(). - During driver unbind, the clocks are disabled thrice, by: 1. renesas_sdhi_clk_disable(), called from tmio_mmc_host_runtime_suspend(), 2. pm_clk_suspend(), called from tmio_mmc_host_remove() through pm_runtime_put_sync(), 3. renesas_sdhi_clk_disable(), called from renesas_sdhi_remove(). In addition, an extra clock unprepare happens in __pm_clk_remove(), called from genpd_remove_device() through pm_clk_destroy(). Note that tmio_mmc_host_runtime_suspend() is never called, so there is definitely an imbalance in the Runtime PM handling.. Without the call to pm_runtime_dont_use_autosuspend(), the clocks are disabled twice during driver unbind, by: 1. renesas_sdhi_clk_disable(), called from renesas_sdhi_remove(), 2. __pm_clk_remove(), called from genpd_remove_device() through pm_clk_destroy(). Hence the warning is gone, by accident, as e.g. the pm_clk_resume() call is never balanced by its counterpart pm_clk_suspend(). Thanks for your comments! Gr{oetje,eeting}s, Geert -- Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx In personal conversations with technical people, I call myself a hacker. But when I'm talking to journalists I just say "programmer" or something like that. -- Linus Torvalds