On 16/03/2023 13:32, Marc Gonzalez wrote: > On 15/03/2023 22:14, Heiner Kallweit wrote: > >> Then you mention "too soon after reset", but add a delay to power-up. >> If the delay would be needed after reset, then shouldn't it be in >> meson_mmc_probe() after the call to device_reset_optional()? > > The first mmc_power_up() is called from: > > [ 0.916446] mmc_power_up+0x2c/0xe4 > [ 0.926186] mmc_start_host+0x94/0xa0 > [ 0.926197] mmc_add_host+0x84/0xf0 > [ 0.926205] meson_mmc_probe+0x374/0x3f4 > [ 0.933086] platform_probe+0x68/0xe0 > [ 0.940159] really_probe+0xbc/0x2f0 > [ 0.948008] __driver_probe_device+0x78/0xe0 > [ 0.955080] driver_probe_device+0xd8/0x160 > [ 0.955087] __driver_attach_async_helper+0x4c/0xc0 > [ 0.965689] async_run_entry_fn+0x34/0xe0 > [ 0.965699] process_one_work+0x1cc/0x320 > [ 0.972847] worker_thread+0x14c/0x450 > [ 0.972855] kthread+0x10c/0x110 > [ 1.018906] ret_from_fork+0x10/0x20 > > > The second mmc_power_up() (prematurely exited) is called from: > > [ 1.061686] mmc_power_up+0x2c/0xe4 > [ 1.065136] mmc_rescan+0x18c/0x310 > [ 1.068586] process_one_work+0x1cc/0x320 > [ 1.072553] worker_thread+0x14c/0x450 > [ 1.076262] kthread+0x10c/0x110 > [ 1.086528] ret_from_fork+0x10/0x20 > > > I'm confused about device_reset_optional() vs mmc_power_up(). > Do they both reset the controller? NB: adding 500 ms delay after device_reset_optional() DOES NOT fix the issue: [ 0.875494] >ffe03000.sd [ 0.875823] meson-gx-mmc ffe03000.sd: allocated mmc-pwrseq [ 1.027111] >ffe07000.mmc [ 1.027430] meson-gx-mmc ffe07000.mmc: allocated mmc-pwrseq [ 1.405170] +mmc2 [ 1.405389] -mmc2 [ 1.405413] <ffe03000.sd [ 1.405500] +mmc2 [ 1.412957] Q [ 1.420317] R [ 1.422590] mmc2: YO no device WTF We enter mmc_power_up at 1.405170 We enter mmc_attach_sdio at 1.412957 Empirically, 8 ms is not enough time for the SDIO bus to be ready. Reverting to default power_delay_ms. [ 0.879842] >ffe03000.sd [ 0.880189] meson-gx-mmc ffe03000.sd: allocated mmc-pwrseq [ 1.031635] >ffe07000.mmc [ 1.031958] meson-gx-mmc ffe07000.mmc: allocated mmc-pwrseq [ 1.405284] +mmc2 [ 1.430499] -mmc2 [ 1.430525] <ffe03000.sd [ 1.430616] +mmc2 [ 1.438065] Q [ 1.445404] R [ 1.447677] mmc2: YO no device WTF We enter mmc_power_up at 1.405284 We enter mmc_attach_sdio at 1.438065 Empirically, 33 ms is not enough time for the SDIO bus to be ready. Increasing power_delay_ms to 15. [ 0.879788] >ffe03000.sd [ 0.880092] meson-gx-mmc ffe03000.sd: allocated mmc-pwrseq [ 1.031437] >ffe07000.mmc [ 1.031767] meson-gx-mmc ffe07000.mmc: allocated mmc-pwrseq [ 1.405262] +mmc2 [ 1.442981] -mmc2 [ 1.443007] <ffe03000.sd [ 1.443099] +mmc2 [ 1.450549] Q [ 1.565950] +mmc1 [ 1.584306] mmc2: new ultra high speed SDR50 SDIO card at address 0001 We enter mmc_power_up at 1.405262 We enter mmc_attach_sdio at 1.450549 Empirically, 45 ms *is* enough time for the SDIO bus to be ready. My gut feeling is that the problem with SDIO devices exists on all boards. (I am well aware that I may be wrong.) Would you be willing to test a small kernel? I can provide a kernel defconfig and/or a buildroot defconfig. Regards
