When shutdown, rk808 driver shutdown first, so io_lock mutex is locked.
At same time, gpu change freq by runtime pm, will locked on rk808_reg_read.
Finally, gpu device shutdown, call pm_runtime_barrier. Deadlock happens.
Fix this bug by make rk808 shutdown later.
task PC stack pid father
init D c06e9a14 0 1 0 0x00000000
[<c06e9a14>] (__schedule+0x48c/0x554) from [<c02e281c>] (__pm_runtime_barrier+0x114/0x138)
[<c02e281c>] (__pm_runtime_barrier+0x114/0x138) from [<c02e3cf4>] (pm_runtime_barrier+0x68/0xb8)
[<c02e3cf4>] (pm_runtime_barrier+0x68/0xb8) from [<c02d85ac>] (device_shutdown+0xa0/0x190)
[<c02d85ac>] (device_shutdown+0xa0/0x190) from [<c0042128>] (kernel_restart+0xc/0x50)
[<c0042128>] (kernel_restart+0xc/0x50) from [<c00422c4>] (SyS_reboot+0x150/0x1b0)
[<c00422c4>] (SyS_reboot+0x150/0x1b0) from [<c000da00>] (ret_fast_syscall+0x0/0x30)
kworker/2:1 D c06e9a14 0 32 2 0x00000000
Workqueue: pm pm_runtime_work
[<c06e9a14>] (__schedule+0x48c/0x554) from [<c06e9ec0>] (schedule_preempt_disabled+0x24/0x34)
[<c06e9ec0>] (schedule_preempt_disabled+0x24/0x34) from [<c06e8bf8>] (__mutex_lock_slowpath+0x1c0/0x22c)
[<c06e8bf8>] (__mutex_lock_slowpath+0x1c0/0x22c) from [<c06e8c94>] (mutex_lock+0x30/0x48)
[<c06e8c94>] (mutex_lock+0x30/0x48) from [<c02f0dbc>] (rk808_reg_read+0x24/0x5c)
[<c02f0dbc>] (rk808_reg_read+0x24/0x5c) from [<c02f0fa8>] (rk808_dcdc_is_enabled+0x28/0x38)
[<c02f0fa8>] (rk808_dcdc_is_enabled+0x28/0x38) from [<c02a6acc>] (_regulator_is_enabled+0x34/0x40)
[<c02a6acc>] (_regulator_is_enabled+0x34/0x40) from [<c02a6c50>] (_regulator_do_set_voltage+0xbc/0x3e4)
[<c02a6c50>] (_regulator_do_set_voltage+0xbc/0x3e4) from [<c02a7c1c>] (regulator_set_voltage+0xc0/0xe0)
[<c02a7c1c>] (regulator_set_voltage+0xc0/0xe0) from [<c0027b7c>] (dvfs_regulator_set_voltage_readback.isra.6+0x14/0x8c)
[<c0027b7c>] (dvfs_regulator_set_voltage_readback.isra.6+0x14/0x8c) from [<c0027d5c>] (dvfs_scale_volt_direct+0xa0/0xc4)
[<c0027d5c>] (dvfs_scale_volt_direct+0xa0/0xc4) from [<c0027f24>] (dvfs_target+0x1a4/0x1d4)
[<c0027f24>] (dvfs_target+0x1a4/0x1d4) from [<c00273e8>] (dvfs_clk_set_rate+0x44/0x80)
[<c00273e8>] (dvfs_clk_set_rate+0x44/0x80) from [<c02c1388>] (mali_dvfs_clk_set+0x28/0x48)
[<c02c1388>] (mali_dvfs_clk_set+0x28/0x48) from [<c02c01f4>] (kbase_platform_dvfs_set_level+0x9c/0xe0)
[<c02c01f4>] (kbase_platform_dvfs_set_level+0x9c/0xe0) from [<c02c032c>] (kbase_platform_dvfs_enable+0xf4/0x110)
[<c02c032c>] (kbase_platform_dvfs_enable+0xf4/0x110) from [<c02bf908>] (pm_callback_runtime_off+0x3c/0x58)
[<c02bf908>] (pm_callback_runtime_off+0x3c/0x58) from [<c02d2650>] (kbase_device_runtime_suspend+0x24/0x34)
[<c02d2650>] (kbase_device_runtime_suspend+0x24/0x34) from [<c02de738>] (pm_generic_runtime_suspend+0x2c/0x40)
[<c02de738>] (pm_generic_runtime_suspend+0x2c/0x40) from [<c02e25c4>] (__rpm_callback+0x38/0x68)
[<c02e25c4>] (__rpm_callback+0x38/0x68) from [<c02e2658>] (rpm_callback+0x64/0x7c)
[<c02e2658>] (rpm_callback+0x64/0x7c) from [<c02e3210>] (rpm_suspend+0x2d8/0x534)
[<c02e3210>] (rpm_suspend+0x2d8/0x534) from [<c02e3e60>] (pm_runtime_work+0x7c/0x94)
[<c02e3e60>] (pm_runtime_work+0x7c/0x94) from [<c004773c>] (process_one_work+0x29c/0x458)
[<c004773c>] (process_one_work+0x29c/0x458) from [<c0047a8c>] (worker_thread+0x194/0x2d4)
[<c0047a8c>] (worker_thread+0x194/0x2d4) from [<c004cf74>] (kthread+0xa0/0xac)
[<c004cf74>] (kthread+0xa0/0xac) from [<c000da98>] (ret_from_fork+0x14/0x3c)
wifi: SDIO use flags to detect card-in or card-remove.If
tagged up wifi and off, then go into deep suspend and resume, we
do not judge if wifi is on or off, this will crash pm.resume for
drivers cannot access register without hclk_mmc.
[ 357.835059] 1192..dw_mci_set_ios: no card. [mmc1]
[ 358.322686] dwmmc_rockchip ff0d0000.rksdmmc: Timeout resetting block (ctrl reset 0x1)
[ 358.322792] dpm_run_callback(): platform_pm_resume+0x0/0x54 returns -19
[ 358.322813] PM: Device ff0d0000.rksdmmc failed to resume: error -19
Fix it by default return success state to pm.resume if wifi off in SDIO slot
[IMPORTANT COMMIT]
(1) Complete HS200 tuning algorithm for better performance & stability
(2) Auto workaround for error emmc clk settingin dts by Linux BSP engineer.
(3) Fixme: BUG on HS-DDR50 eMMC with div = 0
