On Mon, Jun 10, 2019 at 03:44:01PM +0200, Paolo Pisati wrote:
> From time to time, my apq8096-db820c fails to boot to userspace, reset back to
> fastboot and locks up: to easily reproduce the issue, i'm boot looping using a
> cron job with a 1 min reboot entry on the board while leaving a "while 1; do
> fastboot boot boot.img; done" on the host pc.
>
> The issue is present in mainline up to 5.2-rc4, using defconfig and:
>
> CONFIG_SCSI_UFS_QCOM=y
> CONFIG_PHY_QCOM_QMP=y
> CONFIG_PHY_QCOM_UFS=y
>
> but was present in previous releases too (e.g. 4.14., 4.19, etc qcom-lt or
> mainline), where it's even easier to reproduce (e.g. takes way less reboots to
> trigger it).
Hello Paolo,
I have a guess of what is going on.
db820c has 3 PCIe controllers,
that shares a singe QMP block (that has clocks, regulators, and resets).
The QMP block has 3 PCIe PHYs, that have their own clocks and resets.
>
> These are the last lines printed out:
> ...
> [ 7.407209] qcom-qmp-phy 34000.phy: Registered Qcom-QMP phy
> [ 7.448058] qcom-qmp-phy 7410000.phy: Registered Qcom-QMP phy
> [ 7.461859] ufs_qcom_phy_qmp_14nm 627000.phy: invalid resource
> [ 7.535434] qcom-qmp-phy 34000.phy: phy common block init timed-out
^^ here the phy_init() called from pcie-qcom.c
which ends up to a call to qcom_qmp_phy_enable()
which has this code:
ret = qcom_qmp_phy_com_init(qphy);
if (ret)
return ret;
qcom_qmp_phy_com_init() has this code:
if (qmp->init_count++) {
mutex_unlock(&qmp->phy_mutex);
return 0;
}
qcom_qmp_phy_com_init() later fails,
since the common block init time out, so the qmp driver
disables clocks, asserts reset, and disables regulators
> [ 7.538596] phy phy-34000.phy.0: phy init failed --> -110
> [ 7.550891] qcom-pcie: probe of 600000.pcie failed with error -110
^^ here the first PCIe controller instance fails to probe
> [ 7.619008] qcom-pcie 608000.pcie: 608000.pcie supply vddpe-3v3 not found,
> using dummy regulator
^^ here the second PCIe controller is probed.
it will call phy_init()
which will again call qcom_qmp_phy_enable() which will call
qcom_qmp_phy_com_init()
where this code:
if (qmp->init_count++) {
mutex_unlock(&qmp->phy_mutex);
return 0;
}
now will return 0,
so clocks will never be enabled, resets never deasserted, regulators
never enabled.
since qcom_qmp_phy_com_init() returns success in this case,
qcom_qmp_phy_enable() will try to continue with the init,
and writes to disabled hardware is usually not a good idea.
I think the proper fix for this is:
diff --git a/drivers/phy/qualcomm/phy-qcom-qmp.c b/drivers/phy/qualcomm/phy-qcom-qmp.c
index cd91b4179b10..22352e3b0ec5 100644
--- a/drivers/phy/qualcomm/phy-qcom-qmp.c
+++ b/drivers/phy/qualcomm/phy-qcom-qmp.c
@@ -1490,7 +1490,7 @@ static int qcom_qmp_phy_enable(struct phy *phy)
ret = qcom_qmp_phy_com_init(qphy);
if (ret)
- return ret;
+ goto err_lane_rst;
if (cfg->has_lane_rst) {
ret = reset_control_deassert(qphy->lane_rst);
Kind regards,
Niklas
>
> Log Type: B - Since Boot(Power On Reset), D - Delta, S - Statistic
> S - QC_IMAGE_VERSION_STRING=BOOT.XF.1.0-00301
> S - IMAGE_VARIANT_STRING=M8996LAB
> S - OEM_IMAGE_VERSION_STRING=crm-ubuntu68
> S - Boot Interface: UFS
> S - Secure Boot: Off
> ...
>
> Full boot here: https://pastebin.ubuntu.com/p/rtjVrD3yzk/
>
> Any idea what is going on? Am i doing something wrong?
> --
> bye,
> p.
