Re: [PATCH v2 3/3] scsi: ufs: rockchip: init support for UFS

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



在 2024/8/12 12:10, Manivannan Sadhasivam 写道:
On Mon, Aug 12, 2024 at 09:28:26AM +0800, Shawn Lin wrote:
JHi Mani,

在 2024/8/10 17:28, Manivannan Sadhasivam 写道:
On Fri, Aug 09, 2024 at 04:16:41PM +0800, Shawn Lin wrote:

[...]

+static int ufs_rockchip_hce_enable_notify(struct ufs_hba *hba,
+					 enum ufs_notify_change_status status)
+{
+	int err = 0;
+
+	if (status == PRE_CHANGE) {
+		int retry_outer = 3;
+		int retry_inner;
+start:
+		if (ufshcd_is_hba_active(hba))
+			/* change controller state to "reset state" */
+			ufshcd_hba_stop(hba);
+
+		/* UniPro link is disabled at this point */
+		ufshcd_set_link_off(hba);
+
+		/* start controller initialization sequence */
+		ufshcd_writel(hba, CONTROLLER_ENABLE, REG_CONTROLLER_ENABLE);
+
+		usleep_range(100, 200);
+
+		/* wait for the host controller to complete initialization */
+		retry_inner = 50;
+		while (!ufshcd_is_hba_active(hba)) {
+			if (retry_inner) {
+				retry_inner--;
+			} else {
+				dev_err(hba->dev,
+					"Controller enable failed\n");
+				if (retry_outer) {
+					retry_outer--;
+					goto start;
+				}
+				return -EIO;
+			}
+			usleep_range(1000, 1100);
+		}

You just duplicated ufshcd_hba_execute_hce() here. Why? This doesn't make sense.

Since we set UFSHCI_QUIRK_BROKEN_HCE, and we also need to do someting
which is very similar to ufshcd_hba_execute_hce(), before calling
ufshcd_dme_reset(). Similar but not totally the same. I'll try to see if
we can export ufshcd_hba_execute_hce() to make full use of it.


But you are starting the controller using REG_CONTROLLER_ENABLE. Isn't that
supposed to be broken if you set UFSHCI_QUIRK_BROKEN_HCE? Or I am
misunderstanding the quirk?


Our controller doesn't work with exiting code, whether setting
UFSHCI_QUIRK_BROKEN_HCE or not.


Okay. Then this means you do not need this quirk at all.


For UFSHCI_QUIRK_BROKEN_HCE case, it calls ufshcd_dme_reset()first,
but we need to set REG_CONTROLLER_ENABLE first.

For !UFSHCI_QUIRK_BROKEN_HCE case, namly ufshcd_hba_execute_hce, it
sets REG_CONTROLLER_ENABLE  first but never send DMA_RESET and calls
ufshcd_dme_enable.


I don't see where ufshcd_dme_enable() is getting called for
!UFSHCI_QUIRK_BROKEN_HCE case.

So the closet code path is to go through UFSHCI_QUIRK_BROKEN_HCE case,
and set REG_CONTROLLER_ENABLE by adding hce_enable_notify hook.


No, that is abusing the quirk. But I'm confused about why your controller wants
resetting the unipro stack _after_ enabling the controller? Why can't it be
reset before?


It can't be. The DME_RESET to reset the unipro stack will be failed
without enabling REG_CONTROLLER_ENABLE. And the controller does want us
to reset the unipro stack before other coming UICs.

So I considered it's a kind of broken HCE case as well. Should I add a
new quirk or add a new hba_enable hook in ufs_hba_variant_ops? Or just
use UFSHCI_QUIRK_BROKEN_HCE ?


+	} else { /* POST_CHANGE */
+		err = ufshcd_vops_phy_initialization(hba);
+	}
+
+	return err;
+}
+

[...]

+static const struct dev_pm_ops ufs_rockchip_pm_ops = {
+	SET_SYSTEM_SLEEP_PM_OPS(ufs_rockchip_suspend, ufs_rockchip_resume)
+	SET_RUNTIME_PM_OPS(ufs_rockchip_runtime_suspend, ufs_rockchip_runtime_resume, NULL)

Why can't you use ufshcd PM ops as like other vendor drivers?

It doesn't work from the test. We have many use case to power down the
controller and device, so there is no flow to recovery the link. Only
when the first accessing to UFS fails, the ufshcd error handle recovery the
link. This is not what we expect.


What tests? The existing UFS controller drivers are used in production devices
and they never had a usecase to invent their own PM callbacks. So if your
controller is special, then you need to justify it more elaborately. If
something is missing in ufshcd callbacks, then we can add them.


All the register got lost each time as we power down both controller & PHY
and devices in suspend.

Which suspend? runtime or system suspend? I believe system suspend.

Both.


So we have to restore the necessary
registers and link. I didn't see where the code recovery the controller
settings in ufshcd_resume, except ufshcd_err_handler()triggers that.
Am I missing any thing?

Can you explain what is causing the powerdown of the controller and PHY?
Because, ufshcd_suspend() just turns off the clocks and regulators (if
UFSHCD_CAP_AGGR_POWER_COLLAPSE is set) and spm_lvl 3 set by this driver only
puts the device in sleep mode and link in hibern8 state.


For runtime PM case, it's the power-domain driver will power down the
controller and PHY if UFS stack is not active any more(autosuspend).

For system PM case, it's the SoC's firmware to cutting of all the power
for controller/PHY and device.

- Mani

Below is the dump we get if using
SET_SYSTEM_SLEEP_PM_OPS(ufshcd_system_suspend, ufshcd_system_resume).
It can work as ufshcd_err_handler () will fix the link, but we have to
suffer from getting the error log each time. Moreover, we need to gate
26MHz refclk for device when RPM is called. So our own rpm callback is
needed.

[   14.318444] <<GTP-INF>>[gt1x_wakeup_sleep:964] Wakeup by poweron
[   14.439723] ufshcd-rockchip 2a2d0000.ufs: Controller not ready to accept
UIC commands
[   14.439730] ufshcd-rockchip 2a2d0000.ufs: pwr ctrl cmd 0x18 with mode 0x0
uic error -5
[   14.439736] ufshcd-rockchip 2a2d0000.ufs: UFS Host state=1
[   14.439740] ufshcd-rockchip 2a2d0000.ufs: outstanding reqs=0x0 tasks=0x0
[   14.439744] ufshcd-rockchip 2a2d0000.ufs: saved_err=0x0,
saved_uic_err=0x0
[   14.439748] ufshcd-rockchip 2a2d0000.ufs: Device power mode=2, UIC link
state=2
[   14.439753] ufshcd-rockchip 2a2d0000.ufs: PM in progress=1, sys.
suspended=1
[   14.439758] ufshcd-rockchip 2a2d0000.ufs: Auto BKOPS=0, Host self-block=0
[   14.439762] ufshcd-rockchip 2a2d0000.ufs: Clk gate=1
[   14.439766] ufshcd-rockchip 2a2d0000.ufs: last_hibern8_exit_tstamp at 0
us, hibern8_exit_cnt=0
[   14.439770] ufshcd-rockchip 2a2d0000.ufs: last intr at 10807625 us, last
intr status=0x440
[   14.439775] ufshcd-rockchip 2a2d0000.ufs: error handling flags=0x0, req.
abort count=0
[   14.439779] ufshcd-rockchip 2a2d0000.ufs: hba->ufs_version=0x200, Host
capabilities=0x187011f, caps=0x48c
[   14.439785] ufshcd-rockchip 2a2d0000.ufs: quirks=0x2100, dev. quirks=0xc4
[   14.439790] ufshcd-rockchip 2a2d0000.ufs: UFS dev info: SAMSUNG
KLUDG2R1DE-B0F1  rev 0100
[   14.439796] ufshcd-rockchip 2a2d0000.ufs: clk: core, rate: 50000000
[   14.439822] host_regs: 00000000: 0187011f 00000000 00000200 00000000
[   14.439827] host_regs: 00000010: 00000000 000005e6 00000000 00000000
[   14.439831] host_regs: 00000020: 00000000 00000000 00000000 00000000
[   14.439835] host_regs: 00000030: 00000000 00000000 00000000 00000000
[   14.439839] host_regs: 00000040: 00000000 00000000 00000000 00000000
[   14.439843] host_regs: 00000050: 00000000 00000000 00000000 00000000
[   14.439847] host_regs: 00000060: 00000000 00000000 00000000 00000000
[   14.439851] host_regs: 00000070: 00000000 00000000 00000000 00000000
[   14.439855] host_regs: 00000080: 00000000 00000000 00000000 00000000
[   14.439859] host_regs: 00000090: 00000000 00000000 00000000 00000000
[   14.439863] ufshcd-rockchip 2a2d0000.ufs: No record of pa_err
[   14.439867] ufshcd-rockchip 2a2d0000.ufs: No record of dl_err
[   14.439871] ufshcd-rockchip 2a2d0000.ufs: No record of nl_err
[   14.439876] ufshcd-rockchip 2a2d0000.ufs: No record of tl_err
[   14.439880] ufshcd-rockchip 2a2d0000.ufs: No record of dme_err
[   14.439884] ufshcd-rockchip 2a2d0000.ufs: No record of auto_hibern8_err
[   14.439888] ufshcd-rockchip 2a2d0000.ufs: No record of fatal_err
[   14.439892] ufshcd-rockchip 2a2d0000.ufs: No record of link_startup_fail
[   14.439896] ufshcd-rockchip 2a2d0000.ufs: No record of resume_fail
[   14.439900] ufshcd-rockchip 2a2d0000.ufs: No record of suspend_fail
[   14.439905] ufshcd-rockchip 2a2d0000.ufs: dev_reset[0] = 0x0 at 1418763
us
[   14.439910] ufshcd-rockchip 2a2d0000.ufs: dev_reset: total cnt=1
[   14.439914] ufshcd-rockchip 2a2d0000.ufs: No record of host_reset
[   14.439918] ufshcd-rockchip 2a2d0000.ufs: No record of task_abort
[   14.439930] ufshcd-rockchip 2a2d0000.ufs: ufshcd_uic_hibern8_exit:
hibern8 exit failed. ret = -5
[   14.439935] ufshcd-rockchip 2a2d0000.ufs: __ufshcd_wl_resume: hibern8
exit failed -5
[   14.439944] ufs_device_wlun 0:0:0:49488: ufshcd_wl_resume failed: -5
[   14.439950] ufs_device_wlun 0:0:0:49488: PM: dpm_run_callback():
scsi_bus_resume+0x0/0xa8 returns -5
[   14.440003] ufshcd-rockchip 2a2d0000.ufs: ufshcd_err_handler started; HBA
state eh_fatal; powered 1; shutting down 0; saved_err = 0; saved_uic_err =
0; force_reset = 0; link is broken
[   14.440017] ufs_device_wlun 0:0:0:49488: PM: failed to resume async:
error -5

- Mani






[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Index of Archives]     [SCSI Target Devel]     [Linux SCSI Target Infrastructure]     [Kernel Newbies]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Linux IIO]     [Samba]     [Device Mapper]

  Powered by Linux