| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Thu, 19 Dec 2019 15:12:25 +0800
From: cang@...eaurora.org
To: Vinod Koul <vkoul@...nel.org>
Cc: Jeffrey Hugo <jeffrey.l.hugo@...il.com>, asutoshd@...eaurora.org,
nguyenb@...eaurora.org, Rajendra Nayak <rnayak@...eaurora.org>,
linux-scsi@...r.kernel.org, kernel-team@...roid.com,
saravanak@...gle.com, Mark Salyzyn <salyzyn@...gle.com>,
Andy Gross <agross@...nel.org>,
Alim Akhtar <alim.akhtar@...sung.com>,
Avri Altman <avri.altman@....com>,
Pedro Sousa <pedrom.sousa@...opsys.com>,
"James E.J. Bottomley" <jejb@...ux.ibm.com>,
"Martin K. Petersen" <martin.petersen@...cle.com>,
"open list:ARM/QUALCOMM SUPPORT" <linux-arm-msm@...r.kernel.org>,
open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH v5 2/7] scsi: ufs-qcom: Add reset control support for host
controller
On 2019-12-18 12:12, Vinod Koul wrote:
> On 18-12-19, 02:44, cang@...eaurora.org wrote:
>
>> Hi Vinod and Jeffrey,
>>
>> Let me summary here, now the 1000000us timeout works for both 845 and
>> 8998.
>> However, 8150 still fails.
>>
>> > > The bigger question is why is the reset causing the timeout to be
>> > > increased for sdm845 and not to work in case of sm8150! (Vinod)
>>
>> I would not say this patch increases the timeout. With this patch,
>> the PCS polling timeout, per my profiling, the PCS ready usually needs
>> less than 5000us, which is the actual time needed for PCS bit to be
>> ready.
>>
>> The reason why 1000us worked for you is because, w/o the patch, UFS
>> PHY
>> registers are retained from pre-kernel stage (bootloader i.e.), the
>> PCS
>> ready
>> bit was set to 1 in pre-kernel stage, so when kernel driver reads it,
>> it
>> returns
>> 1, not even to be polled at all. It may seem "faster", but not the
>> right
>> thing to do, because kernel stage may need different PHY settings than
>> pre-kernel stage, keeping the settings configured in pre-kernel stage
>> is not
>> always right, so this patch is needed. And increasing 1000us to
>> 1000000us
>> is the right thing to do, but not a hack.
>>
>> As reg for the phy initialization timeout on 8150, I found there is
>> something
>> wrong with its settings in /drivers/phy/qualcomm/phy-qcom-qmp.c
>>
>> static const struct qmp_phy_init_tbl sm8150_ufsphy_serdes_tbl[] = {
>> QMP_PHY_INIT_CFG(QPHY_POWER_DOWN_CONTROL, 0x01),
>> QMP_PHY_INIT_CFG(QSERDES_V4_COM_SYSCLK_EN_SEL, 0xd9),
>>
>> "QMP_PHY_INIT_CFG(QPHY_POWER_DOWN_CONTROL, 0x01)" should NOT appear in
>> the
>> serdes
>> table! I haven't check who made this change, but please have a try
>> after
>> remove
>> this line from sm8150_ufsphy_serdes_tbl.
>
> That is me :) Looks like I made an error while porting from downstream.
> I
> did a quick check to remove this and it doesn't work yet, let me
> recheck
> the settings again ...
>
> Thanks for your help!
Hi Vinod,
Indeed, you need to tweak your settings. I spent some time help you
figure this out. Try below change and please let me know if it can
resolve your problem.
I would not say this is a regression caused by my patch, it is just
my patch reveals something incorrect in the settings.
diff --git a/drivers/phy/qualcomm/phy-qcom-qmp.c
b/drivers/phy/qualcomm/phy-qcom-qmp.c
index 8e642a6..0cc9044 100755
--- a/drivers/phy/qualcomm/phy-qcom-qmp.c
+++ b/drivers/phy/qualcomm/phy-qcom-qmp.c
@@ -66,7 +66,7 @@
/* QPHY_V3_PCS_MISC_CLAMP_ENABLE register bits */
#define CLAMP_EN BIT(0) /* enables i/o
clamp_n */
-#define PHY_INIT_COMPLETE_TIMEOUT 1000
+#define PHY_INIT_COMPLETE_TIMEOUT 1000000
#define POWER_DOWN_DELAY_US_MIN 10
#define POWER_DOWN_DELAY_US_MAX 11
@@ -166,6 +166,7 @@ static const unsigned int
sdm845_ufsphy_regs_layout[] = {
};
static const unsigned int sm8150_ufsphy_regs_layout[] = {
+ [QPHY_SW_RESET] = 0x08,
[QPHY_START_CTRL] = 0x00,
[QPHY_PCS_READY_STATUS] = 0x180,
};
@@ -885,7 +886,6 @@ static const struct qmp_phy_init_tbl
msm8998_usb3_pcs_tbl[] = {
};
static const struct qmp_phy_init_tbl sm8150_ufsphy_serdes_tbl[] = {
- QMP_PHY_INIT_CFG(QPHY_POWER_DOWN_CONTROL, 0x01),
QMP_PHY_INIT_CFG(QSERDES_V4_COM_SYSCLK_EN_SEL, 0xd9),
QMP_PHY_INIT_CFG(QSERDES_V4_COM_HSCLK_SEL, 0x11),
QMP_PHY_INIT_CFG(QSERDES_V4_COM_HSCLK_HS_SWITCH_SEL, 0x00),
@@ -1390,7 +1390,6 @@ static const struct qmp_phy_cfg sm8150_ufsphy_cfg
= {
.pwrdn_ctrl = SW_PWRDN,
.is_dual_lane_phy = true,
- .no_pcs_sw_reset = true,
};
static void qcom_qmp_phy_configure(void __iomem *base,
---
Aside of the phy settings, your DT needs some modifications too,
seems you copied most of them from sdm845.
https://git.kernel.org/pub/scm/linux/kernel/git/qcom/linux.git/commit/?h=for-next&id=3834a2e92229ef26d30de28acb698b2b23d3e397
<--snip-->
> + ufs_mem_phy: phy@...7000 {
> + compatible = "qcom,sm8150-qmp-ufs-phy";
> + reg = <0 0x01d87000 0 0x18c>;
The size 0x18c is wrong, in the code you are even accessing registers
whose offsets are beyond 0x18c, see
#define QSERDES_V4_COM_BIN_VCOCAL_CMP_CODE1_MODE0 0x1ac
#define QSERDES_V4_COM_BIN_VCOCAL_CMP_CODE2_MODE0 0x1b0
#define QSERDES_V4_COM_BIN_VCOCAL_CMP_CODE1_MODE1 0x1b4
#define QSERDES_V4_COM_BIN_VCOCAL_HSCLK_SEL 0x1bc
#define QSERDES_V4_COM_BIN_VCOCAL_CMP_CODE2_MODE1 0x1b8
FYI, the total size of serdes registers is 0x1c0.
<--snip-->
> + ufs_mem_phy_lanes: lanes@...7400 {
> + reg = <0 0x01d87400 0 0x108>,
> + <0 0x01d87600 0 0x1e0>,
> + <0 0x01d87c00 0 0x1dc>,
Same as above, see
#define QPHY_V4_MULTI_LANE_CTRL1 0x1e0
FYI, the total size of PCS registers is 0x200
> + <0 0x01d87800 0 0x108>,
> + <0 0x01d87a00 0 0x1e0>;
> + #phy-cells = <0>;
> + };
<--snip-->
Powered by blists - more mailing lists