linux-kernel - Re: [PATCH] phy: qcom-qmp: Correct READY

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20190612172501.GY4814@minitux>
Date:   Wed, 12 Jun 2019 10:25:01 -0700
From:   Bjorn Andersson <bjorn.andersson@...aro.org>
To:     Marc Gonzalez <marc.w.gonzalez@...e.fr>
Cc:     Kishon Vijay Abraham I <kishon@...com>,
        MSM <linux-arm-msm@...r.kernel.org>,
        LKML <linux-kernel@...r.kernel.org>,
        Evan Green <evgreen@...omium.org>,
        Vivek Gautam <vivek.gautam@...eaurora.org>,
        Niklas Cassel <niklas.cassel@...aro.org>,
        Stanimir Varbanov <svarbanov@...sol.com>
Subject: Re: [PATCH] phy: qcom-qmp: Correct READY_STATUS poll break condition

On Wed 12 Jun 09:24 PDT 2019, Marc Gonzalez wrote:

> On 05/06/2019 01:24, Bjorn Andersson wrote:
> 
> > After issuing a PHY_START request to the QMP, the hardware documentation
> > states that the software should wait for the PCS_READY_STATUS to become
> > 1.
> > 
> > With the introduction of c9b589791fc1 ("phy: qcom: Utilize UFS reset
> > controller") an additional 1ms delay was introduced between the start
> > request and the check of the status bit. This greatly increases the
> > chances for the hardware to actually becoming ready before the status
> > bit is read.
> > 
> > The result can be seen in that UFS PHY enabling is now reported as a
> > failure in 10% of the boots on SDM845, which is a clear regression from
> > the previous rare/occasional failure.
> > 
> > This patch fixes the "break condition" of the poll to check for the
> > correct state of the status bit.
> > 
> > Unfortunately PCIe on 8996 and 8998 does not specify the mask_pcs_ready
> > register, which means that the code checks a bit that's always 0. So the
> > patch also fixes these, in order to not regress these targets.
> > 
> > Cc: stable@...r.kernel.org
> > Cc: Evan Green <evgreen@...omium.org>
> > Cc: Marc Gonzalez <marc.w.gonzalez@...e.fr>
> > Cc: Vivek Gautam <vivek.gautam@...eaurora.org>
> > Fixes: 73d7ec899bd8 ("phy: qcom-qmp: Add msm8998 PCIe QMP PHY support")
> > Fixes: e78f3d15e115 ("phy: qcom-qmp: new qmp phy driver for qcom-chipsets")
> > Signed-off-by: Bjorn Andersson <bjorn.andersson@...aro.org>
> > ---
> > 
> > @Kishon, this is a regression spotted in v5.2-rc1, so please consider applying
> > this towards v5.2.
> > 
> >  drivers/phy/qualcomm/phy-qcom-qmp.c | 4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/phy/qualcomm/phy-qcom-qmp.c b/drivers/phy/qualcomm/phy-qcom-qmp.c
> > index cd91b4179b10..43abdfd0deed 100644
> > --- a/drivers/phy/qualcomm/phy-qcom-qmp.c
> > +++ b/drivers/phy/qualcomm/phy-qcom-qmp.c
> > @@ -1074,6 +1074,7 @@ static const struct qmp_phy_cfg msm8996_pciephy_cfg = {
> >  
> >  	.start_ctrl		= PCS_START | PLL_READY_GATE_EN,
> >  	.pwrdn_ctrl		= SW_PWRDN | REFCLK_DRV_DSBL,
> > +	.mask_pcs_ready		= PHYSTATUS,
> >  	.mask_com_pcs_ready	= PCS_READY,
> >  
> >  	.has_phy_com_ctrl	= true,
> > @@ -1253,6 +1254,7 @@ static const struct qmp_phy_cfg msm8998_pciephy_cfg = {
> >  
> >  	.start_ctrl             = SERDES_START | PCS_START,
> >  	.pwrdn_ctrl		= SW_PWRDN | REFCLK_DRV_DSBL,
> > +	.mask_pcs_ready		= PHYSTATUS,
> >  	.mask_com_pcs_ready	= PCS_READY,
> >  };
> >  
> > @@ -1547,7 +1549,7 @@ static int qcom_qmp_phy_enable(struct phy *phy)
> >  	status = pcs + cfg->regs[QPHY_PCS_READY_STATUS];
> >  	mask = cfg->mask_pcs_ready;
> >  
> > -	ret = readl_poll_timeout(status, val, !(val & mask), 1,
> > +	ret = readl_poll_timeout(status, val, val & mask, 1,
> >  				 PHY_INIT_COMPLETE_TIMEOUT);
> >  	if (ret) {
> >  		dev_err(qmp->dev, "phy initialization timed-out\n");
> 
> Your patch made me realize that:
> msm8998_pciephy_cfg.has_phy_com_ctrl = false
> thus
> msm8998_pciephy_cfg.mask_com_pcs_ready is useless, AFAICT.
> 

While 8998 has a COM block, it does (among other things) not have a
ready bit. So afaict has_phy_com_ctrl = false is correct.

The addition of mask_pcs_ready is part of resolving the regression in
5.2, so I suggest that we remove mask_com_pcs_ready separately.

> (I copied msm8996_pciephy_cfg for msm8998_pciephy_cfg)
> 
> Does msm8996_pciephy_cfg really need both mask_pcs_ready AND
> mask_com_pcs_ready?
> 

8996 has a COM block and it contains both the control bits and the
status bits, so that looks correct.

> I'll test your patch tomorrow.
> 

I appreciate that.

Thanks,
Bjorn