linux-kernel - Re: [PATCH v5 2/2] misc: rtsx: Add power saving functions and fix driving parameter

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: <20200909174424.GA717281@bjorn-Precision-5520>
Date:   Wed, 9 Sep 2020 12:44:24 -0500
From:   Bjorn Helgaas <helgaas@...nel.org>
To:     吳昊澄 Ricky <ricky_wu@...ltek.com>
Cc:     "arnd@...db.de" <arnd@...db.de>,
        "gregkh@...uxfoundation.org" <gregkh@...uxfoundation.org>,
        "bhelgaas@...gle.com" <bhelgaas@...gle.com>,
        "ulf.hansson@...aro.org" <ulf.hansson@...aro.org>,
        "rui_feng@...lsil.com.cn" <rui_feng@...lsil.com.cn>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "puranjay12@...il.com" <puranjay12@...il.com>,
        "linux-pci@...r.kernel.org" <linux-pci@...r.kernel.org>,
        "vailbhavgupta40@...ail.com" <vailbhavgupta40@...ail.com>
Subject: Re: [PATCH v5 2/2] misc: rtsx: Add power saving functions and fix
 driving parameter

On Wed, Sep 09, 2020 at 07:10:19AM +0000, 吳昊澄 Ricky wrote:
> > -----Original Message-----
> > From: Bjorn Helgaas [mailto:helgaas@...nel.org]
> > Sent: Wednesday, September 09, 2020 6:29 AM
> > To: 吳昊澄 Ricky
> > Cc: arnd@...db.de; gregkh@...uxfoundation.org; bhelgaas@...gle.com;
> > ulf.hansson@...aro.org; rui_feng@...lsil.com.cn; linux-kernel@...r.kernel.org;
> > puranjay12@...il.com; linux-pci@...r.kernel.org;
> > vailbhavgupta40@...ail.com
> > Subject: Re: [PATCH v5 2/2] misc: rtsx: Add power saving functions and fix driving
> > parameter
> > 
> > On Mon, Sep 07, 2020 at 06:07:31PM +0800, ricky_wu@...ltek.com wrote:
> > > From: Ricky Wu <ricky_wu@...ltek.com>
> > >
> > > v4:
> > > split power down flow and power saving function to two patch
> > >
> > > v5:
> > > fix up modified change under the --- line
> > 
> > Hehe, this came out *above* the "---" line :)
> > 
> > > Add rts522a L1 sub-state support
> > > Save more power on rts5227 rts5249 rts525a rts5260
> > > Fix rts5260 driving parameter
> > >
> > > Signed-off-by: Ricky Wu <ricky_wu@...ltek.com>
> > > ---
> > >  drivers/misc/cardreader/rts5227.c  | 112 +++++++++++++++++++++-
> > >  drivers/misc/cardreader/rts5249.c  | 145
> > ++++++++++++++++++++++++++++-
> > >  drivers/misc/cardreader/rts5260.c  |  28 +++---
> > >  drivers/misc/cardreader/rtsx_pcr.h |  17 ++++
> > >  4 files changed, 283 insertions(+), 19 deletions(-)
> > >
> > > diff --git a/drivers/misc/cardreader/rts5227.c
> > b/drivers/misc/cardreader/rts5227.c
> > > index 747391e3fb5d..8859011672cb 100644
> > > --- a/drivers/misc/cardreader/rts5227.c
> > > +++ b/drivers/misc/cardreader/rts5227.c
> > > @@ -72,15 +72,80 @@ static void rts5227_fetch_vendor_settings(struct
> > rtsx_pcr *pcr)
> > >
> > >  	pci_read_config_dword(pdev, PCR_SETTING_REG2, &reg);
> > >  	pcr_dbg(pcr, "Cfg 0x%x: 0x%x\n", PCR_SETTING_REG2, reg);
> > > +	if (rtsx_check_mmc_support(reg))
> > > +		pcr->extra_caps |= EXTRA_CAPS_NO_MMC;
> > >  	pcr->sd30_drive_sel_3v3 = rtsx_reg_to_sd30_drive_sel_3v3(reg);
> > >  	if (rtsx_reg_check_reverse_socket(reg))
> > >  		pcr->flags |= PCR_REVERSE_SOCKET;
> > >  }
> > >
> > > +static void rts5227_init_from_cfg(struct rtsx_pcr *pcr)
> > > +{
> > > +	struct pci_dev *pdev = pcr->pci;
> > > +	int l1ss;
> > > +	u32 lval;
> > > +	struct rtsx_cr_option *option = &pcr->option;
> > > +
> > > +	l1ss = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_L1SS);
> > > +	if (!l1ss)
> > > +		return;
> > > +
> > > +	pci_read_config_dword(pdev, l1ss + PCI_L1SS_CTL1, &lval);
> > 
> > This looks a little problematic.  PCI_L1SS_CTL1 is an architected
> > register in the ASPM L1 PM Substates capability, and its value may
> > change at runtime because drivers/pci/pcie/aspm.c manages it.
> > 
> > It looks like the code below does device-specific configuration based
> > on the current PCI_L1SS_CTL1 value.  But what happens if aspm.c
> > changes PCI_L1SS_CTL1 later?
> 
> We are going to make sure and set the best configuration on the
> current time, if host change the capability later, it doesn't affect
> function, only affect a little power saving   

Why don't you unconditionally do the following?

  rtsx_set_dev_flag(pcr, ASPM_L1_1_EN);
  rtsx_set_dev_flag(pcr, ASPM_L1_2_EN);
  rtsx_set_dev_flag(pcr, PM_L1_1_EN);
  rtsx_set_dev_flag(pcr, PM_L1_2_EN);

Let's assume the generic code in aspm.c has cleared all these bits:

  PCI_L1SS_CTL1_ASPM_L1_1
  PCI_L1SS_CTL1_ASPM_L1_2
  PCI_L1SS_CTL1_PCIPM_L1_1
  PCI_L1SS_CTL1_PCIPM_L1_2

in the L1 PM Substates capability.

I think you are saying that if you *also* clear ASPM_L1_1_EN,
ASPM_L1_2_EN, PM_L1_1_EN, and PM_L1_2_EN in your device-specific
registers, it uses less power than if you set those device-specific
bits.  Right?

And moreover, I think you're saying that if aspm.c subsequently *sets*
some of those bits in the L1 PM Substates capability, those substates
*work* even though the device-specific ASPM_L1_1_EN, ASPM_L1_2_EN,
PM_L1_1_EN, and PM_L1_2_EN bits are not set.  Right?

I do not feel good about this as a general strategy.  I think we
should program the device so the behavior is completely predictable,
regardless of the generic enable bits happened to be set at
probe-time.

The current approach means that if we enable L1 substates after the
driver probe, the device is configured differently than if we enabled
L1 substates before probe.  That's not a reliable way to operate it.

> > > +	if (CHK_PCI_PID(pcr, 0x522A)) {
> > > +		if (0 == (lval & 0x0F))
> > > +			rtsx_pci_enable_oobs_polling(pcr);
> > > +		else
> > > +			rtsx_pci_disable_oobs_polling(pcr);
> > > +	}
> > > +
> > > +	if (lval & PCI_L1SS_CTL1_ASPM_L1_1)
> > > +		rtsx_set_dev_flag(pcr, ASPM_L1_1_EN);
> > > +	else
> > > +		rtsx_clear_dev_flag(pcr, ASPM_L1_1_EN);
> > > +
> > > +	if (lval & PCI_L1SS_CTL1_ASPM_L1_2)
> > > +		rtsx_set_dev_flag(pcr, ASPM_L1_2_EN);
> > > +	else
> > > +		rtsx_clear_dev_flag(pcr, ASPM_L1_2_EN);
> > > +
> > > +	if (lval & PCI_L1SS_CTL1_PCIPM_L1_1)
> > > +		rtsx_set_dev_flag(pcr, PM_L1_1_EN);
> > > +	else
> > > +		rtsx_clear_dev_flag(pcr, PM_L1_1_EN);
> > > +
> > > +	if (lval & PCI_L1SS_CTL1_PCIPM_L1_2)
> > > +		rtsx_set_dev_flag(pcr, PM_L1_2_EN);
> > > +	else
> > > +		rtsx_clear_dev_flag(pcr, PM_L1_2_EN);
> > > +
> > > +	if (option->ltr_en) {
> > > +		u16 val;
> > > +
> > > +		pcie_capability_read_word(pcr->pci, PCI_EXP_DEVCTL2, &val);
> > 
> > Same thing here.  I don't think the PCI core currently changes
> > PCI_EXP_DEVCTL2 after boot, but it's not a good idea to assume it's
> > going to be constant.
> > 
> 
> The same reply
> 
> > > +		if (val & PCI_EXP_DEVCTL2_LTR_EN) {
> > > +			option->ltr_enabled = true;
> > > +			option->ltr_active = true;
> > > +			rtsx_set_ltr_latency(pcr, option->ltr_active_latency);
> > > +		} else {
> > > +			option->ltr_enabled = false;
> > > +		}
> > > +	}
> > > +
> > > +	if (rtsx_check_dev_flag(pcr, ASPM_L1_1_EN | ASPM_L1_2_EN
> > > +				| PM_L1_1_EN | PM_L1_2_EN))
> > > +		option->force_clkreq_0 = false;
> > > +	else
> > > +		option->force_clkreq_0 = true;
> > > +
> > > +}
> > 
> > ------Please consider the environment before printing this e-mail.