| 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: Mon, 25 Oct 2021 18:33:50 +0800
From: Kai-Heng Feng <kai.heng.feng@...onical.com>
To: Bjorn Helgaas <helgaas@...nel.org>
Cc: Bjorn Helgaas <bhelgaas@...gle.com>,
Heiner Kallweit <hkallweit1@...il.com>,
Anthony Wong <anthony.wong@...onical.com>,
Linux PCI <linux-pci@...r.kernel.org>,
LKML <linux-kernel@...r.kernel.org>,
Krzysztof Wilczyński <kw@...ux.com>,
Vidya Sagar <vidyas@...dia.com>,
Logan Gunthorpe <logang@...tatee.com>
Subject: Re: [PATCH v2 3/3] PCI/ASPM: Add LTR sysfs attributes
On Thu, Oct 21, 2021 at 11:09 PM Bjorn Helgaas <helgaas@...nel.org> wrote:
>
> On Thu, Oct 21, 2021 at 11:51:59AM +0800, Kai-Heng Feng wrote:
> > Sometimes BIOS may not be able to program ASPM and LTR settings, for
> > instance, the NVMe devices behind Intel VMD bridges. For this case, both
> > ASPM and LTR have to be enabled to have significant power saving.
> >
> > Since we still want POLICY_DEFAULT honor the default BIOS ASPM settings,
> > introduce LTR sysfs knobs so users can set max snoop and max nosnoop
> > latency manually or via udev rules.
>
> How is the user supposed to figure out what "max snoop" and "max
> nosnoop" values to program?
Actually, the only way I know is to get the value from other OS.
>
> If we add this, I'm afraid we'll have people programming random things
> that seem to work but are not actually reliable.
IMO users need to take full responsibility for own doings.
Also, it's already doable by using setpci...
If we don't want to add LTR sysfs, what other options do we have to
enable VMD ASPM and LTR by default since BIOS doesn't do it for us?
1) Enable it in the PCI or VMD driver, however this approach violates
POLICY_DEFAULT.
2) Use `setpci` directly in udev rules to enable VMD ASPM and LTR.
I think 2) is bad, and since 1) isn't so good either, the approach in
this patch may be the best compromise.
Kai-Heng
>
> > [1] https://github.com/systemd/systemd/pull/17899/
> > Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=209789
> > Signed-off-by: Kai-Heng Feng <kai.heng.feng@...onical.com>
> > ---
> > v2:
> > - New patch.
> >
> > drivers/pci/pcie/aspm.c | 59 +++++++++++++++++++++++++++++++++++++++++
> > 1 file changed, 59 insertions(+)
> >
> > diff --git a/drivers/pci/pcie/aspm.c b/drivers/pci/pcie/aspm.c
> > index 1560859ab056..f7dc62936445 100644
> > --- a/drivers/pci/pcie/aspm.c
> > +++ b/drivers/pci/pcie/aspm.c
> > @@ -1299,6 +1299,59 @@ static ssize_t clkpm_store(struct device *dev,
> > return len;
> > }
> >
> > +static ssize_t ltr_attr_show_common(struct device *dev,
> > + struct device_attribute *attr, char *buf, u8 state)
> > +{
> > + struct pci_dev *pdev = to_pci_dev(dev);
> > + int ltr;
> > + u16 val;
> > +
> > + ltr = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_LTR);
> > + if (!ltr)
> > + return -EINVAL;
> > +
> > + pci_read_config_word(pdev, ltr + state, &val);
> > +
> > + return sysfs_emit(buf, "0x%0x\n", val);
> > +}
> > +
> > +static ssize_t ltr_attr_store_common(struct device *dev,
> > + struct device_attribute *attr,
> > + const char *buf, size_t len, u8 state)
> > +{
> > + struct pci_dev *pdev = to_pci_dev(dev);
> > + int ltr;
> > + u16 val;
> > +
> > + ltr = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_LTR);
> > + if (!ltr)
> > + return -EINVAL;
> > +
> > + if (kstrtou16(buf, 16, &val) < 0)
> > + return -EINVAL;
> > +
> > + /* LatencyScale is not permitted to be 110 or 111 */
> > + if ((val >> 10) > 5)
> > + return -EINVAL;
> > +
> > + pci_write_config_word(pdev, ltr + state, val);
> > +
> > + return len;
> > +}
> > +
> > +#define LTR_ATTR(_f, _s) \
> > +static ssize_t _f##_show(struct device *dev, \
> > + struct device_attribute *attr, char *buf) \
> > +{ return ltr_attr_show_common(dev, attr, buf, PCI_LTR_##_s); } \
> > + \
> > +static ssize_t _f##_store(struct device *dev, \
> > + struct device_attribute *attr, \
> > + const char *buf, size_t len) \
> > +{ return ltr_attr_store_common(dev, attr, buf, len, PCI_LTR_##_s); }
> > +
> > +LTR_ATTR(ltr_max_snoop_lat, MAX_SNOOP_LAT);
> > +LTR_ATTR(ltr_max_nosnoop_lat, MAX_NOSNOOP_LAT);
> > +
> > static DEVICE_ATTR_RW(clkpm);
> > static DEVICE_ATTR_RW(l0s_aspm);
> > static DEVICE_ATTR_RW(l1_aspm);
> > @@ -1306,6 +1359,8 @@ static DEVICE_ATTR_RW(l1_1_aspm);
> > static DEVICE_ATTR_RW(l1_2_aspm);
> > static DEVICE_ATTR_RW(l1_1_pcipm);
> > static DEVICE_ATTR_RW(l1_2_pcipm);
> > +static DEVICE_ATTR_RW(ltr_max_snoop_lat);
> > +static DEVICE_ATTR_RW(ltr_max_nosnoop_lat);
> >
> > static struct attribute *aspm_ctrl_attrs[] = {
> > &dev_attr_clkpm.attr,
> > @@ -1315,6 +1370,8 @@ static struct attribute *aspm_ctrl_attrs[] = {
> > &dev_attr_l1_2_aspm.attr,
> > &dev_attr_l1_1_pcipm.attr,
> > &dev_attr_l1_2_pcipm.attr,
> > + &dev_attr_ltr_max_snoop_lat.attr,
> > + &dev_attr_ltr_max_nosnoop_lat.attr,
> > NULL
> > };
> >
> > @@ -1338,6 +1395,8 @@ static umode_t aspm_ctrl_attrs_are_visible(struct kobject *kobj,
> >
> > if (n == 0)
> > return link->clkpm_capable ? a->mode : 0;
> > + else if (n == 7 || n == 8)
> > + return pdev->ltr_path ? a->mode : 0;
> >
> > return link->aspm_capable & aspm_state_map[n - 1] ? a->mode : 0;
> > }
> > --
> > 2.32.0
> >
Powered by blists - more mailing lists