From: Andrew Murray <andrew.murray@arm.com>
To: "Derrick, Jonathan" <jonathan.derrick@intel.com>
Cc: "lorenzo.pieralisi@arm.com" <lorenzo.pieralisi@arm.com>,
"helgaas@kernel.org" <helgaas@kernel.org>,
"Paszkiewicz, Artur" <artur.paszkiewicz@intel.com>,
"Baldysiak, Pawel" <pawel.baldysiak@intel.com>,
"Fugate, David" <david.fugate@intel.com>,
"Shevchenko, Andriy" <andriy.shevchenko@intel.com>,
"linux-pci@vger.kernel.org" <linux-pci@vger.kernel.org>,
"Busch, Keith" <keith.busch@intel.com>
Subject: Re: [PATCH 2/3] PCI: vmd: Expose VMD details from BIOS
Date: Fri, 1 Nov 2019 14:44:18 +0000 [thread overview]
Message-ID: <20191101144417.GI9723@e119886-lin.cambridge.arm.com> (raw)
In-Reply-To: <d67fbc2ba322104110c70606a375facf1d21045a.camel@intel.com>
On Fri, Nov 01, 2019 at 02:24:02PM +0000, Derrick, Jonathan wrote:
> Hi ANdrew,
>
> Thanks for the review
>
> On Fri, 2019-11-01 at 13:16 +0000, Andrew Murray wrote:
> > On Wed, Oct 16, 2019 at 11:04:47AM -0600, Jon Derrick wrote:
> > > When some VMDs are enabled and others are not, it's difficult to
> > > determine which IIO stack corresponds to the enabled VMD.
> > >
> > > To assist userspace with management tasks, VMD BIOS will write the VMD
> > > instance number and socket number into the first enabled root port's IO
> > > Base/Limit registers prior to OS handoff. VMD driver can capture this
> > > information and expose it to userspace.
> > >
> > > Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
> > > Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> > > ---
> > > drivers/pci/controller/vmd.c | 79 ++++++++++++++++++++++++++++++++++++++++++--
> > > 1 file changed, 77 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/drivers/pci/controller/vmd.c b/drivers/pci/controller/vmd.c
> > > index 959c7c7..dbe1bff 100644
> > > --- a/drivers/pci/controller/vmd.c
> > > +++ b/drivers/pci/controller/vmd.c
> > > @@ -98,6 +98,8 @@ struct vmd_dev {
> > > struct irq_domain *irq_domain;
> > > struct pci_bus *bus;
> > > u8 busn_start;
> > > + u8 socket_nr;
> > > + u8 instance_nr;
> > >
> > > struct dma_map_ops dma_ops;
> > > struct dma_domain dma_domain;
> > > @@ -543,6 +545,74 @@ static int vmd_pci_write(struct pci_bus *bus, unsigned int devfn, int reg,
> > > .write = vmd_pci_write,
> > > };
> > >
> > > +/**
> > > + * for_each_vmd_root_port - iterate over all enabled VMD Root Ports
> > > + * @vmd: &struct vmd_dev VMD device descriptor
> > > + * @rp: int iterator cursor
> > > + * @temp: u32 temporary value for config read
> > > + *
> > > + * VMD Root Ports are located in the VMD PCIe Domain at 00:[0-3].0, and config
> > > + * space can be determinately accessed through the VMD Config BAR. Because VMD
> > > + * Root Ports can be individually disabled, it's important to iterate for the
> > > + * first enabled Root Port as determined by reading the Vendor/Device register.
> > > + */
> > > +#define for_each_vmd_root_port(vmd, rp, temp) \
> > > + for (rp = 0; rp < 4; rp++) \
> > > + if (vmd_cfg_read(vmd, 0, PCI_DEVFN(root_port, 0), \
> > > + PCI_VENDOR_ID, 4, &temp) || \
> > > + temp == 0xffffffff) {} else
> >
> > You may want to consider using PCI_ERROR_RESPONSE here instead of 0xffffffff.
> > Though this hasn't yet been merged:
> >
> > https://patchwork.ozlabs.org/project/linux-pci/list/?series=126820
> >
>
> Sure it will fit this case perfectly once it's merged
>
> > > +
> > > +static int vmd_parse_domain(struct vmd_dev *vmd)
> > > +{
> > > + int root_port, ret;
> > > + u32 temp, iobase;
> > > +
> > > + vmd->socket_nr = -1;
> > > + vmd->instance_nr = -1;
> > > +
> > > + for_each_vmd_root_port(vmd, root_port, temp) {
> > > + ret = vmd_cfg_read(vmd, 0, PCI_DEVFN(root_port, 0),
> > > + PCI_IO_BASE, 2, &iobase);
> > > + if (ret)
> > > + return ret;
> > > +
> > > + vmd->socket_nr = (iobase >> 4) & 0xf;
> > > + vmd->instance_nr = (iobase >> 14) & 0x3;
> >
> > I'm not familiar with VMD - however how can you be sure that the VMD BIOS
> > will always populate these values here? Is it possible that earlier BIOS's
> > won't do this and something will go wrong here?
> >
> > Is there any sanity checking that can happen here?
>
> Yes that's entirely possible and would show indeterminate values in
> that case. It would be up to the user to understand if the BIOS
> supports the mode before relying on the data.
>
> I am investigating to see if we can do a dmi_match to verify the data
> before publishing.
I think that would be helpful if possible as it would simplify the
user software - and also prevent the user ever getting garbage data.
>
>
> >
> > > +
> > > + /* First available will be used */
> > > + break;
> > > + }
> > > +
> > > + return 0;
> > > +}
> > > +
> > > +static ssize_t socket_nr_show(struct device *dev,
> > > + struct device_attribute *attr, char *buf)
> > > +{
> > > + struct pci_dev *pdev = to_pci_dev(dev);
> > > + struct vmd_dev *vmd = pci_get_drvdata(pdev);
> > > +
> > > + return sprintf(buf, "%u\n", vmd->socket_nr);
> > > +}
> > > +static DEVICE_ATTR_RO(socket_nr);
> > > +
> > > +static ssize_t instance_nr_show(struct device *dev,
> > > + struct device_attribute *attr, char *buf)
> > > +{
> > > + struct pci_dev *pdev = to_pci_dev(dev);
> > > + struct vmd_dev *vmd = pci_get_drvdata(pdev);
> > > +
> > > + return sprintf(buf, "%u\n", vmd->instance_nr);
> > > +}
> > > +static DEVICE_ATTR_RO(instance_nr);
> > > +
> > > +static struct attribute *vmd_dev_attrs[] = {
> > > + &dev_attr_socket_nr.attr,
> > > + &dev_attr_instance_nr.attr,
> > > + NULL
> > > +};
> > > +ATTRIBUTE_GROUPS(vmd_dev);
> > > +
> > > static void vmd_attach_resources(struct vmd_dev *vmd)
> > > {
> > > vmd->dev->resource[VMD_MEMBAR1].child = &vmd->resources[1];
> > > @@ -582,6 +652,11 @@ static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
> > > resource_size_t offset[2] = {0};
> > > resource_size_t membar2_offset = 0x2000;
> > > struct pci_bus *child;
> > > + int ret;
> > > +
> > > + ret = vmd_parse_domain(vmd);
> > > + if (ret)
> > > + return ret;
> >
> > This always will succeed. But what happens if this function returns yet
> > socket_nr/instance_nr hasn't been written to? Is that OK?
> >
>
> Basically only one possibility that could occur and that's if the VMD
> is enabled without any VMD Root Ports being enabled on the VMD domain.
> It's an odd configuration but is technically valid, although the domain
> becomes useless until the user reboots and enables the VMD Root Ports.
>
> So it's more-or-less implied either socket_nr/instance_nr will have
> data or the domain won't be usable.
Of course in this case, the default value will be -1, which should be
quite obvious to a user that this isn't a valid value.
Thanks,
Andrew Murray
>
> Thanks,
> Jon
>
>
> > Thanks,
> >
> > Andrew Murray
> >
> > >
> > > /*
> > > * Shadow registers may exist in certain VMD device ids which allow
> > > @@ -591,7 +666,6 @@ static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
> > > */
> > > if (features & VMD_FEAT_HAS_MEMBAR_SHADOW) {
> > > u32 vmlock;
> > > - int ret;
> > >
> > > membar2_offset = MB2_SHADOW_OFFSET + MB2_SHADOW_SIZE;
> > > ret = pci_read_config_dword(vmd->dev, PCI_REG_VMLOCK, &vmlock);
> > > @@ -876,7 +950,8 @@ static int vmd_resume(struct device *dev)
> > > .probe = vmd_probe,
> > > .remove = vmd_remove,
> > > .driver = {
> > > - .pm = &vmd_dev_pm_ops,
> > > + .pm = &vmd_dev_pm_ops,
> > > + .dev_groups = vmd_dev_groups,
> > > },
> > > };
> > > module_pci_driver(vmd_drv);
> > > --
> > > 1.8.3.1
> > >
next prev parent reply other threads:[~2019-11-01 14:44 UTC|newest]
Thread overview: 18+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-10-16 17:04 [PATCH 0/3] Expose VMD BIOS domain info Jon Derrick
2019-10-16 17:04 ` [PATCH 1/3] PCI: vmd: Add helpers to access device config space Jon Derrick
2019-10-16 17:04 ` [PATCH 2/3] PCI: vmd: Expose VMD details from BIOS Jon Derrick
2019-10-31 11:37 ` Lorenzo Pieralisi
2019-11-01 13:16 ` Andrew Murray
2019-11-01 14:24 ` Derrick, Jonathan
2019-11-01 14:44 ` Andrew Murray [this message]
2019-11-01 21:53 ` Bjorn Helgaas
2019-11-01 22:16 ` Derrick, Jonathan
2019-11-04 18:07 ` Lorenzo Pieralisi
2019-11-05 10:12 ` Lorenzo Pieralisi
2019-11-05 21:32 ` Derrick, Jonathan
2019-11-05 22:22 ` Keith Busch
2019-11-05 22:38 ` Derrick, Jonathan
2019-11-05 22:45 ` Keith Busch
2020-01-27 10:38 ` Lorenzo Pieralisi
2020-01-27 23:48 ` Derrick, Jonathan
2019-10-16 17:04 ` [PATCH 3/3] PCI: vmd: Restore domain info during resets/unloads Jon Derrick
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20191101144417.GI9723@e119886-lin.cambridge.arm.com \
--to=andrew.murray@arm.com \
--cc=andriy.shevchenko@intel.com \
--cc=artur.paszkiewicz@intel.com \
--cc=david.fugate@intel.com \
--cc=helgaas@kernel.org \
--cc=jonathan.derrick@intel.com \
--cc=keith.busch@intel.com \
--cc=linux-pci@vger.kernel.org \
--cc=lorenzo.pieralisi@arm.com \
--cc=pawel.baldysiak@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).