Re: [RFC PATCH 4/9] cxl/mem: Map memory device registers

[Dan Williams <dan.j.williams@intel.com>]

On Fri, Nov 13, 2020 at 5:12 PM Ben Widawsky <ben.widawsky@intel.com> wrote:
>
> On 20-11-13 12:17:32, Bjorn Helgaas wrote:
> > On Tue, Nov 10, 2020 at 09:43:51PM -0800, Ben Widawsky wrote:
> > > All the necessary bits are initialized in order to find and map the
> > > register space for CXL Memory Devices. This is accomplished by using the
> > > Register Locator DVSEC (CXL 2.0 - 8.1.9.1) to determine which PCI BAR to
> > > use, and how much of an offset from that BAR should be added.
> >
> > "Initialize the necessary bits ..." to use the usual imperative
> > sentence structure, as you did in the subject.
> >
> > > If the memory device registers are found and mapped a new internal data
> > > structure tracking device state is allocated.
> >
> > "Allocate device state if we find device registers" or similar.
> >
> > > Signed-off-by: Ben Widawsky <ben.widawsky@intel.com>
> > > ---
> > >  drivers/cxl/mem.c | 68 +++++++++++++++++++++++++++++++++++++++++++----
> > >  drivers/cxl/pci.h |  6 +++++
> > >  2 files changed, 69 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/drivers/cxl/mem.c b/drivers/cxl/mem.c
> > > index aa7d881fa47b..8d9b9ab6c5ea 100644
> > > --- a/drivers/cxl/mem.c
> > > +++ b/drivers/cxl/mem.c
> > > @@ -7,9 +7,49 @@
> > >  #include "pci.h"
> > >
> > >  struct cxl_mem {
> > > +   struct pci_dev *pdev;
> > >     void __iomem *regs;
> > >  };
> > >
> > > +static struct cxl_mem *cxl_mem_create(struct pci_dev *pdev, u32 reg_lo, u32 reg_hi)
> > > +{
> > > +   struct device *dev = &pdev->dev;
> > > +   struct cxl_mem *cxlm;
> > > +   void __iomem *regs;
> > > +   u64 offset;
> > > +   u8 bar;
> > > +   int rc;
> > > +
> > > +   offset = ((u64)reg_hi << 32) | (reg_lo & 0xffff0000);
> > > +   bar = reg_lo & 0x7;
> > > +
> > > +   /* Basic sanity check that BAR is big enough */
> > > +   if (pci_resource_len(pdev, bar) < offset) {
> > > +           dev_err(dev, "bar%d: %pr: too small (offset: %#llx)\n",
> > > +                           bar, &pdev->resource[bar], (unsigned long long) offset);
> >
> > s/bar/BAR/
> >
> > > +           return ERR_PTR(-ENXIO);
> > > +   }
> > > +
> > > +   rc = pcim_iomap_regions(pdev, 1 << bar, pci_name(pdev));
> > > +   if (rc != 0) {
> > > +           dev_err(dev, "failed to map registers\n");
> > > +           return ERR_PTR(-ENXIO);
> > > +   }
> > > +
> > > +   cxlm = devm_kzalloc(&pdev->dev, sizeof(*cxlm), GFP_KERNEL);
> > > +   if (!cxlm) {
> > > +           dev_err(dev, "No memory available\n");
> > > +           return ERR_PTR(-ENOMEM);
> > > +   }
> > > +
> > > +   regs = pcim_iomap_table(pdev)[bar];
> > > +   cxlm->pdev = pdev;
> > > +   cxlm->regs = regs + offset;
> > > +
> > > +   dev_dbg(dev, "Mapped CXL Memory Device resource\n");
> > > +   return cxlm;
> > > +}
> > > +
> > >  static int cxl_mem_dvsec(struct pci_dev *pdev, int dvsec)
> > >  {
> > >     int pos;
> > > @@ -34,9 +74,9 @@ static int cxl_mem_dvsec(struct pci_dev *pdev, int dvsec)
> > >
> > >  static int cxl_mem_probe(struct pci_dev *pdev, const struct pci_device_id *id)
> > >  {
> > > +   struct cxl_mem *cxlm = ERR_PTR(-ENXIO);
> > >     struct device *dev = &pdev->dev;
> > > -   struct cxl_mem *cxlm;
> >
> > The order was better before ("dev", then "clxm").  Oh, I suppose this
> > is a "reverse Christmas tree" thing.
> >
>
> I don't actually care either way as long as it's consistent. I tend to do
> reverse Christmas tree for no particular reason.

Yeah, reverse Christmas tree for no particular reason.

>
> > > -   int rc, regloc;
> > > +   int rc, regloc, i;
> > >
> > >     rc = cxl_bus_prepared(pdev);
> > >     if (rc != 0) {
> > > @@ -44,15 +84,33 @@ static int cxl_mem_probe(struct pci_dev *pdev, const struct pci_device_id *id)
> > >             return rc;
> > >     }
> > >
> > > +   rc = pcim_enable_device(pdev);
> > > +   if (rc)
> > > +           return rc;
> > > +
> > >     regloc = cxl_mem_dvsec(pdev, PCI_DVSEC_ID_CXL_REGLOC);
> > >     if (!regloc) {
> > >             dev_err(dev, "register location dvsec not found\n");
> > >             return -ENXIO;
> > >     }
> > > +   regloc += 0xc; /* Skip DVSEC + reserved fields */
> > > +
> > > +   for (i = regloc; i < regloc + 0x24; i += 8) {
> > > +           u32 reg_lo, reg_hi;
> > > +
> > > +           pci_read_config_dword(pdev, i, &reg_lo);
> > > +           pci_read_config_dword(pdev, i + 4, &reg_hi);
> > > +
> > > +           if (CXL_REGLOG_IS_MEMDEV(reg_lo)) {
> > > +                   cxlm = cxl_mem_create(pdev, reg_lo, reg_hi);
> > > +                   break;
> > > +           }
> > > +   }
> > > +
> > > +   if (IS_ERR(cxlm))
> > > +           return -ENXIO;
> >
> > I think this would be easier to read if cxl_mem_create() returned NULL
> > on failure (it prints error messages and we throw away
> > -ENXIO/-ENOMEM distinction here anyway) so you could do:
> >
> >   struct cxl_mem *cxlm = NULL;
> >
> >   for (...) {
> >     if (...) {
> >       cxlm = cxl_mem_create(pdev, reg_lo, reg_hi);
> >       break;
> >     }
> >   }
> >
> >   if (!cxlm)
> >     return -ENXIO;  /* -ENODEV might be more natural? */
> >
>
> I agree on both counts. Both of these came from Dan, so I will let him explain.

I'm not attached to differentiating -ENOMEM from -ENXIO and am ok to
drop the ERR_PTR() return. I do tend to use -ENXIO for failure to
perform an initialization action vs failure to even find the device,
but if -ENODEV seems more idiomatic to Bjorn, I won't argue.