> -----Original Message-----
> From: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
> Sent: Tuesday, August 13, 2024 10:11 AM
> To: Ruhl, Michael J <michael.j.ruhl@xxxxxxxxx>
> Cc: intel-xe@xxxxxxxxxxxxxxxxxxxxx; platform-driver-x86@xxxxxxxxxxxxxxx;
> david.e.box@xxxxxxxxxxxxxxx; ilpo.jarvinen@xxxxxxxxxxxxxxx; Brost, Matthew
> <matthew.brost@xxxxxxxxx>; hdegoede@xxxxxxxxxx; Vivi, Rodrigo
> <rodrigo.vivi@xxxxxxxxx>
> Subject: Re: [PATCH v11] drm/xe/vsec: Support BMG devices
>
> On Mon, Aug 12, 2024 at 04:04:22PM -0400, Michael J. Ruhl wrote:
> > The Battlemage (BMG) discrete graphics card supports the Platform,
> > Monitoring Technology (PMT) feature directly on the primary PCI
> > device.
> >
> > Utilize the PMT callback API to add support for the BMG devices.
>
> ...
>
> > +#include <linux/bitfield.h>
> > +#include <linux/bits.h>
> > +#include <linux/cleanup.h>
> > +#include <linux/intel_vsec.h>
> > +#include <linux/module.h>
> > +#include <linux/mutex.h>
> > +#include <linux/pci.h>
>
> ...
>
> > +#define SOC_BASE 0x280000
> > +
> > +#define BMG_PMT_BASE 0xDB000
> > +#define BMG_DISCOVERY_OFFSET (SOC_BASE + BMG_PMT_BASE)
>
> > +#define BMG_TELEMETRY_BASE 0xE0000
> > +#define BMG_TELEMETRY_OFFSET (SOC_BASE + BMG_TELEMETRY_BASE)
>
> This looks like double indirection.
> Wouldn't suffix _BASE_OFFSET be better for PMT and TELEMETRY cases?
I am not sure I understand.
Are you saying rename BMG_PMT_BASE to BMG_PMT_BASE_OFFSET?
>
> ...
>
> > +#define BMG_DEVICE_ID 0xE2F8
>
> Is this defined in any specification? I mean is the format the same as PCI device
> ID?
I think that this is defined in BMG PMT yaml definition. It is provide in the PMT discovery
data, so it is defined by the specific device.
> ...
>
> > +#define GFX_BAR 0
>
> Do you need a separate definition for this?
Guess not. Will remove. 😊
> ...
>
> > +enum record_id {
> > + PUNIT,
> > + OOBMSM_0,
> > + OOBMSM_1
>
> Trailing comma?
>
> > +};
> > +
> > +enum capability {
> > + CRASHLOG,
> > + TELEMETRY,
> > + WATCHER
>
> Ditto?
Will update.
> > +};
>
> ...
>
> > + switch (record_id) {
> > + case PUNIT:
> > + *index = 0;
> > + if (cap_type == TELEMETRY)
> > + *offset = PUNIT_TELEMETRY_OFFSET;
> > + else
> > + *offset = PUNIT_WATCHER_OFFSET;
> > + break;
> > +
> > + case OOBMSM_0:
> > + *index = 1;
> > + if (cap_type == WATCHER)
> > + *offset = OOBMSM_0_WATCHER_OFFSET;
> > + break;
> > +
> > + case OOBMSM_1:
> > + *index = 1;
> > + if (cap_type == TELEMETRY)
> > + *offset = OOBMSM_1_TELEMETRY_OFFSET;
> > + break;
>
> default case?
I validate the record_id and cap_type values at the beginning of the function,
so default would be redundant.
The goal was to validate, then set data.
So adding the default will remove the record_id check from the if. Do you prefer
that path?
> > + }
>
> ...
>
> > +static int xe_pmt_telem_read(struct pci_dev *pdev, u32 guid, u64
> > +*data, u32 count) {
> > + struct xe_device *xe = pdev_to_xe_device(pdev);
> > + void __iomem *telem_addr = xe->mmio.regs +
> BMG_TELEMETRY_OFFSET;
> > + u32 mem_region;
> > + u32 offset;
> > + int ret;
> > +
> > + ret = guid_decode(guid, &mem_region, &offset);
> > + if (ret)
> > + return ret;
> > +
> > + telem_addr += offset;
> > +
> > + guard(mutex)(&xe->pmt.lock);
> > +
> > + /* indicate that we are not at an appropriate power level */
> > + if (!xe_pm_runtime_get_if_active(xe))
> > + return -ENODATA;
> > +
> > + /* set SoC re-mapper index register based on GUID memory region */
> > + xe_mmio_rmw32(xe->tiles[0].primary_gt, SG_REMAP_INDEX1,
> SG_REMAP_BITS,
> > + FIELD_PREP(SG_REMAP_BITS, mem_region));
> > +
> > + memcpy_fromio(data, telem_addr, count);
>
> > + ret = count;
> > + xe_pm_runtime_put(xe);
>
> Does this have a side effect on count? If yes, a comment, if no, you may return
> count directly.
It does not. I was looking at that yesterday (when I updated with the guard()) and
thought about updating. I will update.
Thank you for your comments!
M
>
> > + return ret;
> > +}
>
> --
> With Best Regards,
> Andy Shevchenko
>
