[Intel-xe] [PATCH 3/3] drm/xe: Add sysfs entry to report per gt memory size
Upadhyay, Tejas
tejas.upadhyay at intel.com
Thu May 4 12:50:53 UTC 2023
> -----Original Message-----
> From: Iddamsetty, Aravind <aravind.iddamsetty at intel.com>
> Sent: Thursday, May 4, 2023 6:14 PM
> To: Upadhyay, Tejas <tejas.upadhyay at intel.com>; intel-
> xe at lists.freedesktop.org; Vivi, Rodrigo <rodrigo.vivi at intel.com>
> Subject: Re: [PATCH 3/3] drm/xe: Add sysfs entry to report per gt memory
> size
>
>
>
> On 04-05-2023 16:50, Tejas Upadhyay wrote:
> > Add sysfs entry to read per gt physical memory in total including
> > stolen memory.
> >
> > Cc: Aravind Iddamsetty <aravind.iddamsetty at intel.com>
> > Signed-off-by: Tejas Upadhyay <tejas.upadhyay at intel.com>
> > ---
> > drivers/gpu/drm/xe/xe_device_types.h | 2 +
> > drivers/gpu/drm/xe/xe_gt_sysfs.c | 74 +++++++++++++++++++++++++++-
> > drivers/gpu/drm/xe/xe_gt_sysfs.h | 1 -
> > drivers/gpu/drm/xe/xe_gt_types.h | 7 +++
> > drivers/gpu/drm/xe/xe_mmio.c | 1 +
> > 5 files changed, 82 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/xe/xe_device_types.h
> > b/drivers/gpu/drm/xe/xe_device_types.h
> > index 8898aea4bc2b..f6b68a3c6eeb 100644
> > --- a/drivers/gpu/drm/xe/xe_device_types.h
> > +++ b/drivers/gpu/drm/xe/xe_device_types.h
> > @@ -254,6 +254,8 @@ struct xe_device {
> > /** @gt: graphics tile */
> > struct xe_gt gt[XE_MAX_GT];
> >
> > + struct kobject *sysfs_gt;
> > +
> > /**
> > * @mem_access: keep track of memory access in the device,
> possibly
> > * triggering additional actions when they occur.
> > diff --git a/drivers/gpu/drm/xe/xe_gt_sysfs.c
> > b/drivers/gpu/drm/xe/xe_gt_sysfs.c
> > index c01cc689058c..b133dd7b9d31 100644
> > --- a/drivers/gpu/drm/xe/xe_gt_sysfs.c
> > +++ b/drivers/gpu/drm/xe/xe_gt_sysfs.c
> > @@ -29,12 +29,76 @@ static void gt_sysfs_fini(struct drm_device *drm,
> void *arg)
> > kobject_put(gt->sysfs);
> > }
> >
> > +typedef ssize_t (*show)(struct device *dev, struct device_attribute
> > +*attr, char *buf);
> > +
> > +struct xe_ext_attr {
> > + struct device_attribute attr;
> > + show xe_show;
> > +};
> > +
> > +static ssize_t
> > +xe_sysfs_show(struct device *dev, struct device_attribute *attr, char
> > +*buf) {
> > + ssize_t value;
> > + struct xe_ext_attr *ea = container_of(attr, struct xe_ext_attr,
> > +attr);
> > +
> > + /* FIXME: Wa_16015496043 Hold forcewake on GT0 & GT1 to
> disallow rc6 */
> > + value = ea->xe_show(dev, attr, buf);
> > +
> > + return value;
> > +}
> > +
> > +static ssize_t
> > +addr_range_show(struct device *kdev, struct device_attribute *attr,
> > +char *buf) {
> > + struct xe_gt *gt = kobj_to_gt(&kdev->kobj);
> > +
> > + return sysfs_emit(buf, "%pa\n", >->mem.actual_physical_mem); }
> > +
> > +#define XE_DEVICE_ATTR_RO(_name, _show) \
> > + struct xe_ext_attr dev_attr_##_name = \
> > + { __ATTR(_name, 0444, xe_sysfs_show, NULL), _show}
>
> i do not see the use of this.
> > +
> > +static XE_DEVICE_ATTR_RO(addr_range, addr_range_show);
> > +
> > +static const struct attribute *addr_range_attrs[] = {
> > + /* TODO: Report any other HBM Sparing sysfs per gt? */
> > + &dev_attr_addr_range.attr.attr,
> > + NULL
> > +};
> > +
> > +void xe_gt_sysfs_register_mem(struct xe_gt *gt, struct kobject
> > +*parent) {
> > + /* FIXME: Check has_mem_sparing here */
> > +
> > + if (sysfs_create_files(parent, addr_range_attrs))
> > + drm_err(>_to_xe(gt)->drm,
> > + "Setting up sysfs to read total physical memory per
> tile
> > +failed\n"); }
> > +
> > +static struct kobject *xe_setup_gt_sysfs(struct kobject *parent) {
> > + return kobject_create_and_add("gt", parent); }
> > +
> > int xe_gt_sysfs_init(struct xe_gt *gt) {
> > - struct device *dev = gt_to_xe(gt)->drm.dev;
> > + struct xe_device *xe = gt_to_xe(gt);
> > + struct device *kdev = xe->drm.primary->kdev;
> > struct kobj_gt *kg;
> > int err;
> >
> > + /* Create sysfs dir only for primary GT to avoid dup */
> > + if (!gt->info.id) {
> > + xe->sysfs_gt = xe_setup_gt_sysfs(&kdev->kobj);
>
> i'm not sure if we ever wanted to create a top level GT folder
>
> @rodrigo, any thoughts?
>
> > + if (!xe->sysfs_gt) {
> > + drm_err(&xe->drm,
> > + "failed to register GT sysfs directory\n");
> > + return 0;
> > + }
> > + }
> > +
> > kg = kzalloc(sizeof(*kg), GFP_KERNEL);
> > if (!kg)
> > return -ENOMEM;
> > @@ -42,7 +106,7 @@ int xe_gt_sysfs_init(struct xe_gt *gt)
> > kobject_init(&kg->base, &xe_gt_sysfs_kobj_type);
> > kg->gt = gt;
> >
> > - err = kobject_add(&kg->base, &dev->kobj, "gt%d", gt->info.id);
> > + err = kobject_add(&kg->base, xe->sysfs_gt, "gt%d", gt->info.id);
> > if (err) {
> > kobject_put(&kg->base);
> > return err;
> > @@ -50,9 +114,15 @@ int xe_gt_sysfs_init(struct xe_gt *gt)
> >
> > gt->sysfs = &kg->base;
> >
> > + gt->sysfs_defaults = kobject_create_and_add(".defaults", gt->sysfs);
>
> what is the use of defaults??
No use. There is no use I saw in i915 as well. But may be for some future use it is kept. Or I missed something. If want to remove, can be removed on objections.
Thanks,
Tejas
>
> Thanks,
> Aravind.
> > + if (!gt->sysfs_defaults)
> > + drm_err(&xe->drm, "failed to create gt sysfs .defaults\n");
> > +
> > err = drmm_add_action_or_reset(>_to_xe(gt)->drm, gt_sysfs_fini,
> gt);
> > if (err)
> > return err;
> >
> > + xe_gt_sysfs_register_mem(gt, gt->sysfs);
> > +
> > return 0;
> > }
> > diff --git a/drivers/gpu/drm/xe/xe_gt_sysfs.h
> > b/drivers/gpu/drm/xe/xe_gt_sysfs.h
> > index ecbfcc5c7d42..a8078c5992bd 100644
> > --- a/drivers/gpu/drm/xe/xe_gt_sysfs.h
> > +++ b/drivers/gpu/drm/xe/xe_gt_sysfs.h
> > @@ -15,5 +15,4 @@ kobj_to_gt(struct kobject *kobj) {
> > return container_of(kobj, struct kobj_gt, base)->gt; }
> > -
> > #endif /* _XE_GT_SYSFS_H_ */
> > diff --git a/drivers/gpu/drm/xe/xe_gt_types.h
> > b/drivers/gpu/drm/xe/xe_gt_types.h
> > index 47f059bb8c6d..5615f2d65996 100644
> > --- a/drivers/gpu/drm/xe/xe_gt_types.h
> > +++ b/drivers/gpu/drm/xe/xe_gt_types.h
> > @@ -138,6 +138,11 @@ struct xe_gt {
> > * objects (virtual split)
> > */
> > struct {
> > + /**
> > + * @actual_physical_mem: Track actual LMEM size
> > + * including stolen mem
> > + */
> > + resource_size_t actual_physical_mem;
> > /**
> > * @vram: VRAM info for GT, multiple GTs can point to same
> info
> > * (virtual split), can be subset of global device VRAM @@ -
> 324,6
> > +329,8 @@ struct xe_gt {
> >
> > /** @sysfs: sysfs' kobj used by xe_gt_sysfs */
> > struct kobject *sysfs;
> > + /* sysfs defaults per gt */
> > + struct kobject *sysfs_defaults;
> >
> > /** @mocs: info */
> > struct {
> > diff --git a/drivers/gpu/drm/xe/xe_mmio.c
> > b/drivers/gpu/drm/xe/xe_mmio.c index 2c35a1b79be7..bc05b6c0cefb
> 100644
> > --- a/drivers/gpu/drm/xe/xe_mmio.c
> > +++ b/drivers/gpu/drm/xe/xe_mmio.c
> > @@ -296,6 +296,7 @@ int xe_mmio_probe_vram(struct xe_device *xe)
> > if (err)
> > return err;
> >
> > + gt->mem.actual_physical_mem = tile_size;
> > gt->mem.vram.io_start = xe->mem.vram.io_start +
> tile_offset;
> > gt->mem.vram.io_size = min_t(u64, vram_size, io_size);
> >
More information about the Intel-xe
mailing list