[PATCH 1/4] drm/xe/vf: React to MIGRATED interrupt

Michal Wajdeczko michal.wajdeczko at intel.com
Mon Sep 23 08:39:44 UTC 2024



On 21.09.2024 00:29, Tomasz Lis wrote:
> To properly support VF Save/Restore procedure, fixups need to be
> applied after PF driver finishes its part of VF Restore. Those
> fixups are applied by the VF driver within a VM.
> 
> A VF driver gets informed that it was migrated by receiving an
> interrupt from each GuC. That should be the trigger for fixups.
> 
> The VF can safely do post-migration fixups on resources associated
> to each GuC only after that GuC issued the MIGRATED interrupt.
> 
> This change introduces a worker to be used for post-migration fixups,
> and a mechanism to schedule said worker when all GuCs sent the irq.
> 
> Signed-off-by: Tomasz Lis <tomasz.lis at intel.com>
> ---
>  drivers/gpu/drm/xe/Makefile          |  1 +
>  drivers/gpu/drm/xe/xe_device_types.h |  1 +
>  drivers/gpu/drm/xe/xe_guc.c          |  3 +
>  drivers/gpu/drm/xe/xe_memirq.c       |  3 +
>  drivers/gpu/drm/xe/xe_sriov.c        | 15 +++++
>  drivers/gpu/drm/xe/xe_sriov.h        |  1 +
>  drivers/gpu/drm/xe/xe_sriov_types.h  |  6 ++
>  drivers/gpu/drm/xe/xe_sriov_vf.c     | 82 ++++++++++++++++++++++++++++
>  drivers/gpu/drm/xe/xe_sriov_vf.h     | 17 ++++++
>  9 files changed, 129 insertions(+)
>  create mode 100644 drivers/gpu/drm/xe/xe_sriov_vf.c
>  create mode 100644 drivers/gpu/drm/xe/xe_sriov_vf.h
> 
> diff --git a/drivers/gpu/drm/xe/Makefile b/drivers/gpu/drm/xe/Makefile
> index ae245fbd91ee..aa06644dffd5 100644
> --- a/drivers/gpu/drm/xe/Makefile
> +++ b/drivers/gpu/drm/xe/Makefile
> @@ -123,6 +123,7 @@ xe-y += \
>  	xe_gt_sriov_vf.o \
>  	xe_guc_relay.o \
>  	xe_memirq.o \
> +	xe_sriov_vf.o \
>  	xe_sriov.o
>  
>  xe-$(CONFIG_PCI_IOV) += \
> diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
> index 5ad96d283a71..331b55b457ab 100644
> --- a/drivers/gpu/drm/xe/xe_device_types.h
> +++ b/drivers/gpu/drm/xe/xe_device_types.h
> @@ -374,6 +374,7 @@ struct xe_device {
>  
>  		/** @sriov.pf: PF specific data */
>  		struct xe_device_pf pf;
> +		struct xe_device_vf vf;

missing kernel-doc

>  
>  		/** @sriov.wq: workqueue used by the virtualization workers */
>  		struct workqueue_struct *wq;
> diff --git a/drivers/gpu/drm/xe/xe_guc.c b/drivers/gpu/drm/xe/xe_guc.c
> index b6cd5e941f19..65cfd6bd68f1 100644
> --- a/drivers/gpu/drm/xe/xe_guc.c
> +++ b/drivers/gpu/drm/xe/xe_guc.c
> @@ -1096,6 +1096,9 @@ void xe_guc_irq_handler(struct xe_guc *guc, const u16 iir)
>  {
>  	if (iir & GUC_INTR_GUC2HOST)
>  		xe_guc_ct_irq_handler(&guc->ct);
> +
> +	if (iir & GUC_INTR_SW_INT_0)
> +		xe_sriov_migrated_event_handler(guc);

this likely should be

	xe_gt_sriov_vf_migrated_event_handler(guc_to_gt(guc));

>  }
>  
>  void xe_guc_sanitize(struct xe_guc *guc)
> diff --git a/drivers/gpu/drm/xe/xe_memirq.c b/drivers/gpu/drm/xe/xe_memirq.c
> index 3f8d4ca64302..2d2f40378942 100644
> --- a/drivers/gpu/drm/xe/xe_memirq.c
> +++ b/drivers/gpu/drm/xe/xe_memirq.c
> @@ -435,6 +435,9 @@ static void memirq_dispatch_guc(struct xe_memirq *memirq, struct iosys_map *stat
>  
>  	if (memirq_received(memirq, status, ilog2(GUC_INTR_GUC2HOST), name))
>  		xe_guc_irq_handler(guc, GUC_INTR_GUC2HOST);
> +
> +	if (memirq_received(memirq, status, ilog2(GUC_INTR_SW_INT_0), name))
> +		xe_guc_irq_handler(guc, GUC_INTR_SW_INT_0);
>  }
>  
>  /**
> diff --git a/drivers/gpu/drm/xe/xe_sriov.c b/drivers/gpu/drm/xe/xe_sriov.c
> index 69a066ef20c0..7447d85104e5 100644
> --- a/drivers/gpu/drm/xe/xe_sriov.c
> +++ b/drivers/gpu/drm/xe/xe_sriov.c
> @@ -9,9 +9,11 @@
>  
>  #include "xe_assert.h"
>  #include "xe_device.h"
> +#include "xe_guc.h"
>  #include "xe_mmio.h"
>  #include "xe_sriov.h"
>  #include "xe_sriov_pf.h"
> +#include "xe_sriov_vf.h"
>  
>  /**
>   * xe_sriov_mode_to_string - Convert enum value to string.
> @@ -112,6 +114,9 @@ int xe_sriov_init(struct xe_device *xe)
>  			return err;
>  	}
>  
> +	if (IS_SRIOV_VF(xe))
> +		xe_sriov_vf_init_early(xe);
> +
>  	xe_assert(xe, !xe->sriov.wq);
>  	xe->sriov.wq = alloc_workqueue("xe-sriov-wq", 0, 0);
>  	if (!xe->sriov.wq)
> @@ -150,3 +155,13 @@ const char *xe_sriov_function_name(unsigned int n, char *buf, size_t size)
>  		strscpy(buf, "PF", size);
>  	return buf;
>  }
> +
> +int xe_sriov_migrated_event_handler(struct xe_guc *guc)

if this function is strictly GuC related then it should have prefix

	xe_guc_...

and likely be placed in xe_guc.c file

but if we treat this function is GT related then it should have prefix

	xe_gt_sriov_...

and since it's VF specific it's name should also include VF:

	xe_gt_sriov_vf_...

and be in VF specific file xe_gt_sriov_vf.c

and maybe this function can be void

> +{
> +	struct xe_gt *gt = guc_to_gt(guc);
> +
> +	if (!IS_SRIOV_VF(gt_to_xe(gt)))
> +		return 0;

either this should be xe_gt_assert() or function should be named like

	xe_guc_sw_0_irq_handler(guc)

and then call migration stuff only if VF

> +
> +	return xe_sriov_vf_migrated_event_handler(gt);

if this is GT related function then it should start with:

	xe_gt_sriov_vf_...

> +}
> diff --git a/drivers/gpu/drm/xe/xe_sriov.h b/drivers/gpu/drm/xe/xe_sriov.h
> index 688fbabf08f1..f7575177b75a 100644
> --- a/drivers/gpu/drm/xe/xe_sriov.h
> +++ b/drivers/gpu/drm/xe/xe_sriov.h
> @@ -13,6 +13,7 @@
>  struct drm_printer;
>  
>  const char *xe_sriov_mode_to_string(enum xe_sriov_mode mode);
> +int xe_sriov_migrated_event_handler(struct xe_guc *guc);

this header is not for VF (or PF) specific functions
nor for GuC/GT oriented functions

>  const char *xe_sriov_function_name(unsigned int n, char *buf, size_t len);
>  
>  void xe_sriov_probe_early(struct xe_device *xe);
> diff --git a/drivers/gpu/drm/xe/xe_sriov_types.h b/drivers/gpu/drm/xe/xe_sriov_types.h
> index c7b7ad4af5c8..a0b590bc7ffa 100644
> --- a/drivers/gpu/drm/xe/xe_sriov_types.h
> +++ b/drivers/gpu/drm/xe/xe_sriov_types.h
> @@ -56,4 +56,10 @@ struct xe_device_pf {
>  	struct mutex master_lock;
>  };
>  
> +struct xe_device_vf {

missing kernel-doc for the struct

> +	/** @migration_worker: migration recovery worker */
> +	struct work_struct migration_worker;
> +	unsigned long migration_gt_flags;

missing kernel-doc for the flags

btw, instead of adding 'migration' prefix better option seems to be to
define anonymous struct named 'migration' with .flags and .worker


> +};
> +
>  #endif
> diff --git a/drivers/gpu/drm/xe/xe_sriov_vf.c b/drivers/gpu/drm/xe/xe_sriov_vf.c
> new file mode 100644
> index 000000000000..b068c57b2bdc
> --- /dev/null
> +++ b/drivers/gpu/drm/xe/xe_sriov_vf.c
> @@ -0,0 +1,82 @@
> +// SPDX-License-Identifier: MIT
> +/*
> + * Copyright © 2023-2024 Intel Corporation
> + */
> +
> +#include <drm/drm_managed.h>
> +
> +#include "xe_assert.h"
> +#include "xe_device.h"
> +#include "xe_guc_ct.h"

is this include needed ?

> +#include "xe_module.h"

is this include needed ?

> +#include "xe_sriov.h"
> +#include "xe_sriov_vf.h"
> +#include "xe_sriov_printk.h"
> +
> +static void migration_worker_func(struct work_struct *w);
> +
> +void xe_sriov_vf_init_early(struct xe_device *xe)

add kernel-doc

> +{
> +	INIT_WORK(&xe->sriov.vf.migration_worker, migration_worker_func);
> +}
> +
> +static void vf_post_migration_recovery(struct xe_device *xe)
> +{
> +	drm_dbg(&xe->drm, "migration recovery in progress\n");
> +	/* FIXME: add the recovery steps */
> +	drm_notice(&xe->drm, "migration recovery completed\n");

maybe add this message only after adding those still missing recovery
steps, as otherwise it will be little misleading

> +}
> +
> +static void migration_worker_func(struct work_struct *w)
> +{
> +	struct xe_device *xe = container_of(w, struct xe_device,
> +						     sriov.vf.migration_worker);

looks like this line unaligned to above (

> +
> +	vf_post_migration_recovery(xe);
> +}
> +
> +/**
> + * xe_sriov_vf_start_migration_recovery - Start VF migration recovery.

usually we don't add full kernel-doc to static functions

> + * @xe: the &xe_device to start recovery on
> + *
> + * This function shall be called only by VF.
> + */
> +static void xe_sriov_vf_start_migration_recovery(struct xe_device *xe)
> +{
> +	bool started;
> +
> +	XE_WARN_ON(!IS_SRIOV_VF(xe));

use xe_assert() instead

> +
> +	WRITE_ONCE(xe->sriov.vf.migration_gt_flags, 0);
> +	smp_mb();
> +
> +	started = queue_work(xe->sriov.wq, &xe->sriov.vf.migration_worker);
> +	dev_info(xe->drm.dev, "VF migration recovery %s\n", started ?
> +		 "scheduled" : "already in progress");

in xe.ko we prefer drm_info() over dev_info()
and xe_gt_info() if something is GT related

> +}
> +
> +static bool vf_ready_to_recovery_on_all_tiles(struct xe_device *xe)
> +{
> +	struct xe_gt *gt;
> +	unsigned int id;
> +
> +	for_each_gt(gt, xe, id) {
> +		if (!test_bit(id, &xe->sriov.vf.migration_gt_flags))
> +			return false;
> +	}
> +	return true;
> +}
> +
> +int xe_sriov_vf_migrated_event_handler(struct xe_gt *gt)

since this is a GT-level function, it should be rather placed in the
xe_gt_sriov_vf.c file

and also please add kernel-doc as this is public function

> +{
> +	struct xe_device *xe = gt_to_xe(gt);
> +
> +	set_bit(gt->info.id, &xe->sriov.vf.migration_gt_flags);
> +	smp_mb__after_atomic();
> +	dev_info(xe->drm.dev, "VF migration recovery ready on gt%d\n",
> +		 gt->info.id);

probably it would be better to just use xe_gt_info() here

> +	if (vf_ready_to_recovery_on_all_tiles(xe))

maybe this could be part of the function below?

> +		xe_sriov_vf_start_migration_recovery(xe);
> +
> +	return -EREMOTEIO;
> +}
> diff --git a/drivers/gpu/drm/xe/xe_sriov_vf.h b/drivers/gpu/drm/xe/xe_sriov_vf.h
> new file mode 100644
> index 000000000000..d70cd84747e5
> --- /dev/null
> +++ b/drivers/gpu/drm/xe/xe_sriov_vf.h
> @@ -0,0 +1,17 @@
> +/* SPDX-License-Identifier: MIT */
> +/*
> + * Copyright © 2023-2024 Intel Corporation
> + */
> +
> +#ifndef _XE_SRIOV_VF_H_
> +#define _XE_SRIOV_VF_H_
> +
> +#include <linux/types.h>

likely you don't need this yet here

> +
> +struct xe_device;
> +struct xe_gt;
> +
> +void xe_sriov_vf_init_early(struct xe_device *xe);
> +int xe_sriov_vf_migrated_event_handler(struct xe_gt *gt);
> +
> +#endif



More information about the Intel-xe mailing list