Re: [Intel-xe] [PATCH v2 3/4] drm/xe/gsc: add gsc device support

[Rodrigo Vivi]

On Thu, Sep 14, 2023 at 11:01:37AM +0300, Alexander Usyskin wrote:
> From: Vitaly Lubart <vitaly.lubart@xxxxxxxxx>
> 
> Create mei-gscfi auxiliary device and configure interrupts
> to be consumed by mei-gsc device driver.
> 
> Signed-off-by: Vitaly Lubart <vitaly.lubart@xxxxxxxxx>
> Signed-off-by: Alexander Usyskin <alexander.usyskin@xxxxxxxxx>
> ---
>  drivers/gpu/drm/xe/Kconfig           |   1 +
>  drivers/gpu/drm/xe/Makefile          |   1 +
>  drivers/gpu/drm/xe/xe_device.c       |   4 +
>  drivers/gpu/drm/xe/xe_device_types.h |   4 +
>  drivers/gpu/drm/xe/xe_heci_gsc.c     | 205 +++++++++++++++++++++++++++
>  drivers/gpu/drm/xe/xe_heci_gsc.h     |  35 +++++
>  drivers/gpu/drm/xe/xe_irq.c          |  14 +-
>  7 files changed, 262 insertions(+), 2 deletions(-)
>  create mode 100644 drivers/gpu/drm/xe/xe_heci_gsc.c
>  create mode 100644 drivers/gpu/drm/xe/xe_heci_gsc.h
> 
> diff --git a/drivers/gpu/drm/xe/Kconfig b/drivers/gpu/drm/xe/Kconfig
> index 096bd066afa8..da82084fe236 100644
> --- a/drivers/gpu/drm/xe/Kconfig
> +++ b/drivers/gpu/drm/xe/Kconfig
> @@ -37,6 +37,7 @@ config DRM_XE
>  	select DRM_SCHED
>  	select MMU_NOTIFIER
>  	select WANT_DEV_COREDUMP
> +	select AUXILIARY_BUS
>  	help
>  	  Experimental driver for Intel Xe series GPUs
>  
> diff --git a/drivers/gpu/drm/xe/Makefile b/drivers/gpu/drm/xe/Makefile
> index 9d2311f8141f..fbdb28fa5ace 100644
> --- a/drivers/gpu/drm/xe/Makefile
> +++ b/drivers/gpu/drm/xe/Makefile
> @@ -73,6 +73,7 @@ xe-y += xe_bb.o \
>  	xe_guc_log.o \
>  	xe_guc_pc.o \
>  	xe_guc_submit.o \
> +	xe_heci_gsc.o \
>  	xe_hw_engine.o \
>  	xe_hw_engine_class_sysfs.o \
>  	xe_hw_fence.o \
> diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
> index d6fc06d4c9dc..4d6e2f2b281f 100644
> --- a/drivers/gpu/drm/xe/xe_device.c
> +++ b/drivers/gpu/drm/xe/xe_device.c
> @@ -323,6 +323,8 @@ int xe_device_probe(struct xe_device *xe)
>  			goto err_irq_shutdown;
>  	}
>  
> +	xe_heci_gsc_init(xe);
> +

could we place this call earlier in the flow? maybe right after setting up mmio?
or maybe after pcode init where we confirmed the boot can proceed on discrete?

>  	err = xe_display_init(xe);
>  	if (err)
>  		goto err_irq_shutdown;
> @@ -365,6 +367,8 @@ void xe_device_remove(struct xe_device *xe)
>  
>  	xe_display_fini(xe);
>  
> +	xe_heci_gsc_fini(xe);
> +
>  	xe_irq_shutdown(xe);
>  }
>  
> diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
> index 1d1fe53fc30d..80233c2f0d81 100644
> --- a/drivers/gpu/drm/xe/xe_device_types.h
> +++ b/drivers/gpu/drm/xe/xe_device_types.h
> @@ -13,6 +13,7 @@
>  #include <drm/ttm/ttm_device.h>
>  
>  #include "xe_devcoredump_types.h"
> +#include "xe_heci_gsc.h"
>  #include "xe_gt_types.h"
>  #include "xe_platform_types.h"
>  #include "xe_step_types.h"
> @@ -364,6 +365,9 @@ struct xe_device {
>  	 */
>  	struct task_struct *pm_callback_task;
>  
> +	/** @gsc: graphics security controller */
> +	struct xe_heci_gsc heci_gsc;

documentation doesn't match variable name.

> +
>  	/* private: */
>  
>  #if IS_ENABLED(CONFIG_DRM_XE_DISPLAY)
> diff --git a/drivers/gpu/drm/xe/xe_heci_gsc.c b/drivers/gpu/drm/xe/xe_heci_gsc.c
> new file mode 100644
> index 000000000000..1eca1c46f257
> --- /dev/null
> +++ b/drivers/gpu/drm/xe/xe_heci_gsc.c
> @@ -0,0 +1,205 @@
> +// SPDX-License-Identifier: MIT
> +/*
> + * Copyright(c) 2023, Intel Corporation. All rights reserved.
> + */
> +
> +#include <linux/irq.h>
> +#include <linux/mei_aux.h>
> +#include <linux/pci.h>
> +#include <linux/sizes.h>
> +
> +#include "regs/xe_regs.h"
> +#include "xe_device_types.h"
> +#include "xe_drv.h"
> +#include "xe_heci_gsc.h"
> +#include "xe_platform_types.h"
> +
> +#define GSC_BAR_LENGTH  0x00000FFC
> +
> +static void heci_gsc_irq_mask(struct irq_data *d)
> +{
> +	/* generic irq handling */
> +}
> +
> +static void heci_gsc_irq_unmask(struct irq_data *d)
> +{
> +	/* generic irq handling */
> +}
> +
> +static struct irq_chip heci_gsc_irq_chip = {
> +	.name = "gsc_irq_chip",
> +	.irq_mask = heci_gsc_irq_mask,
> +	.irq_unmask = heci_gsc_irq_unmask,
> +};
> +
> +static int heci_gsc_irq_init(int irq)
> +{
> +	irq_set_chip_and_handler_name(irq, &heci_gsc_irq_chip,
> +				      handle_simple_irq, "heci_gsc_irq_handler");
> +
> +	return irq_set_chip_data(irq, NULL);
> +}
> +
> +/**
> + * struct heci_gsc_def - graphics security controller heci interface definitions
> + *
> + * @name: name of the heci device
> + * @bar: address of the mmio bar
> + * @bar_size: size of the mmio bar
> + * @use_polling: indication of using polling mode for the device
> + * @slow_firmware: indication of whether the device is slow (needs longer timeouts)
> + */
> +struct heci_gsc_def {
> +	const char *name;
> +	unsigned long bar;
> +	size_t bar_size;
> +	bool use_polling;
> +	bool slow_firmware;

will we need the lmem_size here?

or what's the difference on the mei-gsc and mei-gscfi exactly?

> +};
> +
> +/* gsc resources and definitions */
> +static const struct heci_gsc_def heci_gsc_def_dg1 = {
> +	.name = "mei-gscfi",
> +	.bar = DG1_GSC_HECI2_BASE,
> +	.bar_size = GSC_BAR_LENGTH,
> +};
> +
> +static const struct heci_gsc_def heci_gsc_def_dg2 = {
> +	.name = "mei-gscfi",
> +	.bar = DG2_GSC_HECI2_BASE,
> +	.bar_size = GSC_BAR_LENGTH,
> +};
> +
> +static void heci_gsc_release_dev(struct device *dev)
> +{
> +	struct auxiliary_device *aux_dev = to_auxiliary_dev(dev);
> +	struct mei_aux_device *adev = auxiliary_dev_to_mei_aux_dev(aux_dev);
> +
> +	kfree(adev);
> +}
> +
> +void xe_heci_gsc_fini(struct xe_device *xe)
> +{
> +	struct xe_heci_gsc *heci_gsc = &xe->heci_gsc;
> +
> +	if (!HAS_HECI_GSCFI(xe))
> +		return;
> +
> +	if (heci_gsc->adev) {
> +		struct auxiliary_device *aux_dev = &heci_gsc->adev->aux_dev;
> +
> +		auxiliary_device_delete(aux_dev);
> +		auxiliary_device_uninit(aux_dev);
> +		heci_gsc->adev = NULL;
> +	}
> +
> +	if (heci_gsc->irq >= 0)
> +		irq_free_desc(heci_gsc->irq);
> +	heci_gsc->irq = -1;
> +}
> +
> +void xe_heci_gsc_init(struct xe_device *xe)
> +{
> +	struct xe_heci_gsc *heci_gsc = &xe->heci_gsc;
> +	struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
> +	struct mei_aux_device *adev;
> +	struct auxiliary_device *aux_dev;
> +	const struct heci_gsc_def *def;
> +	int ret;
> +
> +	if (!HAS_HECI_GSCFI(xe))
> +		return;
> +
> +	heci_gsc->irq = -1;
> +
> +	if (xe->info.platform == XE_DG1) {
> +		def = &heci_gsc_def_dg1;
> +	} else if (xe->info.platform == XE_DG2) {
> +		def = &heci_gsc_def_dg2;

in general it is better to add the most recent on top of the oldest.

> +	} else {
> +		drm_warn_once(&xe->drm, "Unknown platform\n");
> +		return;
> +	}
> +
> +	if (!def->name) {
> +		drm_warn_once(&xe->drm, "HECI is not implemented!\n");
> +		return;
> +	}
> +
> +	/* skip irq initialization */
> +	if (def->use_polling)
> +		goto add_device;

what about creating a function
heci_gsc_irq_setup(...)

and then
if (!def->use_polling) {
   ret = heci_gsc_irq_setup(...)
   if (ret)
      goto fail;

> +
> +	heci_gsc->irq = irq_alloc_desc(0);
> +	if (heci_gsc->irq < 0) {
> +		drm_err(&xe->drm, "gsc irq error %d\n", heci_gsc->irq);
> +		goto fail;
> +	}
> +
> +	ret = heci_gsc_irq_init(heci_gsc->irq);
> +	if (ret < 0) {
> +		drm_err(&xe->drm, "gsc irq init failed %d\n", ret);
> +		goto fail;
> +	}
> +
> +add_device:

it looks like this add_device always deserve a dedicated function.

> +	adev = kzalloc(sizeof(*adev), GFP_KERNEL);
> +	if (!adev)
> +		goto fail;
> +	adev->irq = heci_gsc->irq;
> +	adev->bar.parent = &pdev->resource[0];
> +	adev->bar.start = def->bar + pdev->resource[0].start;
> +	adev->bar.end = adev->bar.start + def->bar_size - 1;
> +	adev->bar.flags = IORESOURCE_MEM;
> +	adev->bar.desc = IORES_DESC_NONE;
> +	adev->slow_firmware = def->slow_firmware;
> +
> +	aux_dev = &adev->aux_dev;
> +	aux_dev->name = def->name;
> +	aux_dev->id = (pci_domain_nr(pdev->bus) << 16) |
> +		      PCI_DEVID(pdev->bus->number, pdev->devfn);
> +	aux_dev->dev.parent = &pdev->dev;
> +	aux_dev->dev.release = heci_gsc_release_dev;
> +
> +	ret = auxiliary_device_init(aux_dev);
> +	if (ret < 0) {
> +		drm_err(&xe->drm, "gsc aux init failed %d\n", ret);
> +		kfree(adev);
> +		goto fail;
> +	}
> +
> +	heci_gsc->adev = adev; /* needed by the notifier */
> +	ret = auxiliary_device_add(aux_dev);
> +	if (ret < 0) {
> +		drm_err(&xe->drm, "gsc aux add failed %d\n", ret);
> +		heci_gsc->adev = NULL;
> +
> +		/* adev will be freed with the put_device() and .release sequence */
> +		auxiliary_device_uninit(aux_dev);
> +		goto fail;
> +	}
> +
> +	return;
> +fail:
> +	xe_heci_gsc_fini(xe);
> +}
> +
> +void xe_heci_gsc_irq_handler(struct xe_device *xe, u32 iir)
> +{
> +	int ret;
> +
> +	if ((iir & GSC_IRQ_INTF(1)) == 0)
> +		return;
> +
> +	if (!HAS_HECI_GSCFI(xe)) {
> +		drm_warn_once(&xe->drm, "GSC irq: not supported");
> +		return;
> +	}
> +
> +	if (xe->heci_gsc.irq < 0)
> +		return;
> +
> +	ret = generic_handle_irq(xe->heci_gsc.irq);
> +	if (ret)
> +		drm_err_ratelimited(&xe->drm, "error handling GSC irq: %d\n", ret);
> +}
> diff --git a/drivers/gpu/drm/xe/xe_heci_gsc.h b/drivers/gpu/drm/xe/xe_heci_gsc.h
> new file mode 100644
> index 000000000000..9db454478fae
> --- /dev/null
> +++ b/drivers/gpu/drm/xe/xe_heci_gsc.h
> @@ -0,0 +1,35 @@
> +/* SPDX-License-Identifier: MIT */
> +/*
> + * Copyright(c) 2023, Intel Corporation. All rights reserved.
> + */
> +#ifndef __XE_HECI_GSC_DEV_H__
> +#define __XE_HECI_GSC_DEV_H__
> +
> +#include <linux/types.h>
> +
> +struct xe_device;
> +struct mei_aux_device;
> +
> +/*
> + * The HECI1 bit corresponds to bit15 and HECI2 to bit14.
> + * The reason for this is to allow growth for more interfaces in the future.
> + */
> +#define GSC_IRQ_INTF(_x)  BIT(15 - (_x))
> +
> +/**
> + * struct xe_heci_gsc - graphics security controller for xe, HECI interface
> + *
> + * @adev : pointer to mei auxiliary device structure
> + * @irq : irq number
> + *
> + */
> +struct xe_heci_gsc {
> +	struct mei_aux_device *adev;
> +	int irq;
> +};
> +
> +void xe_heci_gsc_init(struct xe_device *xe);
> +void xe_heci_gsc_fini(struct xe_device *xe);
> +void xe_heci_gsc_irq_handler(struct xe_device *xe, u32 iir);
> +
> +#endif /* __XE_HECI_GSC_DEV_H__ */
> diff --git a/drivers/gpu/drm/xe/xe_irq.c b/drivers/gpu/drm/xe/xe_irq.c
> index 1dee3e832eb5..d297e9b8a3be 100644
> --- a/drivers/gpu/drm/xe/xe_irq.c
> +++ b/drivers/gpu/drm/xe/xe_irq.c
> @@ -128,6 +128,7 @@ void xe_irq_enable_hwe(struct xe_gt *gt)
>  	struct xe_device *xe = gt_to_xe(gt);
>  	u32 ccs_mask, bcs_mask;
>  	u32 irqs, dmask, smask;
> +	u32 gsc_mask = GSC_IRQ_INTF(1);
>  
>  	if (xe_device_guc_submission_enabled(xe)) {
>  		irqs = GT_RENDER_USER_INTERRUPT |
> @@ -180,6 +181,9 @@ void xe_irq_enable_hwe(struct xe_gt *gt)
>  		if (xe_hw_engine_mask_per_class(gt, XE_ENGINE_CLASS_OTHER)) {
>  			xe_mmio_write32(gt, GUNIT_GSC_INTR_ENABLE, irqs);
>  			xe_mmio_write32(gt, GUNIT_GSC_INTR_MASK, ~irqs);
> +		} else if (HAS_HECI_GSCFI(xe)) {
> +			xe_mmio_write32(gt, GUNIT_GSC_INTR_ENABLE, gsc_mask);
> +			xe_mmio_write32(gt, GUNIT_GSC_INTR_MASK, ~gsc_mask);

is there any way we could combine this with the upper calls to the same register?
I believe i915 is combining them for instance...

>  		}
>  	}
>  }
> @@ -284,6 +288,11 @@ static void gt_irq_handler(struct xe_tile *tile,
>  			instance = INTR_ENGINE_INSTANCE(identity[bit]);
>  			intr_vec = INTR_ENGINE_INTR(identity[bit]);
>  
> +			if (class == XE_ENGINE_CLASS_OTHER && instance == OTHER_GSC_INSTANCE) {
> +				xe_heci_gsc_irq_handler(xe, intr_vec);
> +				continue;
> +			}
> +
>  			engine_gt = pick_engine_gt(tile, class, instance);
>  
>  			hwe = xe_gt_hw_engine(engine_gt, class, instance, false);
> @@ -470,8 +479,9 @@ static void gt_irq_reset(struct xe_tile *tile)
>  	if (ccs_mask & (BIT(2)|BIT(3)))
>  		xe_mmio_write32(mmio,  CCS2_CCS3_INTR_MASK, ~0);
>  
> -	if (tile->media_gt &&
> -	    xe_hw_engine_mask_per_class(tile->media_gt, XE_ENGINE_CLASS_OTHER)) {
> +	if ((tile->media_gt &&
> +	     xe_hw_engine_mask_per_class(tile->media_gt, XE_ENGINE_CLASS_OTHER)) ||
> +	    HAS_HECI_GSCFI(tile_to_xe(tile))) {
>  		xe_mmio_write32(mmio, GUNIT_GSC_INTR_ENABLE, 0);
>  		xe_mmio_write32(mmio, GUNIT_GSC_INTR_MASK, ~0);
>  	}
> -- 
> 2.34.1
>