intel-gfx.lists.freedesktop.org archive mirror
 help / color / mirror / Atom feed
From: Christoph Hellwig <hch@lst.de>
To: Zhi Wang <zhi.wang.linux@gmail.com>
Cc: intel-gfx@lists.freedesktop.org, linux-kernel@vger.kernel.org,
	Terrence Xu <terrence.xu@intel.com>,
	dri-devel@lists.freedesktop.org, jgg@nvidia.com,
	intel-gvt-dev@lists.freedesktop.org, hch@lst.de
Subject: Re: [Intel-gfx] [PATCH 1/3] i915/gvt: Introduce the mmio_table.c to support VFIO new mdev API
Date: Mon, 31 Jan 2022 07:34:51 +0100	[thread overview]
Message-ID: <20220131063451.GA4390@lst.de> (raw)
In-Reply-To: <20220127120508.11330-1-zhi.a.wang@intel.com>

>  ifeq ($(CONFIG_DRM_I915_GVT),y)
> -i915-y += intel_gvt.o
> +i915-y += intel_gvt.o gvt/mmio_table.o

With the split from my series in mind that builds all of the gvt/
subdirectory into a separate module I'd be tempted to places this new
file into the main i915 directory as e.g. intel_gvt_mmio_table.c, given
that it will have to be built into the main i915 module.

> -static void init_device_info(struct intel_gvt *gvt)

I'd keep this function as-is, as the newly added caller isn't actually
needed (I'll comment on this more on the next patch).

> -/* Describe per-platform limitations. */
> -struct intel_gvt_device_info {
> -	u32 max_support_vgpus;
> -	u32 cfg_space_size;
> -	u32 mmio_size;
> -	u32 mmio_bar;
> -	unsigned long msi_cap_offset;
> -	u32 gtt_start_offset;
> -	u32 gtt_entry_size;
> -	u32 gtt_entry_size_shift;
> -	int gmadr_bytes_in_cmd;
> -	u32 max_surface_size;
> -};

.. and with that there should be no need to move this declaration
as well.

> -struct gvt_mmio_block {
> +struct intel_gvt_mmio_block {

Any good reason for this rename?  It just seems to create a lot of
churn without muchof a reason.

> +static int do_mmio(u32 offset, u16 flags, u32 size, u32 addr_mask,
> +		   u32 ro_mask, u32 device,
> +		   struct intel_gvt_mmio_table_iter *iter)

Nit:  I'd pass the iter first to these kinds of callbacks.
Also the do_ name (including for the method name in the struct itself)
looks odd.  I'd rather use a _cb or _fn postfix.

> +	for (i = start; i < end; i += 4) {
> +		info = kzalloc(sizeof(*info), GFP_KERNEL);
> +		if (!info)
> +			return -ENOMEM;
> +
> +		info->offset = i;
> +		p = intel_gvt_find_mmio_info(gvt, info->offset);
> +		if (p) {
> +			WARN(1, "dup mmio definition offset %x\n",
> +				info->offset);
> +			kfree(info);
> +
> +			/* We return -EEXIST here to make GVT-g load fail.
> +			 * So duplicated MMIO can be found as soon as
> +			 * possible.
> +			 */
> +			return -EEXIST;
> +		}

I'd allocate the new structure only after the lookup to simplify this
a bit.

> +
> +		info->ro_mask = ro_mask;

The r/o mask is only used here, so why not move it into the local
declarations in handlers.c instead of the table built into i915.ko?

> +		info->device = device;
> +		info->read = intel_vgpu_default_mmio_read;
> +		info->write = intel_vgpu_default_mmio_write;

Given that we always initialize ->read and ->write here,
setup_mmio_handler can be simplified a bit and only needs to override
the handlers if actually specified in the table.

> +static int init_mmio_info(struct intel_gvt *gvt)
> +{
> +	struct intel_gvt_mmio_table_iter iter;
> +
> +	iter.i915 = gvt->gt->i915;
> +	iter.data = gvt;
> +	iter.do_mmio = do_mmio;
> +	iter.do_mmio_block = do_mmio_block;

Nit: This and the other caller could initialize the iter structure
statically:

	struct intel_gvt_mmio_table_iter iter = {
		.i915		= gvt->gt->i915,
		.data		= gvt,
		.mmio_cb	= intel_gvt_setup_mmio_cb,
		.mmio_block_cb	= intel_gvt_setup_mmio_block_cb,
	};

> +	block = find_mmio_block(gvt, VGT_PVINFO_PAGE);
> +	block->read = pvinfo_mmio_read;
> +	block->write = pvinfo_mmio_write;

Check for NULL here?

>  	} else if (IS_SKYLAKE(i915) ||
> -		   IS_KABYLAKE(i915) ||
> -		   IS_COFFEELAKE(i915) ||
> -		   IS_COMETLAKE(i915)) {
> -		ret = init_bdw_mmio_info(gvt);
> +			IS_KABYLAKE(i915) ||
> +			IS_COFFEELAKE(i915) ||
> +			IS_COMETLAKE(i915)) {
> +		ret = init_bdw_mmio_handlers(gvt);

Why the spurious reformatting?

> +/**
> + * intel_gvt_get_device_type - return the device flag of a GVT device
> + * @i915: drm i915 private data
> + *
> + * This function will return the device flag of a GVT device.
> + */
> +unsigned long intel_gvt_get_device_type(struct drm_i915_private *i915)
> +{
> +	if (IS_BROADWELL(i915))
> +		return D_BDW;
> +	else if (IS_SKYLAKE(i915))
> +		return D_SKL;
> +	else if (IS_KABYLAKE(i915))
> +		return D_KBL;
> +	else if (IS_BROXTON(i915))
> +		return D_BXT;
> +	else if (IS_COFFEELAKE(i915) || IS_COMETLAKE(i915))
> +		return D_CFL;
> +
> +	return 0;
> +}

I'd move this into intel_gvt.c, next to is_supported_device which
also lists all the supported platforms.  Preferably as a prep patch
that does the move and change of argument before the main MMIO table
patch.

> +++ b/drivers/gpu/drm/i915/gvt/mmio_table.h

Do we really need this new header vs just using intel_gvt.h?

  parent reply	other threads:[~2022-01-31  6:34 UTC|newest]

Thread overview: 25+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-01-27 12:05 [Intel-gfx] [PATCH 1/3] i915/gvt: Introduce the mmio_table.c to support VFIO new mdev API Zhi Wang
2022-01-27 12:05 ` [Intel-gfx] [PATCH 2/3] i915/gvt: save the initial HW state snapshot in i915 Zhi Wang
2022-01-31  6:38   ` Christoph Hellwig
2022-01-27 12:05 ` [Intel-gfx] [PATCH 3/3] i915/gvt: Use the initial HW state snapshot saved " Zhi Wang
2022-01-27 14:52 ` [Intel-gfx] [PATCH 1/3] i915/gvt: Introduce the mmio_table.c to support VFIO new mdev API Christoph Hellwig
2022-01-27 15:17 ` [Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for series starting with [1/3] " Patchwork
2022-01-27 15:18 ` [Intel-gfx] ✗ Fi.CI.SPARSE: " Patchwork
2022-01-27 15:48 ` [Intel-gfx] ✓ Fi.CI.BAT: success " Patchwork
2022-01-27 17:14 ` [Intel-gfx] [PATCH 1/3] " kernel test robot
2022-01-27 17:46 ` kernel test robot
2022-01-27 19:08 ` kernel test robot
2022-01-27 20:50 ` [Intel-gfx] ✗ Fi.CI.IGT: failure for series starting with [1/3] " Patchwork
2022-01-28  5:01 ` [Intel-gfx] [RFC PATCH] i915/gvt: intel_gvt_match_device() can be static kernel test robot
2022-01-28  5:11 ` [Intel-gfx] [PATCH 1/3] i915/gvt: Introduce the mmio_table.c to support VFIO new mdev API kernel test robot
2022-01-31  6:34 ` Christoph Hellwig [this message]
2022-02-07  7:32 ` Christoph Hellwig
2022-02-07  8:28   ` Wang, Zhi A
2022-02-07  8:35     ` Christoph Hellwig
2022-02-07 10:48       ` Jani Nikula
2022-02-07 11:57         ` Zhi Wang
2022-02-07 12:06           ` Christoph Hellwig
2022-02-07 12:13             ` Jani Nikula
2022-02-08  8:04         ` Joonas Lahtinen
2022-02-07 14:47 ` kernel test robot
2022-02-08 14:38 ` kernel test robot

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20220131063451.GA4390@lst.de \
    --to=hch@lst.de \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=intel-gfx@lists.freedesktop.org \
    --cc=intel-gvt-dev@lists.freedesktop.org \
    --cc=jgg@nvidia.com \
    --cc=linux-kernel@vger.kernel.org \
    --cc=terrence.xu@intel.com \
    --cc=zhi.wang.linux@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).