Thank you for your contribution! Sashiko AI review found 3 potential issue(s) 
to consider:
- [Medium] video/hdmi: hdmi_infoframe_pack_only() missing EMP support
- [Medium] video/hdmi: missing bounds and type checks in 
hdmi_emp_infoframe_vtem_pack()
- [Low] video/hdmi: missing @vtem kernel-doc entry in union hdmi_infoframe
--

commit 435f43f92e62fa2ad37ec03ba43478dd2a0e2bba
Author: Derek Foreman <[email protected]>

video/hdmi: Add VTEM EMP packing

This commit adds structures and helpers to initialize, pack, and log HDMI
Video Timing Extended Metadata (VTEM) Extended Metadata Packets (EMPs) for
Variable Refresh Rate support.

> diff --git a/drivers/video/hdmi.c b/drivers/video/hdmi.c
> index 45b42f14a7503..ed9abc7ad2285 100644
> --- a/drivers/video/hdmi.c
> +++ b/drivers/video/hdmi.c
> @@ -983,6 +983,10 @@ hdmi_infoframe_pack(union hdmi_infoframe *frame,
>               length = hdmi_vendor_any_infoframe_pack(&frame->vendor,
>                                                       buffer, size);
>               break;
> +     case HDMI_INFOFRAME_TYPE_EMP:
> +             length = hdmi_emp_infoframe_vtem_pack(&frame->vtem,
> +                                                   buffer, size);
> +             break;
>       default:
>               WARN(1, "Bad infoframe type %d\n", frame->any.type);
>               length = -EINVAL;

[Severity: Medium]
As reported by Sashiko on RFC [1], hdmi_infoframe_pack() handles
HDMI_INFOFRAME_TYPE_EMP, but hdmi_infoframe_pack_only() in
drivers/video/hdmi.c was not updated to handle it:

drivers/video/hdmi.c:hdmi_infoframe_pack_only() {
        switch (frame->any.type) {
        case HDMI_INFOFRAME_TYPE_AVI:
        ...
        case HDMI_INFOFRAME_TYPE_VENDOR:
                length = hdmi_vendor_any_infoframe_pack_only(&frame->vendor,
                                                             buffer, size);
                break;
        default:
                WARN(1, "Bad infoframe type %d\n", frame->any.type);
                length = -EINVAL;
        }
}

If a driver packs an EMP infoframe stored in union hdmi_infoframe via
hdmi_infoframe_pack_only(), won't it hit the default branch, trigger the
WARN(1), and fail with -EINVAL?

Also, since hdmi_emp_infoframe_vtem_pack() does not modify its vtem
argument, should its parameter be const struct hdmi_emp_infoframe_vtem *
(or have a corresponding _pack_only helper) so hdmi_infoframe_pack_only()
can call it with a const union hdmi_infoframe *?

[ ... ]

> @@ -1919,3 +1951,66 @@ int hdmi_infoframe_unpack(union hdmi_infoframe *frame,
>       return ret;
>  }
>  EXPORT_SYMBOL(hdmi_infoframe_unpack);
> +
> +/**
> + * hdmi_emp_infoframe_vtem_init() - initialize an HDMI VTEM EMP
> + * @emp: HDMI VTEM EMP
> + */
> +void hdmi_emp_infoframe_vtem_init(struct hdmi_emp_infoframe_vtem *emp)
> +{
> +     memset(emp, 0, sizeof(*emp));
> +     emp->type = HDMI_INFOFRAME_TYPE_EMP;
> +}
> +EXPORT_SYMBOL(hdmi_emp_infoframe_vtem_init);
> +
> +/**
> + * hdmi_emp_infoframe_vtem_pack() - pack a vtem into a binary buffer
> + * @vtem: pointer to &struct hdmi_emp_infoframe_vtem to pack
> + * @buffer: pointer to pre-allocated output buffer
> + * @size: maximum size in bytes of @buffer
> + *
> + * Pack a VTEM infoframe struct into a binary buffer in a way that's suitable
> + * to be sent out over the wire.
> + *
> + * Returns the number of bytes packed into the binary buffer or a negative
> + * error code on failure.
> + */
> +ssize_t hdmi_emp_infoframe_vtem_pack(struct hdmi_emp_infoframe_vtem *vtem,
> +                                  void *buffer, size_t size)
> +{
> +     u8 *ptr = buffer;
> +
> +     if (size < HDMI_INFOFRAME_SIZE(VTEM))
> +             return -EINVAL;
> +
> +     /* 4 bits max */
> +     if (vtem->fva_factor_m1 > 0xF)
> +             return -EINVAL;
> +
> +     memset(buffer, 0, size);
> +
> +     /* Header Data */
> +     ptr[0] = vtem->type;
> +     ptr[1] = BIT(7) | BIT(6);
> +     ptr[2] = 0;
> +
> +     ptr[3] = BIT(7) | BIT(2);
> +
> +     ptr[4] = 0;
> +     ptr[5] = 1;
> +     ptr[6] = 0;
> +     ptr[7] = 1;
> +     ptr[8] = 0;
> +     ptr[9] = 4;
> +
> +     /* This is where the VTEM part starts */
> +     ptr[10] = (vtem->fva_factor_m1 << 4) |
> +               (vtem->m_const ? BIT(1) : 0) |
> +               (vtem->game_vrr_en ? BIT(0) : 0);
> +     ptr[11] = vtem->base_vfront;
> +     ptr[12] = vtem->base_refresh_rate >> 8 & 0x3;
> +     ptr[13] = vtem->base_refresh_rate & 0xFF;
> +
> +     return 14;
> +}
> +EXPORT_SYMBOL(hdmi_emp_infoframe_vtem_pack);

[Severity: Medium]
As reported by Sashiko on RFC [2], hdmi_emp_infoframe_vtem_pack() checks
that vtem->fva_factor_m1 fits in 4 bits and returns -EINVAL if exceeded,
but silently masks the u16 vtem->base_refresh_rate to 10 bits when writing
ptr[12] and ptr[13] instead of rejecting values above 0x3ff. It also
writes ptr[0] = vtem->type without verifying that vtem->type is
HDMI_INFOFRAME_TYPE_EMP.

In the DRM HDMI state helper path added later in this series,
drm_atomic_helper_connector_hdmi_check() calls hdmi_validate_vrr(), which
returns early when new_crtc_state->vrr_enabled is false and skips its
refresh rate check, while hdmi_generate_emp_infoframe_vtem() still runs
and sets vtem->base_refresh_rate from drm_mode_vrefresh(&crtc_state->mode)
for non-VIC modes before calling hdmi_infoframe_pack().

Should hdmi_emp_infoframe_vtem_pack() validate that
vtem->base_refresh_rate <= 0x3ff and vtem->type == HDMI_INFOFRAME_TYPE_EMP,
returning -EINVAL on invalid input rather than transmitting a truncated or
malformed packet?

> diff --git a/include/linux/hdmi.h b/include/linux/hdmi.h
> index b80a5ee63bb23..71e2b36896afe 100644
> --- a/include/linux/hdmi.h
> +++ b/include/linux/hdmi.h

[ ... ]

> @@ -457,6 +475,7 @@ union hdmi_infoframe {
>       union hdmi_vendor_any_infoframe vendor;
>       struct hdmi_audio_infoframe audio;
>       struct hdmi_drm_infoframe drm;
> +     struct hdmi_emp_infoframe_vtem vtem;
>  };

[Severity: Low]
This isn't a runtime bug, but union hdmi_infoframe in include/linux/hdmi.h
has a kernel-doc comment block above it (included by
Documentation/gpu/drm-kms-helpers.rst):

include/linux/hdmi.h:
/**
 * union hdmi_infoframe - overall union of all abstract infoframe 
representations
 * @any: generic infoframe
 * @avi: avi infoframe
 * @spd: spd infoframe
 * @vendor: union of all vendor infoframes
 * @audio: audio infoframe
 * @drm: Dynamic Range and Mastering infoframe

Should an @vtem entry be added to the kernel-doc comment for union
hdmi_infoframe to avoid a scripts/kernel-doc warning during W=1 or
htmldocs builds?

[1] 
https://lore.kernel.org/r/[email protected]
[2] 
https://lore.kernel.org/r/[email protected]

-- 
Sashiko AI review ยท 
https://sashiko.dev/#/patchset/[email protected]?part=3

Reply via email to