[Bug 117131] vga_switcheroo does not switch IGP -> DIS ( IGP == i915 , DIS == radeon )

[bugzilla-daemon at bugzilla.kernel.org]

https://bugzilla.kernel.org/show_bug.cgi?id=117131

--- Comment #4 from Jason Vas Dias <jason.vas.dias at gmail.com> ---
Aha, so this ACPI call failure looks very fatal to any vga_switcheroo :

/**
 * radeon_atpx_switch_start - notify the sbios of a GPU switch
 *
 * @atpx: atpx info struct
 * @mux_id: mux state (0 = integrated GPU, 1 = discrete GPU)
 *
 * Execute the ATPX_FUNCTION_GRAPHICS_DEVICE_SWITCH_START_NOTIFICATION ATPX
 * function to notify the sbios that a switch between the discrete GPU and
 * integrated GPU has begun (all asics).
 * Returns 0 on success, error on failure.
 */
static int radeon_atpx_switch_start(struct radeon_atpx *atpx, u16 mux_id)
{
    struct acpi_buffer params;
    union acpi_object *info;
    struct atpx_mux input;

    if (atpx->functions.switch_start) {
        input.size = 4;
        input.mux = mux_id;
        params.length = input.size;
        params.pointer = &input;
        info = radeon_atpx_call(atpx->handle,
                    ATPX_FUNCTION_GRAPHICS_DEVICE_SWITCH_START_NOTIFICATION,
                    &params);
        if (!info)
            return -EIO;
        kfree(info);
    }
    return 0;
}

/**
 * radeon_atpx_switch_end - notify the sbios of a GPU switch
 *
 * @atpx: atpx info struct
 * @mux_id: mux state (0 = integrated GPU, 1 = discrete GPU)
 *
 * Execute the ATPX_FUNCTION_GRAPHICS_DEVICE_SWITCH_END_NOTIFICATION ATPX
 * function to notify the sbios that a switch between the discrete GPU and
 * integrated GPU has ended (all asics).
 * Returns 0 on success, error on failure.
 */
static int radeon_atpx_switch_end(struct radeon_atpx *atpx, u16 mux_id)
{
    struct acpi_buffer params;
    union acpi_object *info;
    struct atpx_mux input;

    if (atpx->functions.switch_end) {
        input.size = 4;
        input.mux = mux_id;
        params.length = input.size;
        params.pointer = &input;
        info = radeon_atpx_call(atpx->handle,
                    ATPX_FUNCTION_GRAPHICS_DEVICE_SWITCH_END_NOTIFICATION,
                    &params);
        if (!info)
            return -EIO;
        kfree(info);
    }
    return 0;
}





So if any of these ACPI calls fail, the entire switch fails .

There is obviously some way that the closed source FGLRX driver 
is able to make these calls successfully ; the card goes into
graphics mode OK with fglrx under Linux 3.10 (EL7) .

But it appears that the new open source radeon driver 8970M support 
is not working correctly in Linux 4.5.0 .

I guess this bug should more properly be raised directly against the
drivers/gpu/drm/radeon driver then .

Maybe I should try running the EL7 partition OS in a VM and tracing how
FGLRX is calling these ACPI calls & modifying radeon_atpx_handler to make
them the same way ? a rather cumbersome way to get to the bottom of it ...

I will also try various combinations of the above kernel parameters and
radeon.dpm=1 or radeon.dpm=0 .

-- 
You are receiving this mail because:
You are watching the assignee of the bug.