On 2023/4/5 17:55, Sui Jingfeng wrote:
Hi,
thanks you for the time and effort for reviewing.
On 2023/4/4 19:03, Javier Martinez Canillas wrote:
Javier Martinez Canillas <javierm@xxxxxxxxxx> writes:
[...]
/*
* Remove the device from the device hierarchy. This is the
right thing
- * to do for firmware-based DRM drivers, such as EFI, VESA or
VGA. After
+ * to do for firmware-based fb drivers, such as EFI, VESA or
VGA. After
That sentences is not well phrased. Maybe say 'This is required for
firmware-provided graphics, such as EFI, VESA or VGA.'
Graphic drivers or display drivers would indeed be more accurate
here. But
I think that "fb drivers" is still well pharsed since the are other
places
where either fbdev or DRM drivers for firmware-provided framebuffers
are
named like that.
Sui,
Maybe you could post a follow-up patch to improve the comment as
suggested
by Thomas?
Yes, certainly.
This is the right thing to do for conflicting drivers takes over the
hardware resource required.
While the `drivers` include both drm driver and the real device
dependent framebuffer driver,
They are typically build as kernel module as oppose to the efifb and
simplefb which is built
into kernel kernel. Firmware framebuffer is conflict with the drm
driver is because the memory
region they use overlaps. If there no overlap, then no `taken over`
will be happen.
By the way, I'm asked to made efifb and simplefb works on LoongArch and
Mips platform in the past.
We are using downstream kernel(linux-4.19) at that time. efifb is ony
works for platform with uefi
firmware support. By ensure that framebuffer locate at the address range
of the on-board video ram(VRAM)
and passing screen_info parameters to kernel correctly,
drm_aperture_remove_conflicting_framebuffers
will success. This required the uefi firmware engineer understand this,
for loongson bridge chip, this need
a small trick. Simplefb is only tested on loongson SoC platform by
providing fb parameters in DT which
match the PMON firmware's setting. As the framebuffer may located at
anywhere in the physical address
space, there no aperture for SoC anymore. Should call
aperture_remove_conflicting_devices(0, ~0, false, "drmfb")
to remove them all.
But the comments is actually nearly perfect in overall, it has some
difficulty to improve
the perfection. Below is my personal understanding toward the above
sentence.
efifb and simplefb belong to the class of firmware based framebuffer
driver.
They are generic and platform agnostic, yet they have to relay on the
firmware
to passing fb format, fb size, fb base address, fb resolution and fb
stride etc to the kernel.
Linux kernel using those information to fill the global screen_info
structure.
sysfb_init() then using the global screen_info to create a platform
device,
the device will be claimed by efifb or simplefb driver finally. This
is a hand over solution.
It relay on the firmware setup such a framebuffer and hand over the
state(this is
actually a kind of modeset state) to kernel.
efifb only own the potential hardware resource for a very short time if a
conflicting drm driver probe successfully.
For the platform/graphics without a drm driver, developers may choose to
use efifb driver as a replacement. So, there are no conflicting
happen on
such a case. The `nomodeset` kernel cmd options can also be used for
debugging and testing purpose if the more intelligent drm driver is
broken
due to bugs.
