Skip to contentSkip to conversation

fb/intelfb: Do not depend on EMBEDDED

PATCH Jesse Barnes 4 participants 9 messages Dec 13, 2009 – Dec 16, 2009

lore
Share & tools
lore
#0 lore raw

Right, the logic is that the driver really is for embedded (i.e. very special purpose) use. It should not be selected unless you really know what you're doing or are building a very particular product. If you can think of a better way of preventing users and distros from accidentally selecting this, then please send a patch.
Jesse

Quoted 1 line · Dave Airlie
Quoted 1 line
Quoted 1 line

I am worried that the intelfb driver depends on EMBEDDED. I consider>> this an abuse of the EMBEDDED configuration option, which as I>> understand it was originally meant to expose fine-tuning options,>> rather than to arbitrarily disable drivers when not selected.>>Since we merged a kms driver for Intel hw that supports all intel chipsets>and more importantly all the outputs on Intel chipsets, intelfb should>be considered legacy at the least and broken on > 50% of intel hw.>>We left it in in that most ppl who wanted it were using it in embedded >configs, whereas for most users it just doesn't work, like I don't thinkit>supports LVDS which means loading it on a laptop will trash it.>>Dave>>>> >> So I suggest that we drop this dependency now.>> >> Signed-off-by: Jean Delvare <[email protected]>>> Cc: Jesse Barnes <[email protected]>>> Cc: Dave Airlie <[email protected]>>> --->> Jesse, in the original commit, you wrote that intelfb was "really a>> special purpose embedded driver". It looks like a perfectly standard>> framebuffer driver to me, which means that it may have users beyond>> embedded. For example I always prefer framebuffer over X for my>> servers. Or am I missing something and intelfb is really special?>> >> drivers/video/Kconfig | 2 +->> 1 file changed, 1 insertion(+), 1 deletion(-)>> >> --- linux-2.6.32.orig/drivers/video/Kconfig 2009-12-03 08:48:34.000000000 +0100>> +++ linux-2.6.32/drivers/video/Kconfig 2009-12-11 10:57:43.000000000 +0100>> @@ -1121,7 +1121,7 @@ config FB_CARILLO_RANCH>> >> config FB_INTEL>> tristate "Intel 830M/845G/852GM/855GM/865G/915G/945G/945GM/965G/965GM support (EXPERIMENTAL)">> - depends on EXPERIMENTAL && FB && PCI && X86 && AGP_INTEL && EMBEDDED>> + depends on EXPERIMENTAL && FB && PCI && X86 && AGP_INTEL>> select FB_MODE_HELPERS>> select FB_CFB_FILLRECT>> select FB_CFB_COPYAREA>> >> >���{.n�+�������+%�����ݶ��w��{.n�+����{��G�����{ay�ʇڙ�,j��f���h��������z_�(�階�ݢj"���m������G������?����&���~��iO��z��v�^�m���� ������?�I�

#1 lore raw

Hi Dave, Jesse,

Quoted 8 lines · Dave Airlie
Quoted 4 lines

I am worried that the intelfb driver depends on EMBEDDED. I consider this an abuse of the EMBEDDED configuration option, which as I understand it was originally meant to expose fine-tuning options, rather than to arbitrarily disable drivers when not selected.

Since we merged a kms driver for Intel hw that supports all intel chipsets and more importantly all the outputs on Intel chipsets, intelfb should be considered legacy at the least and broken on > 50% of intel hw.

Are you suggesting that the "kms driver" is a drop-in replacement for intelfb? More specifically, can I use that "kms driver" to get a high-resolution console?

Quoted 3 lines

We left it in in that most ppl who wanted it were using it in embedded configs, whereas for most users it just doesn't work, like I don't think it supports LVDS which means loading it on a laptop will trash it.

The intelfb driver is marked EXPERIMENTAL, so it is already expected to not work perfectly in all cases.

Quoted 3 lines · Jesse Barnes

Right, the logic is that the driver really is for embedded (i.e. very special purpose) use. It should not be selected unless you really know what you're doing or are building a very particular product.

The Kconfig help text doesn't say anything about this.

My understanding is that the intelfb driver was not _designed_ to be useful on embedded designs only. It just happens to be incomplete in such a way that it works only in a few selected cases, which happen to be embedded cases, and it fails in many other cases.

The proper way to handle this is not to make the driver depend on EMBEDDED. The proper way would be to change the intelfb driver so that it no longer binds to devices it will not properly support. If the driver doesn't support LVDS (whatever it is) then it should cleanly fail on systems which have that.

Quoted 2 lines

If you can think of a better way of preventing users and distros from accidentally selecting this,

The reason why I sent a patch in the first place is exactly opposite:
I want to let distros select this driver. My case is as follows: we had a product which included the intelfb driver, which we are in the process of upgrading. Now we find that the intelfb driver is gone (no longer selectable), which causes a problem as far as the upgrade path of our customers is concerned.

So the problem I have to solve is: given a customer who was successfully using the intelfb driver before, what solutions can we offer when said customer upgrades to our new product? My own solution was straightforward: keep including the intelfb driver in the new product. Thus my patch dropping the dependency on EMBEDDED. If another solution exists, please let me know.

Quoted 1 line

then please send a patch.

First of all, I'd need to better understand how the various drivers relate to each other, and what functionality they provide. In my little outdated mind, framebuffer is for high-resolution console without X, while drm is for accelerated X. Apparently this changed more or less recently, but the documentation wasn't updated.

It would help if the DRM option description was updated. It still
reads: "Direct Rendering Manager (XFree86 4.1.0 and higher DRI support)". If the DRM core is now also providing support for framebuffer-like functionality (again, if I understand correctly) then the reference to XFree86 should be dropped. The help text should also be updated to properly describe all that the DRM core offers today.

Honestly, I'm probably not the best person to write this text, as I don't know much about the current state of this graphics stuff.

Thanks,

Signature · 2 lines
Jean Delvare
Suse L3
#2 lore raw
Quoted 8 lines
Quoted 4 lines

Since we merged a kms driver for Intel hw that supports all intel chipsets and more importantly all the outputs on Intel chipsets, intelfb should be considered legacy at the least and broken on > 50% of intel hw.

Are you suggesting that the "kms driver" is a drop-in replacement for intelfb? More specifically, can I use that "kms driver" to get a high-resolution console?

Yes, thats all it does, but you need to make suer the X.org userspace you ship supports which for Intel it has done for a while.

Quoted 4 lines

My understanding is that the intelfb driver was not _designed_ to be useful on embedded designs only. It just happens to be incomplete in such a way that it works only in a few selected cases, which happen to be embedded cases, and it fails in many other cases.

Well I wrote a large chunk of it and it was for an embedded system.

Quoted 5 lines

The proper way to handle this is not to make the driver depend on EMBEDDED. The proper way would be to change the intelfb driver so that it no longer binds to devices it will not properly support. If the driver doesn't support LVDS (whatever it is) then it should cleanly fail on systems which have that.

LVDS are laptops, adding code to fix intelfb to know when its failing is pointless, kernel modesetting supports all the hw properly that intelfb fails on.

Quoted 9 lines
Quoted 2 lines

If you can think of a better way of preventing users and distros from accidentally selecting this,

The reason why I sent a patch in the first place is exactly opposite:
I want to let distros select this driver. My case is as follows: we had a product which included the intelfb driver, which we are in the process of upgrading. Now we find that the intelfb driver is gone (no longer selectable), which causes a problem as far as the upgrade path of our customers is concerned.

We don't want distros using this driver going forward or shipping it, any distro that has a reason to enable it needs to enable EMBEDDED,

We cannot have two drivers for this hardware enabled by default, and having two options in the menus confuses users and distros alike, so EMBEDDED is a clear sign. I'm tempted to make it CONFIG_BROKEN.

Quoted 6 lines

So the problem I have to solve is: given a customer who was successfully using the intelfb driver before, what solutions can we offer when said customer upgrades to our new product? My own solution was straightforward: keep including the intelfb driver in the new product. Thus my patch dropping the dependency on EMBEDDED. If another solution exists, please let me know.

You can do that, just enable CONFIG_EMBEDDED, but that is a distro choice, upsteram this driver is dead, you should see if the KMS solution is suitable for you customer and migrate them to it.

Quoted 6 lines

First of all, I'd need to better understand how the various drivers relate to each other, and what functionality they provide. In my little outdated mind, framebuffer is for high-resolution console without X, while drm is for accelerated X. Apparently this changed more or less recently, but the documentation wasn't updated.

Quoted 7 lines

It would help if the DRM option description was updated. It still
reads: "Direct Rendering Manager (XFree86 4.1.0 and higher DRI support)". If the DRM core is now also providing support for framebuffer-like functionality (again, if I understand correctly) then the reference to XFree86 should be dropped. The help text should also be updated to properly describe all that the DRM core offers today.

We should probably fix that alright, I'll have to think about how we can do that.

Dave.

#3 lore raw

Hi Dave,

Le dimanche 13 décembre 2009 22:53, Dave Airlie a écrit :

Quoted 6 lines
Quoted 3 lines

Are you suggesting that the "kms driver" is a drop-in replacement for intelfb? More specifically, can I use that "kms driver" to get a high-resolution console?

Yes, thats all it does, but you need to make sure the X.org userspace you ship supports which for Intel it has done for a while.

... if and only if I need X.org at all, right? I sure hope that installing X.org hasn't become mandatory just to have a high-resolution console?

Quoted 9 lines
Quoted 6 lines

(...)
The proper way to handle this is not to make the driver depend on EMBEDDED. The proper way would be to change the intelfb driver so that it no longer binds to devices it will not properly support. If the driver doesn't support LVDS (whatever it is) then it should cleanly fail on systems which have that.

LVDS are laptops, adding code to fix intelfb to know when its failing is pointless,

I don't think it is that pointless. Failing gracefully, preferably with a useful error message, is always important.

Quoted 1 line

kernel modesetting supports all the hw properly that intelfb fails on.

True but irrelevant. If I load a driver and it crashes my system, that's bad, even if there is another driver available that would not have crashed my system (and may even have made something useful out of it.) Again, we want drivers to fail gracefully, and point the user to alternatives if applicable.

Quoted 5 lines

(...)
We don't want distros using this driver going forward or shipping it, any distro that has a reason to enable it needs to enable EMBEDDED,

We cannot have two drivers for this hardware enabled by default, and

Actually we sort of can, that's the power of modules. I'm not sure what "make allyesconfig" would do though... does it honnor "default n" in Kconfig?

Quoted 2 lines

having two options in the menus confuses users and distros alike, so EMBEDDED is a clear sign. I'm tempted to make it CONFIG_BROKEN.

The remaining users of the intelfb driver may disagree, but I'd indeed prefer the driver to be marked as broken, if that's what it is. Or even removed from the kernel tree. If they want to keep it in the tree, they should fix it.

Quoted 8 lines
Quoted 6 lines

So the problem I have to solve is: given a customer who was successfully using the intelfb driver before, what solutions can we offer when said customer upgrades to our new product? My own solution was straightforward: keep including the intelfb driver in the new product. Thus my patch dropping the dependency on EMBEDDED. If another solution exists, please let me know.

You can do that, just enable CONFIG_EMBEDDED, but that is a distro choice,

Enabling CONFIG_EMBEDDED would give us access to other tweaking options that we don't want to touch, and I have no idea if any of these is set by default. So thanks but no thanks. I'm much better just dropping the dependency on EMBEDDED for intelfb, at least I control the scope of this change.

Quoted 2 lines

upstream this driver is dead, you should see if the KMS solution is suitable for you customer and migrate them to it.

Except that I have no clue if any of my customers are actually using it, nor who they are, nor what hardware they use. As it stands, I would know if we were to drop intelfb in our next product, because these customers would complain, probably loudly. But obviously this is no solution business-wise.

If I wanted to attempt a transparent migration, where would I start? As I recall, console on framebuffer needs specific boot parameters, right? I don't suppose that kms will transparently pick them and do the right thing, do I? If there documentation available on how one can get a high resolution console using kms?

Thanks,

Signature · 2 lines
Jean Delvare
Suse L3
#4 lore raw
Quoted 5 lines · On Wed, 16 Dec 2009 14:37:31 +0100 Jean Delvare <[email protected]> wrote:

If I wanted to attempt a transparent migration, where would I start? As I recall, console on framebuffer needs specific boot parameters, right? I don't suppose that kms will transparently pick them and do the right thing, do I? If there documentation available on how one can get a high resolution console using kms?

Actually, that's one of the big advantages of the KMS based driver. It *does* transparently pick the native resolution of the detected output(s) and set up a high res console. You can provide boot options to override it, but unlike intelfb, boot options aren't necessary to get things working in the first place.

If you want to run X later, you'll need an updated xf86-video-intel (preferably 2.9.x) or xf86-video-radeon driver as well, since code to support the KMS interfaces landed only slightly before 2.6.29 came out in the case of Intel.

Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center
#5 lore raw
Quoted 5 lines · Jesse Barnes

Actually, that's one of the big advantages of the KMS based driver. It *does* transparently pick the native resolution of the detected output(s) and set up a high res console. You can provide boot options to override it, but unlike intelfb, boot options aren't necessary to get things working in the first place.

How do I set the (default) resolution and timings with non-EDID display (e.g. analog RGB TV)?

fbset seems to only set the "logical" resolution, doesn't like the pixclock.

Signature · 1 line
Krzysztof Halasa
#6 lore raw
Quoted 8 lines
Quoted 7 lines

Actually, that's one of the big advantages of the KMS based driver. It *does* transparently pick the native resolution of the detected output(s) and set up a high res console. You can provide boot options to override it, but unlike intelfb, boot options aren't necessary to get things working in the first place.

Quoted 5 lines

How do I set the (default) resolution and timings with non-EDID display (e.g. analog RGB TV)?

fbset seems to only set the "logical" resolution, doesn't like the pixclock.

Currently with the command line video=, this isn't optimal but still nobody has come up with a way to make fbset useful since it can't deal with multiple heads.

also the old intelfb didn't deal with analog rgb tv at all.

Dave.

#7 lore raw
Quoted 3 lines · Dave Airlie

Currently with the command line video=, this isn't optimal but still nobody has come up with a way to make fbset useful since it can't deal with multiple heads.

Ahh, now I can see it.

Quoted 1 line

also the old intelfb didn't deal with analog rgb tv at all.

Well... actually, it did: intelfb works very well if a correct mode is set (a trivial patch adjusting the clock lower limit to 13.5 MHz may be needed, or one can use the double clock and resolution, or maybe 1:1 pixel aspect ratio etc). It's not about the CH7xxx encoder, the TV is connected directly to VGA output.

E.g.

# PAL 720x576, 50 Hz, Interlaced (13.5 MHz dotclock) #

#                               Horizontal      Vertical
#       Resolution              720             576
#       Scan Frequency          15.625 kHz      50 Hz (I)
#       Sync Width              4.296 us        1.280 ms
#       Front Porch             1.000 us        0.256 ms
#       Back Porch              5.370 us        1.600 ms
#       Active Time             53.333 us       36.864 ms
#       Blank Time              10.667 us       3.136 ms
#       Polarity                Negative        Negative
#
#       H:      720 + 14 +  58 + 72 = 864
#       V:      576 +  4 +  20 + 25 = 625

mode "PAL"

        # D: 13.5 MHz, H: 15.625 kHz, V: 50 Hz
        geometry 720 576 720 576 32
        timings 74074 72 14 25 4 58 20
        laced true
        hsync low
        vsync low

endmode

"fbset PAL" is all one needs to use it.

(I know i915 doesn't have interlaced mode support for analog VGA out, not a problem).

Signature · 1 line
Krzysztof Halasa
#8 lore raw
In reply to Jean Delvare
Quoted 13 lines · On Sun, 13 Dec 2009 12:50:13 +0100 Jean Delvare <[email protected]> wrote:
Quoted 4 lines · Jesse Barnes

Right, the logic is that the driver really is for embedded (i.e. very special purpose) use. It should not be selected unless you really know what you're doing or are building a very particular product.

The Kconfig help text doesn't say anything about this.

My understanding is that the intelfb driver was not _designed_ to be useful on embedded designs only. It just happens to be incomplete in such a way that it works only in a few selected cases, which happen to be embedded cases, and it fails in many other cases.

Quoted 5 lines

The proper way to handle this is not to make the driver depend on EMBEDDED. The proper way would be to change the intelfb driver so that it no longer binds to devices it will not properly support. If the driver doesn't support LVDS (whatever it is) then it should cleanly fail on systems which have that.

Sorry my last message came across as a bit curt (I was writing on a phone and didn't want to type anymore :).

Making intelfb not bind to unsupported devices would require a good chunk of work; it would need to scan available outputs among other things.

Quoted 6 lines

The reason why I sent a patch in the first place is exactly opposite:
I want to let distros select this driver. My case is as follows: we had a product which included the intelfb driver, which we are in the process of upgrading. Now we find that the intelfb driver is gone (no longer selectable), which causes a problem as far as the upgrade path of our customers is concerned.

Hopefully you can migrate your product to the KMS based fb driver instead. It should provide the same functionality but with a better feature set (e.g. suspend/resume support).

Quoted 6 lines

So the problem I have to solve is: given a customer who was successfully using the intelfb driver before, what solutions can we offer when said customer upgrades to our new product? My own solution was straightforward: keep including the intelfb driver in the new product. Thus my patch dropping the dependency on EMBEDDED. If another solution exists, please let me know.

Enabling EMBEDDED for your distro would be another option; if you have a very specific product in mind and want to preserve the same features and bugs, then that might be the best route in this particular case.

Quoted 7 lines

It would help if the DRM option description was updated. It still
reads: "Direct Rendering Manager (XFree86 4.1.0 and higher DRI support)". If the DRM core is now also providing support for framebuffer-like functionality (again, if I understand correctly) then the reference to XFree86 should be dropped. The help text should also be updated to properly describe all that the DRM core offers today.

Yeah, we do need better help text. People still get confused about i830 and i810 as well...

Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center