Skip to contentSkip to conversation

Clevo M5X0JE hangs in ACPI init

Bug 41722 Rogério Brito 9 participants 53 messages Jan 7, 2012 – Feb 7, 2012 2 messages not in the archive

Subsystems (2)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]/

A message that is not in the archive (sent off-list or not yet received): [email protected]

#2 lore raw

Hi, Bjorn and others,

Now that bugzilla is up, I think that I can resume testing things on that problematic notebook that I mentioned in earlier e-mails, so that mom (in CC) can use her notebook with Linux.

For the record, *some* description of the problem is at:

* https://bugzilla.kernel.org/show_bug.cgi?id=41132
* https://bugzilla.kernel.org/show_bug.cgi?id=41722

The first one is a regression that I bisected, while the second one seems to be a bit harder to understand for a layman like me.

Of couse, I can summarize what I found and what I didn't.

Sorry for the top-posting, but it is here to keep the context of the previous message.

Thanks,
Rogério Brito.

Quoted 37 lines · On Tue, Sep 6, 2011 at 19:22, <[email protected]> wrote:

--- Comment #15 from Rogério Brito <[email protected]>  2011-09-06 22:22:44 --- Dear Bjorn,

Sorry for not responding promptly but I had some family members with health problems. Things are settling right now and I think that I can be faster now.

(In reply to comment #14)

Quoted 2 lines

So it looks like 2.4.27 doesn't even try to enable ACPI mode, and 2.6.8 hangs the same way as 3.1.0-rc3.

Yes, exactly.

Quoted 2 lines

Can you try 2.6.28?  If that doesn't boot, try the Ubuntu 9.04 kernel, which was reported to boot on your machine/BIOS.

Unfortunately, I tried to boot with that version of the Ubuntu live CD and it crashed the same way. I may try it once again, paying more attention if it behaves differently in any kind of way, though.

Oh, talking about PCI and ACPI failing, could you take a look at issue #41132 to see if you can help with another machine that stopped working after a commit that I bisected?

Thanks,

Rogério Brito.

--
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#3 lore raw

2012/1/6 Rogério Brito <[email protected]>:

Quoted 4 lines

For the record, *some* description of the problem is at:

* https://bugzilla.kernel.org/show_bug.cgi?id=41132

This one looks very much like the thing that Edward Donovan fixed fairly recently in commit 52553ddffad7 ("genirq: fix regression in irqfixup, irqpoll")

So that should be fixed in 3.2 (and it's marked for stable, so I think it's in the latest stable kernels too)

Quoted 1 line

* https://bugzilla.kernel.org/show_bug.cgi?id=41722

That, however, looks totally insane. I don't even have a clue where to begin. I take it that it works with "acpi=off", and don't have a clue about what might go wrong with ACPI enabled.

Linus

#4 lore raw
Quoted 4 lines · Linus Torvalds

2012/1/6 Rogério Brito <[email protected]>:

Quoted 3 lines

For the record, *some* description of the problem is at:

Quoted 8 lines
Quoted 1 line

* https://bugzilla.kernel.org/show_bug.cgi?id=41132

This one looks very much like the thing that Edward Donovan fixed fairly recently in commit 52553ddffad7 ("genirq: fix regression in irqfixup, irqpoll")

So that should be fixed in 3.2 (and it's marked for stable, so I think it's in the latest stable kernels too)

Yessir, as of 3.0.13 and 3.1.5. Rogério - for all the users I've heard from, the 2.6.39 IRQ regressions are fixed now. I'll be interested to know if any IRQ handled by 2.6.38 isn't handled now, but I expect it will be. I'm sorry if that isn't enough for this laptop, though -

Ed

#5 lore raw

Hi, Edward.

Quoted 3 lines · Edward Donovan

I'll be interested to know if any IRQ handled by 2.6.38 isn't handled now, but I expect it will be. I'm sorry if that isn't enough for this laptop, though -

I bisected the PCI stuff and ended up with a bad commit 12c22d6ef299ccf0955, which was made by Linus:

,----[ git show 12c22d6ef299ccf0955e5756eb57d90d7577ac68 ]

| commit 12c22d6ef299ccf0955e5756eb57d90d7577ac68
| Author: Linus Torvalds <[email protected]>
| Date:   Wed Mar 26 11:22:40 2008 -0700
| 
|     Revert "PCI: remove transparent bridge sizing"
| (...)    

`----

The patch that I am using to make this laptop boot at least is essentially a revert of that commit.

Just for the sake of documentation, I am attaching here the patch that makes things work (at least partially).

Any directions to debug this properly are welcome.

Thanks,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
drivers/pci/setup-bus.c +4 -0
index 86b69f85..84543f5 100644
@@ -849,6 +849,10 @@ void __ref __pci_bus_size_bridges(struct pci_bus *bus,
 		break;
 
 	case PCI_CLASS_BRIDGE_PCI:
+		/* don't size subtractive decoding (transparent)
+		 * PCI-to-PCI bridges */
+		if (bus->self->transparent)
+			break;
 		pci_bridge_check_ranges(bus);
 		if (bus->self->is_hotplug_bridge) {
 			additional_io_size  = pci_hotplug_io_size;
maintainer
#6 lore raw
Quoted 10 lines · Rogério Brito

I bisected the PCI stuff and ended up with a bad commit 12c22d6ef299ccf0955, which was made by Linus:

,----[ git show 12c22d6ef299ccf0955e5756eb57d90d7577ac68 ]

| commit 12c22d6ef299ccf0955e5756eb57d90d7577ac68
| Author: Linus Torvalds <[email protected]>
| Date:   Wed Mar 26 11:22:40 2008 -0700
|
|     Revert "PCI: remove transparent bridge sizing"
| (...)
Quoted 7 lines

`----

The patch that I am using to make this laptop boot at least is essentially a revert of that commit.

Just for the sake of documentation, I am attaching here the patch that makes things work (at least partially).

Your patch avoids sizing transparent bridges. I don't have an opinion on whether that's a good idea in general.

However, I am curious about what breaks on your system when we do size the transparent bridge. It seems like sizing it should *work*, even if it's not strictly necessary, so I wonder if this hang is telling us about some other PCI allocation issue we should fix. Would you mind opening a bugzilla for this, since the current tree is broken for you, and attaching a log or digital photo of the hang?

Bjorn

#7 lore raw

Hi, Bjorn and other people.

2012/1/9 Bjorn Helgaas <[email protected]>:

Quoted 4 lines

However, I am curious about what breaks on your system when we do size the transparent bridge.  It seems like sizing it should *work*, even if it's not strictly necessary, so I wonder if this hang is telling us about some other PCI allocation issue we should fix.

Makes sense.

Quoted 2 lines

Would you mind opening a bugzilla for this, since the current tree is broken for you, and attaching a log or digital photo of the hang?

I have already opened a bugzilla entry for this at:

    https://bugzilla.kernel.org/show_bug.cgi?id=41622

It contains some information already (see the attachments), but I can always post more, of course. Just ask and I will try my best. (I will try to post some pictures as soon as I can get back my bugzilla password to change that bug).

Thanks,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#8 lore raw
In reply to Linus Torvalds

Hi there, Linus.

Quoted 9 lines · Linus Torvalds

2012/1/6 Rogério Brito <[email protected]>:

Quoted 4 lines

For the record, *some* description of the problem is at:

* https://bugzilla.kernel.org/show_bug.cgi?id=41132

This one looks very much like the thing that Edward Donovan fixed fairly recently in commit 52553ddffad7 ("genirq: fix regression in irqfixup, irqpoll")

It does look similar, but not quite the same. And, yes, my main desktop (the one that I'm using right now) was affected by Edward's commit, which, BTW, has my Reported-and-tested-by.

This one is different. Here is a brief summary of the situation:

# Computer description

* Rebranded Clevo M5X0JE
* nForce2 chipset
* AMD Sempron CPU
* NIC with driver forcedeth
* wifi rtl8187

# Booting options

The current working options passed to the kernel are: acpi=off pnpbios=off noapic

It doesn't boot with a vanilla kernel. It only boots when I pass the options above *and* compile the kernel with the following patch applied:

,----[ do_not_size_subtractive_decoding_transparent_pci_to_pci_bridges.patch ]

| diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
| index 86b69f85..84543f5 100644
| --- a/drivers/pci/setup-bus.c
| +++ b/drivers/pci/setup-bus.c
| @@ -849,6 +849,10 @@ void __ref __pci_bus_size_bridges(struct pci_bus *bus,
|  		break;
|  
|  	case PCI_CLASS_BRIDGE_PCI:
| +		/* don't size subtractive decoding (transparent)
| +		 * PCI-to-PCI bridges */
| +		if (bus->self->transparent)
| +			break;
|  		pci_bridge_check_ranges(bus);
|  		if (bus->self->is_hotplug_bridge) {
|  			additional_io_size  = pci_hotplug_io_size;

`----

Otherwise, it hangs right when trying to configure PCI, which, according to the dmesg log when it boots occurs approximately 0.12 seconds after the kernel shows its first message.

Quoted 8 lines

So that should be fixed in 3.2 (and it's marked for stable, so I think it's in the latest stable kernels too)

Quoted 1 line

* https://bugzilla.kernel.org/show_bug.cgi?id=41722

That, however, looks totally insane. I don't even have a clue where to begin. I take it that it works with "acpi=off", and don't have a clue about what might go wrong with ACPI enabled.

Yes, with acpi=off it "works" (well, it seems that without ACPI many things don't get enabled). At least mom was able to visit a site like Gmail to check her email.

She would like say that she is positively impressed with the committment of the community with a case of a layman person and she told me to send you all big thank you.

Thanks from me also,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#9 lore raw
Quoted 20 lines · Rogério Brito

It doesn't boot with a vanilla kernel. It only boots when I pass the options above *and* compile the kernel with the following patch applied:

,----[ do_not_size_subtractive_decoding_transparent_pci_to_pci_bridges.patch ]

| diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
| index 86b69f85..84543f5 100644
| --- a/drivers/pci/setup-bus.c
| +++ b/drivers/pci/setup-bus.c
| @@ -849,6 +849,10 @@ void __ref __pci_bus_size_bridges(struct pci_bus *bus,
|               break;
|
|       case PCI_CLASS_BRIDGE_PCI:
| +             /* don't size subtractive decoding (transparent)
| +              * PCI-to-PCI bridges */
| +             if (bus->self->transparent)
| +                     break;
|               pci_bridge_check_ranges(bus);
|               if (bus->self->is_hotplug_bridge) {
|                       additional_io_size  = pci_hotplug_io_size;
Quoted 1 line

`----

Ahh. I'd forgotten about that particular PCI patch.

That is definitely the right thing to do, and commit 8fa5913d54f3 ("PCI: remove transparent bridge sizing") did exactly that, but then we reverted it in commit 12c22d6ef299 because it caused some odd problems for some people.

I think we should try to re-do that "avoid sizing transparent bridges" commit, because it really should make it much easier to do PCI allocations under some very common situations (there's a *lot* of common intel transparent PCI bridges).

Jesse, Yinghai, Ram - should we try to just re-do that commit in this merge window, and see how that goes? We can always revert it again if it causes problems..

Linus

#10 lore raw
Quoted 30 lines · On Sun, 8 Jan 2012 14:23:29 -0800 Linus Torvalds…
Quoted 20 lines · Rogério Brito

It doesn't boot with a vanilla kernel. It only boots when I pass the options above *and* compile the kernel with the following patch applied:

,----[ do_not_size_subtractive_decoding_transparent_pci_to_pci_bridges.patch ]

| diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
| index 86b69f85..84543f5 100644
| --- a/drivers/pci/setup-bus.c
| +++ b/drivers/pci/setup-bus.c
| @@ -849,6 +849,10 @@ void __ref __pci_bus_size_bridges(struct pci_bus *bus,
|               break;
|
|       case PCI_CLASS_BRIDGE_PCI:
| +             /* don't size subtractive decoding (transparent)
| +              * PCI-to-PCI bridges */
| +             if (bus->self->transparent)
| +                     break;
|               pci_bridge_check_ranges(bus);
|               if (bus->self->is_hotplug_bridge) {
|                       additional_io_size  = pci_hotplug_io_size;
Quoted 1 line

`----

Ahh. I'd forgotten about that particular PCI patch.

That is definitely the right thing to do, and commit 8fa5913d54f3 ("PCI: remove transparent bridge sizing") did exactly that, but then we reverted it in commit 12c22d6ef299 because it caused some odd problems for some people.

Quoted 8 lines

I think we should try to re-do that "avoid sizing transparent bridges" commit, because it really should make it much easier to do PCI allocations under some very common situations (there's a *lot* of common intel transparent PCI bridges).

Jesse, Yinghai, Ram - should we try to just re-do that commit in this merge window, and see how that goes? We can always revert it again if it causes problems..

I don't remember what problems we hit, but if Ram and Yinghai are willing to take a look at them we should go ahead and try again.

Thanks,

Signature · 1 line
Jesse Barnes, Intel Open Source Technology Center
#11 lore raw
Quoted 3 lines · Jesse Barnes

I don't remember what problems we hit, but if Ram and Yinghai are willing to take a look at them we should go ahead and try again.

So the two bugzilla entries quoted in the revert are

http://bugzilla.kernel.org/show_bug.cgi?id=10080
http://bugzilla.kernel.org/show_bug.cgi?id=9961

with at least one real smoking gun in this one:

https://lkml.org/lkml/2008/3/26/94

however at least *part* of the problem for some people was not the transparency itself, as much as the fact that we did bad things with 64-bit resources etc.

So what I think mostly happened was that not sizing up transparent bridges ended up showing up other bugs. We've fixed at least some of those other bugs in the meantime (things like using "unsigned long" for physical addresses in ioremap etc, which broke when 64-bit resources were used on 32-bit architectures).

But it's entirely possible that it will still trigger similar issues. For example, even if a bridge is transparent, maybe it is still limited to 32-bit addresses? If we look at the actual IO windows, we'd get that 32-bit limit right automatically. If we just say "it's transparent", we might allocate child devices with 64-bit resources above the 4GB area, and be screwed. That seems to have been one problem above - even if we now get it right on a software level, there may actually be hardware issues in the same area.

(It's not clear whether the 64-bit resource issue back in 2008 was due to hardware or due to our resource bugs).

Adding Ivan to the cc due to historical issues - although I haven't seen him in email for a year, so he may be gone.

Linus

#12 lore raw
Quoted 10 lines · Linus Torvalds
Quoted 3 lines · Jesse Barnes

I don't remember what problems we hit, but if Ram and Yinghai are willing to take a look at them we should go ahead and try again.

So the two bugzilla entries quoted in the revert are

 http://bugzilla.kernel.org/show_bug.cgi?id=10080
 http://bugzilla.kernel.org/show_bug.cgi?id=9961

Quoted 7 lines

 with at least one real smoking gun in this one:

 https://lkml.org/lkml/2008/3/26/94

however at least *part* of the problem for some people was not the transparency itself, as much as the fact that we did bad things with 64-bit resources etc.

2 cases:
a. transparent bridge has resource allocated from BIOS. b. transparent bridge does not have resource allocated from BIOS.

skipping transparent bridge size checking only affects case b.

when skip code is there, those bridge register will not be probed ( with 0xfff0fff0),
and those bridge bar will not get allocated. and child devices that does not get allocated from
BIOS, kernel will allocate from parent bus resources... up to peer root bus.

Rogério's laptop does not like bus resize, could be because the bridge does not like kernel to use
0xfff0fff0 to probe it...

We could add one quirks etc to skip this kind of probe.

Quoted 7 lines

So what I think mostly happened was that not sizing up transparent bridges ended up showing up other bugs. We've fixed at least some of those other bugs in the meantime (things like using "unsigned long" for physical addresses in ioremap etc, which broke when 64-bit resources were used on 32-bit architectures).

Quoted 8 lines

But it's entirely possible that it will still trigger similar issues. For example, even if a bridge is transparent, maybe it is still limited to 32-bit addresses? If we look at the actual IO windows, we'd get that 32-bit limit right automatically. If we just say "it's transparent", we might allocate child devices with 64-bit resources above the 4GB area, and be screwed. That seems to have been one problem above - even if we now get it right on a software level, there may actually be hardware issues in the same area.

in pci_bus_alloc_resource(), we already have

        /* don't allocate too high if the pref mem doesn't support 64bit*/
        if (!(res->flags & IORESOURCE_MEM_64))
                max = PCIBIOS_MAX_MEM_32;

So it should allocate to range below 4G to unassigned children devices.

but need to make sure (peer) root bus does have valid resource range at first.

Rogério,

Do you have bootlog with "debug ignore_loglevel pci=earlydump" ?

Thanks

Yinghai

#13 lore raw

Dear Yinghai,

Quoted 6 lines · Yinghai Lu

in pci_bus_alloc_resource(),  we already have

       /* don't allocate too high if the pref mem doesn't support 64bit*/ if (!(res->flags & IORESOURCE_MEM_64))
               max = PCIBIOS_MAX_MEM_32;

Quoted 7 lines

So it should allocate to range below 4G to unassigned children devices.

but need to make sure (peer) root bus does have valid resource range at first.

Rogério,

Do you have bootlog with "debug ignore_loglevel pci=earlydump" ?

I have posted some information to bugzilla (see [0]), but not with pci=earlydump. It seems that bugzilla doesn't want to mail me a reset e-mail, but while I am waiting, I posted this on my homepage (see [1]).

Please, note that this boot log was taken with the patch that I e-mailed earlier (i.e., the one that makes things work). If you want me to send a log with a vanilla kernel, just let me know and I will grab more information.

Thanks,
Rogério.

[0]: https://bugzilla.kernel.org/show_bug.cgi?id=41622
[1]: http://www.ime.usp.br/~rbrito/linux/clevo/clevo-dmesg-with-earlydump-2012-01-10.txt

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#14 lore raw
Quoted 5 lines · Rogério Brito

I have posted some information to bugzilla (see [0]), but not with pci=earlydump. It seems that bugzilla doesn't want to mail me a reset e-mail, but while I am waiting, I posted this on my homepage (see [1]).

cool.
your system have bridge 00:10.0, and under it there is 05:07.0 and it is one card bus bridge

for pci bridge, BIOS only allocate small range to mmio, and does not allocate mmio_pref to it.

looks like bridge 00:10.0 does not like kernel probe its mmio pref register.

with the transparent bridge skip patch, kernel will not touch mmio pref register,

and cardbus bridge will get allocation from root bus resource range...

[    0.162791] pci 0000:05:07.0: BAR 10: assigned [mem 0x84000000-0x87ffffff]
[    0.162863] pci 0000:05:07.0: BAR 9: assigned [mem
0x88000000-0x8bffffff pref]
[    0.162946] pci 0000:05:07.0: BAR 8: assigned [io  0x1000-0x10ff]
[    0.163016] pci 0000:05:07.0: BAR 7: assigned [io  0x1400-0x14ff]
[    0.163083] pci 0000:05:07.0: CardBus bridge to [bus 06-09]
[    0.163150] pci 0000:05:07.0:   bridge window [io  0x1400-0x14ff]
[    0.163220] pci 0000:05:07.0:   bridge window [io  0x1000-0x10ff]
[    0.163290] pci 0000:05:07.0:   bridge window [mem
0x88000000-0x8bffffff pref]
[    0.163331] pci 0000:05:07.0:   bridge window [mem 0x84000000-0x87ffffff]
[    0.163402] pci 0000:00:10.0: PCI bridge to [bus 05-06]
[    0.163471] pci 0000:00:10.0:   bridge window [mem 0xb3200000-0xb32fffff]

I have some pending patches that may fix bridge resource resizing.

Can you try them at

git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git
for-pci2

also please boot with "debug ignore_loglevel"

BTW, did you try boot with "pci=use_crs" and not with "acpi=off" ?

Thanks

Yinghai

#15 lore raw

Hi, Yinghai.

Quoted 9 lines · Yinghai Lu

On Mon, Jan 9, 2012 at 6:41 PM, Rogério Brito <[email protected]> wrote:
I have some pending patches that may fix bridge resource resizing.

Can you try them at

     git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git
for-pci2

also please boot with "debug ignore_loglevel"

Grabbing it right now. Will report it after I get some sleep (5AM here).

Quoted 1 line

BTW, did you try boot with "pci=use_crs" and not with "acpi=off" ?

I tried, but I got a hang (both with Linux---many versions---and with FreeBSD). With Linux I get:

http://www.ime.usp.br/~rbrito/linux/clevo/boot-without-acpi-off.jpg

Suggestions?

Thanks,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#16 lore raw
Quoted 6 lines · Rogério Brito

Hi, Yinghai.

Quoted 3 lines · Yinghai Lu

On Mon, Jan 9, 2012 at 6:41 PM, Rogério Brito <[email protected]> wrote:
I have some pending patches that may fix bridge resource resizing.

Quoted 8 lines
Quoted 6 lines

Can you try them at

     git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git
for-pci2

also please boot with "debug ignore_loglevel"

Grabbing it right now. Will report it after I get some sleep (5AM here).

please try attached patch, it will fix one bios wrong setting.

Quoted 8 lines
Quoted 2 lines

BTW, did you try boot with "pci=use_crs" and not with "acpi=off" ?

I tried, but I got a hang (both with Linux---many versions---and with FreeBSD). With Linux I get:

  http://www.ime.usp.br/~rbrito/linux/clevo/boot-without-acpi-off.jpg

need to work on that later.

Yinghai

Subject: [PATCH] PCI: Disable cardbus bridge MEM1 pref CTL

Some BIOS enable both pref for MEM0 and MEM1.

but we assume MEM1 is non-pref...

---
 drivers/pci/setup-bus.c |    8 ++++++++
 1 file changed, 8 insertions(+)
drivers/pci/setup-bus.c +8 -0
Index: linux-2.6/drivers/pci/setup-bus.c
===================================================================
@@ -774,6 +774,14 @@ static void pci_bus_size_cardbus(struct
 	if (realloc_head)
 		add_to_list(realloc_head, bridge, b_res+1, pci_cardbus_io_size, 0 /* dont care */);
 
+	/* MEM1 must not be pref mmio */
+	pci_read_config_word(bridge, PCI_CB_BRIDGE_CONTROL, &ctrl);
+	if (ctrl & PCI_CB_BRIDGE_CTL_PREFETCH_MEM1) {
+		ctrl &= ~PCI_CB_BRIDGE_CTL_PREFETCH_MEM1;
+		pci_write_config_word(bridge, PCI_CB_BRIDGE_CONTROL, ctrl);
+		pci_read_config_word(bridge, PCI_CB_BRIDGE_CONTROL, &ctrl);
+	}
+
 	/*
 	 * Check whether prefetchable memory is supported
 	 * by this bridge.
#17 lore raw

Dear Yinghai,

Quoted 10 lines · Yinghai Lu
Quoted 7 lines · Rogério Brito
Quoted 4 lines · Yinghai Lu

git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git
for-pci2

also please boot with "debug ignore_loglevel"

Grabbing it right now. Will report it after I get some sleep (5AM here).

please try attached patch, it will fix one bios wrong setting.

Should I use this patch agains which tree?

Right now, I have Linus's tree, yours (cited above) and I am, up to now, I had been compiling everything from Linus's v3.2 tag: both a vanilla kernel, as asked by Bjorn and with the patch that I mentioned to Linus ("the one that makes things boot").

If it applies (and newer kernels don't have things which your patch needs), I will keep working with Linus's v3.2 tag.

When I compiled from your branch for-pci2 from linux-yinghai.git (without any extra patches), it *did* boot, but noticed one thing weird: using htop and in single user mode, it reported 100% of CPU in use. With regular top, a more fine grained output revealed that no processes were using the CPU, but that almost all of the time was spent in kernel mode.

Anyway, as you asked, I put the log of what I see at

    http://www.ime.usp.br/~rbrito/linux/clevo/2012-01-13-0951-yanghai-for-pci2-branch/
Quoted 9 lines
Quoted 7 lines
Quoted 1 line

BTW, did you try boot with "pci=use_crs" and not with "acpi=off" ?

I tried, but I got a hang (both with Linux---many versions---and with FreeBSD). With Linux I get:

http://www.ime.usp.br/~rbrito/linux/clevo/boot-without-acpi-off.jpg

need to work on that later.

Thanks for your (everybody else's) kindness/helpfulness with these multiple issues.

Thanks,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#18 lore raw
Quoted 16 lines · Rogério Brito

Dear Yinghai,

Quoted 10 lines · Yinghai Lu
Quoted 7 lines · Rogério Brito
Quoted 4 lines · Yinghai Lu

  git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git
for-pci2

also please boot with "debug ignore_loglevel"

Grabbing it right now. Will report it after I get some sleep (5AM here).

please try attached patch, it will fix one bios wrong setting.

Should I use this patch agains which tree?

Quoted 7 lines

Right now, I have Linus's tree, yours (cited above) and I am, up to now, I had been compiling everything from Linus's v3.2 tag: both a vanilla kernel, as asked by Bjorn and with the patch that I mentioned to Linus ("the one that makes things boot").

If it applies (and newer kernels don't have things which your patch needs), I will keep working with Linus's v3.2 tag.

No problem, I will put that patch into for-pci2 for your conveniences.

Quoted 10 lines

When I compiled from your branch for-pci2 from linux-yinghai.git (without any extra patches), it *did* boot, but noticed one thing weird: using htop and in single user mode, it reported 100% of CPU in use. With regular top, a more fine grained output revealed that no processes were using the CPU, but that almost all of the time was spent in kernel mode.

Anyway, as you asked, I put the log of what I see at

   http://www.ime.usp.br/~rbrito/linux/clevo/2012-01-13-0951-yanghai-for-pci2-branch/

[    0.165615] pci 0000:05:07.0: CardBus bridge to [bus 06-09]
[    0.165682] pci 0000:05:07.0:   bridge window [io  0x0001-0x0000]
[    0.165751] pci 0000:05:07.0:   bridge window [io  0x0001-0x0000]
[    0.165821] pci 0000:05:07.0:   bridge window [mem
0x00000001-0x00000000 pref]
[    0.165904] pci 0000:05:07.0:   bridge window [mem 0x00000001-0x00000000]
[    0.165975] pci 0000:00:10.0: PCI bridge to [bus 05-06]
[    0.166043] pci 0000:00:10.0:   bridge window [mem 0xb3200000-0xb32fffff]
[    0.166259] pci 0000:00:10.0: setting latency timer to 64
[    0.166329] pci 0000:05:07.0: device not available (can't reserve
[io  0x0001-0x0000])
[    0.166412] pci 0000:05:07.0: Error enabling bridge (-22), continuing

We have some problems with handling cardbus bridge as optional resources.

will produce some patches and update for-pci2 branch for your test.

Thanks

Yinghai

#19 lore raw
Quoted 10 lines · Yinghai Lu
Quoted 7 lines · Rogério Brito

Right now, I have Linus's tree, yours (cited above) and I am, up to now, I had been compiling everything from Linus's v3.2 tag: both a vanilla kernel, as asked by Bjorn and with the patch that I mentioned to Linus ("the one that makes things boot").

If it applies (and newer kernels don't have things which your patch needs), I will keep working with Linus's v3.2 tag.

No problem, I will put that patch into for-pci2 for your conveniences.

please check for-pci2 again, i put two cardbus related fixes there.

or you can apply attached two patches.

Thanks

Yinghai

Subject: [PATCH] PCI: Disable cardbus bridge MEM1 pref CTL

Some BIOS enable both pref for MEM0 and MEM1.

but we assume MEM1 is non-pref...

---
 drivers/pci/setup-bus.c |    8 ++++++++
 1 file changed, 8 insertions(+)
drivers/pci/setup-bus.c +8 -0
Index: linux-2.6/drivers/pci/setup-bus.c
===================================================================
@@ -774,6 +774,14 @@ static void pci_bus_size_cardbus(struct
 	if (realloc_head)
 		add_to_list(realloc_head, bridge, b_res+1, pci_cardbus_io_size, 0 /* dont care */);
 
+	/* MEM1 must not be pref mmio */
+	pci_read_config_word(bridge, PCI_CB_BRIDGE_CONTROL, &ctrl);
+	if (ctrl & PCI_CB_BRIDGE_CTL_PREFETCH_MEM1) {
+		ctrl &= ~PCI_CB_BRIDGE_CTL_PREFETCH_MEM1;
+		pci_write_config_word(bridge, PCI_CB_BRIDGE_CONTROL, ctrl);
+		pci_read_config_word(bridge, PCI_CB_BRIDGE_CONTROL, &ctrl);
+	}
+
 	/*
 	 * Check whether prefetchable memory is supported
 	 * by this bridge.

Subject: [PATCH] PCI: Fix cardbus bridge resources as optional size handling

We should not set the requested size to -2.

that will confuse the resource list sorting with align when SIZEALIGN is used.

Change to STARTALIGN and pass align from start.

We are safe to do that just as we do that regular pci bridge.

In the long run, we should just treat cardbus like regular pci bridge.

Also fix when realloc is not passed, should keep the requested size.

---
 drivers/pci/setup-bus.c |   71 +++++++++++++++++++++++++++---------------------
 1 file changed, 40 insertions(+), 31 deletions(-)
drivers/pci/setup-bus.c +40 -31

89 lines not on this page · raw message

#20 lore raw
In reply to Yinghai Lu

Hi, Yinghai.

Quoted 5 lines · Yinghai Lu
Quoted 2 lines · Rogério Brito

If it applies (and newer kernels don't have things which your patch needs), I will keep working with Linus's v3.2 tag.

No problem, I will put that patch into for-pci2 for your conveniences.

Thanks. That's definitely easier for me. I'm grabbing it to compile and will report in a moment.

Quoted 23 lines
Quoted 9 lines

When I compiled from your branch for-pci2 from linux-yinghai.git (without any extra patches), it *did* boot, but noticed one thing weird: using htop and in single user mode, it reported 100% of CPU in use. With regular top, a more fine grained output revealed that no processes were using the CPU, but that almost all of the time was spent in kernel mode.

Anyway, as you asked, I put the log of what I see at

   http://www.ime.usp.br/~rbrito/linux/clevo/2012-01-13-0951-yanghai-for-pci2-branch/

[    0.165615] pci 0000:05:07.0: CardBus bridge to [bus 06-09]
[    0.165682] pci 0000:05:07.0:   bridge window [io  0x0001-0x0000] [    0.165751] pci 0000:05:07.0:   bridge window [io  0x0001-0x0000] [    0.165821] pci 0000:05:07.0:   bridge window [mem 0x00000001-0x00000000 pref]
[    0.165904] pci 0000:05:07.0:   bridge window [mem 0x00000001-0x00000000] [    0.165975] pci 0000:00:10.0: PCI bridge to [bus 05-06]
[    0.166043] pci 0000:00:10.0:   bridge window [mem 0xb3200000-0xb32fffff] [    0.166259] pci 0000:00:10.0: setting latency timer to 64
[    0.166329] pci 0000:05:07.0: device not available (can't reserve [io  0x0001-0x0000])
[    0.166412] pci 0000:05:07.0: Error enabling bridge (-22), continuing

Quoted 3 lines

We have some problems with handling cardbus bridge as optional resources.

will produce some patches and update for-pci2 branch for your test.

OK. Will try that as soon as my download completes.

Thanks again,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#21 lore raw

Dear Yinghai and other people,

Quoted 9 lines · Rogério Brito
Quoted 5 lines · Yinghai Lu
Quoted 2 lines · Rogério Brito

If it applies (and newer kernels don't have things which your patch needs), I will keep working with Linus's v3.2 tag.

No problem, I will put that patch into for-pci2 for your conveniences.

Thanks. That's definitely easier for me. I'm grabbing it to compile and will report in a moment.

Just to clarify, my "definitely easier for me" means "I know what to do (or, at least, I'm closer this way not doing wrong things)".

Quoted 23 lines
Quoted 23 lines
Quoted 9 lines

When I compiled from your branch for-pci2 from linux-yinghai.git (without any extra patches), it *did* boot, but noticed one thing weird: using htop and in single user mode, it reported 100% of CPU in use. With regular top, a more fine grained output revealed that no processes were using the CPU, but that almost all of the time was spent in kernel mode.

Anyway, as you asked, I put the log of what I see at

   http://www.ime.usp.br/~rbrito/linux/clevo/2012-01-13-0951-yanghai-for-pci2-branch/

[    0.165615] pci 0000:05:07.0: CardBus bridge to [bus 06-09]
[    0.165682] pci 0000:05:07.0:   bridge window [io  0x0001-0x0000] [    0.165751] pci 0000:05:07.0:   bridge window [io  0x0001-0x0000] [    0.165821] pci 0000:05:07.0:   bridge window [mem 0x00000001-0x00000000 pref]
[    0.165904] pci 0000:05:07.0:   bridge window [mem 0x00000001-0x00000000] [    0.165975] pci 0000:00:10.0: PCI bridge to [bus 05-06]
[    0.166043] pci 0000:00:10.0:   bridge window [mem 0xb3200000-0xb32fffff] [    0.166259] pci 0000:00:10.0: setting latency timer to 64
[    0.166329] pci 0000:05:07.0: device not available (can't reserve [io  0x0001-0x0000])
[    0.166412] pci 0000:05:07.0: Error enabling bridge (-22), continuing

Quoted 5 lines
Quoted 3 lines

We have some problems with handling cardbus bridge as optional resources.

will produce some patches and update for-pci2 branch for your test.

OK. Will try that as soon as my download completes.

Just for the record, I have grabbed your for-pci2 branch with HEAD being 5db4211 and this is what I get:

    http://www.ime.usp.br/~rbrito/linux/clevo/boot-yinghai-for-pci2-g5db4211.m4v

A warning, though is that there were some options regarding PCI buses which I did not know exactly how to answer. I can post my config here if you are interested in that. Well, actually, I can give you ssh access to this notebook if you want to hack on it. Just let me know what times you (or anyone else) are online and I will try to arrange everything.

Any guidance is welcome.

Thanks,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#22 lore raw
Quoted 48 lines · Rogério Brito

Dear Yinghai and other people,

Quoted 9 lines · Rogério Brito
Quoted 5 lines · Yinghai Lu
Quoted 2 lines · Rogério Brito

If it applies (and newer kernels don't have things which your patch needs), I will keep working with Linus's v3.2 tag.

No problem, I will put that patch into for-pci2 for your conveniences.

Thanks. That's definitely easier for me. I'm grabbing it to compile and will report in a moment.

Just to clarify, my "definitely easier for me" means "I know what to do (or, at least, I'm closer this way not doing wrong things)".

Quoted 23 lines
Quoted 23 lines
Quoted 9 lines

When I compiled from your branch for-pci2 from linux-yinghai.git (without any extra patches), it *did* boot, but noticed one thing weird: using htop and in single user mode, it reported 100% of CPU in use. With regular top, a more fine grained output revealed that no processes were using the CPU, but that almost all of the time was spent in kernel mode.

Anyway, as you asked, I put the log of what I see at

   http://www.ime.usp.br/~rbrito/linux/clevo/2012-01-13-0951-yanghai-for-pci2-branch/

[    0.165615] pci 0000:05:07.0: CardBus bridge to [bus 06-09]
[    0.165682] pci 0000:05:07.0:   bridge window [io  0x0001-0x0000] [    0.165751] pci 0000:05:07.0:   bridge window [io  0x0001-0x0000] [    0.165821] pci 0000:05:07.0:   bridge window [mem 0x00000001-0x00000000 pref]
[    0.165904] pci 0000:05:07.0:   bridge window [mem 0x00000001-0x00000000] [    0.165975] pci 0000:00:10.0: PCI bridge to [bus 05-06]
[    0.166043] pci 0000:00:10.0:   bridge window [mem 0xb3200000-0xb32fffff] [    0.166259] pci 0000:00:10.0: setting latency timer to 64
[    0.166329] pci 0000:05:07.0: device not available (can't reserve [io  0x0001-0x0000])
[    0.166412] pci 0000:05:07.0: Error enabling bridge (-22), continuing

Quoted 5 lines
Quoted 3 lines

We have some problems with handling cardbus bridge as optional resources.

will produce some patches and update for-pci2 branch for your test.

OK. Will try that as soon as my download completes.

Just for the record, I have grabbed your for-pci2 branch with HEAD being 5db4211 and this is what I get:

Quoted 9 lines

   http://www.ime.usp.br/~rbrito/linux/clevo/boot-yinghai-for-pci2-g5db4211.m4v

A warning, though is that there were some options regarding PCI buses which I did not know exactly how to answer. I can post my config here if you are interested in that. Well, actually, I can give you ssh access to this notebook if you want to hack on it. Just let me know what times you (or anyone else) are online and I will try to arrange everything.

Please enable CONFIG_BOOT_PRINTK_DELAY in your .config

and boot with "boot_delay=1000" or even "boot_delay=5000"

then video capture could help.

Thanks

Yinghai

maintainer
#23 lore raw

Hi Rogério,

1) Your M5X0JE is horribly broken, but I couldn't find any similar reports on the web, so I wonder if there's a BIOS setting or something that's unique to your system. Can you (a) capture the current BIOS settings (photos of SETUP screens or something) and (b) reset to factory defaults and see if anything changes? If things work better with the defaults, knowing what the difference is might help us fix Linux.

2) Can you use a Windows tool like AIDA64 (free trial version at
http://www.aida64.com/downloads) to collect information about how Windows configures the box? Last time I used it, there was a way to save a full configuration report. Maybe there's a clue in the differences between Windows and Linux.

#24 lore raw
In reply to Yinghai Lu

Hi, people.

Sorry about the delay, but I have gathered a bunch of information about the machine.

For the record, I tried to put some order on the subdirectory at:

    http://www.ime.usp.br/~rbrito/linux/clevo/
Quoted 7 lines · Yinghai Lu
Quoted 6 lines · Rogério Brito

A warning, though is that there were some options regarding PCI buses which I did not know exactly how to answer. I can post my config here if you are interested in that. Well, actually, I can give you ssh access to this notebook if you want to hack on it. Just let me know what times you (or anyone else) are online and I will try to arrange everything.

Regarding these options, Yinghai, the options which I enabled once I typed "make oldconfig" were:

CONFIG_PCI_PRI
CONFIG_PCI_PASID

Quoted 5 lines

Please enable CONFIG_BOOT_PRINTK_DELAY in your .config

and boot with "boot_delay=1000" or even "boot_delay=5000"

then video capture could help.

I did just that and uploaded a video from Linus's git tree with the options above to:

    http://youtu.be/_dYhkWHfep0

It took a long time to boot (about 7 minutes), but all details are there, I hope. I can repeat the same steps for your for-pci2 branch, if you want me to.

Quoted 3 lines · Bjorn

1) Your M5X0JE is horribly broken, but I couldn't find any similar reports on the web, so I wonder if there's a BIOS setting or something that's unique to your system.

After searching, and searching, and searching, I could find this thread (in Portuguese) of some people with this computer, all of them having problems with multiple distributions (which is not a surprise, as the kernel has problems with these machines):

http://www.forumdebian.com.br/topico-notebook-amazon-pc-amz-a-101-resolvido

Just for the record, this Clevo was rebranded by a company called Amazon PC (the model is A101). Amazon PC has disappeared from Earth.

Quoted 5 lines

Can you (a) capture the current BIOS
settings (photos of SETUP screens or something) and (b) reset to factory defaults and see if anything changes? If things work better with the defaults, knowing what the difference is might help us fix Linux.

I took pictures of the few setup screens, and I reset everything to factory defaults, but still no luck. You can see the photos at:

http://www.ime.usp.br/~rbrito/linux/clevo/photos/

Quoted 3 lines

2) Can you use a Windows tool like AIDA64 (free trial version at
http://www.aida64.com/downloads) to collect information about how Windows configures the box?

Sure, I downloaded it an generated some reports. Two of them with this tool that you mentioned and one with lshw. They are at:

http://www.ime.usp.br/~rbrito/linux/clevo/hardware-description/

--- Comment #16 from Zhang Rui <[email protected]> 2012-01-18 05:39:52 --- hmm, what's the status of this bug?
does the problem still exist in the latest upstream kernel?

Yes, Zhang. Both bugs:

    https://bugzilla.kernel.org/show_bug.cgi?id=41622
    https://bugzilla.kernel.org/show_bug.cgi?id=41722

are still reproducible with the latest kernels. Please, don't close them.

OK. I'm going to bed right now, but I will be happy to provide any further information or test any patches/trees.

Thanks for all the help,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#25 lore raw

2012/1/18 Rogério Brito <[email protected]>:

Quoted 6 lines

Regarding these options, Yinghai, the options which I enabled once I typed "make oldconfig" were:

CONFIG_PCI_PRI
CONFIG_PCI_PASID

should be not related.

Quoted 7 lines
Quoted 6 lines

Please enable CONFIG_BOOT_PRINTK_DELAY in your .config

and boot with "boot_delay=1000" or even "boot_delay=5000"

then video capture could help.

Quoted 8 lines

I did just that and uploaded a video from Linus's git tree with the options above to:

   http://youtu.be/_dYhkWHfep0

It took a long time to boot (about 7 minutes), but all details are there, I hope.  I can repeat the same steps for your for-pci2 branch, if you want me to.

yes, please esp need patch.

http://git.kernel.org/?p=linux/kernel/git/yinghai/linux-yinghai.git;a=commitdiff;h=e3a7a7d4ea1b6a2be82ef454096aff7856c14de5

Thanks

Yinghai

#26 lore raw

Hi, Yinghai.

Quoted 8 lines · Yinghai Lu

2012/1/18 Rogério Brito <[email protected]>:

Quoted 5 lines

Regarding these options, Yinghai, the options which I enabled once I typed "make oldconfig" were:

CONFIG_PCI_PRI
CONFIG_PCI_PASID

should be not related.

Thanks. It would be good if, once all these problems were fixed, if the help of those options were fleshed out a little bit for lusers (e.g., me). Heck, I can even try to contribute some patches to document some options, once I understand them.

Quoted 5 lines
Quoted 5 lines

I did just that and uploaded a video from Linus's git tree with the options above to:

   http://youtu.be/_dYhkWHfep0

Quoted 7 lines
Quoted 3 lines

It took a long time to boot (about 7 minutes), but all details are there, I hope.  I can repeat the same steps for your for-pci2 branch, if you want me to.

yes, please esp need patch.

http://git.kernel.org/?p=linux/kernel/git/yinghai/linux-yinghai.git;a=commitdiff;h=e3a7a7d4ea1b6a2be82ef454096aff7856c14de5

With CONFIG_BOOT_PRINTK_DELAY enabled and with "boot_delay=500" on the command line, I checked out the for-pci2 branch of your tree and this is what I get:

    http://youtu.be/bWEme7iFD7g

If you want me to, say, recompile Linus's vanilla tree with only a few selected patches, please let me know. I will now try to test some other combinations of the options passed to the kernel.

Regards,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#27 lore raw

2012/1/19 Rogério Brito <[email protected]>:

Quoted 8 lines
Quoted 3 lines

yes, please esp need patch.

http://git.kernel.org/?p=linux/kernel/git/yinghai/linux-yinghai.git;a=commitdiff;h=e3a7a7d4ea1b6a2be82ef454096aff7856c14de5

With CONFIG_BOOT_PRINTK_DELAY enabled and with "boot_delay=500" on the command line, I checked out the for-pci2 branch of your tree and this is what I get:

Quoted 5 lines

   http://youtu.be/bWEme7iFD7g

If you want me to, say, recompile Linus's vanilla tree with only a few selected patches, please let me know. I will now try to test some other combinations of the options passed to the kernel.

resource allocating for transparent bridge resizing works well and as expected.

looks like quirk for usb hand off cause problem.

can you try another branch in my tree?

git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git
usb_smi_disable_early

it will do usb hand off very early.

you can do:

mkdir linux-2.6 || exit -1
cd linux-2.6

git init-db
# Add Linus's tree as a remote
git remote add linus
git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git

git remote add yinghai
git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git

git remote update

later will only do:

git remote update

git checkout -b rb_2012_01_19a linus/master
git merge yinghai/usb_smi_disable_early

Thanks

Yinghai

#28 lore raw

Hi, all.

Quoted 6 lines · Yinghai Lu

2012/1/19 Rogério Brito <[email protected]>:

Quoted 3 lines

If you want me to, say, recompile Linus's vanilla tree with only a few selected patches, please let me know. I will now try to test some other combinations of the options passed to the kernel.

resource allocating for transparent bridge resizing works well and as expected.

OK.

Quoted 6 lines

looks like quirk for usb hand off cause problem.

can you try another branch in my tree?

git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git
usb_smi_disable_early

Sure, will do that right now.

Quoted 15 lines

it will do usb hand off very early.

you can do:

mkdir linux-2.6 || exit -1
cd linux-2.6

git init-db
# Add Linus's tree as a remote
git remote add linus
git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6.git

git remote add yinghai
git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git

Quoted 8 lines

git remote update

later will only do:

git remote update

git checkout -b rb_2012_01_19a linus/master
git merge yinghai/usb_smi_disable_early

Will report ASAP. Thanks.

Regards,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#29 lore raw

Hi, all.

Quoted 7 lines · Rogério Brito
Quoted 6 lines · Yinghai Lu

looks like quirk for usb hand off cause problem.

can you try another branch in my tree?

git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git
usb_smi_disable_early

OK, now it seems that we are getting farther. The system boots, even mounts the root filesystem, udev starts and, then, when it seems that some USB messages are going to appear on the screen, the notebook crashes with the same distortion that happened before (with some of the last messages appearing being related to OHCI).

If needed, I can try to record what I see and upload to youtube. Do you want me to record what I see?

So, even though this is a partial progress, I am happier.

Thanks,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#30 lore raw

2012/1/19 Rogério Brito <[email protected]>:

Quoted 6 lines

OK, now it seems that we are getting farther. The system boots, even mounts the root filesystem, udev starts and, then, when it seems that some USB messages are going to appear on the screen, the notebook crashes with the same distortion that happened before (with some of the last messages appearing being related to OHCI).

in that case, Can you try to disable OHCI in .config

CONFIG_USB_OHCI_HCD

OHCI controller 00:0b.0 is using
Memory at b0004000 (32-bit, non-prefetchable) [size=4K]

VGA controller 00:05.0

	Memory at b2000000 (32-bit, non-prefetchable) [size=16M]
	Memory at c0000000 (64-bit, prefetchable) [size=256M]
	Memory at b1000000 (64-bit, non-prefetchable) [size=16M]
	[virtual] Expansion ROM at 80000000 [disabled] [size=128K]

it seems not related. or could be ioremap etc have problem?

So can you try to boot with nopat?

Thanks

Yinghai

#31 lore raw
Quoted 16 lines · Yinghai Lu

2012/1/19 Rogério Brito <[email protected]>:

Quoted 6 lines

OK, now it seems that we are getting farther. The system boots, even mounts the root filesystem, udev starts and, then, when it seems that some USB messages are going to appear on the screen, the notebook crashes with the same distortion that happened before (with some of the last messages appearing being related to OHCI).

in that case, Can you try to disable OHCI in .config

CONFIG_USB_OHCI_HCD

OHCI controller 00:0b.0 is using
Memory at b0004000 (32-bit, non-prefetchable) [size=4K]

VGA controller 00:05.0

Quoted 8 lines
	Memory at b2000000 (32-bit, non-prefetchable) [size=16M]
	Memory at c0000000 (64-bit, prefetchable) [size=256M]
	Memory at b1000000 (64-bit, non-prefetchable) [size=16M]
	[virtual] Expansion ROM at 80000000 [disabled] [size=128K]

it seems not related. or could be ioremap etc have problem?

So can you try to boot with nopat?

Yinghai/Bjorn,

After carefully examination of the allocations done by the vanilla upstream
kernel; the one without the 'skip transparent bridge while sizing' code, I find that the allocation to one of the Bar; BAR[16], of the CardBus Bridge at 05:07.0 does not overlap that of the parent transparent bridge at 00:10.0

Here is what I see in the video at
http://www.youtube.com/watch?v=_dYhkWHfep0&feature=youtu.be

[268.002554] pci 0000:05:07.0: Cardbus bridge to  [bus 06-09]
[269.002582] pci 0000:05:07.0:  bridge window [io  0x1400-0x14ff]
[270.009090] pci 0000:05:07.0:  bridge window [io  0x1000-0x10ff]
[271.012358] pci 0000:05:07.0:  bridge window [mem  0x84000000-0x87ffffff pref]
[272.015626] pci 0000:05:07.0:  bridge window [mem  0x88000000-0x8bffffff]
		                              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ offending allocation
[273.018894] pci 0000:00:10.0: PCI bridge to [bus  05-06]
[274.022162] pci 0000:00:10.0:  bridge window [io  0x1000-0x1fff]
[275.025430] pci 0000:00:10.0:  bridge window [mem  0xb3200000-0xb32fffff]
[276.028698] pci 0000:00:10.0:  bridge window [mem  0xb8400000-0x87ffffff pref]

To me, the issue seems to be that due to some reason the OS either fails to Or does not try to reallocate the addresses assigned to the the non-pref mem BAR of the the transparent bridge at 00:10.0.

In any case since you have deeper knowledge in this area, I will let you verify my theory. RP

Quoted 4 lines

Thanks

Yinghai

#32 lore raw

2012/1/21 Ram Pai <[email protected]>:

Quoted 49 lines · Yinghai Lu
Quoted 21 lines

2012/1/19 Rogério Brito <[email protected]>:

Quoted 6 lines

OK, now it seems that we are getting farther. The system boots, even mounts the root filesystem, udev starts and, then, when it seems that some USB messages are going to appear on the screen, the notebook crashes with the same distortion that happened before (with some of the last messages appearing being related to OHCI).

in that case, Can you try to disable OHCI in .config

CONFIG_USB_OHCI_HCD

OHCI controller 00:0b.0 is using
Memory at b0004000 (32-bit, non-prefetchable) [size=4K]

VGA controller 00:05.0
      Memory at b2000000 (32-bit, non-prefetchable) [size=16M] Memory at c0000000 (64-bit, prefetchable) [size=256M] Memory at b1000000 (64-bit, non-prefetchable) [size=16M] [virtual] Expansion ROM at 80000000 [disabled] [size=128K]

Quoted 3 lines

it seems not related. or could be ioremap etc have problem?

So can you try to boot with nopat?

Yinghai/Bjorn,

       After carefully examination of the allocations done by the vanilla upstream kernel; the one without the 'skip transparent bridge while sizing' code, I find that the allocation to one of the Bar; BAR[16], of the CardBus Bridge at 05:07.0 does not overlap that of the parent transparent bridge at 00:10.0

       Here is what I see in the video at
       http://www.youtube.com/watch?v=_dYhkWHfep0&feature=youtu.be

[268.002554] pci 0000:05:07.0: Cardbus bridge to  [bus 06-09]
[269.002582] pci 0000:05:07.0:  bridge window [io  0x1400-0x14ff]
[270.009090] pci 0000:05:07.0:  bridge window [io  0x1000-0x10ff]
[271.012358] pci 0000:05:07.0:  bridge window [mem  0x84000000-0x87ffffff pref]
[272.015626] pci 0000:05:07.0:  bridge window [mem  0x88000000-0x8bffffff]

                                             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ offending allocation

[273.018894] pci 0000:00:10.0: PCI bridge to [bus  05-06]
[274.022162] pci 0000:00:10.0:  bridge window [io  0x1000-0x1fff]
[275.025430] pci 0000:00:10.0:  bridge window [mem  0xb3200000-0xb32fffff]
[276.028698] pci 0000:00:10.0:  bridge window [mem  0xb8400000-0x87ffffff pref]
Quoted 3 lines

To me, the issue seems to be that due to some reason the OS either fails to Or does not try to reallocate the addresses assigned to the the non-pref mem BAR of the  the transparent bridge at 00:10.0.

the allocation looks right. and even with transparent bus.

assume that skip resizing transparent bridge, just make some allocation different, then happen to not trigger the overwriting to VGA controller.

anyway, let's wait for test about not using OHCI driver.

Thanks

Yinghai

#33 lore raw
In reply to Yinghai Lu

Dear Yinghai, Bjorn, Ram, Linus and other people,

Quoted 21 lines · Yinghai Lu

2012/1/19 Rogério Brito <[email protected]>:

Quoted 6 lines

OK, now it seems that we are getting farther. The system boots, even mounts the root filesystem, udev starts and, then, when it seems that some USB messages are going to appear on the screen, the notebook crashes with the same distortion that happened before (with some of the last messages appearing being related to OHCI).

in that case, Can you try to disable OHCI in .config

CONFIG_USB_OHCI_HCD

OHCI controller 00:0b.0 is using
Memory at b0004000 (32-bit, non-prefetchable) [size=4K]

VGA controller 00:05.0
       Memory at b2000000 (32-bit, non-prefetchable) [size=16M] Memory at c0000000 (64-bit, prefetchable) [size=256M] Memory at b1000000 (64-bit, non-prefetchable) [size=16M] [virtual] Expansion ROM at 80000000 [disabled] [size=128K]

Quoted 3 lines

it seems not related. or could be ioremap etc have problem?

So can you try to boot with nopat?

I have experimented with Linux 3.3.0-rc1-00080-g16111ea, which is a merge of:

* yinghai's usb_smi_disable_early branch with HEAD = a3a6c096
* linus's master branch with HEAD = 4a7cbb56

I have booted the kernel as above, configured, as Yinghai asked, with OHCI disabled, but no other changes from the previous times. I registered in video two boots with only a slight change in options passed to the kernel, with:

1 - http://youtu.be/fKYubKaNhuc being with options
`acpi=off pnpbios=off noapic debug ignore_loglevel pci=earlydump
boot_delay=10`

2 - http://youtu.be/Gll9RLfXS_c being with options
`acpi=off pnpbios=off noapic debug ignore_loglevel pci=earlydump
nopat boot_delay=10`

The only change is the nopat option, as suggested by Yinghai.

As you can see in the two videos, some things seem to have progressed:
the kernel actually gets to boot, gets to mount/load the initramfs, passes things to init and, when it is loading things, I get a hang with a messed screen, as before.

Among other things, one thing took my attention, which were some messages like:

ACPI Exception: AE_BAD_PARAMETER, Thread 9340 could not acquire
Mutex [0x1] (20120111/utmutex-276)

which seems strange, particularly as I passed the option acpi=off to my kernel.

I don't know if that's a red herring or not. I can perform other tests, of course.

Thanks as usual,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
maintainer
#34 lore raw
Quoted 1 line · Rogério Brito

The only change is the nopat option, as suggested by Yinghai.

I like this idea. Last August, when we first looked at the hang when enabling ACPI, Len Brown suggested that the outb() that does the transition triggers SMM, and the hang is likely in SMM (which is basically invisible and undebuggable from Linux). And he said "we've had problems in the past with SMM relying on something that Windows just happened to do -- such as different MTRRs."

I couldn't see any MTRR info in your video (probably goes by too fast as it's very early). You might be able to get that info by adding another call to print_mtrr_state() later in boot. Then it would be interesting to try to get the corresponding info from Windows.

Quoted 5 lines

   ACPI Exception: AE_BAD_PARAMETER, Thread 9340 could not acquire Mutex [0x1] (20120111/utmutex-276)

which seems strange, particularly as I passed the option acpi=off to my kernel.

Definitely odd. I don't think that should happen. I see a couple places where we might be able to get there even with ACPI disabled, e.g., eeepc_wmi_check_atkd() -> acpi_get_devices() -> acpi_ut_acquire_mutex(). You could find out by adding a dump_stack() where that error is printed. It'd be nice to fix it, but I don't think it's related to the real problems on your box.

Bjorn

#35 lore raw
In reply to Rogério Brito

2012/1/24 Rogério Brito <[email protected]>:

Quoted 22 lines
Quoted 3 lines

it seems not related. or could be ioremap etc have problem?

So can you try to boot with nopat?

I have experimented with Linux 3.3.0-rc1-00080-g16111ea, which is a merge of:

* yinghai's usb_smi_disable_early branch with HEAD = a3a6c096
* linus's master branch with HEAD = 4a7cbb56

I have booted the kernel as above, configured, as Yinghai asked, with OHCI disabled, but no other changes from the previous times.  I registered in video two boots with only a slight change in options passed to the kernel, with:

1 - http://youtu.be/fKYubKaNhuc being with options acpi=off pnpbios=off noapic debug ignore_loglevel pci=earlydump boot_delay=10

2 - http://youtu.be/Gll9RLfXS_c being with options acpi=off pnpbios=off noapic debug ignore_loglevel pci=earlydump nopat boot_delay=10

Quoted 6 lines

The only change is the nopat option, as suggested by Yinghai.

As you can see in the two videos, some things seem to have progressed:
the kernel actually gets to boot, gets to mount/load the initramfs, passes things to init and, when it is loading things, I get a hang with a messed screen, as before.

Your bios does not have RAM page aligned...

[    0.000000]  BIOS-e820: 0000000000000000 - 000000000009dc00 (usable)
[    0.000000]  BIOS-e820: 000000000009dc00 - 00000000000a0000 (reserved)

assume that partial page is used SMI. let's leave the whole page alone for SMI

Can you please apply attached patch about e820 ram page alignement?

Thanks

Yinghai

Subject: [PATCH] x86: Align e820 ram range to page

Some bios provided e820 ram entries are not page aligned.

To make later processing simple. We could round up those range to to page aligned.

---
 arch/x86/kernel/e820.c |   44 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 44 insertions(+)
arch/x86/kernel/e820.c +44 -0
Index: linux-2.6/arch/x86/kernel/e820.c
===================================================================
@@ -880,6 +880,47 @@ static int __init parse_memmap_opt(char
 }
 early_param("memmap", parse_memmap_opt);
 
+static void __init e820_align_ram_page(void)
+{
+	int i;
+	bool changed = false;;
+
+	for (i = 0; i < e820.nr_map; i++) {
+		struct e820entry *entry = &e820.map[i];
+		u64 start, end;
+		u64 start_aligned, end_aligned;
+
+		if (entry->type != E820_RAM)
+			continue;
+
+		start = entry->addr;
+		end = start + entry->size;
+
+		start_aligned = round_up(start, PAGE_SIZE);
+		end_aligned = round_down(end, PAGE_SIZE);
+
+		if (end_aligned <= start_aligned) {
+			e820_update_range(start, end - start, E820_RAM, E820_RESERVED);
+			changed = true;
+			continue;
+		}
+		if (start < start_aligned) {
+			e820_update_range(start, start_aligned - start, E820_RAM, E820_RESERVED);
+			changed = true;
+		}
+		if (end_aligned < end) {
+			e820_update_range(end_aligned, end - end_aligned, E820_RAM, E820_RESERVED);
+			changed = true;
+		}
+	}
+
+	if (changed) {
+		sanitize_e820_map(e820.map, ARRAY_SIZE(e820.map), &e820.nr_map);
+		printk(KERN_INFO "aligned physical RAM map:\n");
+		e820_print_map("aligned");
+	}
+}
+
 void __init finish_e820_parsing(void)
 {
 	if (userdef) {
@@ -892,6 +933,9 @@ void __init finish_e820_parsing(void)
 		printk(KERN_INFO "user-defined physical RAM map:\n");
 		e820_print_map("user");
 	}
+
+	/* In case, We have RAM entres that are not PAGE aligned */
+	e820_align_ram_page();
 }
 
 static inline const char *e820_type_to_string(int e820_type)
#36 lore raw

2012/1/24 Yinghai Lu <[email protected]>:

Quoted 2 lines

Your bios does not have RAM page aligned...

That's pretty common. BIOS tables etc are seldom page-aligned, and afaik we handle it all correctly. Look for PFN_UP(start) and PFN_DOWN(start+length) patterns in both e820.c and memblock.c.

Sure, it is _possible_ that we have some place we screw up in, but I think it's unlikely. As mentioned, non-page-aligned memory is quite common, especially in the low BIOS region just under the 640kB mark.

Linus

#37 lore raw
In reply to Yinghai Lu

2012/1/24 Yinghai Lu <[email protected]>:

Quoted 14 lines

2012/1/24 Rogério Brito <[email protected]>:

Quoted 13 lines
Quoted 3 lines

it seems not related. or could be ioremap etc have problem?

So can you try to boot with nopat?

I have experimented with Linux 3.3.0-rc1-00080-g16111ea, which is a merge of:

* yinghai's usb_smi_disable_early branch with HEAD = a3a6c096
* linus's master branch with HEAD = 4a7cbb56

I have booted the kernel as above, configured, as Yinghai asked, with OHCI disabled, but no other changes from the previous times.  I registered in video two boots with only a slight change in options passed to the kernel, with:

you really need disable_cardbus_mem1_pref.patch that i sent out before.

also please try to boot with
pci=nobios

it will disable pci bios32 probing.

Thanks

Yinghai

#38 lore raw

Hi, people.

Quoted 11 lines · Yinghai Lu

2012/1/24 Yinghai Lu <[email protected]>:

Quoted 10 lines

2012/1/24 Rogério Brito <[email protected]>:

Quoted 9 lines
Quoted 3 lines

it seems not related. or could be ioremap etc have problem?

So can you try to boot with nopat?

I have experimented with Linux 3.3.0-rc1-00080-g16111ea, which is a merge of:

* yinghai's usb_smi_disable_early branch with HEAD = a3a6c096
* linus's master branch with HEAD = 4a7cbb56

Quoted 6 lines
Quoted 4 lines
Quoted 4 lines

I have booted the kernel as above, configured, as Yinghai asked, with OHCI disabled, but no other changes from the previous times.  I registered in video two boots with only a slight change in options passed to the kernel, with:

you really need disable_cardbus_mem1_pref.patch that i sent out before.

I thought that it was already in your git branch, but I applied it this time.

Quoted 4 lines

also please try to boot with
pci=nobios

it will disable pci bios32 probing.

Did that (with both the disable_cardbus... and with e820_ram_aligned patches applied), but I still got the same results. This was booting with nopat and pci=nobios.

The next thing that I'm doing is recording a video of how the computer boots when that transparent sizing bridge patch is reverted, so that you can have that as a reference. Then, I will rip apart the ACPI stuff from the configuration of the kernel to see if it is ACPI that is getting in the way of something here.

I'm open to other suggestions.

Thanks,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#39 lore raw

Hi there, Yinghai, Linus, Bjorn and others.

Quoted 5 lines · Rogério Brito

The next thing that I'm doing is recording a video of how the computer boots when that transparent sizing bridge patch is reverted, so that you can have that as a reference. Then, I will rip apart the ACPI stuff from the configuration of the kernel to see if it is ACPI that is getting in the way of something here.

I started ripping things apart, first with the whole ACPI subsystem, then with PM, with sound and I noticed that things always hung with the nvidia watchdog/TCO module. So, I disabled it and the boot doesn't hang there anymore. In fact, it goes further, and, for the first time since I got my mother's notebook, the nouveau module was loaded.

Unfortunately, it died with a kernel panic and didn't reach the point of X firing up.

So, there is something fishy indeed with these things.

Still open to sugestions. I will now compile a vanilla kernel from Linus's tree gradually putting things in, but from what I already talked with Bjorn, I think that ACPI will be a barrier (and it is sad that so many things *don't* work with ACPI disabled).

Regards,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#40 lore raw
Quoted 8 lines · Rogério Brito

I started ripping things apart, first with the whole ACPI subsystem, then with PM, with sound and I noticed that things always hung with the nvidia watchdog/TCO module. So, I disabled it and the boot doesn't hang there anymore. In fact, it goes further, and, for the first time since I got my mother's notebook, the nouveau module was loaded.

Unfortunately, it died with a kernel panic and didn't reach the point of X firing up.

Update: X *does* start and the panic occurs when GNOME is loading. I will try next to run a kernel with the nvidia tco thing blacklisted *but* with the transparent bridging resizing thing on to see if the notebook is at least usable (as I suspect that nouveau may be having a hard time with the current status of the PCI rework).

Regards,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#41 lore raw

Another update:

Quoted 9 lines · Rogério Brito
Quoted 2 lines · Rogério Brito

Unfortunately, it died with a kernel panic and didn't reach the point of X firing up.

Update: X *does* start and the panic occurs when GNOME is loading. I will try next to run a kernel with the nvidia tco thing blacklisted *but* with the transparent bridging resizing thing on to see if the notebook is at least usable (as I suspect that nouveau may be having a hard time with the current status of the PCI rework).

The kernel panic is indeed there when I compile nouveau in, *after* some time that X is loaded.

Wild thoughts: I just saw that nouveau calls some ACPI stuff to get EDID information and, as ACPI is broken on this notebook, could that be a potential reason for the panic that I'm seeing? (Yes, I can take a photo of the screen).

I'm simply curious about what knowledgeable people would think.

Regards,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#42 lore raw

2012/1/25 Rogério Brito <[email protected]>:

Quoted 6 lines

Wild thoughts: I just saw that nouveau calls some ACPI stuff to get EDID information and, as ACPI is broken on this notebook, could that be a potential reason for the panic that I'm seeing? (Yes, I can take a photo of the screen).

Which 2.6.x kernel works on your laptop without disabling acpi?

looks like we should make linux without disabling acpi work on your system at first.

BTW, can you post you .config?

Thanks

Yinghai

#43 lore raw
Quoted 8 lines · Yinghai Lu

2012/1/25 Rogério Brito <[email protected]>:

Quoted 5 lines

Wild thoughts: I just saw that nouveau calls some ACPI stuff to get EDID information and, as ACPI is broken on this notebook, could that be a potential reason for the panic that I'm seeing? (Yes, I can take a photo of the screen).

Which 2.6.x kernel works on your laptop without disabling acpi?

Unfortunately, no 2.6 kernel has worked on this laptop without acpi=off. I also once booted FreeBSD here just to see how it would behave and it also hang with ACPI enabled.

Quoted 1 line

From what I see on this notebook, if ACPI is not turned off, the

computer simply hangs once the ACPI is started (it gives one layman like me the impression of entering an infinite loop). But, again, this is just a layman observation of what the system appears to be doing.

Quoted 2 lines

looks like we should make linux without disabling acpi work on your system at first.

I'm OK with whatever you people want me to do to get things right.

Quoted 1 line

BTW, can you post you .config?

Sure, it is here:

    http://www.ime.usp.br/~rbrito/linux/clevo/linux-configs/config-3.2.0-rc5-12270-g6fe13a6-yinghai

If you want me to change anything, please let me know and I'll try to do my best.

Regards,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
maintainer
#44 lore raw
Quoted 10 lines · Rogério Brito
Quoted 8 lines · Yinghai Lu

2012/1/25 Rogério Brito <[email protected]>:

Quoted 5 lines

Wild thoughts: I just saw that nouveau calls some ACPI stuff to get EDID information and, as ACPI is broken on this notebook, could that be a potential reason for the panic that I'm seeing? (Yes, I can take a photo of the screen).

Which 2.6.x kernel works on your laptop without disabling acpi?

Quoted 8 lines

Unfortunately, no 2.6 kernel has worked on this laptop without acpi=off. I also once booted FreeBSD here just to see how it would behave and it also hang with ACPI enabled.

From what I see on this notebook, if ACPI is not turned off, the computer simply hangs once the ACPI is started (it gives one layman like me the impression of entering an infinite loop). But, again, this is just a layman observation of what the system appears to be doing.

My guess is that most of these problems are related, so if we fixed whatever causes the ACPI hang, the other problems would probably go away, too.

Did you figure out anything about the MTRRs?

Bjorn

maintainer
#45 lore raw
Quoted 20 lines · Bjorn Helgaas
Quoted 10 lines · Rogério Brito
Quoted 8 lines · Yinghai Lu

2012/1/25 Rogério Brito <[email protected]>:

Quoted 5 lines

Wild thoughts: I just saw that nouveau calls some ACPI stuff to get EDID information and, as ACPI is broken on this notebook, could that be a potential reason for the panic that I'm seeing? (Yes, I can take a photo of the screen).

Which 2.6.x kernel works on your laptop without disabling acpi?

Quoted 8 lines

Unfortunately, no 2.6 kernel has worked on this laptop without acpi=off. I also once booted FreeBSD here just to see how it would behave and it also hang with ACPI enabled.

From what I see on this notebook, if ACPI is not turned off, the computer simply hangs once the ACPI is started (it gives one layman like me the impression of entering an infinite loop). But, again, this is just a layman observation of what the system appears to be doing.

Quoted 5 lines

My guess is that most of these problems are related, so if we fixed whatever causes the ACPI hang, the other problems would probably go away, too.

Did you figure out anything about the MTRRs?

Ping :) Let me know if you'd like a patch to print out the MTRR info later in boot where it will be slow enough to capture in a video.

Bjorn

#46 lore raw

Hi there.

Quoted 21 lines · Bjorn Helgaas
Quoted 20 lines · Bjorn Helgaas
Quoted 18 lines · Rogério Brito
Quoted 8 lines · Yinghai Lu

2012/1/25 Rogério Brito <[email protected]>:

Quoted 5 lines

Wild thoughts: I just saw that nouveau calls some ACPI stuff to get EDID information and, as ACPI is broken on this notebook, could that be a potential reason for the panic that I'm seeing? (Yes, I can take a photo of the screen).

Which 2.6.x kernel works on your laptop without disabling acpi?

Unfortunately, no 2.6 kernel has worked on this laptop without acpi=off. I also once booted FreeBSD here just to see how it would behave and it also hang with ACPI enabled.

From what I see on this notebook, if ACPI is not turned off, the computer simply hangs once the ACPI is started (it gives one layman like me the impression of entering an infinite loop). But, again, this is just a layman observation of what the system appears to be doing.

Quoted 8 lines
Quoted 5 lines

My guess is that most of these problems are related, so if we fixed whatever causes the ACPI hang, the other problems would probably go away, too.

Did you figure out anything about the MTRRs?

Ping :) Let me know if you'd like a patch to print out the MTRR info later in boot where it will be slow enough to capture in a video.

It's quite late here and I'm going to bed, but tomorrow I will send the things that I collected so far (not much more, but still).

Thanks for the reminder,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
maintainer
#48 lore raw
In reply to Rogério Brito
Quoted 4 lines · Rogério Brito

Please, note that this boot log was taken with the patch that I e-mailed earlier (i.e., the one that makes things work). If you want me to send a log with a vanilla kernel, just let me know and I will grab more information.

I'd definitely be interested in a log with a vanilla kernel.

#49 lore raw

Hi, Bjorn.

2012/1/10 Bjorn Helgaas <[email protected]>:

Quoted 8 lines · Rogério Brito
Quoted 5 lines

Please, note that this boot log was taken with the patch that I e-mailed earlier (i.e., the one that makes things work). If you want me to send a log with a vanilla kernel, just let me know and I will grab more information.

I'd definitely be interested in a log with a vanilla kernel.

Sure, here you go:

    http://www.ime.usp.br/~rbrito/linux/clevo/boot-with-vanilla-3.2.m4v

Thanks for any insight,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#50 lore raw
Quoted 10 lines · Rogério Brito

Hi, Bjorn.

2012/1/10 Bjorn Helgaas <[email protected]>:

Quoted 7 lines · Rogério Brito
Quoted 5 lines

Please, note that this boot log was taken with the patch that I e-mailed earlier (i.e., the one that makes things work). If you want me to send a log with a vanilla kernel, just let me know and I will grab more information.

Quoted 6 lines
Quoted 1 line

I'd definitely be interested in a log with a vanilla kernel.

Sure, here you go:

    http://www.ime.usp.br/~rbrito/linux/clevo/boot-with-vanilla-3.2.m4v

I cannot tell what exactly is the problem, but looks like the transparent bridge gets assigned a positive-decoded address, which seems to create some conflict later somehow..

RP

#51 lore raw
In reply to Linus Torvalds

Hi, Linus.

Quoted 23 lines · Linus Torvalds
Quoted 20 lines · Rogério Brito

It doesn't boot with a vanilla kernel. It only boots when I pass the options above *and* compile the kernel with the following patch applied:

,----[ do_not_size_subtractive_decoding_transparent_pci_to_pci_bridges.patch ]

| diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
| index 86b69f85..84543f5 100644
| --- a/drivers/pci/setup-bus.c
| +++ b/drivers/pci/setup-bus.c
| @@ -849,6 +849,10 @@ void __ref __pci_bus_size_bridges(struct pci_bus *bus,
|               break;
|
|       case PCI_CLASS_BRIDGE_PCI:
| +             /* don't size subtractive decoding (transparent)
| +              * PCI-to-PCI bridges */
| +             if (bus->self->transparent)
| +                     break;
|               pci_bridge_check_ranges(bus);
|               if (bus->self->is_hotplug_bridge) {
|                       additional_io_size  = pci_hotplug_io_size;
Quoted 1 line

`----

Quoted 1 line

Ahh. I'd forgotten about that particular PCI patch.

Thanks for acknowledging that issue.

Quoted 4 lines

I think we should try to re-do that "avoid sizing transparent bridges" commit, because it really should make it much easier to do PCI allocations under some very common situations (there's a *lot* of common intel transparent PCI bridges).

All this PCI stuff is alien to me, but I am learning some by osmosis here, which is not bad at all. :)

I only fear that if the patch above is applied and proves to have ill effects if reverting it will leave this computer (and, potentially, some others) without booting. As I always say, I can test anything that you guys want me to do, given some guidance.

Thanks,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#52 lore raw
In reply to Rogério Brito

Hi Rogério -

I'm glad to see you're getting help from a lot of skilled people, here.

Quoted 7 lines · Rogério Brito
Quoted 6 lines · Linus Torvalds

2012/1/6 Rogério Brito <[email protected]>:

Quoted 4 lines

For the record, *some* description of the problem is at:

* https://bugzilla.kernel.org/show_bug.cgi?id=41132

Quoted 7 lines
Quoted 3 lines

This one looks very much like the thing that Edward Donovan fixed fairly recently in commit 52553ddffad7 ("genirq: fix regression in irqfixup, irqpoll")

It does look similar, but not quite the same. And, yes, my main desktop (the one that I'm using right now) was affected by Edward's commit, which, BTW, has my Reported-and-tested-by.

The one area I have experience with is the IRQ regression you cite in 41132. I was bitten by the regressions in spurious IRQ handling, like you, and this has been my first trip inside the kernel. The patches I sent are the sum of my kernel wisdom to date, and I don't want to impersonate an experienced developer, even by accident. :)

At any rate: I can't tell, yet, what IRQ trouble is happening on this notebook. I couldn't pick any out from the description below. And so I can't tell, either :) , how it is similar to the bug above, but not quite the same.

Is there an IRQ disabled, "nobody cared", problem, even with the newest kernels?

If you have info on IRQ problems, I can look at it, hopefully soon. Right now, it might make sense to let this PCI effort roll forward, before opening a second front, but I can't say that for sure. I certainly hope we can make your mom's notebook a happy Linux box.

Thanks,

Ed

Quoted 39 lines

This one is different. Here is a brief summary of the situation:

# Computer description

* Rebranded Clevo M5X0JE
* nForce2 chipset
* AMD Sempron CPU
* NIC with driver forcedeth
* wifi rtl8187

# Booting options

The current working options passed to the kernel are: acpi=off pnpbios=off noapic

It doesn't boot with a vanilla kernel. It only boots when I pass the options above *and* compile the kernel with the following patch applied:

,----[ do_not_size_subtractive_decoding_transparent_pci_to_pci_bridges.patch ]

| diff --git a/drivers/pci/setup-bus.c b/drivers/pci/setup-bus.c
| index 86b69f85..84543f5 100644
| --- a/drivers/pci/setup-bus.c
| +++ b/drivers/pci/setup-bus.c
| @@ -849,6 +849,10 @@ void __ref __pci_bus_size_bridges(struct pci_bus *bus,
|               break;
|
|       case PCI_CLASS_BRIDGE_PCI:
| +             /* don't size subtractive decoding (transparent)
| +              * PCI-to-PCI bridges */
| +             if (bus->self->transparent)
| +                     break;
|               pci_bridge_check_ranges(bus);
|               if (bus->self->is_hotplug_bridge) {
|                       additional_io_size  = pci_hotplug_io_size;

`----

Otherwise, it hangs right when trying to configure PCI, which, according to the dmesg log when it boots occurs approximately 0.12 seconds after the kernel shows its first message.

#53 lore raw

Hi, Edward.

2012/1/10 Edward Donovan <[email protected]>:

Quoted 1 line

I'm glad to see you're getting help from a lot of skilled people, here.

This is great. I'm keeping mom in CC and she is pleasantly surprised that her case got attention from developers. She wasn't expecting one single case to be treated seriously by important people (in her words---and mine too).

Quoted 9 lines
Quoted 3 lines

It does look similar, but not quite the same. And, yes, my main desktop (the one that I'm using right now) was affected by Edward's commit, which, BTW, has my Reported-and-tested-by.

The one area I have experience with is the IRQ regression you cite in 41132.  I was bitten by the regressions in spurious IRQ handling, like you, and this has been my first trip inside the kernel.  The patches I sent are the sum of my kernel wisdom to date, and I don't want to impersonate an experienced developer, even by accident. :)

Ooops. Sorry I listed two issues from different computers (yes, I am seeing many issues with the computers that I have at my disposal). The ones that I should have had listed relevant to this notebook are:

* https://bugzilla.kernel.org/show_bug.cgi?id=41622
* https://bugzilla.kernel.org/show_bug.cgi?id=41722

Issue

* https://bugzilla.kernel.org/show_bug.cgi?id=41132

should not have been mentioned here. Too many issues and I got confused. Sorry.

Quoted 7 lines

Is there an IRQ disabled, "nobody cared", problem, even with the newest kernels?

If you have info on IRQ problems, I can look at it, hopefully soon. Right now, it might make sense to let this PCI effort roll forward, before opening a second front, but I can't say that for sure. I certainly hope we can make your mom's notebook a happy Linux box.

Well, as you mention, yes, I have one case of "nobody cared" in this very desktop (this one I didn't file on bugzilla) that I am using right now (and I have to boot with irqpoll), but I guess that this is another issue entirely. If you could help me with this "nobody cared" thing, then I will be glad to provide copious amounts of data. :)

Thanks,

Signature · 3 lines
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br
#54 lore raw
In reply to Rogério Brito

Dear sirs
I greatly appreciate the interest in helping me get my notebook using linux I was surprised by the commitment of you to help
forever grateful
Marcia

Em 6 de janeiro de 2012 23:31, Rogério Brito <[email protected]> escreveu:

Quoted 67 lines

Hi, Bjorn and others,

Now that bugzilla is up, I think that I can resume testing things on that problematic notebook that I mentioned in earlier e-mails, so that mom (in CC) can use her notebook with Linux.

For the record, *some* description of the problem is at:

* https://bugzilla.kernel.org/show_bug.cgi?id=41132
* https://bugzilla.kernel.org/show_bug.cgi?id=41722

The first one is a regression that I bisected, while the second one seems to be a bit harder to understand for a layman like me.

Of couse, I can summarize what I found and what I didn't.

Sorry for the top-posting, but it is here to keep the context of the previous message.

Thanks,
Rogério Brito.

Quoted 29 lines · On Tue, Sep 6, 2011 at 19:22,  <[email protected]> wrote:

--- Comment #15 from Rogério Brito <[email protected]>  2011-09-06 22:22:44 --- Dear Bjorn,

Sorry for not responding promptly but I had some family members with health problems. Things are settling right now and I think that I can be faster now.

(In reply to comment #14)

Quoted 2 lines

So it looks like 2.4.27 doesn't even try to enable ACPI mode, and 2.6.8 hangs the same way as 3.1.0-rc3.

Yes, exactly.

Quoted 2 lines

Can you try 2.6.28?  If that doesn't boot, try the Ubuntu 9.04 kernel, which was reported to boot on your machine/BIOS.

Unfortunately, I tried to boot with that version of the Ubuntu live CD and it crashed the same way. I may try it once again, paying more attention if it behaves differently in any kind of way, though.

Oh, talking about PCI and ACPI failing, could you take a look at issue #41132 to see if you can help with another machine that stopped working after a commit that I bisected?

Quoted 8 lines

Thanks,

Rogério Brito.

--
Configure bugmail: https://bugzilla.kernel.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.

--
Rogério Brito : rbrito@{ime.usp.br,gmail.com} : GPG key 4096R/BCFCAAAA
http://rb.doesntexist.org : Packages for LaTeX : algorithms.berlios.de
DebianQA: http://qa.debian.org/developer.php?login=rbrito%40ime.usp.br