Re: Linux Support for Thunderbolt using Apple Monitor

Re: Linux Support for Thunderbolt using Apple Monitor

[Greg KH]

On Fri, May 04, 2012 at 02:19:57PM +1000, Francois Rigaut wrote:
> Greg, Nico,
> 
> Thanks Greg for giving us some attention. I went ahead and collected
> a number of information that could be useful to track down our
> hotplug issue.
> 
> 1) Coldplug: Booting with the thunderbolt display attached. This
> works. Several devices (additional to what I get without the display
> attached) are listed. Here is the relevant info:
> 
> http://maumae.net/thunderbolt/coldplug_efi/log  -> kernel.log
> http://maumae.net/thunderbolt/coldplug_efi/lspci -> lspci and lspci -vv
> http://maumae.net/thunderbolt/coldplug_efi/lsmod
> http://maumae.net/thunderbolt/coldplug_efi/config_excerpt.txt
> full config @ http://maumae.net/thunderbolt/config.gz
> http://maumae.net/thunderbolt/coldplug_efi/disconnect -> message
> when disconnecting (pciehp loaded just before)
> 
> note that an attempt to reconnect leads to the same issue as
> described before, i.e. nothing happens.
> 
> 2) Booting with the display/device not plugged, just for reference:
> 
> http://maumae.net/thunderbolt/notplugged_efi/log
> http://maumae.net/thunderbolt/notplugged_efi/lsmod
> http://maumae.net/thunderbolt/notplugged_efi/misc
> http://maumae.net/thunderbolt/notplugged_efi/lspci

Ok, it looks like these are the devices that we are missing while
"hotplugging":
37:00.0 PCI bridge: Intel Corporation Device 1513
38:00.0 PCI bridge: Intel Corporation Device 1513
38:01.0 PCI bridge: Intel Corporation Device 1513
38:02.0 PCI bridge: Intel Corporation Device 1513
38:03.0 PCI bridge: Intel Corporation Device 1513
38:04.0 PCI bridge: Intel Corporation Device 1513
38:05.0 PCI bridge: Intel Corporation Device 1513
39:00.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
3a:03.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
3b:00.0 USB controller: Pericom Semiconductor Device 400e (rev 01)
3b:00.1 USB controller: Pericom Semiconductor Device 400e (rev 01)
3b:00.2 USB controller: Pericom Semiconductor Device 400f (rev 01)
3c:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM57761 Gigabit Ethernet PCIe (rev 10)
3d:00.0 FireWire (IEEE 1394): nee Agere Systems nee Lucent
Microelectronics FW643 PCI Express 1394b Controller (PHY/Link) (rev 08)

Which is good, I was worried that there was a video controller out there
on the end of the thunderbolt link.

So, the problem really is two here:
  - your video card doesn't detect the new display
  - your pci devices do not show up

The first one probably needs to be reported to the Intel video driver
people, they can help you out the best.

The second one means that something is up with the pci hotplug
controller.  I've cced the linux-pci list, the people there should be
able to help out better than I, as I no longer have any PCI hotplug
hardware to work with anymore.

> 3) Again, I have tried to load pciehp and acpiphp with both EFI boot
> and BIOS emulation boot:
> - pciehp loads ok but as I said earlier, no message whatsoever in
> everything.log when I try to connect the thunderbolt display

If you load it with debugging enabled, do you get extra messages in the
log?

> - acpiphp: I was wrong yesterday. In fact, trying to load acpiphp
> gives me a message:
> [root@poliahu thunderbolt]# modprobe acpiphp
> ERROR: could not insert 'acpiphp': No such device

Again, can you load it with debugging (I think both load with "debug=1"
as a module option to do this).

> Not sure what to make of it. I have tried to also load acpiphp after
> a coldplug boot; same error message.
> 
> I've read somewhere that this error message may mean that the h/w
> doesn't support hotplugging. However, I think the coldplug lspci -vv
> above tells us the h/w does support pcie hotplug: all the pci
> bridges appear to have D3hot+ capability (and it certainly does, as
> this works under osx on the same machine).

Yes, it "should" work, but figuring out what the pci hotplug method that
is going on is going to be the main thing to solve here.

Oh, what exact machine is this?  Maybe I can ask around and see if I can
dig up one myself, and hey, I always need an excuse to get a new display
:)

thanks,

greg k-h

Re: Linux Support for Thunderbolt using Apple Monitor

[Yinghai Lu]

On Fri, May 4, 2012 at 11:48 AM, Greg KH <gregkh@linuxfoundation.org> wrote:
> Ok, it looks like these are the devices that we are missing while
> "hotplugging":
> 37:00.0 PCI bridge: Intel Corporation Device 1513
> 38:00.0 PCI bridge: Intel Corporation Device 1513
> 38:01.0 PCI bridge: Intel Corporation Device 1513
> 38:02.0 PCI bridge: Intel Corporation Device 1513
> 38:03.0 PCI bridge: Intel Corporation Device 1513
> 38:04.0 PCI bridge: Intel Corporation Device 1513
> 38:05.0 PCI bridge: Intel Corporation Device 1513
> 39:00.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
> 3a:03.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
> 3b:00.0 USB controller: Pericom Semiconductor Device 400e (rev 01)
> 3b:00.1 USB controller: Pericom Semiconductor Device 400e (rev 01)
> 3b:00.2 USB controller: Pericom Semiconductor Device 400f (rev 01)
> 3c:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM57761 Gigabit Ethernet PCIe (rev 10)
> 3d:00.0 FireWire (IEEE 1394): nee Agere Systems nee Lucent
> Microelectronics FW643 PCI Express 1394b Controller (PHY/Link) (rev 08)
>
> Which is good, I was worried that there was a video controller out there
> on the end of the thunderbolt link.
>
> So, the problem really is two here:
>  - your video card doesn't detect the new display
>  - your pci devices do not show up
>
> The first one probably needs to be reported to the Intel video driver
> people, they can help you out the best.
>
> The second one means that something is up with the pci hotplug
> controller.  I've cced the linux-pci list, the people there should be
> able to help out better than I, as I no longer have any PCI hotplug
> hardware to work with anymore.
>

those devices are under 04:04.0, but that does not have slot cap
enabled by BIOS. So pciehp is going to to help.

04:04.0 PCI bridge: Intel Corporation Device 151a (rev 01) (prog-if 00
[Normal decode])
	Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr-
Stepping- SERR- FastB2B- DisINTx+
	Status: Cap+ 66MHz- UDF- FastB2B- ParErr- DEVSEL=fast >TAbort-
<TAbort- <MAbort- >SERR- <PERR- INTx-
	Latency: 0, Cache Line Size: 256 bytes
	Bus: primary=04, secondary=37, subordinate=67, sec-latency=0
	Memory behind bridge: a0900000-a09fffff
	Secondary status: 66MHz- FastB2B- ParErr- DEVSEL=fast >TAbort-
<TAbort- <MAbort- <SERR- <PERR-
	BridgeCtl: Parity- SERR- NoISA- VGA- MAbort- >Reset- FastB2B-
		PriDiscTmr- SecDiscTmr- DiscTmrStat- DiscTmrSERREn-
	Capabilities: [80] Power Management version 3
		Flags: PMEClk- DSI- D1+ D2+ AuxCurrent=0mA PME(D0+,D1+,D2+,D3hot+,D3cold+)
		Status: D0 NoSoftRst+ PME-Enable- DSel=0 DScale=1 PME-
	Capabilities: [88] MSI: Enable+ Count=1/1 Maskable- 64bit+
		Address: 00000000fee0f00c  Data: 4189
	Capabilities: [c0] Express (v2) Downstream Port (Slot-), MSI 00
		DevCap:	MaxPayload 128 bytes, PhantFunc 0, Latency L0s <64ns, L1 <1us
			ExtTag+ RBE+ FLReset-
		DevCtl:	Report errors: Correctable- Non-Fatal- Fatal- Unsupported-
			RlxdOrd+ ExtTag- PhantFunc- AuxPwr- NoSnoop+
			MaxPayload 128 bytes, MaxReadReq 512 bytes
		DevSta:	CorrErr- UncorrErr- FatalErr- UnsuppReq- AuxPwr+ TransPend-
		LnkCap:	Port #4, Speed 2.5GT/s, Width x4, ASPM L0s L1, Latency L0
unlimited, L1 unlimited
			ClockPM- Surprise- LLActRep+ BwNot+
		LnkCtl:	ASPM Disabled; Disabled- Retrain- CommClk-
			ExtSynch- ClockPM- AutWidDis- BWInt- AutBWInt-
		LnkSta:	Speed 2.5GT/s, Width x4, TrErr- Train- SlotClk+ DLActive-
BWMgmt- ABWMgmt-
		DevCap2: Completion Timeout: Not Supported, TimeoutDis- ARIFwd-
		DevCtl2: Completion Timeout: 50us to 50ms, TimeoutDis- ARIFwd-
		LnkCtl2: Target Link Speed: 2.5GT/s, EnterCompliance- SpeedDis-,
Selectable De-emphasis: -6dB
			 Transmit Margin: Normal Operating Range, EnterModifiedCompliance-
ComplianceSOS-
			 Compliance De-emphasis: -6dB
		LnkSta2: Current De-emphasis Level: -3.5dB, EqualizationComplete-,
EqualizationPhase1-
			 EqualizationPhase2-, EqualizationPhase3-, LinkEqualizationRequest-

Re: Linux Support for Thunderbolt using Apple Monitor

[Yinghai Lu]

On Fri, May 4, 2012 at 12:37 PM, Yinghai Lu <yinghai@kernel.org> wrote:
> On Fri, May 4, 2012 at 11:48 AM, Greg KH <gregkh@linuxfoundation.org> wrote:
>> Ok, it looks like these are the devices that we are missing while
>> "hotplugging":
>> 37:00.0 PCI bridge: Intel Corporation Device 1513
>> 38:00.0 PCI bridge: Intel Corporation Device 1513
>> 38:01.0 PCI bridge: Intel Corporation Device 1513
>> 38:02.0 PCI bridge: Intel Corporation Device 1513
>> 38:03.0 PCI bridge: Intel Corporation Device 1513
>> 38:04.0 PCI bridge: Intel Corporation Device 1513
>> 38:05.0 PCI bridge: Intel Corporation Device 1513
>> 39:00.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
>> 3a:03.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
>> 3b:00.0 USB controller: Pericom Semiconductor Device 400e (rev 01)
>> 3b:00.1 USB controller: Pericom Semiconductor Device 400e (rev 01)
>> 3b:00.2 USB controller: Pericom Semiconductor Device 400f (rev 01)
>> 3c:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM57761 Gigabit Ethernet PCIe (rev 10)
>> 3d:00.0 FireWire (IEEE 1394): nee Agere Systems nee Lucent
>> Microelectronics FW643 PCI Express 1394b Controller (PHY/Link) (rev 08)
>>
>> Which is good, I was worried that there was a video controller out there
>> on the end of the thunderbolt link.
>>
>> So, the problem really is two here:
>>  - your video card doesn't detect the new display
>>  - your pci devices do not show up
>>
>> The first one probably needs to be reported to the Intel video driver
>> people, they can help you out the best.
>>
>> The second one means that something is up with the pci hotplug
>> controller.  I've cced the linux-pci list, the people there should be
>> able to help out better than I, as I no longer have any PCI hotplug
>> hardware to work with anymore.
>>
>
> those devices are under 04:04.0, but that does not have slot cap
> enabled by BIOS. So pciehp is going to to help.

should be: pciehp is not going to help.

so acpiphp are involved?

Yinghai

Re: Linux Support for Thunderbolt using Apple Monitor

[Francois Rigaut]

On 05/05/12 05:38, Yinghai Lu wrote:
> On Fri, May 4, 2012 at 12:37 PM, Yinghai Lu<yinghai@kernel.org>  wrote:
>> On Fri, May 4, 2012 at 11:48 AM, Greg KH<gregkh@linuxfoundation.org>  wrote:
>>> Ok, it looks like these are the devices that we are missing while
>>> "hotplugging":
>>> 37:00.0 PCI bridge: Intel Corporation Device 1513
>>> 38:00.0 PCI bridge: Intel Corporation Device 1513
>>> 38:01.0 PCI bridge: Intel Corporation Device 1513
>>> 38:02.0 PCI bridge: Intel Corporation Device 1513
>>> 38:03.0 PCI bridge: Intel Corporation Device 1513
>>> 38:04.0 PCI bridge: Intel Corporation Device 1513
>>> 38:05.0 PCI bridge: Intel Corporation Device 1513
>>> 39:00.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
>>> 3a:03.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
>>> 3b:00.0 USB controller: Pericom Semiconductor Device 400e (rev 01)
>>> 3b:00.1 USB controller: Pericom Semiconductor Device 400e (rev 01)
>>> 3b:00.2 USB controller: Pericom Semiconductor Device 400f (rev 01)
>>> 3c:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM57761 Gigabit Ethernet PCIe (rev 10)
>>> 3d:00.0 FireWire (IEEE 1394): nee Agere Systems nee Lucent
>>> Microelectronics FW643 PCI Express 1394b Controller (PHY/Link) (rev 08)
>>>
>>> Which is good, I was worried that there was a video controller out there
>>> on the end of the thunderbolt link.
>>>
>>> So, the problem really is two here:
>>>   - your video card doesn't detect the new display
>>>   - your pci devices do not show up
>>>
>>> The first one probably needs to be reported to the Intel video driver
>>> people, they can help you out the best.
no it's definitely not the first one. As you saw, there'sa whole bunch 
of other devices (3 USB, firewire, network) that do not show up either.

The new info from Yinghai (thanks Yinghai) explains why hotpluging with 
pciehp did not work/detect the PCIE hotplug and did not result in *any* 
kernel messages.

So the whole thing I guess boils down to making acpiphp work. And why it 
outputs the "no such device" error message and does not load:

I tried with debug=1 and got a bit more info in the log files:

[root@poliahu frigaut]# lsmod | grep -e acpi -e pci
acpi_cpufreq            5941  0
mperf                   1275  1 acpi_cpufreq
processor              26144  1 acpi_cpufreq
[root@poliahu frigaut]# modprobe acpiphp debug=1
ERROR: could not insert 'acpiphp': No such device

at this point this message showed up in everything.log:
May  5 14:09:55 localhost kernel: [28329.700022] pci_hotplug: PCI Hot 
Plug PCI Core version: 0.5
May  5 14:09:55 localhost kernel: [28329.702041] acpiphp: ACPI Hot Plug 
PCI Controller Driver version: 0.5
May  5 14:09:55 localhost kernel: [28329.702484] acpiphp_glue: found 
PCI-to-PCI bridge at PCI 0000:04:00.0
May  5 14:09:55 localhost kernel: [28329.702715] acpiphp_glue: Bus 
0000:05 has 0 slots
May  5 14:09:55 localhost kernel: [28329.702720] acpiphp_glue: Total 0 slots

then:
[root@poliahu frigaut]# lsmod | grep -e acpi -e pci
pci_hotplug            25324  0
acpi_cpufreq            5941  0
mperf                   1275  1 acpi_cpufreq
processor              26144  1 acpi_cpufreq
[root@poliahu frigaut]#

shows that acpiphp was indeed not loaded (I assume the error occurs 
after it has written out the above messages in the log file), but 
pci_hotplug was as a dep of acpiphp I assume (thus, correct behavior).

So we're back to the issue of the acpiphp load error.

Greg, I see you're one of the authors of acpiphp:
#define DRIVER_VERSION    "0.5"
#define DRIVER_AUTHOR    "Greg Kroah-Hartman <gregkh@us.ibm.com>, 
Takayoshi Kochi <t-kochi@bq.jp.nec.com>, Matthew Wilcox <willy@hp.com>"
#define DRIVER_DESC    "ACPI Hot Plug PCI Controller Driver"

but that might have been a while back?

Cheers, and thanks,
Francois



>>>
>>> The second one means that something is up with the pci hotplug
>>> controller.  I've cced the linux-pci list, the people there should be
>>> able to help out better than I, as I no longer have any PCI hotplug
>>> hardware to work with anymore.
>>>
>> those devices are under 04:04.0, but that does not have slot cap
>> enabled by BIOS. So pciehp is going to to help.
> should be: pciehp is not going to help.
>
> so acpiphp are involved?
>
> Yinghai

Re: Linux Support for Thunderbolt using Apple Monitor

[Jiang Liu]

This log message below implies that your BIOS doesn't provide the ACPI interfaces
needed by acpiphp driver. Maybe an ACPI dump file may help to identify the issue.
You may generate acpidump file by:
acpidump > acpidump.bin

>May  5 14:09:55 localhost kernel: [28329.702720] acpiphp_glue: Total 0 slots 

On 05/05/2012 12:19 PM, Francois Rigaut wrote:
> May  5 14:09:55 localhost kernel: [28329.702720] acpiphp_glue: Total 0 slots

Re: Linux Support for Thunderbolt using Apple Monitor

[Francois Rigaut]

http://maumae.net/thunderbolt/acpidump.bin
Thanks.
Francois


On 05/05/12 18:18, Jiang Liu wrote:
> This log message below implies that your BIOS doesn't provide the ACPI interfaces
> needed by acpiphp driver. Maybe an ACPI dump file may help to identify the issue.
> You may generate acpidump file by:
> acpidump>  acpidump.bin
>
>> May  5 14:09:55 localhost kernel: [28329.702720] acpiphp_glue: Total 0 slots
> On 05/05/2012 12:19 PM, Francois Rigaut wrote:
>> May  5 14:09:55 localhost kernel: [28329.702720] acpiphp_glue: Total 0 slots

Re: Linux Support for Thunderbolt using Apple Monitor

[Jiang Liu]

There's another mysterious _DSM function under \_SB.PCI0.P0P2.UPSB.
Seems it's related to Thunderbolt technology.

                    Method (_DSM, 4, NotSerialized)
                    {
                        If (LEqual (Arg0, Buffer (0x10)
                                {
                                    /* 0000 */    0xC6, 0xB7, 0xB5, 0xA0, 0x18, 0x13, 0x1C, 0x44,
                                    /* 0008 */    0xB0, 0xC9, 0xFE, 0x69, 0x5E, 0xAF, 0x94, 0x9B
                                }))
                        {
                            Store (Package (0x02)
                                {
                                    "PCI-Thunderbolt",
                                    0x01
                                }, Local0)
                            DTGP (Arg0, Arg1, Arg2, Arg3, RefOf (Local0))
                            Return (Local0)
                        }

                        Return (0x80000002)
                    }


On 05/05/2012 04:47 PM, Francois Rigaut wrote:
> http://maumae.net/thunderbolt/acpidump.bin
> Thanks.
> Francois
> 
> 
> On 05/05/12 18:18, Jiang Liu wrote:
>> This log message below implies that your BIOS doesn't provide the ACPI interfaces
>> needed by acpiphp driver. Maybe an ACPI dump file may help to identify the issue.
>> You may generate acpidump file by:
>> acpidump>  acpidump.bin
>>
>>> May  5 14:09:55 localhost kernel: [28329.702720] acpiphp_glue: Total 0 slots
>> On 05/05/2012 12:19 PM, Francois Rigaut wrote:
>>> May  5 14:09:55 localhost kernel: [28329.702720] acpiphp_glue: Total 0 slots

Re: Linux Support for Thunderbolt using Apple Monitor

[Jiang Liu]

Hi Francois,
	It seems that the PCI topology related to the Thunderbird is:
00:01.0 PCI bridge: Intel Corporation Xeon E3-1200/2nd Generation Core Processor Family PCI Express Root Port (rev 09)
	03:00.0 PCI bridge: Intel Corporation Device 151a (rev 01)
	04:00.0 PCI bridge: Intel Corporation Device 151a (rev 01)
	04:03.0 PCI bridge: Intel Corporation Device 151a (rev 01)
	04:04.0 PCI bridge: Intel Corporation Device 151a (rev 01)
		37:00.0 PCI bridge: Intel Corporation Device 1513
		38:00.0 PCI bridge: Intel Corporation Device 1513
			39:00.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
			3a:03.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
				3b:00.0 USB controller: Pericom Semiconductor Device 400e (rev 01)
				3b:00.1 USB controller: Pericom Semiconductor Device 400e (rev 01)
				3b:00.2 USB controller: Pericom Semiconductor Device 400f (rev 01)
		38:01.0 PCI bridge: Intel Corporation Device 1513
			3c:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM57761 Gigabit Ethernet PCIe (rev 10)
		38:02.0 PCI bridge: Intel Corporation Device 1513
			3d:00.0 FireWire (IEEE 1394): nee Agere Systems nee Lucent Microelectronics FW643 PCI Express 1394b Controller (PHY/Link) (rev 08)
		38:03.0 PCI bridge: Intel Corporation Device 1513
		38:04.0 PCI bridge: Intel Corporation Device 1513
		38:05.0 PCI bridge: Intel Corporation Device 1513

It's really a complex topology. The longest path has root port + 3 PCIe switches + endpoint.
The device with id "Intel Corporation Device 151a", "Intel Corporation Device 1513" and
"Pericom Semiconductor Device 400c" should be 3 PCIe switches. I guess the thunderbird
cable is used to connect 04:04.0 to the second PCIe switch with device ID "Intel Corporation
Device 1513". 

According to the acpidump sent out by you, I guess the relationship among PCIe and ACPI is as below:
        Device (PCI0)	// PCIe host bridge
        {
            Name (_HID, EisaId ("PNP0A08"))
            Device (P0P2)	//00:01.0
            {
                Name (_ADR, 0x00010000)
                Device (UPSB)	//03:00.0
                {
                    Name (_ADR, 0x00)
                    Device (DSB2)	//04:04.0
                    {
                        Name (_ADR, 0x00040000)
                        Method (_RMV, 0, NotSerialized)
                        {
                            Return (OSDW ()) // OSDW-> OS is Darwin
                        }
                        Device (UPS2)	//37:00.0
                        {
                            Name (_ADR, 0x00)
			}
		    }
		}
	    }
	}

But _RMV method only returns 1 on iOS or Max OS X. That means ACPI based hotplug is disabled
on Linux for your system. And as Yinghai has pointed out, all the PCIe downstream ports on 
the system don't support PCIe hotplug. So need BIOS or hardware changes to support Thunderbolt
cable hotplug on your system.

On 05/05/2012 12:19 PM, Francois Rigaut wrote:
> 
> On 05/05/12 05:38, Yinghai Lu wrote:
>> On Fri, May 4, 2012 at 12:37 PM, Yinghai Lu<yinghai@kernel.org>  wrote:
>>> On Fri, May 4, 2012 at 11:48 AM, Greg KH<gregkh@linuxfoundation.org>  wrote:
>>>> Ok, it looks like these are the devices that we are missing while
>>>> "hotplugging":
>>>> 37:00.0 PCI bridge: Intel Corporation Device 1513
>>>> 38:00.0 PCI bridge: Intel Corporation Device 1513
>>>> 38:01.0 PCI bridge: Intel Corporation Device 1513
>>>> 38:02.0 PCI bridge: Intel Corporation Device 1513
>>>> 38:03.0 PCI bridge: Intel Corporation Device 1513
>>>> 38:04.0 PCI bridge: Intel Corporation Device 1513
>>>> 38:05.0 PCI bridge: Intel Corporation Device 1513
>>>> 39:00.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
>>>> 3a:03.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
>>>> 3b:00.0 USB controller: Pericom Semiconductor Device 400e (rev 01)
>>>> 3b:00.1 USB controller: Pericom Semiconductor Device 400e (rev 01)
>>>> 3b:00.2 USB controller: Pericom Semiconductor Device 400f (rev 01)
>>>> 3c:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM57761 Gigabit Ethernet PCIe (rev 10)
>>>> 3d:00.0 FireWire (IEEE 1394): nee Agere Systems nee Lucent
>>>> Microelectronics FW643 PCI Express 1394b Controller (PHY/Link) (rev 08)
>>>>
>>>> Which is good, I was worried that there was a video controller out there
>>>> on the end of the thunderbolt link.
>>>>
>>>> So, the problem really is two here:
>>>>   - your video card doesn't detect the new display
>>>>   - your pci devices do not show up
>>>>
>>>> The first one probably needs to be reported to the Intel video driver
>>>> people, they can help you out the best.
> no it's definitely not the first one. As you saw, there'sa whole bunch of other devices (3 USB, firewire, network) that do not show up either.
> 
> The new info from Yinghai (thanks Yinghai) explains why hotpluging with pciehp did not work/detect the PCIE hotplug and did not result in *any* kernel messages.
> 
> So the whole thing I guess boils down to making acpiphp work. And why it outputs the "no such device" error message and does not load:
> 
> I tried with debug=1 and got a bit more info in the log files:
> 
> [root@poliahu frigaut]# lsmod | grep -e acpi -e pci
> acpi_cpufreq            5941  0
> mperf                   1275  1 acpi_cpufreq
> processor              26144  1 acpi_cpufreq
> [root@poliahu frigaut]# modprobe acpiphp debug=1
> ERROR: could not insert 'acpiphp': No such device
> 
> at this point this message showed up in everything.log:
> May  5 14:09:55 localhost kernel: [28329.700022] pci_hotplug: PCI Hot Plug PCI Core version: 0.5
> May  5 14:09:55 localhost kernel: [28329.702041] acpiphp: ACPI Hot Plug PCI Controller Driver version: 0.5
> May  5 14:09:55 localhost kernel: [28329.702484] acpiphp_glue: found PCI-to-PCI bridge at PCI 0000:04:00.0
> May  5 14:09:55 localhost kernel: [28329.702715] acpiphp_glue: Bus 0000:05 has 0 slots
> May  5 14:09:55 localhost kernel: [28329.702720] acpiphp_glue: Total 0 slots
> 
> then:
> [root@poliahu frigaut]# lsmod | grep -e acpi -e pci
> pci_hotplug            25324  0
> acpi_cpufreq            5941  0
> mperf                   1275  1 acpi_cpufreq
> processor              26144  1 acpi_cpufreq
> [root@poliahu frigaut]#
> 
> shows that acpiphp was indeed not loaded (I assume the error occurs after it has written out the above messages in the log file), but pci_hotplug was as a dep of acpiphp I assume (thus, correct behavior).
> 
> So we're back to the issue of the acpiphp load error.
> 
> Greg, I see you're one of the authors of acpiphp:
> #define DRIVER_VERSION    "0.5"
> #define DRIVER_AUTHOR    "Greg Kroah-Hartman <gregkh@us.ibm.com>, Takayoshi Kochi <t-kochi@bq.jp.nec.com>, Matthew Wilcox <willy@hp.com>"
> #define DRIVER_DESC    "ACPI Hot Plug PCI Controller Driver"
> 
> but that might have been a while back?
> 
> Cheers, and thanks,
> Francois
> 
> 
> 
>>>>
>>>> The second one means that something is up with the pci hotplug
>>>> controller.  I've cced the linux-pci list, the people there should be
>>>> able to help out better than I, as I no longer have any PCI hotplug
>>>> hardware to work with anymore.
>>>>
>>> those devices are under 04:04.0, but that does not have slot cap
>>> enabled by BIOS. So pciehp is going to to help.
>> should be: pciehp is not going to help.
>>
>> so acpiphp are involved?
>>
>> Yinghai
> -- 
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Re: Linux Support for Thunderbolt using Apple Monitor

[Nico Schottelius]

Hey Jiang,

Jiang Liu [Sat, May 05, 2012 at 11:18:09PM +0800]:
> 	It seems that the PCI topology related to the Thunderbird is:

Thanks for digging into this, I've the feeling we could somehow get
this running in the end...

> [...] 
> According to the acpidump sent out by you, I guess the relationship among PCIe and ACPI is as below:
>         Device (PCI0)	// PCIe host bridge
>         {
>             Name (_HID, EisaId ("PNP0A08"))
>             Device (P0P2)	//00:01.0
>             {
>                 Name (_ADR, 0x00010000)
>                 Device (UPSB)	//03:00.0
>                 {
>                     Name (_ADR, 0x00)
>                     Device (DSB2)	//04:04.0
>                     {
>                         Name (_ADR, 0x00040000)
>                         Method (_RMV, 0, NotSerialized)
>                         {
>                             Return (OSDW ()) // OSDW-> OS is Darwin
>                         }
>                         Device (UPS2)	//37:00.0
>                         {
>                             Name (_ADR, 0x00)
> 			}
> 		    }
> 		}
> 	    }
> 	}
> 
> But _RMV method only returns 1 on iOS or Max OS X. That means ACPI based hotplug is disabled
> on Linux for your system. And as Yinghai has pointed out, all the PCIe downstream ports on 
> the system don't support PCIe hotplug. So need BIOS or hardware changes to support Thunderbolt
> cable hotplug on your system.

I'm not deep into how acpi works, but wondering: Can't we replace that part,
change the tree or emulate a different OSDW() method similar to LD_PRELOAD?

Cheers,

Nico

-- 
PGP key: 7ED9 F7D3 6B10 81D7 0EC5  5C09 D7DC C8E4 3187 7DF0

Re: Linux Support for Thunderbolt using Apple Monitor

[Jiang Liu]

I'm afraid it won't work by just changing the OSDW() return value.
After analysis the ACPI table, I guess the BIOS ACPI implementation doesn't
confirm to the ACPI specification for hotplug, it needs special support in
OS to enable Thunderbolt cable hotplug on that system. Maybe that's the
reason why it's only enabled on Apple's OSes, not sure it's due to technical
or business considerations.

This PCIe device "05:00.0 System peripheral: Intel Corporation Device 151a
(rev 01)" may play a role in the Thunderbolt cable hotplug operations, but
haven't figured it out yet.

On 05/06/2012 12:48 AM, Nico Schottelius wrote:
> Hey Jiang,
> 
> Jiang Liu [Sat, May 05, 2012 at 11:18:09PM +0800]:
>> 	It seems that the PCI topology related to the Thunderbird is:
> 
> Thanks for digging into this, I've the feeling we could somehow get
> this running in the end...
> 
>> [...] 
>> According to the acpidump sent out by you, I guess the relationship among PCIe and ACPI is as below:
>>         Device (PCI0)	// PCIe host bridge
>>         {
>>             Name (_HID, EisaId ("PNP0A08"))
>>             Device (P0P2)	//00:01.0
>>             {
>>                 Name (_ADR, 0x00010000)
>>                 Device (UPSB)	//03:00.0
>>                 {
>>                     Name (_ADR, 0x00)
>>                     Device (DSB2)	//04:04.0
>>                     {
>>                         Name (_ADR, 0x00040000)
>>                         Method (_RMV, 0, NotSerialized)
>>                         {
>>                             Return (OSDW ()) // OSDW-> OS is Darwin
>>                         }
>>                         Device (UPS2)	//37:00.0
>>                         {
>>                             Name (_ADR, 0x00)
>> 			}
>> 		    }
>> 		}
>> 	    }
>> 	}
>>
>> But _RMV method only returns 1 on iOS or Max OS X. That means ACPI based hotplug is disabled
>> on Linux for your system. And as Yinghai has pointed out, all the PCIe downstream ports on 
>> the system don't support PCIe hotplug. So need BIOS or hardware changes to support Thunderbolt
>> cable hotplug on your system.
> 
> I'm not deep into how acpi works, but wondering: Can't we replace that part,
> change the tree or emulate a different OSDW() method similar to LD_PRELOAD?
> 
> Cheers,
> 
> Nico
> 

Re: Linux Support for Thunderbolt using Apple Monitor

[Greg KH]

On Sun, May 06, 2012 at 11:22:11AM +0800, Jiang Liu wrote:
> I'm afraid it won't work by just changing the OSDW() return value.
> After analysis the ACPI table, I guess the BIOS ACPI implementation doesn't
> confirm to the ACPI specification for hotplug, it needs special support in
> OS to enable Thunderbolt cable hotplug on that system. Maybe that's the
> reason why it's only enabled on Apple's OSes, not sure it's due to technical
> or business considerations.
> 
> This PCIe device "05:00.0 System peripheral: Intel Corporation Device 151a
> (rev 01)" may play a role in the Thunderbolt cable hotplug operations, but
> haven't figured it out yet.

Thanks for looking at this.

So either we need a BIOS update for this platform so that other
operating systems can work properly (i.e. full ACPI hotplug support), or
we write a new hotplug driver to handle this hardware properly.

Intel was rumored to have this working in their labs, I'll poke some
people to see if they have anything they can release...

thanks,

greg k-h

Re: Linux Support for Thunderbolt using Apple Monitor

[Francois Rigaut]

Greg,

I went ahead and contacted John Ronciak at intel.
He gave a presentation on thunderbolt at the 2011 linux plumber conference.
Here is his answer:

"You are correct that most if not all issues related to Thunderbolt on 
Linux is due to BIOS problems. We have not seen anything other BIOS 
issues regarding Thunderbolt issues on Linux. When the BIOS is working 
correctly the hot-plug PCIe works fine wrt the kernel. These are things 
that cannot be fixed in the drivers or in kernel infrastructure. So 
newer systems are going to be better than older ones due to these BIOS 
issues."

Francois

On 10/05/12 02:05, Greg KH wrote:
> On Sun, May 06, 2012 at 11:22:11AM +0800, Jiang Liu wrote:
>> I'm afraid it won't work by just changing the OSDW() return value.
>> After analysis the ACPI table, I guess the BIOS ACPI implementation doesn't
>> confirm to the ACPI specification for hotplug, it needs special support in
>> OS to enable Thunderbolt cable hotplug on that system. Maybe that's the
>> reason why it's only enabled on Apple's OSes, not sure it's due to technical
>> or business considerations.
>>
>> This PCIe device "05:00.0 System peripheral: Intel Corporation Device 151a
>> (rev 01)" may play a role in the Thunderbolt cable hotplug operations, but
>> haven't figured it out yet.
> Thanks for looking at this.
>
> So either we need a BIOS update for this platform so that other
> operating systems can work properly (i.e. full ACPI hotplug support), or
> we write a new hotplug driver to handle this hardware properly.
>
> Intel was rumored to have this working in their labs, I'll poke some
> people to see if they have anything they can release...
>
> thanks,
>
> greg k-h

Re: Linux Support for Thunderbolt using Apple Monitor

[Greg KH]

On Thu, May 10, 2012 at 11:49:35AM +1000, Francois Rigaut wrote:
> Greg,
> 
> I went ahead and contacted John Ronciak at intel.
> He gave a presentation on thunderbolt at the 2011 linux plumber conference.
> Here is his answer:
> 
> "You are correct that most if not all issues related to Thunderbolt
> on Linux is due to BIOS problems. We have not seen anything other
> BIOS issues regarding Thunderbolt issues on Linux. When the BIOS is
> working correctly the hot-plug PCIe works fine wrt the kernel. These
> are things that cannot be fixed in the drivers or in kernel
> infrastructure. So newer systems are going to be better than older
> ones due to these BIOS issues."

Thanks for letting us know, so there's nothing we can do for these
machines, and we just have to hope the next-generation machines get this
correct.

greg k-h

Re: Linux Support for Thunderbolt using Apple Monitor

[Francois Rigaut]

Hi Jiang Liu,

I guess this confirms a presentation at the last Linux Plumber 
conference (September 2011) by John Ronciak (intel).
The presentation is entitled "Thunderbolt Technology: What is it and 
what are the Open Source Implications"
You can get it here http://maumae.net/thunderbolt/Plumbers-Thunderbolt.odp

In particular, slide 9 "Thunderbolt Technology: What does this mean for SW"
states:
"Hot plug PCIe (BIOS/ACPI/OS) - BIOS issues being worked"
so I guess someone at intel is working in the same issue and has indeed 
identified the same BIOS limitations.

I'll try to contact John Ronciak (unless someone here knows him and 
volunteer to ask) to inquire the state of progress of this work.

Thanks to anyone that have provided insight !
Francois


On 06/05/12 01:18, Jiang Liu wrote:
> Hi Francois,
> 	It seems that the PCI topology related to the Thunderbird is:
> 00:01.0 PCI bridge: Intel Corporation Xeon E3-1200/2nd Generation Core Processor Family PCI Express Root Port (rev 09)
> 	03:00.0 PCI bridge: Intel Corporation Device 151a (rev 01)
> 	04:00.0 PCI bridge: Intel Corporation Device 151a (rev 01)
> 	04:03.0 PCI bridge: Intel Corporation Device 151a (rev 01)
> 	04:04.0 PCI bridge: Intel Corporation Device 151a (rev 01)
> 		37:00.0 PCI bridge: Intel Corporation Device 1513
> 		38:00.0 PCI bridge: Intel Corporation Device 1513
> 			39:00.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
> 			3a:03.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
> 				3b:00.0 USB controller: Pericom Semiconductor Device 400e (rev 01)
> 				3b:00.1 USB controller: Pericom Semiconductor Device 400e (rev 01)
> 				3b:00.2 USB controller: Pericom Semiconductor Device 400f (rev 01)
> 		38:01.0 PCI bridge: Intel Corporation Device 1513
> 			3c:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM57761 Gigabit Ethernet PCIe (rev 10)
> 		38:02.0 PCI bridge: Intel Corporation Device 1513
> 			3d:00.0 FireWire (IEEE 1394): nee Agere Systems nee Lucent Microelectronics FW643 PCI Express 1394b Controller (PHY/Link) (rev 08)
> 		38:03.0 PCI bridge: Intel Corporation Device 1513
> 		38:04.0 PCI bridge: Intel Corporation Device 1513
> 		38:05.0 PCI bridge: Intel Corporation Device 1513
>
> It's really a complex topology. The longest path has root port + 3 PCIe switches + endpoint.
> The device with id "Intel Corporation Device 151a", "Intel Corporation Device 1513" and
> "Pericom Semiconductor Device 400c" should be 3 PCIe switches. I guess the thunderbird
> cable is used to connect 04:04.0 to the second PCIe switch with device ID "Intel Corporation
> Device 1513".
>
> According to the acpidump sent out by you, I guess the relationship among PCIe and ACPI is as below:
>          Device (PCI0)	// PCIe host bridge
>          {
>              Name (_HID, EisaId ("PNP0A08"))
>              Device (P0P2)	//00:01.0
>              {
>                  Name (_ADR, 0x00010000)
>                  Device (UPSB)	//03:00.0
>                  {
>                      Name (_ADR, 0x00)
>                      Device (DSB2)	//04:04.0
>                      {
>                          Name (_ADR, 0x00040000)
>                          Method (_RMV, 0, NotSerialized)
>                          {
>                              Return (OSDW ()) // OSDW->  OS is Darwin
>                          }
>                          Device (UPS2)	//37:00.0
>                          {
>                              Name (_ADR, 0x00)
> 			}
> 		    }
> 		}
> 	    }
> 	}
>
> But _RMV method only returns 1 on iOS or Max OS X. That means ACPI based hotplug is disabled
> on Linux for your system. And as Yinghai has pointed out, all the PCIe downstream ports on
> the system don't support PCIe hotplug. So need BIOS or hardware changes to support Thunderbolt
> cable hotplug on your system.
>
> On 05/05/2012 12:19 PM, Francois Rigaut wrote:
>> On 05/05/12 05:38, Yinghai Lu wrote:
>>> On Fri, May 4, 2012 at 12:37 PM, Yinghai Lu<yinghai@kernel.org>   wrote:
>>>> On Fri, May 4, 2012 at 11:48 AM, Greg KH<gregkh@linuxfoundation.org>   wrote:
>>>>> Ok, it looks like these are the devices that we are missing while
>>>>> "hotplugging":
>>>>> 37:00.0 PCI bridge: Intel Corporation Device 1513
>>>>> 38:00.0 PCI bridge: Intel Corporation Device 1513
>>>>> 38:01.0 PCI bridge: Intel Corporation Device 1513
>>>>> 38:02.0 PCI bridge: Intel Corporation Device 1513
>>>>> 38:03.0 PCI bridge: Intel Corporation Device 1513
>>>>> 38:04.0 PCI bridge: Intel Corporation Device 1513
>>>>> 38:05.0 PCI bridge: Intel Corporation Device 1513
>>>>> 39:00.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
>>>>> 3a:03.0 PCI bridge: Pericom Semiconductor Device 400c (rev 02)
>>>>> 3b:00.0 USB controller: Pericom Semiconductor Device 400e (rev 01)
>>>>> 3b:00.1 USB controller: Pericom Semiconductor Device 400e (rev 01)
>>>>> 3b:00.2 USB controller: Pericom Semiconductor Device 400f (rev 01)
>>>>> 3c:00.0 Ethernet controller: Broadcom Corporation NetXtreme BCM57761 Gigabit Ethernet PCIe (rev 10)
>>>>> 3d:00.0 FireWire (IEEE 1394): nee Agere Systems nee Lucent
>>>>> Microelectronics FW643 PCI Express 1394b Controller (PHY/Link) (rev 08)
>>>>>
>>>>> Which is good, I was worried that there was a video controller out there
>>>>> on the end of the thunderbolt link.
>>>>>
>>>>> So, the problem really is two here:
>>>>>    - your video card doesn't detect the new display
>>>>>    - your pci devices do not show up
>>>>>
>>>>> The first one probably needs to be reported to the Intel video driver
>>>>> people, they can help you out the best.
>> no it's definitely not the first one. As you saw, there'sa whole bunch of other devices (3 USB, firewire, network) that do not show up either.
>>
>> The new info from Yinghai (thanks Yinghai) explains why hotpluging with pciehp did not work/detect the PCIE hotplug and did not result in *any* kernel messages.
>>
>> So the whole thing I guess boils down to making acpiphp work. And why it outputs the "no such device" error message and does not load:
>>
>> I tried with debug=1 and got a bit more info in the log files:
>>
>> [root@poliahu frigaut]# lsmod | grep -e acpi -e pci
>> acpi_cpufreq            5941  0
>> mperf                   1275  1 acpi_cpufreq
>> processor              26144  1 acpi_cpufreq
>> [root@poliahu frigaut]# modprobe acpiphp debug=1
>> ERROR: could not insert 'acpiphp': No such device
>>
>> at this point this message showed up in everything.log:
>> May  5 14:09:55 localhost kernel: [28329.700022] pci_hotplug: PCI Hot Plug PCI Core version: 0.5
>> May  5 14:09:55 localhost kernel: [28329.702041] acpiphp: ACPI Hot Plug PCI Controller Driver version: 0.5
>> May  5 14:09:55 localhost kernel: [28329.702484] acpiphp_glue: found PCI-to-PCI bridge at PCI 0000:04:00.0
>> May  5 14:09:55 localhost kernel: [28329.702715] acpiphp_glue: Bus 0000:05 has 0 slots
>> May  5 14:09:55 localhost kernel: [28329.702720] acpiphp_glue: Total 0 slots
>>
>> then:
>> [root@poliahu frigaut]# lsmod | grep -e acpi -e pci
>> pci_hotplug            25324  0
>> acpi_cpufreq            5941  0
>> mperf                   1275  1 acpi_cpufreq
>> processor              26144  1 acpi_cpufreq
>> [root@poliahu frigaut]#
>>
>> shows that acpiphp was indeed not loaded (I assume the error occurs after it has written out the above messages in the log file), but pci_hotplug was as a dep of acpiphp I assume (thus, correct behavior).
>>
>> So we're back to the issue of the acpiphp load error.
>>
>> Greg, I see you're one of the authors of acpiphp:
>> #define DRIVER_VERSION    "0.5"
>> #define DRIVER_AUTHOR    "Greg Kroah-Hartman<gregkh@us.ibm.com>, Takayoshi Kochi<t-kochi@bq.jp.nec.com>, Matthew Wilcox<willy@hp.com>"
>> #define DRIVER_DESC    "ACPI Hot Plug PCI Controller Driver"
>>
>> but that might have been a while back?
>>
>> Cheers, and thanks,
>> Francois
>>
>>
>>
>>>>> The second one means that something is up with the pci hotplug
>>>>> controller.  I've cced the linux-pci list, the people there should be
>>>>> able to help out better than I, as I no longer have any PCI hotplug
>>>>> hardware to work with anymore.
>>>>>
>>>> those devices are under 04:04.0, but that does not have slot cap
>>>> enabled by BIOS. So pciehp is going to to help.
>>> should be: pciehp is not going to help.
>>>
>>> so acpiphp are involved?
>>>
>>> Yinghai
>> -- 
>> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>> Please read the FAQ at  http://www.tux.org/lkml/

Re: Linux Support for Thunderbolt using Apple Monitor

[Francois Rigaut]

On 05/05/12 04:48, Greg KH wrote:
> Oh, what exact machine is this? Maybe I can ask around and see if I 
> can dig up one myself, and hey, I always need an excuse to get a new 
> display :) thanks, greg k-h 
The machine is a macbook air 13" mid 2011, with a thunderbolt port 
(physically identical to a mini display port). I hear Linus is using a 
MBA 11", so I don't feel alone here running linux on this thing.
The display is a 27" Apple thunderbolt display. It's a gorgeous display 
with deep colors and lots of dynamics, and not (much) more expensive 
than other 27" display. But it doesn't hot plug (in linux, yet) and the 
only input is through the thunderbolt cable so no luck in using 
alternatives like HDMI (neither VGA, but VGA doesn't support the display 
resolution 2560x1440 anyway).

Good luck,
Francois

Re: Linux Support for Thunderbolt using Apple Monitor

[Greg KH]

On Sat, May 05, 2012 at 10:28:51AM +1000, Francois Rigaut wrote:
> 
> 
> On 05/05/12 04:48, Greg KH wrote:
> >Oh, what exact machine is this? Maybe I can ask around and see if
> >I can dig up one myself, and hey, I always need an excuse to get a
> >new display :) thanks, greg k-h
> The machine is a macbook air 13" mid 2011, with a thunderbolt port
> (physically identical to a mini display port). I hear Linus is using
> a MBA 11", so I don't feel alone here running linux on this thing.
> The display is a 27" Apple thunderbolt display. It's a gorgeous
> display with deep colors and lots of dynamics, and not (much) more
> expensive than other 27" display. But it doesn't hot plug (in linux,
> yet) and the only input is through the thunderbolt cable so no luck
> in using alternatives like HDMI (neither VGA, but VGA doesn't
> support the display resolution 2560x1440 anyway).

Has anyone tried a Thunderbolt hard drive yet?  Not that I think that
would work better, just curious.

Thanks for the device type, I'll see if I can track one down...

greg k-h