Hello, I have 3 ATSC tuner cards installed on a server machine (with a server motherboard) and I'm having problems getting all 3 to work.
I have a Hauppauge WinTV-HVR-1600 (PCI card, not PCIe card) installed, and I get the following dmesg error upon booting:  "cx18-0: Failed to register irq -22".  The same card installe in another non-server (desktop motherboard) machine is detected correctly and seems to work.
I also have 2 PCIe (not PCI) Hauppauge WinTV-HVR-1250 cards installed and they are not recognized at all (no mention at all in dmesg, lspci, lshw, etc.).  So, I'd would like to know if both of these problems are related in some way and how to fix both issues.
I also tried installing each card individually on that server machine, and yet, I get the same problem.  I also tried adding "pci=bios" and "pci=biosirq" kernel options upon boot, and that did not fix the problem either.
The HVR-1250 cards are handled by the cx23885 driver, the HVR-1600 cards are handled by the cx18 driver; so the problems are most likely  unrelated.
The "cx18-0: Failed to register irq -22" message is because the kernel denied the cx18 driver request for an IRQ line for you first HVR-1600 card.
This is likely because the cx18 driver is aksing for, and can handle, as shared IRQ using the IRQF_SHARED flag, ...
but some other driver loaded first, got that IRQ line and doesn't not handle sharing the PCI irq:
http://lxr.free-electrons.com/source/kernel/irq/manage.c#L1121
right, I am aware of that, but i'm not sure its a problem with the individual tuner device drivers, but the lower level PCI detection.   since this is a server machine (my very first server machine on which I have tried GNU/Linux), I wonder if there is some sort of manual configuration of PCI IRQ in the BIOS that is necessary or if any kernel options need to be added that would fix the problem. 
It could be a device driver for something on the server moptherboard, or a device driver for something in a conventional PCI slot, or some legacy ISA slot.
Well, you might wan't to check the dmesg or turn on some kernel debug to see which PCI devices are getting assigned which legacy PCI interrupt lines (INTA, INTB, INTC, INTD) and what PCI slot they are in.
The PCI INTx lines get rotated between slots and most PCI cards can only use the INTA line.
Which is why I would blame a device on the motherboard and its driver not wanting to share the line.
Which maybe you could fix in the BIOS, if it has some PCI INTx line routing options.
IDK.
YOu could also tell the kernel to disable the offending device driver for the other device, if you can figure out which driver it is.
between cat /proc/interrupts and lspci -vvv, you may be able to figure it out.
Maybe.
hmmm...  how do I enable the debugging for the cx18?  also, keep in mind that I'm not getting any evidence of any detection of the Hauppauge WinTV-HVR-1250 PCIe cards in any of the logs (dmesg, lspci, lshw, etc.).   
I should mention that those Hauppauge WinTV-HVR-1250 PCIe 1x cards are half-height cards that I have insalled in the the PCIe 16x full-height slots by removing the plate on the front of the card.  I am told that by removing the front plate, it should not in any way affect the functionality of the card
The cx23885 driver is a different issue.
if lspci shows the CX2388[578] chip and lsmod doesn't show the cx23885 driver, that's a problem
so if you modinfo cx18, you'll get information on the debug flags for the cx18 debug= module parameter.
Right, what could be causing the complete non-detection of the Hauppauge WinTV-HVR-1250 PCIe cards, because that is the more serious problem... I figure since the Hauppauge 1600 PCI is at least detected, I would eventually be able to get that to work, by finding some solution...  but I'm more concerned about the complete non-detection of the Hauppauge WinTV-HVR-1250 PCIe cards which is supposed to have full support under the lin
That won't help with the IRQ allocation problem with the kernel.
Simple, does lspci show the cards?
no
Then the kernel won't load the driver, nor can it configure the hardware that it can't see. :)
:(
lspci does not show either of the 2 Hauppauge WinTV-HVR-1250 PCIe cards, but does show the Hauppauge WinTV-HVR-1600, with dmesg showing the relevant errors for the Hauppauge WinTV-HVR-1600 initialization.
SO you've got a BIOS setting problem, a power problem, or a motherboard problem when it comes to not seeing the CX2388[578] chips of the HVR-1250's.
Or the HVR-1250's are all defective. (probably unlikely)
wait, but I read that Hauppauge WinTV-HVR-1250 PCIe cards are fully supported:  http://linuxtv.org/wiki/index.php/Hauppauge_WinTV-HVR-1250  
lspci should always show the devices, even if the kernel doesn't have the driver compiled.
If the kernel can't see the hardware electrically, how do you propose it run the hardware?
You have a problem with your PCIe bus or your HVR-1250's
so, could it be that some motherboard/BIOS configuration is preventing the kernel for detecting any PCIe cards?  or worse, is it possible that this server motherboard has features (for example, chipsets, etc.) that are not supported by the linux kernel?
I should mention that this is a brand new server machine (and a fairly recent mid-2014 release) as well.
Yes, to both.
I am currently running a desktop distribution of linux (mythbuntu).  does the server versions of ubuntu/derivitives have more drivers for better server motherboard support?
Also 5 add in cards consume a bit of power.  Make sure your power supply can provided suffcient power.
Kernel is usually the same.
Especially when it comes to PCI chipsets.
Oh, i'm fairly certain that it is not an issue with power, because this is a fairly bare-bones system, with only one HDD and 1 optical drive installed and a farily hefty 650 watt PSU.
1 don't have 5 add-in cards, 1 only have 3.
2 1250 PCIe, and 1 1600 PCI
 also, it has only 1 (of 12 total) memory module and only 1 (of 2) Xeon physical processors  installed.
but let me at least confirm 2 things for now:    (1)  it is perfectly ok to install (and get working) a PCIe 1x card, into a PCIe 8x slot on the motherboard - is this correct?
(2)  It is ok to install a half-height PCIe card into a full-height PCIe form-factor computer case, by removing the front metal plate on the PCIe card itself (since there is a simple screw that attaches that front plate in place) - is this correct?
awalls:  for some reason, by changing the position of one of the WinTV-HVR-1250s to the lowest possible PCIe slot, I was able to get only that card to be loaded/initialized successfully.  I will try next to move the other PCIe WinTV-HVR-1250 card the next consecutive slot right above the first one, to see if it will make that work as well.  So, it seems that the problem is not with the lack of motherboard/chipset support by the 
still no success with the PCI WinTV-HVR-1600 though, but since that is at least recognized by the lspci, I hope that I will evetnually figure that out as well...
qwebirc45607: random guess from my side: PCI(e) lanes are provided by the CPU - since you're using a dual socket board you might have put the card into a slot which is connected to the second CPU slot
xdarklight:  wow.  are you suggesting that since it is a dual-socket (dual physical CPU) motherboard, that each physical CPU only has access to dedicated PCIe slots - that is, probably half of the PCIe slots dedicated to one physical CPU and the other half of the PCIe slots dedicated to the other physical CPU?
I never had a dual socket mainboard, but that's what I assume (that's why I said "random guess")
random google search shows this image: http://superuser.com/questions/529975/limitations-due-to-running-dual-socket-mainboard-with-one-cpu - which seems to confirm my speculations
Yeah.  that seems to be make sense, although I should probably check the documentation on this hardware to make sure....  I knew that the memory banks were segregated in half in that manner, but I didn't even stop to think that it would be the same case with the PCIe slots.  The curious thing is there are an odd number of PCIe slots - 5.   4 of 8xPCIe and 1 of 16xPCIe.  So one processor seems to have access to one more than the 
oh actually, to be sure, there an even number of PCI+PCIe slots,  1 PCI slot, 4 PCIe8x and 1 PCIe16x.  So that would suggest that they are probably divided evenly between the 2 CPUs.
yes that's common
ok, my next question is:  how many PCI and/or PCIe ATSC tuner cards can be successfully installed in any machine with a recent linux kernel version (3.x OR 4.x) and thus be used simultaneously to record different channels simultaneously.  Does the kernel and/or driver(s) and/or chipset/hardware have any limits (besides the total number of physical slots available and the segregated allocation to separate CPUs, that is).
(and by ATSC, I really mean both ATSC and QAM)