[linux-dvb] HVR-1800 - can't find the card
awalls at radix.net
Sat Sep 20 00:52:46 CEST 2008
On Fri, 2008-09-19 at 13:34 -0400, Steven Toth wrote:
> Bill McCartney wrote:
> > Well, I'm not sure if it is a DVB problem. The hvr-1800 shows up in my
> > lspci, but driver doesn't load. I have tried kernels version 2.26.24 and
> > 184.108.40.206 <http://220.127.116.11>, removed all other cards from the system -
> > and still have the same problem.
> > Output when I try to install the driver
> > cx23885 driver version 0.0.1 loaded
> > cx23885: can't get MMIO memory @ 0x0
> > CORE cx23885 No more PCIe resources for subsystem: 0070:7801
> > cx23885: probe of 0000:03:00.0 failed with error -22
> > The output of my lspci -v (of the card)
> > 03:00.0 Multimedia video controller: Conexant Unknown device 8880 (rev 0f)
> > Subsystem: Hauppauge computer works Inc. Unknown device 7801
> > Flags: bus master, fast devsel, latency 0, IRQ 10
> > Memory at <ignored> (64-bit, non-prefetchable)
> > Capabilities:  Express Endpoint IRQ 0
> > Capabilities:  Power Management version 2
> > Capabilities:  Vital Product Data
> > Capabilities: [a0] Message Signalled Interrupts: 64bit+
> > Queue=0/0 Enable-
> > In my kernel logs I see this from the bootlogs:
> > PCI: Cannot allocate resource region 0 of device 0000:03:00.0
> > I've tried several kernel options as far as pci configuration goes --
> > does this mean that I have bad hardware? Should I just return it to the
> > store? Is it a conflict with my motherboard?
> Returning this to the store isn't going to fix the problem. You have a
> limited about of memory allocated to PCI devices and you have no more
> free for the HVR1800.
> From memory the kernel allocates 512MB (?) for PCI address space, which
> gets allocated out to devices. The HVR1800 needs 64MB(?) of address
> space, and the kernel doesn't have that free. (My numbers are sketchy
> but the same principle stands).
> Check lspci -vn and see which of your devices need large amounts of ram
> (typically video cards).
If you're trying to troubleshoot vmalloc address space exhaustion,
before you pull hardware, you may want to try these commands:
$ cat /proc/iomem
will show you what vmalloc allocations are in use for memory mapped IO
by what drivers for what PCI bus id. You'll have to compute sizes
yourself, if you need to know.
$ grep Vmalloc /proc/meminfo
will show you how much vmalloc address space the kernel sets aside and
how much is currently in use.
I know the cx18 driver, with which I am familiar, needs to grab a 64 MiB
continuous chunk for every card. Since the vmalloc address space is
dynamically allocated and deallocated, it can fragment and it can be the
case that a large chunk doesn't exist.
I normally tell cx18 users with ENOMEM problems to add a 'vmalloc=xxxM'
option to their kernel command line that is at least 128 MiB higher than
what their kernel sets aside by default. You might want to try the
vmalloc allocations are also used for module executable code and
probably other things aside from just memory mapped IO.
> For test purposes, removing another card to free some PCI space, this
> will allow you to test the hardware, loading the driver after this
> should be fine. (Again, for test purposes only).
> Increasing the amount of PCI ram available for allocation may be a
> kernel boot-time setting. I haven't look to be honest, as I don't
> experience this issue.... although occasionally other user have (with
> different cards and different driver trees).
> I'm traveling this week so I have very little access to source code and
> hardware numbers, so if anyone has better numbers or can increase the
> PCI allocation then I'd be interested to hear their comments.
> - Steve
More information about the linux-dvb