__you_cannot_kmalloc_that_much

__you_cannot_kmalloc_that_much

[Luck, Tony]

While trying to compile the latest pull from
the lia64.bkbits.net:8080/linux-ia64-2.5
repository I hit the "you_cannot_kmalloc_that_much"
message in the final link.

The culprit is in

	drivers/usb/host/uhci-debug.c

which contains:

#define MAX_OUTPUT      (PAGE_SIZE * 16)

	...

	up->data = kmalloc(MAX_OUTPUT, GFP_KERNEL);

With the default 16k page size we are trying to
allocate 256K of memory, which is bigger that the
largest size pool in include/linux/kmalloc_sizes.h


Do we need a larger pool (I don't think so) or is
this allocation excessive?

-Tony

Re: __you_cannot_kmalloc_that_much

[David Mosberger]

>>>>> On Thu, 19 Jun 2003 11:33:59 -0700, "Luck, Tony" <tony.luck@intel.com> said:

  Tony> While trying to compile the latest pull from
  Tony> the lia64.bkbits.net:8080/linux-ia64-2.5
  Tony> repository I hit the "you_cannot_kmalloc_that_much"
  Tony> message in the final link.

  Tony> The culprit is in

  Tony> drivers/usb/host/uhci-debug.c

  Tony> which contains:

  Tony> #define MAX_OUTPUT      (PAGE_SIZE * 16)

  Tony> ...

  up-> data = kmalloc(MAX_OUTPUT, GFP_KERNEL);

  Tony> With the default 16k page size we are trying to
  Tony> allocate 256K of memory, which is bigger that the
  Tony> largest size pool in include/linux/kmalloc_sizes.h


  Tony> Do we need a larger pool (I don't think so) or is
  Tony> this allocation excessive?

Clearly the latter.  If it works with 4KB page size on x86, certainly
we don't need 4 or 8 times more space on ia64.

Can you take this up directly with the author of the offending line?

Thanks,

	--david

RE: __you_cannot_kmalloc_that_much

[Luck, Tony]

>   Tony> #define MAX_OUTPUT      (PAGE_SIZE * 16)
> 
>   Tony> ...
> 
>   up-> data = kmalloc(MAX_OUTPUT, GFP_KERNEL);
> 
>   Tony> With the default 16k page size we are trying to
>   Tony> allocate 256K of memory, which is bigger that the
>   Tony> largest size pool in include/linux/kmalloc_sizes.h
> 
> 
>   Tony> Do we need a larger pool (I don't think so) or is
>   Tony> this allocation excessive?
> 
> Clearly the latter.  If it works with 4KB page size on x86, certainly
> we don't need 4 or 8 times more space on ia64.
> 
> Can you take this up directly with the author of the offending line?

Ok, I sent an e-mail to Johannes Erdfelt (with a Cc: to the
linux-usb-users mailing list) with a patch to switch the size
to a fixed 64k.

But I'm still not out of the woods ... with this patch applied
I get a:

undefined reference to `local symbols in discarded section .exit.text'

message.  Is this related to the warning about using an old linker?

gcc: 3.2
as: GNU assembler version 2.13.90
ld: GNU ld version 2.13.90.0.2 20020802

-Tony

RE: __you_cannot_kmalloc_that_much

[David Mosberger]

>>>>> On Thu, 19 Jun 2003 12:03:43 -0700, "Luck, Tony" <tony.luck@intel.com> said:

  Tony> undefined reference to `local symbols in discarded section .exit.text'

  Tony> message.  Is this related to the warning about using an old linker?

Not likely.  It sounds like a driver is failing to use __exit_p() (see
linux/init.h).

	--david