Re: panic: kmem_malloc(131072): kmem_map too small (AMD64)

On Sat, Sep 22, 2007 at 11:37:37PM +0200, Kris Kennaway wrote:
> Darren Reed wrote:
>> Kris Kennaway wrote:
>>> Larry Rosenman wrote:
>>>> On Fri, 21 Sep 2007, Kris Kennaway wrote:
>>>> 
>>>>> Larry Rosenman wrote:
>>>>>> I'm a heavy ZFS user, and got the following panic on 2007-09-18 
>>>>>> source/world:
>>>>> 
>>>>> This is a FAQ, please see the archives (you need to increase the 
>>>>> vm.kmem_size to provide more memory to ZFS).
>>>> 
>>>> I thought that was only for i386, and it hadn't been an issue before.
>>> 
>>> Nope.  It is also load-dependent.
>> So I just received this courtesy of ZFS:
>> panic: kmem_malloc(131072): kmem_map too small: 343027712 total allocated
>> cpuid = 0
>> KDB: enter: panic
>> This was with these settings in loader.conf:
>> vm.kmem_size=419430400
>> vm.kmem_size_max=419430400

Setting both doesn't make sense; kmem_size will take over.
kmem_size_max only limits the automatically computed value
of kmem_size, which is:

/*
 * How many physical pages per KVA page allocated.
 * min(max(max(VM_KMEM_SIZE, Physical memory/VM_KMEM_SIZE_SCALE),
 *     VM_KMEM_SIZE_MIN), VM_KMEM_SIZE_MAX)
 * is the total KVA space allocated for kmem_map.
 */

>> vfs.zfs.arc_max=409715200
>> (That's 400M, 400M and 40M, respectively.)
>> Stupid question, perhaps, but is vm.kmem_size/vm.kmem_size_max limited by 
>> physical RAM?
> 
> Yes.
> 
To be precise, it's actually limited by 2 * sizeof(physical RAM).
It's still size of a _virtual_ memory map (kmem_map), after all:

:	/*
: 	 * Limit kmem virtual size to twice the physical memory.
: 	 * This allows for kmem map sparseness, but limits the size
: 	 * to something sane. Be careful to not overflow the 32bit
: 	 * ints while doing the check.
: 	 */
: 	if (((vm_kmem_size / 2) / PAGE_SIZE) > cnt.v_page_count)
: 		vm_kmem_size = 2 * cnt.v_page_count * PAGE_SIZE;


Cheers,
-- 
Ruslan Ermilov
ru_at_FreeBSD.org
FreeBSD committer