Re: Patch to split kmalloc in sd.c in 2.4.18+

In article <[e-mail]> you wrote:
> Pete Zaitcev wrote:
>> Hello:
>> 
>> One problem I see when trying to use a box with 128 SCSI disks
>> is that sd_mod sometimes refuses to load. Earlier kernels simply
>> oopsed when it happened, but that is fixed in 2.4.18. The root
>> of the evil is the enormous array sd[] that sd_init allocates.
>> Alan suggested to split the allocation, which is what I did.
>> 
>> Arjan said that it may be easier to use vmalloc, and sure it is.
>> However, I heard that vmalloc space is not too big, so it may
>> make sense to conserve it (especially on non-x86 32-bitters).
> 
> kmalloc is spare - the vmalloc space is *HUUUUUGE*.

64Mb (effective usable size) is HUGE. sure. but you share it
with all other vmalloc users. I agree with Pete that kmalloc is the
superior solution; I mentioned vmalloc to him before because that would
have been the minimal fix; Pete decided to fix it for real... 
(also vmalloc is also rather slow for hotpaths due to tlb effects)