Change in mmap driver performance from 2..6.38 to 2.6.39

Change in mmap driver performance from 2..6.38 to 2.6.39

[Philip Balister]

I have a device drier that uses a mmap'd section of memory to expose dma 
buffers to user space. 
(https://github.com/balister/linux-omap-philip/blob/e100-2.6.38/drivers/misc/usrp_e.c).

I have it running on the 2.6.38 and 2.6.39 kernel, but it looks like 
read performance is much slower in the .39 version than the .28 version.

Does anyone know what changed between these two kernel versions and how 
it might fix it?

Thanks,

Philip

Re: Change in mmap driver performance from 2..6.38 to 2.6.39

[Tony Lindgren]

* Philip Balister <philip@balister.org> [110711 07:23]:
> I have a device drier that uses a mmap'd section of memory to expose
> dma buffers to user space. (https://github.com/balister/linux-omap-philip/blob/e100-2.6.38/drivers/misc/usrp_e.c).
> 
> I have it running on the 2.6.38 and 2.6.39 kernel, but it looks like
> read performance is much slower in the .39 version than the .28
> version.
> 
> Does anyone know what changed between these two kernel versions and
> how it might fix it?

Maybe try git bisect it to some commit? You can use quilt to keep
applying the temporary patches needed in between test builds for
your driver.. Or maybe git allows that somehow too.

Tony