On Fri, Oct 18, 2019 at 12:33:27AM +0200, Alexander Bluhm wrote:
> Can we replace the mutex that protects the mbuf allocation limit
> by an atomic operation?
Now without unneccessary variable and no extra check in the slow path.
- Code gets shorter
- Avoid mutex, makes it less complex
- it is
On Fri, Oct 18, 2019 at 03:55:39PM +1000, David Gwynne wrote:
> why? is it significantly faster? page allocation should be in the slow path.
At least it was not slower. Performance were slightly higher, but
changes are well below meassurement tolerance.
Usually it is a good idea to avoid locks
why? is it significantly faster? page allocation should be in the slow path.
dlg
> On 18 Oct 2019, at 08:33, Alexander Bluhm wrote:
>
> Hi,
>
> Can we replace the mutex that protects the mbuf allocation limit
> by an atomic operation?
>
> ok?
>
> bluhm
>
> Index: kern/uipc_mbuf.c
>