Thread (32 messages) flat view 32 messages, 6 authors, 2020-07-23

Re: [PATCH v4 09/10] net: eth: altera: add msgdma prefetcher

From: Westergreen, Dalon <hidden>
Date: 2020-07-14 19:07:57
Also in: lkml


On Tue, 2020-07-14 at 09:29 -0700, Jakub Kicinski wrote:
On Tue, 14 Jul 2020 15:58:53 +0000 Westergreen, Dalon wrote:
quoted
On Tue, 2020-07-14 at 08:55 -0700, Jakub Kicinski wrote:
quoted
On Tue, 14 Jul 2020 14:35:16 +0000 Ooi, Joyce wrote:  
quoted
quoted
I'm no device tree expert but these look like config options rather
than
HW
description. They also don't appear to be documented in the next
patch.    
The poll_freq are part of the msgdma prefetcher IP, whereby it
specifies the frequency of descriptor polling operation. I can add
the poll_freq description in the next patch.  
Is the value decided at the time of synthesis or can the driver choose 
the value it wants?  
It is not controlled at synthesis, this parameter should likely not be a
devicetree parameter, perhaps just make it a module parameter with a default
value.
Let's see if I understand the feature - instead of using a doorbell the
HW periodically checks the contents of the next-to-use descriptor to
see if it contains a valid tx frame or rx buffer?
Yes, it checks the next-to-use descriptor to see if the descriptor is valid.

the value specifies the number of cycles to wait before checking again
I've seen vendors abuse fields of ethtool --coalesce to configure
similar settings. tx-usecs-irq and rx-usecs-irq, I think. Since this
part of ethtool API has been ported to netlink, could we perhaps add 
a new field to ethtool --coalesce?
I don't think this is necessary, i think just having a module parameter
meets our needs.  I don't see a need for the value to change on a per
interface basis.  This was primarily used during testing / bringup.

-dalon
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help