Thread (82 messages) read the whole thread 82 messages, 5 authors, 2017-07-19

Re: [PATCH v4 1/7] service cores: header and implementation

From: Jerin Jacob <hidden>
Date: 2017-07-11 12:45:23

-----Original Message-----
Date: Tue, 11 Jul 2017 12:32:58 +0000
From: "Van Haaren, Harry" <redacted>
To: Thomas Monjalon <redacted>, Jerin Jacob
 [off-list ref]
CC: "dev@dpdk.org" <redacted>, "Wiles, Keith" <redacted>,
 "Richardson, Bruce" [off-list ref]
Subject: RE: [PATCH v4 1/7] service cores: header and implementation
quoted
From: Thomas Monjalon [mailto:thomas@monjalon.net]
Sent: Tuesday, July 11, 2017 10:55 AM
To: Jerin Jacob <redacted>; Van Haaren, Harry
[off-list ref]
Cc: dev@dpdk.org; Wiles, Keith <redacted>; Richardson, Bruce
[off-list ref]
Subject: Re: [PATCH v4 1/7] service cores: header and implementation

11/07/2017 10:29, Jerin Jacob:
quoted
IMO, We don't need to expose rte_service_private.h to application. If
you agree, add the following or similar change
If it must not be exposed, the file should not have the prefix rte_
In doc/api/doxy-api.conf, every files with rte_ prefix will be processed
for doxygen documentation:
	FILE_PATTERNS = rte_*.h

The service registration API should be exposed to the application.

Imagine a use case where the application wants to run services *and* an application specific function on the same core.  In the current implementation this is possible, as the application can register a service. The app then configures all services (including its own "app-service") to run on a service lcore.

If we hide the service registration from the application, we make it impossible for the application to multiplex services and application specific workloads on a single core.
Then we could move the registration functions to service.h.
IMO, It does not look correct if we expose _prviate.h to application or
we could rename to service_component.h or something like that.

I strongly prefer of leaving the header as is. Given we have EXPERIMENTAL tag, ABI/API are not a concern until later - we have time to figure out if the service-registration API is good enough in current form, before we commit to it.

I'll send v5 asap with headers as is.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help