Re: [PATCH] powercap / RAPL : remove dependency on iosf_mbi

From: Austin S Hemmelgarn
Date: Thu Sep 24 2015 - 11:10:22 EST


On 2015-09-24 06:03, Pengyu Ma wrote:


On 09/23/2015 01:01 AM, Jacob Pan wrote:
On Tue, 22 Sep 2015 11:11:36 +0800
Pengyu Ma <pengyu.ma@xxxxxxxxxxxxx> wrote:


On 09/22/2015 05:36 AM, Jacob Pan wrote:
On Mon, 21 Sep 2015 11:48:14 +0800
Pengyu Ma <pengyu.ma@xxxxxxxxxxxxx> wrote:

On 09/18/2015 11:43 PM, Jacob Pan wrote:
On Fri, 18 Sep 2015 02:09:55 +0200
"Rafael J. Wysocki" <rjw@xxxxxxxxxxxxx> wrote:

On Thursday, September 17, 2015 03:31:41 PM Pengyu Ma wrote:
iosf_mbi is supported on Quark, Braswell, Baytrail and some Atom
SoC, but RAPL is not limited to these SoC, it supports almost
Intel CPUs. Remove this dependece to make RAPL support more
Intel CPUs.

Please select IOSF_MBI on Atom SoCs.

Unlike Quark, I don't think we want to or do differentiate Atom
from other x86 at compile time. IOSF driver can be compiled as a
module also, therefore RAPL driver needs this explicit dependency
at compile time.
As commit had exported iosf_mbi to let user use it.

commit aa8e4f22ab7773352ba3895597189b8097f2c307
Author: David E. Box <david.e.box@xxxxxxxxxxxxxxx>
Date: Wed Aug 27 14:40:39 2014 -0700

x86/iosf: Add Kconfig prompt for IOSF_MBI selection


While selecting IOSF_MBI is preferred, it does mean carrying extra
code on non-SoC architectures.

We can NOT force user to build in iosf_mbi if they want use RAPL on
haswell/broadwell/skylake.
And RAPL can be compiled and worked well on
haswell/broadwell/skylake without IOSF_MBI.
RAPL is really NOT depended on IOSF_MBI.

True for haswell/broadwell/skylake platforms. But if we want binary
compatibility for Atom and Core, I can' see how simply removing the
dependency would work, unless we have runtime detection of IOSF.
If you want use iosf_mbi on atom, please select it on generic x86
config. But not force it depend on another feature that not related
on it with other boards.
I don't care how iosf_mbi is added to kernel config, but why should I
be forced to add it if I want use RAPL?
It doesn't make any sense.

I understand your concern about wasting code. But let's look at all the
cases of config options here. (without Kconfig dependency as you
suggested)

RAPL\IOSF Y M N
___________________________________________________
Y OK DC* Warn on Atom**
M OK OK Warn on Atom
N OK OK OK
___________________________________________________

Notes:
* DC: don't compile
** Warn on Atom is runtime if I add the following code to RAPL driver,
but this case is ok.

--- a/drivers/powercap/intel_rapl.c
+++ b/drivers/powercap/intel_rapl.c
@@ -982,6 +982,11 @@ static void set_floor_freq_atom(struct rapl_domain
*rd, bool enable) static u32 power_ctrl_orig_val;
u32 mdata;
+ if (!iosf_mbi_available()) {
+ pr_warn("No IOSF MBI access to set floor frequency\n");
+ return;
+ }
+

So the problematic case is when RAPL=Y IOSF=M
Since real IOSF functions are available when
#if IS_ENABLED(CONFIG_IOSF_MBI)
There will be no dummy functions for RAPL to reference in this case.
iosf_mbi_write/read will warn itself.
Since IOSF is a driver, making it a module is a reasonable requirement.
As I mentioned before, I don't think we want to have a CONFIG_ATOM
option for X86.
Actually there is a CONFIG_MATOM already in Kconfig.cpu
That's for code optimization (it changes compiler flags), not determining what system we're actually building for, and on top of that it's for older atom processors, not the new ones. It's fully possible to build a kernel for an Atom processor without selecting this.

Pengyu

+David, HPA

Jacob

Pengyu

Pengyu
Signed-off-by: Pengyu Ma <pengyu.ma@xxxxxxxxxxxxx>
Jacob?

---
drivers/powercap/Kconfig | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/powercap/Kconfig b/drivers/powercap/Kconfig
index 85727ef..a7c81b5 100644
--- a/drivers/powercap/Kconfig
+++ b/drivers/powercap/Kconfig
@@ -17,7 +17,7 @@ if POWERCAP
# Client driver configurations go here.
config INTEL_RAPL
tristate "Intel RAPL Support"
- depends on X86 && IOSF_MBI
+ depends on X86
default n
---help---
This enables support for the Intel Running Average
Power Limit (RAPL)

[Jacob Pan]
[Jacob Pan]
[Jacob Pan]

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/


Attachment: smime.p7s
Description: S/MIME Cryptographic Signature