Re: [PATCH v4 3/6] ip_pipeline: fix lcore mapping for varying SMT threads as in ppc64

On Friday 12 August 2016 03:45 PM, Chao Zhu wrote:

Another comment is, comment out lcore_socket_id check will influence other architectures. If possible, I would like to make this change to Power specific.

Hi Chao,
I am revisiting cpu_core_map_init() fn. I realize, all we handle is max values of socket/core/ht.
So, ideally the function should not break in any arch.

In my quick test w/o adjusting prev max values (4,32,4,0), only for ht > 1, creating cpu map panics
in ppc i.e (4,32,1,0) is able to create core map and app runs. I continue to debug this and fix
this example app separately.

So, I will be sending powerpc specific enablement of lpm,acl,port,table and sched in v5.
Example ip_pipeline will be fixed in separate patch.

Thanks,
Gowrishankar

-----Original Message-----
From: gowrishankar muthukrishnan [mailto:gowrishankar.m@linux.vnet.ibm.com]
Sent: 2016年8月12日 17:00
To: Chao Zhu <redacted>
Cc: dev@dpdk.org; 'Bruce Richardson' <redacted>; 'Konstantin Ananyev' <redacted>; 'Thomas Monjalon' <redacted>; 'Cristian Dumitrescu' <redacted>; 'Pradeep' <redacted>
Subject: Re: [PATCH v4 3/6] ip_pipeline: fix lcore mapping for varying SMT threads as in ppc64

On Friday 12 August 2016 02:14 PM, Chao Zhu wrote:

quoted

Gowrishankar,

I suggest to set the following value:

n_max_cores_per_socket = 8
n_max_ht_per_core = 8

This will cover most of the Power8 servers.
Any comments?

Sure Chao. I will include this change in v5. If there are no other comments, I can spin out v5, with changes in this patch.

Regards,
Gowrishankar

quoted

-----Original Message-----
From: gowrishankar muthukrishnan
[mailto:gowrishankar.m@linux.vnet.ibm.com]
Sent: 2016年8月11日 20:02
To: Chao Zhu <redacted>
Cc: dev@dpdk.org; 'Bruce Richardson' <redacted>;
'Konstantin Ananyev' [off-list ref]; 'Thomas Monjalon'
[off-list ref]; 'Cristian Dumitrescu'
[off-list ref]; 'Pradeep' [off-list ref]
Subject: Re: [PATCH v4 3/6] ip_pipeline: fix lcore mapping for varying
SMT threads as in ppc64

On Thursday 11 August 2016 03:59 PM, Chao Zhu wrote:

quoted

Gowrishankar,

Thanks for the detail.
If my understanding is correct, Power8 has different chips. Some of the OpenPOWER chips have 8 cores per socket. And the max threads per core is 8. Should we support this in cpu_core_map_init()?

Here's a dump from the OpenPOWER system.
======================================
# lscpu
Architecture:          ppc64le
Byte Order:            Little Endian
CPU(s):                64
On-line CPU(s) list:   0,8,16,24,32,40,48,56
Off-line CPU(s) list:  1-7,9-15,17-23,25-31,33-39,41-47,49-55,57-63
Thread(s) per core:    1
Core(s) per socket:    8
Socket(s):             1
NUMA node(s):          1
Model:                 unknown
L1d cache:             64K
L1i cache:             32K
L2 cache:              512K
L3 cache:              8192K
NUMA node0 CPU(s):     0,8,16,24,32,40,48,56
=========================================

quoted

+#if defined(RTE_ARCH_PPC_64)
+	app->core_map = cpu_core_map_init(2, 5, 1, 0); #else

This value seems quite strange. Can you give more detail?

Based on config of tested server (as below output),

CPU(s):                80
On-line CPU(s) list:   0,8,16,24,32,40,48,56,64,72
Off-line CPU(s) list:
1-7,9-15,17-23,25-31,33-39,41-47,49-55,57-63,65-71,73-79
Thread(s) per core:    1  <<<
Core(s) per socket:    5   <<<
Socket(s):             2       <<<
NUMA node(s):          2

cpu_core_map_init parameters (2,5,1,0) were prepared. Instead, we can cap max sockets/core/ht counts to possible maximum supported today.

Regards,
Gowrishankar

quoted

quoted

     	app->core_map = cpu_core_map_init(4, 32, 4, 0);
+#endif

-----Original Message-----
From: gowrishankar muthukrishnan
[mailto:gowrishankar.m@linux.vnet.ibm.com]
Sent: 2016年8月9日 19:14
To: Chao Zhu <redacted>; dev@dpdk.org
Cc: 'Bruce Richardson' <redacted>; 'Konstantin
Ananyev' [off-list ref]; 'Thomas Monjalon'
[off-list ref]; 'Cristian Dumitrescu'
[off-list ref]; 'Pradeep' [off-list ref]
Subject: Re: [PATCH v4 3/6] ip_pipeline: fix lcore mapping for
varying SMT threads as in ppc64

Hi Chao,
Sure. Please find below one.

This patch fixes ip_pipeline panic in app_init_core_map while preparing cpu core map in powerpc with SMT off. cpu_core_map_compute_linux currently prepares core mapping based on file existence in sysfs ie.

/sys/devices/system/cpu/cpu<LCORE_NUM>/topology/physical_package_id
      /sys/devices/system/cpu/cpu<LCORE_NUM>/topology/core_id

These files do not exist for lcores which are offline for any reason (as in powerpc, while SMT is off). In this situation, this function should further continue preparing map for other online lcores instead of returning with -1 for a first unavailable lcore.

Also, in SMT=off scenario for powerpc, lcore ids can not be always
indexed from
0 upto 'number of cores present' (/sys/devices/system/cpu/present).
For eg, for an online lcore 32, core_id returned in sysfs is 112
where online lcores are
10 (as in one configuration), hence sysfs lcore id can not be checked with indexing lcore number before positioning lcore map array.

Thanks,
Gowrishankar

On Tuesday 09 August 2016 02:37 PM, Chao Zhu wrote:

quoted

Gowrishankar,

Can you give more description about this patch?
Thank you!

-----Original Message-----
From: Gowrishankar Muthukrishnan
[mailto:gowrishankar.m@linux.vnet.ibm.com]
Sent: 2016年8月6日 20:33
To: dev@dpdk.org
Cc: Chao Zhu <redacted>; Bruce Richardson
[off-list ref]; Konstantin Ananyev
[off-list ref]; Thomas Monjalon
[off-list ref]; Cristian Dumitrescu
[off-list ref]; Pradeep [off-list ref];
gowrishankar [off-list ref]
Subject: [PATCH v4 3/6] ip_pipeline: fix lcore mapping for varying
SMT threads as in ppc64

From: gowrishankar <redacted>

offline lcore would still refer to original core id and this has to
be considered while creating cpu core mask.

Signed-off-by: Gowrishankar <redacted>
---
     config/defconfig_ppc_64-power8-linuxapp-gcc |  3 ---
     examples/ip_pipeline/cpu_core_map.c         | 12 +-----------
     examples/ip_pipeline/init.c                 |  4 ++++
     3 files changed, 5 insertions(+), 14 deletions(-)

diff --git a/config/defconfig_ppc_64-power8-linuxapp-gcc

b/config/defconfig_ppc_64-power8-linuxapp-gcc
index dede34f..a084672 100644

--- a/config/defconfig_ppc_64-power8-linuxapp-gcc
+++ b/config/defconfig_ppc_64-power8-linuxapp-gcc

@@ -58,6 +58,3 @@ CONFIG_RTE_LIBRTE_FM10K_PMD=n

     # This following libraries are not available on Power. So

they're turned off.
     CONFIG_RTE_LIBRTE_SCHED=n
-CONFIG_RTE_LIBRTE_PORT=n
-CONFIG_RTE_LIBRTE_TABLE=n
-CONFIG_RTE_LIBRTE_PIPELINE=n

diff --git a/examples/ip_pipeline/cpu_core_map.c

b/examples/ip_pipeline/cpu_core_map.c
index cb088b1..482e68e 100644

--- a/examples/ip_pipeline/cpu_core_map.c
+++ b/examples/ip_pipeline/cpu_core_map.c

@@ -351,9 +351,6 @@ cpu_core_map_compute_linux(struct cpu_core_map *map)
     			int lcore_socket_id =
     				cpu_core_map_get_socket_id_linux(lcore_id);

-			if (lcore_socket_id < 0)
-				return -1;
-
     			if (((uint32_t) lcore_socket_id) == socket_id)
     				n_detected++;
     		}

@@ -368,18 +365,11 @@ cpu_core_map_compute_linux(struct cpu_core_map *map)
     					cpu_core_map_get_socket_id_linux(
     					lcore_id);

-				if (lcore_socket_id < 0)
-					return -1;
-

Why to remove the lcore_socket_id check?

     				int lcore_core_id =
     					cpu_core_map_get_core_id_linux(
     						lcore_id);

-				if (lcore_core_id < 0)
-					return -1;
-
-				if (((uint32_t) lcore_socket_id ==
socket_id) &&
-					((uint32_t) lcore_core_id ==
core_id)) {
+				if ((uint32_t) lcore_socket_id == socket_id)
{
     					uint32_t pos = cpu_core_map_pos(map,
     						socket_id,
     						core_id_contig,

diff --git a/examples/ip_pipeline/init.c

b/examples/ip_pipeline/init.c index cd167f6..60c931f 100644

--- a/examples/ip_pipeline/init.c
+++ b/examples/ip_pipeline/init.c

@@ -61,7 +61,11 @@ static void
     app_init_core_map(struct app_params *app)  {
     	APP_LOG(app, HIGH, "Initializing CPU core map ...");
+#if defined(RTE_ARCH_PPC_64)
+	app->core_map = cpu_core_map_init(2, 5, 1, 0); #else

This value seems quite strange. Can you give more detail?

     	app->core_map = cpu_core_map_init(4, 32, 4, 0);
+#endif

     	if (app->core_map == NULL)
     		rte_panic("Cannot create CPU core map\n");
--
1.9.1