Re: [PATCH v1 2/2] powerpc/pseries: Add debugfs interface to retrieve VPHN info

Started by Naveen N. Rao on Dec 13, 2018 · 2 messages from 2 authors · Last reply Dec 14, 2018

open the first message on its own page

Naveen N. Rao 

Hi Michael,

Naveen N. Rao wrote:
Add debugfs interface to retrieve associativity information for lpar
vcpus (debugfs/vphn/lpar) and the hypervisor cpus (debugfs/vphn/hyp).
This information is useful to derive various metrics, including the vcpu
dispatch statistics in a SPLPAR environment.
Any thoughts on this approach vs. adding a tracepoint?


Thanks,
Naveen
quoted hunk
Signed-off-by: Naveen N. Rao <redacted>
---
 arch/powerpc/mm/numa.c | 105 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 105 insertions(+)

diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
index 6677a578f18d..f0b0e87016e6 100644
--- a/arch/powerpc/mm/numa.c
+++ b/arch/powerpc/mm/numa.c
@@ -40,6 +40,7 @@
 #include <asm/setup.h>
 #include <asm/vdso.h>
 #include <asm/drmem.h>
+#include <asm/debugfs.h>
 
 static int numa_enabled = 1;
 
@@ -1089,6 +1090,107 @@ static long hcall_vphn(unsigned long cpu, u64 flags, __be32 *associativity)
 	return rc;
 }
 
+#ifdef CONFIG_DEBUG_FS
+static ssize_t vphn_lpar_cpu_file_read(struct file *filp, char __user *buf,
+		size_t len, loff_t *pos)
+{
+	int cpu = (long)filp->private_data;
+	__be32 associativity[VPHN_ASSOC_BUFSIZE] = {0};
+	int hwcpu = get_hard_smp_processor_id(cpu);
+	long int rc;
+
+	if (len != sizeof(associativity))
+		return -EINVAL;
+
+	rc = hcall_vphn(hwcpu, 1, associativity);
+	if (rc)
+		return -EFAULT;
+
+	rc = copy_to_user(buf, &associativity, sizeof(associativity));
+	if (rc)

101 more quoted lines






  

    

      
      Michael Ellerman      replied to Naveen N. Rao      
      
    

  

  
Hi Naveen,

"Naveen N. Rao" [off-list ref] writes:
Hi Michael,

Naveen N. Rao wrote:
quoted
Add debugfs interface to retrieve associativity information for lpar
vcpus (debugfs/vphn/lpar) and the hypervisor cpus (debugfs/vphn/hyp).
This information is useful to derive various metrics, including the vcpu
dispatch statistics in a SPLPAR environment.
Any thoughts on this approach vs. adding a tracepoint?
Sorry I've been unresponsive on this stuff, I don't know this area that
well.

I guess I'm not opposed to adding some stuff to debugfs, but only if
it's for debugging.

We don't want customers running tools or scripts that rely on this stuff
in debugfs.

If we need to expose more information to be used by production tools
then I think we need to look at sysfs or something like taskstats as a
proper API.

cheers







    
  

  
    

      

        Keyboard shortcuts
      

      

        
          
hback out one level

          
jnext message in thread

          
kprevious message in thread

          
ldrill in

          
Escclose help / fold thread tree

          
?toggle this help