Re: [PATCH v8 2/5] powerpc/hv-24x7: Add rtas call in hv-24x7 driver to get processor details

On 5/12/20 2:37 AM, Nathan Lynch wrote:

Hi,

Kajol Jain [off-list ref] writes:

quoted

diff --git a/arch/powerpc/perf/hv-24x7.c b/arch/powerpc/perf/hv-24x7.c
index 48e8f4b17b91..8cf242aad98f 100644
--- a/arch/powerpc/perf/hv-24x7.c
+++ b/arch/powerpc/perf/hv-24x7.c

@@ -20,6 +20,7 @@
 #include <asm/io.h>
 #include <linux/byteorder/generic.h>
 
+#include <asm/rtas.h>
 #include "hv-24x7.h"
 #include "hv-24x7-catalog.h"
 #include "hv-common.h"

@@ -57,6 +58,75 @@ static bool is_physical_domain(unsigned domain)
 	}
 }
 
+/*
+ * The Processor Module Information system parameter allows transferring
+ * of certain processor module information from the platform to the OS.
+ * Refer PAPR+ document to get parameter token value as '43'.
+ */
+
+#define PROCESSOR_MODULE_INFO   43
+#define PROCESSOR_MAX_LENGTH	(8 * 1024)
+
+DEFINE_SPINLOCK(rtas_local_data_buf_lock);
+EXPORT_SYMBOL(rtas_local_data_buf_lock);

This should be static and not exported, correct?

quoted

+
+static u32 phys_sockets;	/* Physical sockets */
+static u32 phys_chipspersocket;	/* Physical chips per socket*/
+static u32 phys_coresperchip; /* Physical cores per chip */
+
+/*
+ * Function read_sys_info_pseries() make a rtas_call which require
+ * data buffer of size 8K. As standard 'rtas_data_buf' is of size
+ * 4K, we are adding new local buffer 'rtas_local_data_buf'.

Sorry if this has been covered before but I don't understand why it
would require a larger buffer; by my reading this call will return *ten
bytes* of output. Also, current versions of PAPR+ limit the output
length to 4002 bytes. I feel like I'm missing something.

Hi Nathan,
	Thanks for reviewing the patch. Actually when I was testing this patch in
both power8 and power9 machine, I got some issue in power9 because of buffer size.
And I checked the buffer size used in util_linux which is 8192. So, I increase the
buffer size.I will again test it as I did couple of changes after that with 4002 size. 

quoted

+ */
+static __be16 rtas_local_data_buf[PROCESSOR_MAX_LENGTH] __cacheline_aligned;
+
+/*
+ * read_sys_info_pseries()
+ * Retrieve the number of sockets and chips per socket and cores per
+ * chip details through the get-system-parameter rtas call.
+ */
+void read_sys_info_pseries(void)
+{
+	int call_status, len, ntypes;
+
+	/*
+	 * Making system parameter: chips and sockets and cores per chip
+	 * default to 1.
+	 */
+	phys_sockets = 1;
+	phys_chipspersocket = 1;
+	phys_coresperchip = 1;
+	memset(rtas_local_data_buf, 0, PROCESSOR_MAX_LENGTH * sizeof(__be16));

Modifying global state outside of any critical section...? How do
you prevent readers from seeing inconsistent results?

Yes right, Will update.

quoted

+	spin_lock(&rtas_local_data_buf_lock);
+
+	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
+				NULL,
+				PROCESSOR_MODULE_INFO,
+				__pa(rtas_local_data_buf),
+				PROCESSOR_MAX_LENGTH);
+
+	spin_unlock(&rtas_local_data_buf_lock);

Using this lock this way fails to provide any protection to the data
buffer or the phys_* variables.

quoted

+
+	if (call_status != 0) {
+		pr_info("Error calling get-system-parameter (0x%x)\n",
+			call_status);

To be robust, this should handle busy (-2) and extended delay (990x)
statuses. And if it's going to log errors it should use pr_err() and use
decimal, not hex, to report the RTAS call status, since that's how
they're specified in PAPR+.

Thanks for pointing it, Will update.

Thanks,
Kajol Jain