Subject: Lock Perfmon context before calling hardware-sampling handler

From: Kevin Corry <kevcorry@us.ibm.com>
In handle_trace_buffer_interrupts(), we need to lock the context before calling
the hardware-sampling handler to prevent another CPU from operating on the
context. This has to be done here, instead of in the handler itself, because
pfm_cell_disable_counters(), which also calls the handler, is always called
with the context already locked.

Signed-off-by: Kevin Corry <kevcorry@us.ibm.com>
Signed-off-by: Carl Love <carll@us.ibm.com>
Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>

Index: linux-2.6/arch/powerpc/perfmon/perfmon_cell.c
===================================================================
--- linux-2.6.orig/arch/powerpc/perfmon/perfmon_cell.c
+++ linux-2.6/arch/powerpc/perfmon/perfmon_cell.c
@@ -479,8 +479,11 @@ static void handle_trace_buffer_interrup
 	struct pfm_smpl_fmt *smpl_fmt;
 
 	if (pm_status & CBE_PM_TRACE_BUFFER_FULL_INTR) {
-		/* The trace-buffer is full. Get the
-		 * sampling-buffer address and call the handler.
+		/* The trace-buffer is full. Get the sampling-buffer address
+		 * and call the handler. Need to lock the context here instead
+		 * of in the handler, because the handler can also be called
+		 * by the pfm_cell_disable_counters() routine, which is called
+		 * with the context already locked.
 		 */
 		smpl_fmt = ctx->smpl_fmt;
 
@@ -488,8 +491,10 @@ static void handle_trace_buffer_interrup
 		    !strcmp(smpl_fmt->fmt_name, PFM_CELL_HW_SMPL_NAME)) {
 			ctx->ovfl_arg.ovfl_pmd = PFM_CELL_HW_SMPL_OVFL_PMD;
 			ctx->ovfl_arg.active_set = ctx->active_set->id;
+			spin_lock(&ctx->lock);
 			smpl_fmt->fmt_handler(ctx->smpl_addr, &ctx->ovfl_arg,
 					      iip, 0, regs);
+			spin_unlock(&ctx->lock);
 		}
 	}