Re: [PATCH v3 3/4] drm/amd/display: Add control mechanism for FPU utilization

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Am 20.07.21 um 02:49 schrieb Rodrigo Siqueira:
DC invokes DC_FPU_START/END in multiple parts of the code; this can
create a situation where we invoke this FPU operation in a nested way or
exit too early. For avoiding this situation, this commit adds a
mechanism where dc_fpu_begin/end manages the access to
kernel_fpu_begin/end.

Change since V2:
- Christian: Do not use this_cpu_* between get/put_cpu_ptr().

Change since V1:
- Use a better variable names
- Use get_cpu_ptr and put_cpu_ptr to better balance preemption enable
and disable

Cc: Harry Wentland <harry.wentland@xxxxxxx>
Cc: Anson Jacob <Anson.Jacob@xxxxxxx>
Cc: Christian König <christian.koenig@xxxxxxx>
Cc: Hersen Wu <hersenxs.wu@xxxxxxx>
Cc: Aric Cyr <aric.cyr@xxxxxxx>
Signed-off-by: Rodrigo Siqueira <Rodrigo.Siqueira@xxxxxxx>
---
  .../amd/display/amdgpu_dm/amdgpu_dm_trace.h   | 13 ++++---
  .../gpu/drm/amd/display/amdgpu_dm/dc_fpu.c    | 36 ++++++++++++++++---
  drivers/gpu/drm/amd/display/dc/dc_trace.h     |  4 +--
  3 files changed, 42 insertions(+), 11 deletions(-)

diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_trace.h b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_trace.h
index 230bb12c405e..fdcaea22b456 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_trace.h
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/amdgpu_dm_trace.h
@@ -638,23 +638,26 @@ TRACE_EVENT(amdgpu_refresh_rate_track,
  );
TRACE_EVENT(dcn_fpu,
-	    TP_PROTO(bool begin, const char *function, const int line),
-	    TP_ARGS(begin, function, line),
+	    TP_PROTO(bool begin, const char *function, const int line, const int recursion_depth),
+	    TP_ARGS(begin, function, line, recursion_depth),
TP_STRUCT__entry(
  			     __field(bool, begin)
  			     __field(const char *, function)
  			     __field(int, line)
+			     __field(int, recursion_depth)
  	    ),
  	    TP_fast_assign(
  			   __entry->begin = begin;
  			   __entry->function = function;
  			   __entry->line = line;
+			   __entry->recursion_depth = recursion_depth;
  	    ),
-	    TP_printk("%s()+%d: %s",
+	    TP_printk("%s: recursion_depth: %d: %s()+%d:",
+		      __entry->begin ? "begin" : "end",
+		      __entry->recursion_depth,
  		      __entry->function,
-		      __entry->line,
-		      __entry->begin ? "begin" : "end"
+		      __entry->line
  	    )
  );
diff --git a/drivers/gpu/drm/amd/display/amdgpu_dm/dc_fpu.c b/drivers/gpu/drm/amd/display/amdgpu_dm/dc_fpu.c
index d5d156a4517e..d0d3e8a34db5 100644
--- a/drivers/gpu/drm/amd/display/amdgpu_dm/dc_fpu.c
+++ b/drivers/gpu/drm/amd/display/amdgpu_dm/dc_fpu.c
@@ -28,6 +28,19 @@
#include <asm/fpu/api.h> +/**
+ * DOC: DC FPU manipulation overview
+ *
+ * DC core uses FPU operations in multiple parts of the code, which requires a
+ * more specialized way to manage these areas' entrance. To fulfill this
+ * requirement, we created some wrapper functions that encapsulate
+ * kernel_fpu_begin/end to better fit our need in the display component. In
+ * summary, in this file, you can find functions related to FPU operation
+ * management.
+ */
+
+static DEFINE_PER_CPU(int, fpu_recursion_depth);
+
  /**
   * dc_fpu_begin - Enables FPU protection
   * @function_name: A string containing the function name for debug purposes
@@ -43,8 +56,16 @@
   */
  void dc_fpu_begin(const char *function_name, const int line)
  {
-	TRACE_DCN_FPU(true, function_name, line);
-	kernel_fpu_begin();
+	int *pcpu;
+
+	pcpu = get_cpu_ptr(&fpu_recursion_depth);
+	*pcpu += 1;
+

This needs:

#if defined(CONFIG_X86)

+	if (*pcpu == 1)
+		kernel_fpu_begin();

#elif defined(CONFIG_PPC64)
....


Should I pick up the patches and help a bit with the non x86 support?

Regards,
Christian.

+
+	TRACE_DCN_FPU(true, function_name, line, *pcpu);
+	put_cpu_ptr(&fpu_recursion_depth);
  }
/**
@@ -59,6 +80,13 @@ void dc_fpu_begin(const char *function_name, const int line)
   */
  void dc_fpu_end(const char *function_name, const int line)
  {
-	TRACE_DCN_FPU(false, function_name, line);
-	kernel_fpu_end();
+	int *pcpu;
+
+	pcpu = get_cpu_ptr(&fpu_recursion_depth);
+	*pcpu -= 1;
+	if (*pcpu <= 0)
+		kernel_fpu_end();
+
+	TRACE_DCN_FPU(false, function_name, line, *pcpu);
+	put_cpu_ptr(&fpu_recursion_depth);
  }
diff --git a/drivers/gpu/drm/amd/display/dc/dc_trace.h b/drivers/gpu/drm/amd/display/dc/dc_trace.h
index d598ba697e45..c711797e5c9e 100644
--- a/drivers/gpu/drm/amd/display/dc/dc_trace.h
+++ b/drivers/gpu/drm/amd/display/dc/dc_trace.h
@@ -38,5 +38,5 @@
  #define TRACE_DCN_CLOCK_STATE(dcn_clocks) \
  	trace_amdgpu_dm_dc_clocks_state(dcn_clocks)
-#define TRACE_DCN_FPU(begin, function, line) \
-	trace_dcn_fpu(begin, function, line)
+#define TRACE_DCN_FPU(begin, function, line, ref_count) \
+	trace_dcn_fpu(begin, function, line, ref_count)




[Index of Archives]     [Linux DRI Users]     [Linux Intel Graphics]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]     [XFree86]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Linux Kernel]     [Linux SCSI]     [XFree86]
  Powered by Linux