The dcn20_validate_bandwidth function would have code touching the
incorrect registers emitted outside of the boundaries of the
DC_FP_START/END macros, at least on ppc64le. Work around the
problem by wrapping the whole function instead.
Signed-off-by: Daniel Kolesa <dan...@octaforge.org>
---
.../drm/amd/display/dc/dcn20/dcn20_resource.c | 31 ++++++++++++++-----
1 file changed, 23 insertions(+), 8 deletions(-)
diff --git a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
index e310d67..1b0bca9 100644
--- a/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
+++ b/drivers/gpu/drm/amd/display/dc/dcn20/dcn20_resource.c
@@ -3034,25 +3034,32 @@ static bool dcn20_validate_bandwidth_internal(struct dc
*dc, struct dc_state *co
return out;
}
-
-bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
- bool fast_validate)
+/*
+ * This must be noinline to ensure anything that deals with FP registers
+ * is contained within this call; previously our compiling with hard-float
+ * would result in fp instructions being emitted outside of the boundaries
+ * of the DC_FP_START/END macros, which makes sense as the compiler has no
+ * idea about what is wrapped and what is not
+ *
+ * This is largely just a workaround to avoid breakage introduced with 5.6,
+ * ideally all fp-using code should be moved into its own file, only that
+ * should be compiled with hard-float, and all code exported from there
+ * should be strictly wrapped with DC_FP_START/END
+ */
+static noinline bool dcn20_validate_bandwidth_fp(struct dc *dc,
+ struct dc_state *context, bool fast_validate)
{
bool voltage_supported = false;
bool full_pstate_supported = false;
bool dummy_pstate_supported = false;
double p_state_latency_us;
- DC_FP_START();
p_state_latency_us =
context->bw_ctx.dml.soc.dram_clock_change_latency_us;
context->bw_ctx.dml.soc.disable_dram_clock_change_vactive_support =
dc->debug.disable_dram_clock_change_vactive_support;
if (fast_validate) {
- voltage_supported = dcn20_validate_bandwidth_internal(dc,
context, true);
-
- DC_FP_END();
- return voltage_supported;
+ return dcn20_validate_bandwidth_internal(dc, context, true);
}
// Best case, we support full UCLK switch latency
@@ -3081,7 +3088,15 @@ bool dcn20_validate_bandwidth(struct dc *dc, struct
dc_state *context,
restore_dml_state:
context->bw_ctx.dml.soc.dram_clock_change_latency_us =
p_state_latency_us;
+ return voltage_supported;
+}
+bool dcn20_validate_bandwidth(struct dc *dc, struct dc_state *context,
+ bool fast_validate)
+{
+ bool voltage_supported = false;
+ DC_FP_START();
+ voltage_supported = dcn20_validate_bandwidth_fp(dc, context,
fast_validate);
DC_FP_END();
return voltage_supported;
}
--
2.26.2
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
