drm/radeon/kms/avivo: add support for new pll selection algo

Supported on all AVIVO-based asics.
Can be disabled via the new_pll module parameter:
new_pll=0 - disable
new_pll=1 - enable
enabled by default

[airlied: fixed to use do_div]
Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
Signed-off-by: Dave Airlie <airlied@linux.ie>

drivers/gpu/drm/radeon/atombios_crtc.c

@@ -499,8 +499,18 @@ void atombios_crtc_set_pll(struct drm_crtc *crtc, struct drm_display_mode *mode)
 	else
 		pll = &rdev->clock.p2pll;
 
-	radeon_compute_pll(pll, adjusted_clock, &pll_clock, &fb_div, &frac_fb_div,
-			   &ref_div, &post_div, pll_flags);
+	if (ASIC_IS_AVIVO(rdev)) {
+		if (radeon_new_pll)
+			radeon_compute_pll_avivo(pll, adjusted_clock, &pll_clock,
+						 &fb_div, &frac_fb_div,
+						 &ref_div, &post_div, pll_flags);
+		else
+			radeon_compute_pll(pll, adjusted_clock, &pll_clock,
+					   &fb_div, &frac_fb_div,
+					   &ref_div, &post_div, pll_flags);
+	} else
+		radeon_compute_pll(pll, adjusted_clock, &pll_clock, &fb_div, &frac_fb_div,
+				   &ref_div, &post_div, pll_flags);
 
 	index = GetIndexIntoMasterTable(COMMAND, SetPixelClock);
 	atom_parse_cmd_header(rdev->mode_info.atom_context, index, &frev,

drivers/gpu/drm/radeon/radeon.h

@@ -88,6 +88,7 @@ extern int radeon_benchmarking;
 extern int radeon_testing;
 extern int radeon_connector_table;
 extern int radeon_tv;
+extern int radeon_new_pll;
 
 /*
  * Copy from radeon_drv.h so we don't have to include both and have conflicting

drivers/gpu/drm/radeon/radeon_atombios.c

@@ -871,7 +871,8 @@ bool radeon_atom_get_clock_info(struct drm_device *dev)
 			 * pre-DCE 3.0 r6xx hardware.  This might need to be adjusted per
 			 * family.
 			 */
-			p1pll->pll_out_min = 64800;
+			if (!radeon_new_pll)
+				p1pll->pll_out_min = 64800;
 		}
 
 		p1pll->pll_in_min =

drivers/gpu/drm/radeon/radeon_display.c

@@ -560,6 +560,98 @@ void radeon_compute_pll(struct radeon_pll *pll,
 	*post_div_p = best_post_div;
 }
 
+void radeon_compute_pll_avivo(struct radeon_pll *pll,
+			      uint64_t freq,
+			      uint32_t *dot_clock_p,
+			      uint32_t *fb_div_p,
+			      uint32_t *frac_fb_div_p,
+			      uint32_t *ref_div_p,
+			      uint32_t *post_div_p,
+			      int flags)
+{
+	fixed20_12 m, n, frac_n, p, f_vco, f_pclk, best_freq;
+	fixed20_12 pll_out_max, pll_out_min;
+	fixed20_12 pll_in_max, pll_in_min;
+	fixed20_12 reference_freq;
+	fixed20_12 error, ffreq, a, b;
+
+	pll_out_max.full = rfixed_const(pll->pll_out_max);
+	pll_out_min.full = rfixed_const(pll->pll_out_min);
+	pll_in_max.full = rfixed_const(pll->pll_in_max);
+	pll_in_min.full = rfixed_const(pll->pll_in_min);
+	reference_freq.full = rfixed_const(pll->reference_freq);
+	do_div(freq, 10);
+	ffreq.full = rfixed_const(freq);
+	error.full = rfixed_const(100 * 100);
+
+	/* max p */
+	p.full = rfixed_div(pll_out_max, ffreq);
+	p.full = rfixed_floor(p);
+
+	/* min m */
+	m.full = rfixed_div(reference_freq, pll_in_max);
+	m.full = rfixed_ceil(m);
+
+	while (1) {
+		n.full = rfixed_div(ffreq, reference_freq);
+		n.full = rfixed_mul(n, m);
+		n.full = rfixed_mul(n, p);
+
+		f_vco.full = rfixed_div(n, m);
+		f_vco.full = rfixed_mul(f_vco, reference_freq);
+
+		f_pclk.full = rfixed_div(f_vco, p);
+
+		if (f_pclk.full > ffreq.full)
+			error.full = f_pclk.full - ffreq.full;
+		else
+			error.full = ffreq.full - f_pclk.full;
+		error.full = rfixed_div(error, f_pclk);
+		a.full = rfixed_const(100 * 100);
+		error.full = rfixed_mul(error, a);
+
+		a.full = rfixed_mul(m, p);
+		a.full = rfixed_div(n, a);
+		best_freq.full = rfixed_mul(reference_freq, a);
+
+		if (rfixed_trunc(error) < 25)
+			break;
+
+		a.full = rfixed_const(1);
+		m.full = m.full + a.full;
+		a.full = rfixed_div(reference_freq, m);
+		if (a.full >= pll_in_min.full)
+			continue;
+
+		m.full = rfixed_div(reference_freq, pll_in_max);
+		m.full = rfixed_ceil(m);
+		a.full= rfixed_const(1);
+		p.full = p.full - a.full;
+		a.full = rfixed_mul(p, ffreq);
+		if (a.full >= pll_out_min.full)
+			continue;
+		else {
+			DRM_ERROR("Unable to find pll dividers\n");
+			break;
+		}
+	}
+
+	a.full = rfixed_const(10);
+	b.full = rfixed_mul(n, a);
+
+	frac_n.full = rfixed_floor(n);
+	frac_n.full = rfixed_mul(frac_n, a);
+	frac_n.full = b.full - frac_n.full;
+
+	*dot_clock_p = rfixed_trunc(best_freq);
+	*fb_div_p = rfixed_trunc(n);
+	*frac_fb_div_p = rfixed_trunc(frac_n);
+	*ref_div_p = rfixed_trunc(m);
+	*post_div_p = rfixed_trunc(p);
+
+	DRM_DEBUG("%u %d.%d, %d, %d\n", *dot_clock_p * 10, *fb_div_p, *frac_fb_div_p, *ref_div_p, *post_div_p);
+}
+
 static void radeon_user_framebuffer_destroy(struct drm_framebuffer *fb)
 {
 	struct radeon_framebuffer *radeon_fb = to_radeon_framebuffer(fb);

drivers/gpu/drm/radeon/radeon_drv.c

@@ -86,6 +86,7 @@ int radeon_benchmarking = 0;
 int radeon_testing = 0;
 int radeon_connector_table = 0;
 int radeon_tv = 1;
+int radeon_new_pll = 1;
 
 MODULE_PARM_DESC(no_wb, "Disable AGP writeback for scratch registers");
 module_param_named(no_wb, radeon_no_wb, int, 0444);
@@ -120,6 +121,9 @@ module_param_named(connector_table, radeon_connector_table, int, 0444);
 MODULE_PARM_DESC(tv, "TV enable (0 = disable)");
 module_param_named(tv, radeon_tv, int, 0444);
 
+MODULE_PARM_DESC(r4xx_atom, "Select new PLL code for AVIVO chips");
+module_param_named(new_pll, radeon_new_pll, int, 0444);
+
 static int radeon_suspend(struct drm_device *dev, pm_message_t state)
 {
 	drm_radeon_private_t *dev_priv = dev->dev_private;

drivers/gpu/drm/radeon/radeon_mode.h

@@ -437,6 +437,15 @@ extern void radeon_compute_pll(struct radeon_pll *pll,
 			       uint32_t *post_div_p,
 			       int flags);
 
+extern void radeon_compute_pll_avivo(struct radeon_pll *pll,
+				     uint64_t freq,
+				     uint32_t *dot_clock_p,
+				     uint32_t *fb_div_p,
+				     uint32_t *frac_fb_div_p,
+				     uint32_t *ref_div_p,
+				     uint32_t *post_div_p,
+				     int flags);
+
 extern void radeon_setup_encoder_clones(struct drm_device *dev);
 
 struct drm_encoder *radeon_encoder_legacy_lvds_add(struct drm_device *dev, int bios_index);