Commit cd84a27d authored by Alex Deucher's avatar Alex Deucher

drm/radeon/dce8: add support for display watermark setup

Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
parent bc19f597
......@@ -5017,3 +5017,540 @@ void cik_fini(struct radeon_device *rdev)
kfree(rdev->bios);
rdev->bios = NULL;
}
/* display watermark setup */
/**
* dce8_line_buffer_adjust - Set up the line buffer
*
* @rdev: radeon_device pointer
* @radeon_crtc: the selected display controller
* @mode: the current display mode on the selected display
* controller
*
* Setup up the line buffer allocation for
* the selected display controller (CIK).
* Returns the line buffer size in pixels.
*/
static u32 dce8_line_buffer_adjust(struct radeon_device *rdev,
struct radeon_crtc *radeon_crtc,
struct drm_display_mode *mode)
{
u32 tmp;
/*
* Line Buffer Setup
* There are 6 line buffers, one for each display controllers.
* There are 3 partitions per LB. Select the number of partitions
* to enable based on the display width. For display widths larger
* than 4096, you need use to use 2 display controllers and combine
* them using the stereo blender.
*/
if (radeon_crtc->base.enabled && mode) {
if (mode->crtc_hdisplay < 1920)
tmp = 1;
else if (mode->crtc_hdisplay < 2560)
tmp = 2;
else if (mode->crtc_hdisplay < 4096)
tmp = 0;
else {
DRM_DEBUG_KMS("Mode too big for LB!\n");
tmp = 0;
}
} else
tmp = 1;
WREG32(LB_MEMORY_CTRL + radeon_crtc->crtc_offset,
LB_MEMORY_CONFIG(tmp) | LB_MEMORY_SIZE(0x6B0));
if (radeon_crtc->base.enabled && mode) {
switch (tmp) {
case 0:
default:
return 4096 * 2;
case 1:
return 1920 * 2;
case 2:
return 2560 * 2;
}
}
/* controller not enabled, so no lb used */
return 0;
}
/**
* cik_get_number_of_dram_channels - get the number of dram channels
*
* @rdev: radeon_device pointer
*
* Look up the number of video ram channels (CIK).
* Used for display watermark bandwidth calculations
* Returns the number of dram channels
*/
static u32 cik_get_number_of_dram_channels(struct radeon_device *rdev)
{
u32 tmp = RREG32(MC_SHARED_CHMAP);
switch ((tmp & NOOFCHAN_MASK) >> NOOFCHAN_SHIFT) {
case 0:
default:
return 1;
case 1:
return 2;
case 2:
return 4;
case 3:
return 8;
case 4:
return 3;
case 5:
return 6;
case 6:
return 10;
case 7:
return 12;
case 8:
return 16;
}
}
struct dce8_wm_params {
u32 dram_channels; /* number of dram channels */
u32 yclk; /* bandwidth per dram data pin in kHz */
u32 sclk; /* engine clock in kHz */
u32 disp_clk; /* display clock in kHz */
u32 src_width; /* viewport width */
u32 active_time; /* active display time in ns */
u32 blank_time; /* blank time in ns */
bool interlaced; /* mode is interlaced */
fixed20_12 vsc; /* vertical scale ratio */
u32 num_heads; /* number of active crtcs */
u32 bytes_per_pixel; /* bytes per pixel display + overlay */
u32 lb_size; /* line buffer allocated to pipe */
u32 vtaps; /* vertical scaler taps */
};
/**
* dce8_dram_bandwidth - get the dram bandwidth
*
* @wm: watermark calculation data
*
* Calculate the raw dram bandwidth (CIK).
* Used for display watermark bandwidth calculations
* Returns the dram bandwidth in MBytes/s
*/
static u32 dce8_dram_bandwidth(struct dce8_wm_params *wm)
{
/* Calculate raw DRAM Bandwidth */
fixed20_12 dram_efficiency; /* 0.7 */
fixed20_12 yclk, dram_channels, bandwidth;
fixed20_12 a;
a.full = dfixed_const(1000);
yclk.full = dfixed_const(wm->yclk);
yclk.full = dfixed_div(yclk, a);
dram_channels.full = dfixed_const(wm->dram_channels * 4);
a.full = dfixed_const(10);
dram_efficiency.full = dfixed_const(7);
dram_efficiency.full = dfixed_div(dram_efficiency, a);
bandwidth.full = dfixed_mul(dram_channels, yclk);
bandwidth.full = dfixed_mul(bandwidth, dram_efficiency);
return dfixed_trunc(bandwidth);
}
/**
* dce8_dram_bandwidth_for_display - get the dram bandwidth for display
*
* @wm: watermark calculation data
*
* Calculate the dram bandwidth used for display (CIK).
* Used for display watermark bandwidth calculations
* Returns the dram bandwidth for display in MBytes/s
*/
static u32 dce8_dram_bandwidth_for_display(struct dce8_wm_params *wm)
{
/* Calculate DRAM Bandwidth and the part allocated to display. */
fixed20_12 disp_dram_allocation; /* 0.3 to 0.7 */
fixed20_12 yclk, dram_channels, bandwidth;
fixed20_12 a;
a.full = dfixed_const(1000);
yclk.full = dfixed_const(wm->yclk);
yclk.full = dfixed_div(yclk, a);
dram_channels.full = dfixed_const(wm->dram_channels * 4);
a.full = dfixed_const(10);
disp_dram_allocation.full = dfixed_const(3); /* XXX worse case value 0.3 */
disp_dram_allocation.full = dfixed_div(disp_dram_allocation, a);
bandwidth.full = dfixed_mul(dram_channels, yclk);
bandwidth.full = dfixed_mul(bandwidth, disp_dram_allocation);
return dfixed_trunc(bandwidth);
}
/**
* dce8_data_return_bandwidth - get the data return bandwidth
*
* @wm: watermark calculation data
*
* Calculate the data return bandwidth used for display (CIK).
* Used for display watermark bandwidth calculations
* Returns the data return bandwidth in MBytes/s
*/
static u32 dce8_data_return_bandwidth(struct dce8_wm_params *wm)
{
/* Calculate the display Data return Bandwidth */
fixed20_12 return_efficiency; /* 0.8 */
fixed20_12 sclk, bandwidth;
fixed20_12 a;
a.full = dfixed_const(1000);
sclk.full = dfixed_const(wm->sclk);
sclk.full = dfixed_div(sclk, a);
a.full = dfixed_const(10);
return_efficiency.full = dfixed_const(8);
return_efficiency.full = dfixed_div(return_efficiency, a);
a.full = dfixed_const(32);
bandwidth.full = dfixed_mul(a, sclk);
bandwidth.full = dfixed_mul(bandwidth, return_efficiency);
return dfixed_trunc(bandwidth);
}
/**
* dce8_dmif_request_bandwidth - get the dmif bandwidth
*
* @wm: watermark calculation data
*
* Calculate the dmif bandwidth used for display (CIK).
* Used for display watermark bandwidth calculations
* Returns the dmif bandwidth in MBytes/s
*/
static u32 dce8_dmif_request_bandwidth(struct dce8_wm_params *wm)
{
/* Calculate the DMIF Request Bandwidth */
fixed20_12 disp_clk_request_efficiency; /* 0.8 */
fixed20_12 disp_clk, bandwidth;
fixed20_12 a, b;
a.full = dfixed_const(1000);
disp_clk.full = dfixed_const(wm->disp_clk);
disp_clk.full = dfixed_div(disp_clk, a);
a.full = dfixed_const(32);
b.full = dfixed_mul(a, disp_clk);
a.full = dfixed_const(10);
disp_clk_request_efficiency.full = dfixed_const(8);
disp_clk_request_efficiency.full = dfixed_div(disp_clk_request_efficiency, a);
bandwidth.full = dfixed_mul(b, disp_clk_request_efficiency);
return dfixed_trunc(bandwidth);
}
/**
* dce8_available_bandwidth - get the min available bandwidth
*
* @wm: watermark calculation data
*
* Calculate the min available bandwidth used for display (CIK).
* Used for display watermark bandwidth calculations
* Returns the min available bandwidth in MBytes/s
*/
static u32 dce8_available_bandwidth(struct dce8_wm_params *wm)
{
/* Calculate the Available bandwidth. Display can use this temporarily but not in average. */
u32 dram_bandwidth = dce8_dram_bandwidth(wm);
u32 data_return_bandwidth = dce8_data_return_bandwidth(wm);
u32 dmif_req_bandwidth = dce8_dmif_request_bandwidth(wm);
return min(dram_bandwidth, min(data_return_bandwidth, dmif_req_bandwidth));
}
/**
* dce8_average_bandwidth - get the average available bandwidth
*
* @wm: watermark calculation data
*
* Calculate the average available bandwidth used for display (CIK).
* Used for display watermark bandwidth calculations
* Returns the average available bandwidth in MBytes/s
*/
static u32 dce8_average_bandwidth(struct dce8_wm_params *wm)
{
/* Calculate the display mode Average Bandwidth
* DisplayMode should contain the source and destination dimensions,
* timing, etc.
*/
fixed20_12 bpp;
fixed20_12 line_time;
fixed20_12 src_width;
fixed20_12 bandwidth;
fixed20_12 a;
a.full = dfixed_const(1000);
line_time.full = dfixed_const(wm->active_time + wm->blank_time);
line_time.full = dfixed_div(line_time, a);
bpp.full = dfixed_const(wm->bytes_per_pixel);
src_width.full = dfixed_const(wm->src_width);
bandwidth.full = dfixed_mul(src_width, bpp);
bandwidth.full = dfixed_mul(bandwidth, wm->vsc);
bandwidth.full = dfixed_div(bandwidth, line_time);
return dfixed_trunc(bandwidth);
}
/**
* dce8_latency_watermark - get the latency watermark
*
* @wm: watermark calculation data
*
* Calculate the latency watermark (CIK).
* Used for display watermark bandwidth calculations
* Returns the latency watermark in ns
*/
static u32 dce8_latency_watermark(struct dce8_wm_params *wm)
{
/* First calculate the latency in ns */
u32 mc_latency = 2000; /* 2000 ns. */
u32 available_bandwidth = dce8_available_bandwidth(wm);
u32 worst_chunk_return_time = (512 * 8 * 1000) / available_bandwidth;
u32 cursor_line_pair_return_time = (128 * 4 * 1000) / available_bandwidth;
u32 dc_latency = 40000000 / wm->disp_clk; /* dc pipe latency */
u32 other_heads_data_return_time = ((wm->num_heads + 1) * worst_chunk_return_time) +
(wm->num_heads * cursor_line_pair_return_time);
u32 latency = mc_latency + other_heads_data_return_time + dc_latency;
u32 max_src_lines_per_dst_line, lb_fill_bw, line_fill_time;
u32 tmp, dmif_size = 12288;
fixed20_12 a, b, c;
if (wm->num_heads == 0)
return 0;
a.full = dfixed_const(2);
b.full = dfixed_const(1);
if ((wm->vsc.full > a.full) ||
((wm->vsc.full > b.full) && (wm->vtaps >= 3)) ||
(wm->vtaps >= 5) ||
((wm->vsc.full >= a.full) && wm->interlaced))
max_src_lines_per_dst_line = 4;
else
max_src_lines_per_dst_line = 2;
a.full = dfixed_const(available_bandwidth);
b.full = dfixed_const(wm->num_heads);
a.full = dfixed_div(a, b);
b.full = dfixed_const(mc_latency + 512);
c.full = dfixed_const(wm->disp_clk);
b.full = dfixed_div(b, c);
c.full = dfixed_const(dmif_size);
b.full = dfixed_div(c, b);
tmp = min(dfixed_trunc(a), dfixed_trunc(b));
b.full = dfixed_const(1000);
c.full = dfixed_const(wm->disp_clk);
b.full = dfixed_div(c, b);
c.full = dfixed_const(wm->bytes_per_pixel);
b.full = dfixed_mul(b, c);
lb_fill_bw = min(tmp, dfixed_trunc(b));
a.full = dfixed_const(max_src_lines_per_dst_line * wm->src_width * wm->bytes_per_pixel);
b.full = dfixed_const(1000);
c.full = dfixed_const(lb_fill_bw);
b.full = dfixed_div(c, b);
a.full = dfixed_div(a, b);
line_fill_time = dfixed_trunc(a);
if (line_fill_time < wm->active_time)
return latency;
else
return latency + (line_fill_time - wm->active_time);
}
/**
* dce8_average_bandwidth_vs_dram_bandwidth_for_display - check
* average and available dram bandwidth
*
* @wm: watermark calculation data
*
* Check if the display average bandwidth fits in the display
* dram bandwidth (CIK).
* Used for display watermark bandwidth calculations
* Returns true if the display fits, false if not.
*/
static bool dce8_average_bandwidth_vs_dram_bandwidth_for_display(struct dce8_wm_params *wm)
{
if (dce8_average_bandwidth(wm) <=
(dce8_dram_bandwidth_for_display(wm) / wm->num_heads))
return true;
else
return false;
}
/**
* dce8_average_bandwidth_vs_available_bandwidth - check
* average and available bandwidth
*
* @wm: watermark calculation data
*
* Check if the display average bandwidth fits in the display
* available bandwidth (CIK).
* Used for display watermark bandwidth calculations
* Returns true if the display fits, false if not.
*/
static bool dce8_average_bandwidth_vs_available_bandwidth(struct dce8_wm_params *wm)
{
if (dce8_average_bandwidth(wm) <=
(dce8_available_bandwidth(wm) / wm->num_heads))
return true;
else
return false;
}
/**
* dce8_check_latency_hiding - check latency hiding
*
* @wm: watermark calculation data
*
* Check latency hiding (CIK).
* Used for display watermark bandwidth calculations
* Returns true if the display fits, false if not.
*/
static bool dce8_check_latency_hiding(struct dce8_wm_params *wm)
{
u32 lb_partitions = wm->lb_size / wm->src_width;
u32 line_time = wm->active_time + wm->blank_time;
u32 latency_tolerant_lines;
u32 latency_hiding;
fixed20_12 a;
a.full = dfixed_const(1);
if (wm->vsc.full > a.full)
latency_tolerant_lines = 1;
else {
if (lb_partitions <= (wm->vtaps + 1))
latency_tolerant_lines = 1;
else
latency_tolerant_lines = 2;
}
latency_hiding = (latency_tolerant_lines * line_time + wm->blank_time);
if (dce8_latency_watermark(wm) <= latency_hiding)
return true;
else
return false;
}
/**
* dce8_program_watermarks - program display watermarks
*
* @rdev: radeon_device pointer
* @radeon_crtc: the selected display controller
* @lb_size: line buffer size
* @num_heads: number of display controllers in use
*
* Calculate and program the display watermarks for the
* selected display controller (CIK).
*/
static void dce8_program_watermarks(struct radeon_device *rdev,
struct radeon_crtc *radeon_crtc,
u32 lb_size, u32 num_heads)
{
struct drm_display_mode *mode = &radeon_crtc->base.mode;
struct dce8_wm_params wm;
u32 pixel_period;
u32 line_time = 0;
u32 latency_watermark_a = 0, latency_watermark_b = 0;
u32 tmp, wm_mask;
if (radeon_crtc->base.enabled && num_heads && mode) {
pixel_period = 1000000 / (u32)mode->clock;
line_time = min((u32)mode->crtc_htotal * pixel_period, (u32)65535);
wm.yclk = rdev->pm.current_mclk * 10;
wm.sclk = rdev->pm.current_sclk * 10;
wm.disp_clk = mode->clock;
wm.src_width = mode->crtc_hdisplay;
wm.active_time = mode->crtc_hdisplay * pixel_period;
wm.blank_time = line_time - wm.active_time;
wm.interlaced = false;
if (mode->flags & DRM_MODE_FLAG_INTERLACE)
wm.interlaced = true;
wm.vsc = radeon_crtc->vsc;
wm.vtaps = 1;
if (radeon_crtc->rmx_type != RMX_OFF)
wm.vtaps = 2;
wm.bytes_per_pixel = 4; /* XXX: get this from fb config */
wm.lb_size = lb_size;
wm.dram_channels = cik_get_number_of_dram_channels(rdev);
wm.num_heads = num_heads;
/* set for high clocks */
latency_watermark_a = min(dce8_latency_watermark(&wm), (u32)65535);
/* set for low clocks */
/* wm.yclk = low clk; wm.sclk = low clk */
latency_watermark_b = min(dce8_latency_watermark(&wm), (u32)65535);
/* possibly force display priority to high */
/* should really do this at mode validation time... */
if (!dce8_average_bandwidth_vs_dram_bandwidth_for_display(&wm) ||
!dce8_average_bandwidth_vs_available_bandwidth(&wm) ||
!dce8_check_latency_hiding(&wm) ||
(rdev->disp_priority == 2)) {
DRM_DEBUG_KMS("force priority to high\n");
}
}
/* select wm A */
wm_mask = RREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset);
tmp = wm_mask;
tmp &= ~LATENCY_WATERMARK_MASK(3);
tmp |= LATENCY_WATERMARK_MASK(1);
WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, tmp);
WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
(LATENCY_LOW_WATERMARK(latency_watermark_a) |
LATENCY_HIGH_WATERMARK(line_time)));
/* select wm B */
tmp = RREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset);
tmp &= ~LATENCY_WATERMARK_MASK(3);
tmp |= LATENCY_WATERMARK_MASK(2);
WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, tmp);
WREG32(DPG_PIPE_LATENCY_CONTROL + radeon_crtc->crtc_offset,
(LATENCY_LOW_WATERMARK(latency_watermark_b) |
LATENCY_HIGH_WATERMARK(line_time)));
/* restore original selection */
WREG32(DPG_WATERMARK_MASK_CONTROL + radeon_crtc->crtc_offset, wm_mask);
}
/**
* dce8_bandwidth_update - program display watermarks
*
* @rdev: radeon_device pointer
*
* Calculate and program the display watermarks and line
* buffer allocation (CIK).
*/
void dce8_bandwidth_update(struct radeon_device *rdev)
{
struct drm_display_mode *mode = NULL;
u32 num_heads = 0, lb_size;
int i;
radeon_update_display_priority(rdev);
for (i = 0; i < rdev->num_crtc; i++) {
if (rdev->mode_info.crtcs[i]->base.enabled)
num_heads++;
}
for (i = 0; i < rdev->num_crtc; i++) {
mode = &rdev->mode_info.crtcs[i]->base.mode;
lb_size = dce8_line_buffer_adjust(rdev, rdev->mode_info.crtcs[i], mode);
dce8_program_watermarks(rdev, rdev->mode_info.crtcs[i], lb_size, num_heads);
}
}
......@@ -259,6 +259,17 @@
#define SDMA0 (1 << 10)
#define SDMA1 (1 << 11)
/* 0x6b04, 0x7704, 0x10304, 0x10f04, 0x11b04, 0x12704 */
#define LB_MEMORY_CTRL 0x6b04
#define LB_MEMORY_SIZE(x) ((x) << 0)
#define LB_MEMORY_CONFIG(x) ((x) << 20)
#define DPG_WATERMARK_MASK_CONTROL 0x6cc8
# define LATENCY_WATERMARK_MASK(x) ((x) << 8)
#define DPG_PIPE_LATENCY_CONTROL 0x6ccc
# define LATENCY_LOW_WATERMARK(x) ((x) << 0)
# define LATENCY_HIGH_WATERMARK(x) ((x) << 16)
/* 0x6b24, 0x7724, 0x10324, 0x10f24, 0x11b24, 0x12724 */
#define LB_VLINE_STATUS 0x6b24
# define VLINE_OCCURRED (1 << 0)
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment