diff options
Diffstat (limited to 'drivers/gpu/drm/i915/display/intel_dsb.c')
-rw-r--r-- | drivers/gpu/drm/i915/display/intel_dsb.c | 270 |
1 files changed, 198 insertions, 72 deletions
diff --git a/drivers/gpu/drm/i915/display/intel_dsb.c b/drivers/gpu/drm/i915/display/intel_dsb.c index 9fc4003d1579..53d8ae3a70e9 100644 --- a/drivers/gpu/drm/i915/display/intel_dsb.c +++ b/drivers/gpu/drm/i915/display/intel_dsb.c @@ -4,13 +4,14 @@ * */ +#include <drm/drm_print.h> #include <drm/drm_vblank.h> -#include "i915_drv.h" -#include "i915_irq.h" -#include "i915_reg.h" +#include "i915_utils.h" #include "intel_crtc.h" #include "intel_de.h" +#include "intel_display_regs.h" +#include "intel_display_rpm.h" #include "intel_display_types.h" #include "intel_dsb.h" #include "intel_dsb_buffer.h" @@ -92,6 +93,10 @@ struct intel_dsb { /* see DSB_REG_VALUE_MASK */ #define DSB_OPCODE_POLL 0xA /* see DSB_REG_VALUE_MASK */ +#define DSB_OPCODE_GOSUB 0xC /* ptl+ */ +#define DSB_GOSUB_HEAD_SHIFT 26 +#define DSB_GOSUB_TAIL_SHIFT 0 +#define DSB_GOSUB_CONVERT_ADDR(x) ((x) >> 6) static bool pre_commit_is_vrr_active(struct intel_atomic_state *state, struct intel_crtc *crtc) @@ -142,10 +147,10 @@ static int dsb_vtotal(struct intel_atomic_state *state, static int dsb_dewake_scanline_start(struct intel_atomic_state *state, struct intel_crtc *crtc) { + struct intel_display *display = to_intel_display(state); const struct intel_crtc_state *crtc_state = intel_pre_commit_crtc_state(state, crtc); - struct drm_i915_private *i915 = to_i915(state->base.dev); - unsigned int latency = skl_watermark_max_latency(i915, 0); + unsigned int latency = skl_watermark_max_latency(display, 0); return intel_mode_vdisplay(&crtc_state->hw.adjusted_mode) - intel_usecs_to_scanlines(&crtc_state->hw.adjusted_mode, latency); @@ -203,6 +208,15 @@ static bool assert_dsb_has_room(struct intel_dsb *dsb) crtc->base.base.id, crtc->base.name, dsb->id); } +static bool assert_dsb_tail_is_aligned(struct intel_dsb *dsb) +{ + struct intel_crtc *crtc = dsb->crtc; + struct intel_display *display = to_intel_display(crtc->base.dev); + + return !drm_WARN_ON(display->drm, + !IS_ALIGNED(dsb->free_pos * 4, CACHELINE_BYTES)); +} + static void intel_dsb_dump(struct intel_dsb *dsb) { struct intel_crtc *crtc = dsb->crtc; @@ -227,13 +241,40 @@ static bool is_dsb_busy(struct intel_display *display, enum pipe pipe, return intel_de_read_fw(display, DSB_CTRL(pipe, dsb_id)) & DSB_STATUS_BUSY; } +unsigned int intel_dsb_size(struct intel_dsb *dsb) +{ + return dsb->free_pos * 4; +} + +unsigned int intel_dsb_head(struct intel_dsb *dsb) +{ + return intel_dsb_buffer_ggtt_offset(&dsb->dsb_buf); +} + +static unsigned int intel_dsb_tail(struct intel_dsb *dsb) +{ + return intel_dsb_buffer_ggtt_offset(&dsb->dsb_buf) + intel_dsb_size(dsb); +} + +static void intel_dsb_ins_align(struct intel_dsb *dsb) +{ + /* + * Every instruction should be 8 byte aligned. + * + * The only way to get unaligned free_pos is via + * intel_dsb_reg_write_indexed() which already + * makes sure the next dword is zeroed, so no need + * to clear it here. + */ + dsb->free_pos = ALIGN(dsb->free_pos, 2); +} + static void intel_dsb_emit(struct intel_dsb *dsb, u32 ldw, u32 udw) { if (!assert_dsb_has_room(dsb)) return; - /* Every instruction should be 8 byte aligned. */ - dsb->free_pos = ALIGN(dsb->free_pos, 2); + intel_dsb_ins_align(dsb); dsb->ins_start_offset = dsb->free_pos; dsb->ins[0] = ldw; @@ -491,6 +532,8 @@ static void intel_dsb_align_tail(struct intel_dsb *dsb) { u32 aligned_tail, tail; + intel_dsb_ins_align(dsb); + tail = dsb->free_pos * 4; aligned_tail = ALIGN(tail, CACHELINE_BYTES); @@ -501,20 +544,90 @@ static void intel_dsb_align_tail(struct intel_dsb *dsb) dsb->free_pos = aligned_tail / 4; } -void intel_dsb_finish(struct intel_dsb *dsb) +static void intel_dsb_gosub_align(struct intel_dsb *dsb) +{ + u32 aligned_tail, tail; + + intel_dsb_ins_align(dsb); + + tail = dsb->free_pos * 4; + aligned_tail = ALIGN(tail, CACHELINE_BYTES); + + /* + * Wa_16024917128 + * "Ensure GOSUB is not placed in cacheline QW slot 6 or 7 (numbered 0-7)" + */ + if (aligned_tail - tail <= 2 * 8) + intel_dsb_buffer_memset(&dsb->dsb_buf, dsb->free_pos, 0, + aligned_tail - tail); + + dsb->free_pos = aligned_tail / 4; +} + +void intel_dsb_gosub(struct intel_dsb *dsb, + struct intel_dsb *sub_dsb) { struct intel_crtc *crtc = dsb->crtc; + struct intel_display *display = to_intel_display(crtc->base.dev); + unsigned int head, tail; + u64 head_tail; + + if (drm_WARN_ON(display->drm, dsb->id != sub_dsb->id)) + return; + + if (!assert_dsb_tail_is_aligned(sub_dsb)) + return; + + intel_dsb_gosub_align(dsb); + + head = intel_dsb_head(sub_dsb); + tail = intel_dsb_tail(sub_dsb); /* - * DSB_FORCE_DEWAKE remains active even after DSB is - * disabled, so make sure to clear it (if set during - * intel_dsb_commit()). And clear DSB_ENABLE_DEWAKE as - * well for good measure. + * The GOSUB instruction has the following memory layout. + * + * +------------------------------------------------------------+ + * | Opcode | Rsvd | Head Ptr | Tail Ptr | + * | 0x0c | | | | + * +------------------------------------------------------------+ + * |<- 8bits->|<- 4bits ->|<-- 26bits -->|<-- 26bits -->| + * + * We have only 26 bits each to represent the head and tail + * pointers even though the addresses itself are of 32 bit. However, this + * is not a problem because the addresses are 64 bit aligned and therefore + * the last 6 bits are always Zero's. Therefore, we right shift the address + * by 6 before embedding it into the GOSUB instruction. */ - intel_dsb_reg_write(dsb, DSB_PMCTRL(crtc->pipe, dsb->id), 0); - intel_dsb_reg_write_masked(dsb, DSB_PMCTRL_2(crtc->pipe, dsb->id), - DSB_FORCE_DEWAKE, 0); + head_tail = ((u64)(DSB_GOSUB_CONVERT_ADDR(head)) << DSB_GOSUB_HEAD_SHIFT) | + ((u64)(DSB_GOSUB_CONVERT_ADDR(tail)) << DSB_GOSUB_TAIL_SHIFT); + + intel_dsb_emit(dsb, lower_32_bits(head_tail), + (DSB_OPCODE_GOSUB << DSB_OPCODE_SHIFT) | + upper_32_bits(head_tail)); + + /* + * "NOTE: the instructions within the cacheline + * FOLLOWING the GOSUB instruction must be NOPs." + */ + intel_dsb_align_tail(dsb); +} + +void intel_dsb_gosub_finish(struct intel_dsb *dsb) +{ + intel_dsb_align_tail(dsb); + + /* + * Wa_16024917128 + * "Ensure that all subroutines called by GOSUB end with a cacheline of NOPs" + */ + intel_dsb_noop(dsb, 8); + + intel_dsb_buffer_flush_map(&dsb->dsb_buf); +} + +void intel_dsb_finish(struct intel_dsb *dsb) +{ intel_dsb_align_tail(dsb); intel_dsb_buffer_flush_map(&dsb->dsb_buf); @@ -537,6 +650,9 @@ static u32 dsb_error_int_status(struct intel_display *display) if (DISPLAY_VER(display) >= 14) errors |= DSB_ATS_FAULT_INT_STATUS; + if (DISPLAY_VER(display) >= 30) + errors |= DSB_GOSUB_INT_STATUS; + return errors; } @@ -551,17 +667,46 @@ static u32 dsb_error_int_en(struct intel_display *display) if (DISPLAY_VER(display) >= 14) errors |= DSB_ATS_FAULT_INT_EN; + /* + * Wa_16024917128 + * "Disable nested GOSUB interrupt (DSB_INTERRUPT bit 21)" + */ + if (0 && DISPLAY_VER(display) >= 30) + errors |= DSB_GOSUB_INT_EN; + return errors; } +/* + * FIXME calibrate these sensibly, ideally compute based on + * the number of regisetrs to be written. But that requires + * measuring the actual DSB execution speed on each platform + * (and the speed also depends on CDCLK and memory clock)... + */ +static int intel_dsb_noarm_exec_time_us(void) +{ + return 80; +} + +static int intel_dsb_arm_exec_time_us(void) +{ + return 20; +} + +int intel_dsb_exec_time_us(void) +{ + return intel_dsb_noarm_exec_time_us() + + intel_dsb_arm_exec_time_us(); +} + void intel_dsb_vblank_evade(struct intel_atomic_state *state, struct intel_dsb *dsb) { struct intel_crtc *crtc = dsb->crtc; const struct intel_crtc_state *crtc_state = intel_pre_commit_crtc_state(state, crtc); - /* FIXME calibrate sensibly */ - int latency = intel_usecs_to_scanlines(&crtc_state->hw.adjusted_mode, 20); + int latency = intel_usecs_to_scanlines(&crtc_state->hw.adjusted_mode, + intel_dsb_arm_exec_time_us()); int start, end; /* @@ -603,13 +748,11 @@ static void _intel_dsb_chain(struct intel_atomic_state *state, struct intel_display *display = to_intel_display(state->base.dev); struct intel_crtc *crtc = dsb->crtc; enum pipe pipe = crtc->pipe; - u32 tail; if (drm_WARN_ON(display->drm, dsb->id == chained_dsb->id)) return; - tail = chained_dsb->free_pos * 4; - if (drm_WARN_ON(display->drm, !IS_ALIGNED(tail, CACHELINE_BYTES))) + if (!assert_dsb_tail_is_aligned(chained_dsb)) return; intel_dsb_reg_write(dsb, DSB_CTRL(pipe, chained_dsb->id), @@ -629,13 +772,15 @@ static void _intel_dsb_chain(struct intel_atomic_state *state, intel_dsb_reg_write(dsb, DSB_PMCTRL(pipe, chained_dsb->id), DSB_ENABLE_DEWAKE | DSB_SCANLINE_FOR_DEWAKE(hw_dewake_scanline)); + } else { + intel_dsb_reg_write(dsb, DSB_PMCTRL(pipe, chained_dsb->id), 0); } intel_dsb_reg_write(dsb, DSB_HEAD(pipe, chained_dsb->id), - intel_dsb_buffer_ggtt_offset(&chained_dsb->dsb_buf)); + intel_dsb_head(chained_dsb)); intel_dsb_reg_write(dsb, DSB_TAIL(pipe, chained_dsb->id), - intel_dsb_buffer_ggtt_offset(&chained_dsb->dsb_buf) + tail); + intel_dsb_tail(chained_dsb)); if (ctrl & DSB_WAIT_FOR_VBLANK) { /* @@ -650,6 +795,13 @@ static void _intel_dsb_chain(struct intel_atomic_state *state, intel_dsb_wait_scanline_out(state, dsb, dsb_dewake_scanline_start(state, crtc), dsb_dewake_scanline_end(state, crtc)); + + /* + * DSB_FORCE_DEWAKE remains active even after DSB is + * disabled, so make sure to clear it. + */ + intel_dsb_reg_write_masked(dsb, DSB_PMCTRL_2(crtc->pipe, dsb->id), + DSB_FORCE_DEWAKE, 0); } } @@ -674,16 +826,19 @@ void intel_dsb_wait_vblank_delay(struct intel_atomic_state *state, intel_dsb_wait_usec(dsb, usecs); } -static void _intel_dsb_commit(struct intel_dsb *dsb, u32 ctrl, - int hw_dewake_scanline) +/** + * intel_dsb_commit() - Trigger workload execution of DSB. + * @dsb: DSB context + * + * This function is used to do actual write to hardware using DSB. + */ +void intel_dsb_commit(struct intel_dsb *dsb) { struct intel_crtc *crtc = dsb->crtc; struct intel_display *display = to_intel_display(crtc->base.dev); enum pipe pipe = crtc->pipe; - u32 tail; - tail = dsb->free_pos * 4; - if (drm_WARN_ON(display->drm, !IS_ALIGNED(tail, CACHELINE_BYTES))) + if (!assert_dsb_tail_is_aligned(dsb)) return; if (is_dsb_busy(display, pipe, dsb->id)) { @@ -693,7 +848,7 @@ static void _intel_dsb_commit(struct intel_dsb *dsb, u32 ctrl, } intel_de_write_fw(display, DSB_CTRL(pipe, dsb->id), - ctrl | DSB_ENABLE); + DSB_ENABLE); intel_de_write_fw(display, DSB_CHICKEN(pipe, dsb->id), dsb->chicken); @@ -702,45 +857,13 @@ static void _intel_dsb_commit(struct intel_dsb *dsb, u32 ctrl, dsb_error_int_status(display) | DSB_PROG_INT_STATUS | dsb_error_int_en(display) | DSB_PROG_INT_EN); - intel_de_write_fw(display, DSB_HEAD(pipe, dsb->id), - intel_dsb_buffer_ggtt_offset(&dsb->dsb_buf)); - - if (hw_dewake_scanline >= 0) { - int diff, position; + intel_de_write_fw(display, DSB_PMCTRL(pipe, dsb->id), 0); - intel_de_write_fw(display, DSB_PMCTRL(pipe, dsb->id), - DSB_ENABLE_DEWAKE | - DSB_SCANLINE_FOR_DEWAKE(hw_dewake_scanline)); - - /* - * Force DEwake immediately if we're already past - * or close to racing past the target scanline. - */ - position = intel_de_read_fw(display, PIPEDSL(display, pipe)) & PIPEDSL_LINE_MASK; - - diff = hw_dewake_scanline - position; - intel_de_write_fw(display, DSB_PMCTRL_2(pipe, dsb->id), - (diff >= 0 && diff < 5 ? DSB_FORCE_DEWAKE : 0) | - DSB_BLOCK_DEWAKE_EXTENSION); - } + intel_de_write_fw(display, DSB_HEAD(pipe, dsb->id), + intel_dsb_head(dsb)); intel_de_write_fw(display, DSB_TAIL(pipe, dsb->id), - intel_dsb_buffer_ggtt_offset(&dsb->dsb_buf) + tail); -} - -/** - * intel_dsb_commit() - Trigger workload execution of DSB. - * @dsb: DSB context - * @wait_for_vblank: wait for vblank before executing - * - * This function is used to do actual write to hardware using DSB. - */ -void intel_dsb_commit(struct intel_dsb *dsb, - bool wait_for_vblank) -{ - _intel_dsb_commit(dsb, - wait_for_vblank ? DSB_WAIT_FOR_VBLANK : 0, - wait_for_vblank ? dsb->hw_dewake_scanline : -1); + intel_dsb_tail(dsb)); } void intel_dsb_wait(struct intel_dsb *dsb) @@ -795,22 +918,22 @@ struct intel_dsb *intel_dsb_prepare(struct intel_atomic_state *state, enum intel_dsb_id dsb_id, unsigned int max_cmds) { - struct drm_i915_private *i915 = to_i915(state->base.dev); - intel_wakeref_t wakeref; + struct intel_display *display = to_intel_display(state); + struct ref_tracker *wakeref; struct intel_dsb *dsb; unsigned int size; - if (!HAS_DSB(i915)) + if (!HAS_DSB(display)) return NULL; - if (!i915->display.params.enable_dsb) + if (!display->params.enable_dsb) return NULL; dsb = kzalloc(sizeof(*dsb), GFP_KERNEL); if (!dsb) goto out; - wakeref = intel_runtime_pm_get(&i915->runtime_pm); + wakeref = intel_display_rpm_get(display); /* ~1 qword per instruction, full cachelines */ size = ALIGN(max_cmds * 8, CACHELINE_BYTES); @@ -818,7 +941,7 @@ struct intel_dsb *intel_dsb_prepare(struct intel_atomic_state *state, if (!intel_dsb_buffer_create(crtc, &dsb->dsb_buf, size)) goto out_put_rpm; - intel_runtime_pm_put(&i915->runtime_pm, wakeref); + intel_display_rpm_put(display, wakeref); dsb->id = dsb_id; dsb->crtc = crtc; @@ -831,10 +954,10 @@ struct intel_dsb *intel_dsb_prepare(struct intel_atomic_state *state, return dsb; out_put_rpm: - intel_runtime_pm_put(&i915->runtime_pm, wakeref); + intel_display_rpm_put(display, wakeref); kfree(dsb); out: - drm_info_once(&i915->drm, + drm_info_once(display->drm, "[CRTC:%d:%s] DSB %d queue setup failed, will fallback to MMIO for display HW programming\n", crtc->base.base.id, crtc->base.name, dsb_id); @@ -893,4 +1016,7 @@ void intel_dsb_irq_handler(struct intel_display *display, if (errors & DSB_POLL_ERR_INT_STATUS) drm_err(display->drm, "[CRTC:%d:%s] DSB %d poll error\n", crtc->base.base.id, crtc->base.name, dsb_id); + if (errors & DSB_GOSUB_INT_STATUS) + drm_err(display->drm, "[CRTC:%d:%s] DSB %d GOSUB programming error\n", + crtc->base.base.id, crtc->base.name, dsb_id); } |