Quellcodebibliothek Statistik Leitseite products/Sources/formale Sprachen/C/Linux/drivers/gpu/drm/i915/display/   (Open Source Betriebssystem Version 6.17.9©)  Datei vom 24.10.2025 mit Größe 19 kB image not shown  

Quelle  intel_pmdemand.c   Sprache: C

 
// SPDX-License-Identifier: MIT
/*
 * Copyright © 2023 Intel Corporation
 */


#include <linux/bitops.h>

#include <drm/drm_print.h>

#include "i915_utils.h"
#include "intel_atomic.h"
#include "intel_bw.h"
#include "intel_cdclk.h"
#include "intel_de.h"
#include "intel_display_regs.h"
#include "intel_display_trace.h"
#include "intel_pmdemand.h"
#include "intel_step.h"
#include "skl_watermark.h"

struct pmdemand_params {
 u16 qclk_gv_bw;
 u8 voltage_index;
 u8 qclk_gv_index;
 u8 active_pipes;
 u8 active_dbufs; /* pre-Xe3 only */
 /* Total number of non type C active phys from active_phys_mask */
 u8 active_phys;
 u8 plls;
 u16 cdclk_freq_mhz;
 /* max from ddi_clocks[] */
 u16 ddiclk_max;
 u8 scalers;  /* pre-Xe3 only */
};

struct intel_pmdemand_state {
 struct intel_global_state base;

 /* Maintain a persistent list of port clocks across all crtcs */
 int ddi_clocks[I915_MAX_PIPES];

 /* Maintain a persistent list of non type C phys mask */
 u16 active_combo_phys_mask;

 /* Parameters to be configured in the pmdemand registers */
 struct pmdemand_params params;
};

struct intel_pmdemand_state *to_intel_pmdemand_state(struct intel_global_state *obj_state)
{
 return container_of(obj_state, struct intel_pmdemand_state, base);
}

static struct intel_global_state *
intel_pmdemand_duplicate_state(struct intel_global_obj *obj)
{
 struct intel_pmdemand_state *pmdemand_state;

 pmdemand_state = kmemdup(obj->state, sizeof(*pmdemand_state), GFP_KERNEL);
 if (!pmdemand_state)
  return NULL;

 return &pmdemand_state->base;
}

static void intel_pmdemand_destroy_state(struct intel_global_obj *obj,
      struct intel_global_state *state)
{
 kfree(state);
}

static const struct intel_global_state_funcs intel_pmdemand_funcs = {
 .atomic_duplicate_state = intel_pmdemand_duplicate_state,
 .atomic_destroy_state = intel_pmdemand_destroy_state,
};

static struct intel_pmdemand_state *
intel_atomic_get_pmdemand_state(struct intel_atomic_state *state)
{
 struct intel_display *display = to_intel_display(state);
 struct intel_global_state *pmdemand_state =
  intel_atomic_get_global_obj_state(state,
        &display->pmdemand.obj);

 if (IS_ERR(pmdemand_state))
  return ERR_CAST(pmdemand_state);

 return to_intel_pmdemand_state(pmdemand_state);
}

static struct intel_pmdemand_state *
intel_atomic_get_old_pmdemand_state(struct intel_atomic_state *state)
{
 struct intel_display *display = to_intel_display(state);
 struct intel_global_state *pmdemand_state =
  intel_atomic_get_old_global_obj_state(state,
            &display->pmdemand.obj);

 if (!pmdemand_state)
  return NULL;

 return to_intel_pmdemand_state(pmdemand_state);
}

static struct intel_pmdemand_state *
intel_atomic_get_new_pmdemand_state(struct intel_atomic_state *state)
{
 struct intel_display *display = to_intel_display(state);
 struct intel_global_state *pmdemand_state =
  intel_atomic_get_new_global_obj_state(state,
            &display->pmdemand.obj);

 if (!pmdemand_state)
  return NULL;

 return to_intel_pmdemand_state(pmdemand_state);
}

int intel_pmdemand_init(struct intel_display *display)
{
 struct intel_pmdemand_state *pmdemand_state;

 pmdemand_state = kzalloc(sizeof(*pmdemand_state), GFP_KERNEL);
 if (!pmdemand_state)
  return -ENOMEM;

 intel_atomic_global_obj_init(display, &display->pmdemand.obj,
         &pmdemand_state->base,
         &intel_pmdemand_funcs);

 if (IS_DISPLAY_VERx100_STEP(display, 1400, STEP_A0, STEP_C0))
  /* Wa_14016740474 */
  intel_de_rmw(display, XELPD_CHICKEN_DCPR_3, 0, DMD_RSP_TIMEOUT_DISABLE);

 return 0;
}

void intel_pmdemand_init_early(struct intel_display *display)
{
 mutex_init(&display->pmdemand.lock);
 init_waitqueue_head(&display->pmdemand.waitqueue);
}

void
intel_pmdemand_update_phys_mask(struct intel_display *display,
    struct intel_encoder *encoder,
    struct intel_pmdemand_state *pmdemand_state,
    bool set_bit)
{
 enum phy phy;

 if (DISPLAY_VER(display) < 14)
  return;

 if (!encoder)
  return;

 if (intel_encoder_is_tc(encoder))
  return;

 phy = intel_encoder_to_phy(encoder);

 if (set_bit)
  pmdemand_state->active_combo_phys_mask |= BIT(phy);
 else
  pmdemand_state->active_combo_phys_mask &= ~BIT(phy);
}

void
intel_pmdemand_update_port_clock(struct intel_display *display,
     struct intel_pmdemand_state *pmdemand_state,
     enum pipe pipe, int port_clock)
{
 if (DISPLAY_VER(display) < 14)
  return;

 pmdemand_state->ddi_clocks[pipe] = port_clock;
}

static void
intel_pmdemand_update_max_ddiclk(struct intel_display *display,
     struct intel_atomic_state *state,
     struct intel_pmdemand_state *pmdemand_state)
{
 int max_ddiclk = 0;
 const struct intel_crtc_state *new_crtc_state;
 struct intel_crtc *crtc;
 int i;

 for_each_new_intel_crtc_in_state(state, crtc, new_crtc_state, i)
  intel_pmdemand_update_port_clock(display, pmdemand_state,
       crtc->pipe,
       new_crtc_state->port_clock);

 for (i = 0; i < ARRAY_SIZE(pmdemand_state->ddi_clocks); i++)
  max_ddiclk = max(pmdemand_state->ddi_clocks[i], max_ddiclk);

 pmdemand_state->params.ddiclk_max = DIV_ROUND_UP(max_ddiclk, 1000);
}

static void
intel_pmdemand_update_connector_phys(struct intel_display *display,
         struct intel_atomic_state *state,
         struct drm_connector_state *conn_state,
         bool set_bit,
         struct intel_pmdemand_state *pmdemand_state)
{
 struct intel_encoder *encoder = to_intel_encoder(conn_state->best_encoder);
 struct intel_crtc *crtc = to_intel_crtc(conn_state->crtc);
 struct intel_crtc_state *crtc_state;

 if (!crtc)
  return;

 if (set_bit)
  crtc_state = intel_atomic_get_new_crtc_state(state, crtc);
 else
  crtc_state = intel_atomic_get_old_crtc_state(state, crtc);

 if (!crtc_state->hw.active)
  return;

 intel_pmdemand_update_phys_mask(display, encoder, pmdemand_state,
     set_bit);
}

static void
intel_pmdemand_update_active_non_tc_phys(struct intel_display *display,
      struct intel_atomic_state *state,
      struct intel_pmdemand_state *pmdemand_state)
{
 struct drm_connector_state *old_conn_state;
 struct drm_connector_state *new_conn_state;
 struct drm_connector *connector;
 int i;

 for_each_oldnew_connector_in_state(&state->base, connector,
        old_conn_state, new_conn_state, i) {
  if (!intel_connector_needs_modeset(state, connector))
   continue;

  /* First clear the active phys in the old connector state */
  intel_pmdemand_update_connector_phys(display, state,
           old_conn_state, false,
           pmdemand_state);

  /* Then set the active phys in new connector state */
  intel_pmdemand_update_connector_phys(display, state,
           new_conn_state, true,
           pmdemand_state);
 }

 pmdemand_state->params.active_phys =
  min_t(u16, hweight16(pmdemand_state->active_combo_phys_mask),
        7);
}

static bool
intel_pmdemand_encoder_has_tc_phy(struct intel_display *display,
      struct intel_encoder *encoder)
{
 return encoder && intel_encoder_is_tc(encoder);
}

static bool
intel_pmdemand_connector_needs_update(struct intel_atomic_state *state)
{
 struct intel_display *display = to_intel_display(state);
 struct drm_connector_state *old_conn_state;
 struct drm_connector_state *new_conn_state;
 struct drm_connector *connector;
 int i;

 for_each_oldnew_connector_in_state(&state->base, connector,
        old_conn_state, new_conn_state, i) {
  struct intel_encoder *old_encoder =
   to_intel_encoder(old_conn_state->best_encoder);
  struct intel_encoder *new_encoder =
   to_intel_encoder(new_conn_state->best_encoder);

  if (!intel_connector_needs_modeset(state, connector))
   continue;

  if (old_encoder == new_encoder ||
      (intel_pmdemand_encoder_has_tc_phy(display, old_encoder) &&
       intel_pmdemand_encoder_has_tc_phy(display, new_encoder)))
   continue;

  return true;
 }

 return false;
}

static bool intel_pmdemand_needs_update(struct intel_atomic_state *state)
{
 const struct intel_crtc_state *new_crtc_state, *old_crtc_state;
 struct intel_crtc *crtc;
 int i;

 if (intel_bw_pmdemand_needs_update(state))
  return true;

 if (intel_dbuf_pmdemand_needs_update(state))
  return true;

 if (intel_cdclk_pmdemand_needs_update(state))
  return true;

 for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state,
         new_crtc_state, i)
  if (new_crtc_state->port_clock != old_crtc_state->port_clock)
   return true;

 return intel_pmdemand_connector_needs_update(state);
}

int intel_pmdemand_atomic_check(struct intel_atomic_state *state)
{
 struct intel_display *display = to_intel_display(state);
 const struct intel_bw_state *new_bw_state;
 const struct intel_cdclk_state *new_cdclk_state;
 const struct intel_dbuf_state *new_dbuf_state;
 struct intel_pmdemand_state *new_pmdemand_state;

 if (DISPLAY_VER(display) < 14)
  return 0;

 if (!intel_pmdemand_needs_update(state))
  return 0;

 new_pmdemand_state = intel_atomic_get_pmdemand_state(state);
 if (IS_ERR(new_pmdemand_state))
  return PTR_ERR(new_pmdemand_state);

 new_bw_state = intel_atomic_get_bw_state(state);
 if (IS_ERR(new_bw_state))
  return PTR_ERR(new_bw_state);

 /* firmware will calculate the qclk_gv_index, requirement is set to 0 */
 new_pmdemand_state->params.qclk_gv_index = 0;
 new_pmdemand_state->params.qclk_gv_bw = intel_bw_qgv_point_peakbw(new_bw_state);

 new_dbuf_state = intel_atomic_get_dbuf_state(state);
 if (IS_ERR(new_dbuf_state))
  return PTR_ERR(new_dbuf_state);

 if (DISPLAY_VER(display) < 30) {
  new_pmdemand_state->params.active_dbufs =
   min_t(u8, intel_dbuf_num_enabled_slices(new_dbuf_state), 3);
  new_pmdemand_state->params.active_pipes =
   min_t(u8, intel_dbuf_num_active_pipes(new_dbuf_state), 3);
 } else {
  new_pmdemand_state->params.active_pipes =
   min_t(u8, intel_dbuf_num_active_pipes(new_dbuf_state), INTEL_NUM_PIPES(display));
 }

 new_cdclk_state = intel_atomic_get_cdclk_state(state);
 if (IS_ERR(new_cdclk_state))
  return PTR_ERR(new_cdclk_state);

 new_pmdemand_state->params.voltage_index =
  intel_cdclk_actual_voltage_level(new_cdclk_state);
 new_pmdemand_state->params.cdclk_freq_mhz =
  DIV_ROUND_UP(intel_cdclk_actual(new_cdclk_state), 1000);

 intel_pmdemand_update_max_ddiclk(display, state, new_pmdemand_state);

 intel_pmdemand_update_active_non_tc_phys(display, state, new_pmdemand_state);

 /*
 * Active_PLLs starts with 1 because of CDCLK PLL.
 * TODO: Missing to account genlock filter when it gets used.
 */

 new_pmdemand_state->params.plls =
  min_t(u16, new_pmdemand_state->params.active_phys + 1, 7);

 /*
 * Setting scalers to max as it can not be calculated during flips and
 * fastsets without taking global states locks.
 */

 new_pmdemand_state->params.scalers = 7;

 if (state->base.allow_modeset)
  return intel_atomic_serialize_global_state(&new_pmdemand_state->base);
 else
  return intel_atomic_lock_global_state(&new_pmdemand_state->base);
}

static bool intel_pmdemand_check_prev_transaction(struct intel_display *display)
{
 return !(intel_de_wait_for_clear(display,
      XELPDP_INITIATE_PMDEMAND_REQUEST(1),
      XELPDP_PMDEMAND_REQ_ENABLE, 10) ||
   intel_de_wait_for_clear(display,
      GEN12_DCPR_STATUS_1,
      XELPDP_PMDEMAND_INFLIGHT_STATUS, 10));
}

void
intel_pmdemand_init_pmdemand_params(struct intel_display *display,
        struct intel_pmdemand_state *pmdemand_state)
{
 u32 reg1, reg2;

 if (DISPLAY_VER(display) < 14)
  return;

 mutex_lock(&display->pmdemand.lock);
 if (drm_WARN_ON(display->drm,
   !intel_pmdemand_check_prev_transaction(display))) {
  memset(&pmdemand_state->params, 0,
         sizeof(pmdemand_state->params));
  goto unlock;
 }

 reg1 = intel_de_read(display, XELPDP_INITIATE_PMDEMAND_REQUEST(0));

 reg2 = intel_de_read(display, XELPDP_INITIATE_PMDEMAND_REQUEST(1));

 pmdemand_state->params.qclk_gv_bw =
  REG_FIELD_GET(XELPDP_PMDEMAND_QCLK_GV_BW_MASK, reg1);
 pmdemand_state->params.voltage_index =
  REG_FIELD_GET(XELPDP_PMDEMAND_VOLTAGE_INDEX_MASK, reg1);
 pmdemand_state->params.qclk_gv_index =
  REG_FIELD_GET(XELPDP_PMDEMAND_QCLK_GV_INDEX_MASK, reg1);
 pmdemand_state->params.active_phys =
  REG_FIELD_GET(XELPDP_PMDEMAND_PHYS_MASK, reg1);

 pmdemand_state->params.cdclk_freq_mhz =
  REG_FIELD_GET(XELPDP_PMDEMAND_CDCLK_FREQ_MASK, reg2);
 pmdemand_state->params.ddiclk_max =
  REG_FIELD_GET(XELPDP_PMDEMAND_DDICLK_FREQ_MASK, reg2);

 if (DISPLAY_VER(display) >= 30) {
  pmdemand_state->params.active_pipes =
   REG_FIELD_GET(XE3_PMDEMAND_PIPES_MASK, reg1);
 } else {
  pmdemand_state->params.active_pipes =
   REG_FIELD_GET(XELPDP_PMDEMAND_PIPES_MASK, reg1);
  pmdemand_state->params.active_dbufs =
   REG_FIELD_GET(XELPDP_PMDEMAND_DBUFS_MASK, reg1);

  pmdemand_state->params.scalers =
   REG_FIELD_GET(XELPDP_PMDEMAND_SCALERS_MASK, reg2);
 }

unlock:
 mutex_unlock(&display->pmdemand.lock);
}

static bool intel_pmdemand_req_complete(struct intel_display *display)
{
 return !(intel_de_read(display, XELPDP_INITIATE_PMDEMAND_REQUEST(1)) &
   XELPDP_PMDEMAND_REQ_ENABLE);
}

static void intel_pmdemand_poll(struct intel_display *display)
{
 const unsigned int timeout_ms = 10;
 u32 status;
 int ret;

 ret = intel_de_wait_custom(display, XELPDP_INITIATE_PMDEMAND_REQUEST(1),
       XELPDP_PMDEMAND_REQ_ENABLE, 0,
       50, timeout_ms, &status);

 if (ret == -ETIMEDOUT)
  drm_err(display->drm,
   "timed out waiting for Punit PM Demand Response within %ums (status 0x%08x)\n",
   timeout_ms, status);
}

static void intel_pmdemand_wait(struct intel_display *display)
{
 /* Wa_14024400148 For lnl use polling method */
 if (DISPLAY_VER(display) == 20) {
  intel_pmdemand_poll(display);
 } else {
  if (!wait_event_timeout(display->pmdemand.waitqueue,
     intel_pmdemand_req_complete(display),
     msecs_to_jiffies_timeout(10)))
   drm_err(display->drm,
    "timed out waiting for Punit PM Demand Response\n");
 }
}

/* Required to be programmed during Display Init Sequences. */
void intel_pmdemand_program_dbuf(struct intel_display *display,
     u8 dbuf_slices)
{
 u32 dbufs = min_t(u32, hweight8(dbuf_slices), 3);

 /* PM Demand only tracks active dbufs on pre-Xe3 platforms */
 if (DISPLAY_VER(display) >= 30)
  return;

 mutex_lock(&display->pmdemand.lock);
 if (drm_WARN_ON(display->drm,
   !intel_pmdemand_check_prev_transaction(display)))
  goto unlock;

 intel_de_rmw(display, XELPDP_INITIATE_PMDEMAND_REQUEST(0),
       XELPDP_PMDEMAND_DBUFS_MASK,
       REG_FIELD_PREP(XELPDP_PMDEMAND_DBUFS_MASK, dbufs));
 intel_de_rmw(display, XELPDP_INITIATE_PMDEMAND_REQUEST(1), 0,
       XELPDP_PMDEMAND_REQ_ENABLE);

 intel_pmdemand_wait(display);

unlock:
 mutex_unlock(&display->pmdemand.lock);
}

static void
intel_pmdemand_update_params(struct intel_display *display,
        const struct intel_pmdemand_state *new,
        const struct intel_pmdemand_state *old,
        u32 *reg1, u32 *reg2, bool serialized)
{
 /*
 * The pmdemand parameter updates happens in two steps. Pre plane and
 * post plane updates. During the pre plane, as DE might still be
 * handling with some old operations, to avoid unexpected performance
 * issues, program the pmdemand parameters with higher of old and new
 * values. And then after once settled, use the new parameter values
 * as part of the post plane update.
 *
 * If the pmdemand params update happens without modeset allowed, this
 * means we can't serialize the updates. So that implies possibility of
 * some parallel atomic commits affecting the pmdemand parameters. In
 * that case, we need to consider the current values from the register
 * as well. So in pre-plane case, we need to check the max of old, new
 * and current register value if not serialized. In post plane update
 * we need to consider max of new and current register value if not
 * serialized
 */


#define update_reg(reg, field, mask) do { \
 u32 current_val = serialized ? 0 : REG_FIELD_GET((mask), *(reg)); \
 u32 old_val = old ? old->params.field : 0; \
 u32 new_val = new->params.field; \
\
 *(reg) &= ~(mask); \
 *(reg) |= REG_FIELD_PREP((mask), max3(old_val, new_val, current_val)); \
while (0)

 /* Set 1*/
 update_reg(reg1, qclk_gv_bw, XELPDP_PMDEMAND_QCLK_GV_BW_MASK);
 update_reg(reg1, voltage_index, XELPDP_PMDEMAND_VOLTAGE_INDEX_MASK);
 update_reg(reg1, qclk_gv_index, XELPDP_PMDEMAND_QCLK_GV_INDEX_MASK);
 update_reg(reg1, active_phys, XELPDP_PMDEMAND_PHYS_MASK);

 /* Set 2*/
 update_reg(reg2, cdclk_freq_mhz, XELPDP_PMDEMAND_CDCLK_FREQ_MASK);
 update_reg(reg2, ddiclk_max, XELPDP_PMDEMAND_DDICLK_FREQ_MASK);
 update_reg(reg2, plls, XELPDP_PMDEMAND_PLLS_MASK);

 if (DISPLAY_VER(display) >= 30) {
  update_reg(reg1, active_pipes, XE3_PMDEMAND_PIPES_MASK);
 } else {
  update_reg(reg1, active_pipes, XELPDP_PMDEMAND_PIPES_MASK);
  update_reg(reg1, active_dbufs, XELPDP_PMDEMAND_DBUFS_MASK);

  update_reg(reg2, scalers, XELPDP_PMDEMAND_SCALERS_MASK);
 }

#undef update_reg
}

static void
intel_pmdemand_program_params(struct intel_display *display,
         const struct intel_pmdemand_state *new,
         const struct intel_pmdemand_state *old,
         bool serialized)
{
 bool changed = false;
 u32 reg1, mod_reg1;
 u32 reg2, mod_reg2;

 mutex_lock(&display->pmdemand.lock);
 if (drm_WARN_ON(display->drm,
   !intel_pmdemand_check_prev_transaction(display)))
  goto unlock;

 reg1 = intel_de_read(display, XELPDP_INITIATE_PMDEMAND_REQUEST(0));
 mod_reg1 = reg1;

 reg2 = intel_de_read(display, XELPDP_INITIATE_PMDEMAND_REQUEST(1));
 mod_reg2 = reg2;

 intel_pmdemand_update_params(display, new, old, &mod_reg1, &mod_reg2,
         serialized);

 if (reg1 != mod_reg1) {
  intel_de_write(display, XELPDP_INITIATE_PMDEMAND_REQUEST(0),
          mod_reg1);
  changed = true;
 }

 if (reg2 != mod_reg2) {
  intel_de_write(display, XELPDP_INITIATE_PMDEMAND_REQUEST(1),
          mod_reg2);
  changed = true;
 }

 /* Initiate pm demand request only if register values are changed */
 if (!changed)
  goto unlock;

 drm_dbg_kms(display->drm,
      "initiate pmdemand request values: (0x%x 0x%x)\n",
      mod_reg1, mod_reg2);

 intel_de_rmw(display, XELPDP_INITIATE_PMDEMAND_REQUEST(1), 0,
       XELPDP_PMDEMAND_REQ_ENABLE);

 intel_pmdemand_wait(display);

unlock:
 mutex_unlock(&display->pmdemand.lock);
}

static bool
intel_pmdemand_state_changed(const struct intel_pmdemand_state *new,
        const struct intel_pmdemand_state *old)
{
 return memcmp(&new->params, &old->params, sizeof(new->params)) != 0;
}

void intel_pmdemand_pre_plane_update(struct intel_atomic_state *state)
{
 struct intel_display *display = to_intel_display(state);
 const struct intel_pmdemand_state *new_pmdemand_state =
  intel_atomic_get_new_pmdemand_state(state);
 const struct intel_pmdemand_state *old_pmdemand_state =
  intel_atomic_get_old_pmdemand_state(state);

 if (DISPLAY_VER(display) < 14)
  return;

 if (!new_pmdemand_state ||
     !intel_pmdemand_state_changed(new_pmdemand_state,
       old_pmdemand_state))
  return;

 WARN_ON(!new_pmdemand_state->base.changed);

 intel_pmdemand_program_params(display, new_pmdemand_state,
          old_pmdemand_state,
          intel_atomic_global_state_is_serialized(state));
}

void intel_pmdemand_post_plane_update(struct intel_atomic_state *state)
{
 struct intel_display *display = to_intel_display(state);
 const struct intel_pmdemand_state *new_pmdemand_state =
  intel_atomic_get_new_pmdemand_state(state);
 const struct intel_pmdemand_state *old_pmdemand_state =
  intel_atomic_get_old_pmdemand_state(state);

 if (DISPLAY_VER(display) < 14)
  return;

 if (!new_pmdemand_state ||
     !intel_pmdemand_state_changed(new_pmdemand_state,
       old_pmdemand_state))
  return;

 WARN_ON(!new_pmdemand_state->base.changed);

 intel_pmdemand_program_params(display, new_pmdemand_state, NULL,
          intel_atomic_global_state_is_serialized(state));
}

Messung V0.5
C=97 H=96 G=96

¤ Dauer der Verarbeitung: 0.5 Sekunden  ¤

*© Formatika GbR, Deutschland






Wurzel

Suchen

Beweissystem der NASA

Beweissystem Isabelle

NIST Cobol Testsuite

Cephes Mathematical Library

Wiener Entwicklungsmethode

Haftungshinweis

Die Informationen auf dieser Webseite wurden nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit, noch Qualität der bereit gestellten Informationen zugesichert.

Bemerkung:

Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.