void gm20b_ltc_isr(struct gk20a *g)
{
u32 mc_intr, ltc_intr;
int ltc, slice;
mc_intr = gk20a_readl(g, mc_intr_ltc_r());
gk20a_err(dev_from_gk20a(g), "mc_ltc_intr: %08x",
mc_intr);
for (ltc = 0; ltc < g->ltc_count; ltc++) {
if ((mc_intr & 1 << ltc) == 0)
continue;
for (slice = 0; slice < g->gr.slices_per_ltc; slice++) {
ltc_intr = gk20a_readl(g, ltc_ltc0_lts0_intr_r() +
proj_ltc_stride_v() * ltc +
proj_lts_stride_v() * slice);
gk20a_err(dev_from_gk20a(g), "ltc%d, slice %d: %08x",
ltc, slice, ltc_intr);
gk20a_writel(g, ltc_ltc0_lts0_intr_r() +
proj_ltc_stride_v() * ltc +
proj_lts_stride_v() * slice,
ltc_intr);
}
}
}
/*
* Performs a full flush of the L2 cache.
*/
void gm20b_flush_ltc(struct gk20a *g)
{
unsigned long timeout;
int ltc;
#define __timeout_init() \
do { \
timeout = jiffies + HZ; \
} while (0)
#define __timeout_check() \
do { \
if (tegra_platform_is_silicon() && \
time_after(jiffies, timeout)) { \
gk20a_err(dev_from_gk20a(g), "L2 flush timeout!"); \
break; \
} \
} while (0)
/* Clean... */
gk20a_writel(g, ltc_ltcs_ltss_tstg_cmgmt1_r(),
ltc_ltcs_ltss_tstg_cmgmt1_clean_pending_f() |
ltc_ltcs_ltss_tstg_cmgmt1_max_cycles_between_cleans_3_f() |
ltc_ltcs_ltss_tstg_cmgmt1_clean_wait_for_fb_to_pull_true_f() |
ltc_ltcs_ltss_tstg_cmgmt1_clean_evict_last_class_true_f() |
ltc_ltcs_ltss_tstg_cmgmt1_clean_evict_normal_class_true_f() |
ltc_ltcs_ltss_tstg_cmgmt1_clean_evict_first_class_true_f());
/* Wait on each LTC individually. */
for (ltc = 0; ltc < g->ltc_count; ltc++) {
u32 op_pending;
__timeout_init();
do {
int cmgmt1 = ltc_ltc0_ltss_tstg_cmgmt1_r() +
ltc * proj_ltc_stride_v();
op_pending = gk20a_readl(g, cmgmt1);
__timeout_check();
} while (op_pending &
ltc_ltc0_ltss_tstg_cmgmt1_clean_pending_f());
}
/* And invalidate. */
gk20a_writel(g, ltc_ltcs_ltss_tstg_cmgmt0_r(),
ltc_ltcs_ltss_tstg_cmgmt0_invalidate_pending_f() |
ltc_ltcs_ltss_tstg_cmgmt0_max_cycles_between_invalidates_3_f() |
ltc_ltcs_ltss_tstg_cmgmt0_invalidate_evict_last_class_true_f() |
ltc_ltcs_ltss_tstg_cmgmt0_invalidate_evict_normal_class_true_f() |
ltc_ltcs_ltss_tstg_cmgmt0_invalidate_evict_first_class_true_f());
/* Wait on each LTC individually. */
for (ltc = 0; ltc < g->ltc_count; ltc++) {
u32 op_pending;
__timeout_init();
do {
int cmgmt0 = ltc_ltc0_ltss_tstg_cmgmt0_r() +
ltc * proj_ltc_stride_v();
op_pending = gk20a_readl(g, cmgmt0);
__timeout_check();
} while (op_pending &
ltc_ltc0_ltss_tstg_cmgmt0_invalidate_pending_f());
}
}
int gm20b_ltc_cbc_ctrl(struct gk20a *g, enum gk20a_cbc_op op,
u32 min, u32 max)
{
int err = 0;
struct gr_gk20a *gr = &g->gr;
u32 ltc, slice, ctrl1, val, hw_op = 0;
s32 retry = 200;
u32 slices_per_ltc = ltc_ltcs_ltss_cbc_param_slices_per_ltc_v(
gk20a_readl(g, ltc_ltcs_ltss_cbc_param_r()));
gk20a_dbg_fn("");
trace_gk20a_ltc_cbc_ctrl_start(g->dev->name, op, min, max);
if (gr->compbit_store.mem.size == 0)
return 0;
mutex_lock(&g->mm.l2_op_lock);
if (op == gk20a_cbc_op_clear) {
gk20a_writel(g, ltc_ltcs_ltss_cbc_ctrl2_r(),
ltc_ltcs_ltss_cbc_ctrl2_clear_lower_bound_f(min));
gk20a_writel(g, ltc_ltcs_ltss_cbc_ctrl3_r(),
ltc_ltcs_ltss_cbc_ctrl3_clear_upper_bound_f(max));
hw_op = ltc_ltcs_ltss_cbc_ctrl1_clear_active_f();
} else if (op == gk20a_cbc_op_clean) {
hw_op = ltc_ltcs_ltss_cbc_ctrl1_clean_active_f();
} else if (op == gk20a_cbc_op_invalidate) {
hw_op = ltc_ltcs_ltss_cbc_ctrl1_invalidate_active_f();
} else {
BUG_ON(1);
}
gk20a_writel(g, ltc_ltcs_ltss_cbc_ctrl1_r(),
gk20a_readl(g, ltc_ltcs_ltss_cbc_ctrl1_r()) | hw_op);
for (ltc = 0; ltc < g->ltc_count; ltc++) {
for (slice = 0; slice < slices_per_ltc; slice++) {
ctrl1 = ltc_ltc0_lts0_cbc_ctrl1_r() +
ltc * proj_ltc_stride_v() +
slice * proj_lts_stride_v();
retry = 200;
do {
val = gk20a_readl(g, ctrl1);
if (!(val & hw_op))
break;
retry--;
udelay(5);
} while (retry >= 0 ||
!tegra_platform_is_silicon());
if (retry < 0 && tegra_platform_is_silicon()) {
gk20a_err(dev_from_gk20a(g),
"comp tag clear timeout\n");
err = -EBUSY;
goto out;
}
}
}
out:
trace_gk20a_ltc_cbc_ctrl_done(g->dev->name);
mutex_unlock(&g->mm.l2_op_lock);
return err;
}
static int gk20a_ltc_cbc_ctrl(struct gk20a *g, enum gk20a_cbc_op op,
u32 min, u32 max)
{
int err = 0;
struct gr_gk20a *gr = &g->gr;
u32 fbp, slice, ctrl1, val, hw_op = 0;
unsigned long end_jiffies = jiffies +
msecs_to_jiffies(gk20a_get_gr_idle_timeout(g));
u32 delay = GR_IDLE_CHECK_DEFAULT;
u32 slices_per_fbp =
ltc_ltcs_ltss_cbc_param_slices_per_fbp_v(
gk20a_readl(g, ltc_ltcs_ltss_cbc_param_r()));
gk20a_dbg_fn("");
if (gr->compbit_store.size == 0)
return 0;
mutex_lock(&g->mm.l2_op_lock);
if (op == gk20a_cbc_op_clear) {
gk20a_writel(g, ltc_ltcs_ltss_cbc_ctrl2_r(),
ltc_ltcs_ltss_cbc_ctrl2_clear_lower_bound_f(min));
gk20a_writel(g, ltc_ltcs_ltss_cbc_ctrl3_r(),
ltc_ltcs_ltss_cbc_ctrl3_clear_upper_bound_f(max));
hw_op = ltc_ltcs_ltss_cbc_ctrl1_clear_active_f();
} else if (op == gk20a_cbc_op_clean) {
hw_op = ltc_ltcs_ltss_cbc_ctrl1_clean_active_f();
} else if (op == gk20a_cbc_op_invalidate) {
hw_op = ltc_ltcs_ltss_cbc_ctrl1_invalidate_active_f();
} else {
BUG_ON(1);
}
gk20a_writel(g, ltc_ltcs_ltss_cbc_ctrl1_r(),
gk20a_readl(g, ltc_ltcs_ltss_cbc_ctrl1_r()) | hw_op);
for (fbp = 0; fbp < gr->num_fbps; fbp++) {
for (slice = 0; slice < slices_per_fbp; slice++) {
delay = GR_IDLE_CHECK_DEFAULT;
ctrl1 = ltc_ltc0_lts0_cbc_ctrl1_r() +
fbp * proj_ltc_stride_v() +
slice * proj_lts_stride_v();
do {
val = gk20a_readl(g, ctrl1);
if (!(val & hw_op))
break;
usleep_range(delay, delay * 2);
delay = min_t(u32, delay << 1,
GR_IDLE_CHECK_MAX);
} while (time_before(jiffies, end_jiffies) ||
!tegra_platform_is_silicon());
if (!time_before(jiffies, end_jiffies)) {
gk20a_err(dev_from_gk20a(g),
"comp tag clear timeout\n");
err = -EBUSY;
goto out;
}
}
}
out:
mutex_unlock(&g->mm.l2_op_lock);
return 0;
}