[PATCH v2 4/4] drm/msm/dpu: shift IRQ indices by 1
Dmitry Baryshkov
dmitry.baryshkov at linaro.org
Fri Jul 28 15:03:35 UTC 2023
On Thu, 27 Jul 2023 at 22:58, Marijn Suijten
<marijn.suijten at somainline.org> wrote:
>
> Title nit: How about making this "UP by 1"?
>
> On 2023-07-27 18:04:55, Dmitry Baryshkov wrote:
> > In order to simplify IRQ declarations, shift IRQ indices by 1. This
>
> Same here, UP by one.
>
> > makes 0 the 'no IRQ' value. Thanks to this change, we do no longer have
> > to explicitly set the 'no interrupt' fields in catalog structures.
>
> "explicitly set unused interrupts to -1" or
> "explicitly signifiy 'no interrupt' fields with -1"?
>
> On the other hand this might be confusing since -1 thanks to this patch
> no longer means 'no interrupt'. Also okay to leave it as-is.
>
> > Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov at linaro.org>
> > ---
> > .../msm/disp/dpu1/catalog/dpu_3_0_msm8998.h | 4 --
> > .../msm/disp/dpu1/catalog/dpu_4_0_sdm845.h | 4 --
> > .../msm/disp/dpu1/catalog/dpu_5_0_sm8150.h | 8 ---
> > .../msm/disp/dpu1/catalog/dpu_5_1_sc8180x.h | 10 ----
> > .../msm/disp/dpu1/catalog/dpu_5_4_sm6125.h | 3 --
> > .../msm/disp/dpu1/catalog/dpu_6_0_sm8250.h | 8 ---
> > .../msm/disp/dpu1/catalog/dpu_6_2_sc7180.h | 3 --
> > .../msm/disp/dpu1/catalog/dpu_6_3_sm6115.h | 1 -
> > .../msm/disp/dpu1/catalog/dpu_6_4_sm6350.h | 3 --
> > .../msm/disp/dpu1/catalog/dpu_6_5_qcm2290.h | 1 -
> > .../msm/disp/dpu1/catalog/dpu_6_9_sm6375.h | 1 -
> > .../msm/disp/dpu1/catalog/dpu_7_0_sm8350.h | 8 ---
> > .../msm/disp/dpu1/catalog/dpu_7_2_sc7280.h | 6 ---
> > .../msm/disp/dpu1/catalog/dpu_8_0_sc8280xp.h | 13 -----
> > .../msm/disp/dpu1/catalog/dpu_8_1_sm8450.h | 12 -----
> > .../msm/disp/dpu1/catalog/dpu_9_0_sm8550.h | 12 -----
> > drivers/gpu/drm/msm/disp/dpu1/dpu_core_irq.h | 6 +--
> > drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c | 2 +-
> > .../gpu/drm/msm/disp/dpu1/dpu_encoder_phys.h | 2 +-
> > .../gpu/drm/msm/disp/dpu1/dpu_hw_catalog.h | 14 +++---
> > .../gpu/drm/msm/disp/dpu1/dpu_hw_interrupts.c | 49 +++++++++----------
> > .../gpu/drm/msm/disp/dpu1/dpu_hw_interrupts.h | 6 +--
> > 22 files changed, 39 insertions(+), 137 deletions(-)
> >
>
> <cut out all the repeated catalog changes>
>
> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_core_irq.h b/drivers/gpu/drm/msm/disp/dpu1/dpu_core_irq.h
> > index ba06312cbb16..7c286bafb948 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_core_irq.h
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_core_irq.h
> > @@ -37,7 +37,7 @@ irqreturn_t dpu_core_irq(struct msm_kms *kms);
> > */
> > u32 dpu_core_irq_read(
> > struct dpu_kms *dpu_kms,
> > - int irq_idx);
> > + unsigned int irq_idx);
> >
> > /**
> > * dpu_core_irq_register_callback - For registering callback function on IRQ
> > @@ -52,7 +52,7 @@ u32 dpu_core_irq_read(
> > */
> > int dpu_core_irq_register_callback(
> > struct dpu_kms *dpu_kms,
> > - int irq_idx,
> > + unsigned int irq_idx,
> > void (*irq_cb)(void *arg),
> > void *irq_arg);
> >
> > @@ -67,7 +67,7 @@ int dpu_core_irq_register_callback(
> > */
> > int dpu_core_irq_unregister_callback(
> > struct dpu_kms *dpu_kms,
> > - int irq_idx);
> > + unsigned int irq_idx);
> >
> > /**
> > * dpu_debugfs_core_irq_init - register core irq debugfs
> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c
> > index 051447a3620c..8ccfeb002b5f 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c
> > @@ -2555,7 +2555,7 @@ void dpu_encoder_phys_init(struct dpu_encoder_phys *phys_enc,
> > phys_enc->enable_state = DPU_ENC_DISABLED;
> >
> > for (i = 0; i < ARRAY_SIZE(phys_enc->irq); i++)
> > - phys_enc->irq[i] = -EINVAL;
> > + phys_enc->irq[i] = 0;
>
> phys_enc seems to always be kzalloc'ed before it is passed here, so
> maybe we don't need to change the initialization value at all unless you
> want to be super-specific. And could memset() otherwise create faster
> code?
Sure, I should probably drop this.
>
> >
> > atomic_set(&phys_enc->vblank_refcount, 0);
> > atomic_set(&phys_enc->pending_kickoff_cnt, 0);
> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder_phys.h b/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder_phys.h
> > index f91661a69888..e203f3775ed3 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder_phys.h
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_encoder_phys.h
> > @@ -193,7 +193,7 @@ struct dpu_encoder_phys {
> > atomic_t pending_ctlstart_cnt;
> > atomic_t pending_kickoff_cnt;
> > wait_queue_head_t pending_kickoff_wq;
> > - int irq[INTR_IDX_MAX];
> > + unsigned int irq[INTR_IDX_MAX];
> > bool has_intf_te;
> > };
> >
> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.h b/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.h
> > index c72ed0e35dce..945b88c5ab58 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.h
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.h
> > @@ -524,7 +524,7 @@ struct dpu_ubwc_cfg {
> > */
> > struct dpu_ctl_cfg {
> > DPU_HW_BLK_INFO;
> > - s32 intr_start;
> > + unsigned int intr_start;
> > };
> >
> > /**
> > @@ -587,8 +587,8 @@ struct dpu_dspp_cfg {
> > struct dpu_pingpong_cfg {
> > DPU_HW_BLK_INFO;
> > u32 merge_3d;
> > - s32 intr_done;
> > - s32 intr_rdptr;
> > + unsigned int intr_done;
> > + unsigned int intr_rdptr;
> > const struct dpu_pingpong_sub_blks *sblk;
> > };
> >
> > @@ -635,9 +635,9 @@ struct dpu_intf_cfg {
> > u32 type; /* interface type*/
> > u32 controller_id;
> > u32 prog_fetch_lines_worst_case;
> > - s32 intr_underrun;
> > - s32 intr_vsync;
> > - s32 intr_tear_rd_ptr;
> > + unsigned int intr_underrun;
> > + unsigned int intr_vsync;
> > + unsigned int intr_tear_rd_ptr;
> > };
> >
> > /**
> > @@ -656,7 +656,7 @@ struct dpu_wb_cfg {
> > u8 vbif_idx;
> > u32 maxlinewidth;
> > u32 xin_id;
> > - s32 intr_wb_done;
> > + unsigned int intr_wb_done;
> > const u32 *format_list;
> > u32 num_formats;
> > enum dpu_clk_ctrl_type clk_ctrl;
> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_interrupts.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_interrupts.c
> > index ede7161ae904..47f6912ac39e 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_interrupts.c
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_interrupts.c
> > @@ -200,7 +200,7 @@ static const struct dpu_intr_reg dpu_intr_set_7xxx[] = {
> > #define DPU_IRQ_MASK(irq_idx) (BIT(DPU_IRQ_OFFSET(irq_idx)))
> >
> > static inline struct dpu_hw_intr_entry *dpu_core_irq_get_entry(struct dpu_kms *dpu_kms,
> > - int irq_idx)
> > + unsigned int irq_idx)
> > {
> > return &dpu_kms->hw_intr->irq_tbl[irq_idx];
>
> Shouldn't this be -1'd, or should the struct size of this trailing array
> be incremented by one in the nirq variable in dpu_hw_intr_init()?
>
> Note that total_irqs = nirq.
Ugh, yes.
>
> > }
> > @@ -210,7 +210,7 @@ static inline struct dpu_hw_intr_entry *dpu_core_irq_get_entry(struct dpu_kms *d
> > * @dpu_kms: Pointer to DPU's KMS structure
> > * @irq_idx: interrupt index
> > */
> > -static void dpu_core_irq_callback_handler(struct dpu_kms *dpu_kms, int irq_idx)
> > +static void dpu_core_irq_callback_handler(struct dpu_kms *dpu_kms, unsigned int irq_idx)
> > {
> > struct dpu_hw_intr_entry *irq_entry = dpu_core_irq_get_entry(dpu_kms, irq_idx);
> >
> > @@ -232,7 +232,7 @@ irqreturn_t dpu_core_irq(struct msm_kms *kms)
> > struct dpu_kms *dpu_kms = to_dpu_kms(kms);
> > struct dpu_hw_intr *intr = dpu_kms->hw_intr;
> > int reg_idx;
> > - int irq_idx;
> > + unsigned int irq_idx;
> > u32 irq_status;
> > u32 enable_mask;
> > int bit;
> > @@ -288,7 +288,8 @@ irqreturn_t dpu_core_irq(struct msm_kms *kms)
> > return IRQ_HANDLED;
> > }
> >
> > -static int dpu_hw_intr_enable_irq_locked(struct dpu_hw_intr *intr, int irq_idx)
> > +static int dpu_hw_intr_enable_irq_locked(struct dpu_hw_intr *intr,
> > + unsigned int irq_idx)
> > {
> > int reg_idx;
> > const struct dpu_intr_reg *reg;
> > @@ -298,7 +299,7 @@ static int dpu_hw_intr_enable_irq_locked(struct dpu_hw_intr *intr, int irq_idx)
> > if (!intr)
> > return -EINVAL;
> >
> > - if (irq_idx < 0 || irq_idx >= intr->total_irqs) {
> > + if (!irq_idx || irq_idx > intr->total_irqs) {
> > pr_err("invalid IRQ index: [%d]\n", irq_idx);
>
> Logs like this might be harder to interpret (and compare) when the
> numbering is different. In addition, all the IRQs in
> /d/dri/0/debug/core_irq are shifted up by 1 making them harder to
> compare to downstream.
> (Which I hope to not have to do again for a while, now that my INTF TE
> series is finalized and merged)
I hesitated here. Maybe we should log the register and index instead
of logging the raw index.
As for the core_irq vs downstream, that's a good question. I don't
like the idea of adding -1 there. Maybe I'll change that again to
register + index.
>
> Rest looks good, nice cleanup and lightening of the source.
>
> - Marijn
>
> > return -EINVAL;
> > }
> > @@ -341,7 +342,8 @@ static int dpu_hw_intr_enable_irq_locked(struct dpu_hw_intr *intr, int irq_idx)
> > return 0;
> > }
> >
> > -static int dpu_hw_intr_disable_irq_locked(struct dpu_hw_intr *intr, int irq_idx)
> > +static int dpu_hw_intr_disable_irq_locked(struct dpu_hw_intr *intr,
> > + unsigned int irq_idx)
> > {
> > int reg_idx;
> > const struct dpu_intr_reg *reg;
> > @@ -351,7 +353,7 @@ static int dpu_hw_intr_disable_irq_locked(struct dpu_hw_intr *intr, int irq_idx)
> > if (!intr)
> > return -EINVAL;
> >
> > - if (irq_idx < 0 || irq_idx >= intr->total_irqs) {
> > + if (!irq_idx || irq_idx > intr->total_irqs) {
> > pr_err("invalid IRQ index: [%d]\n", irq_idx);
> > return -EINVAL;
> > }
> > @@ -426,7 +428,8 @@ static void dpu_disable_all_irqs(struct dpu_kms *dpu_kms)
> > wmb();
> > }
> >
> > -u32 dpu_core_irq_read(struct dpu_kms *dpu_kms, int irq_idx)
> > +u32 dpu_core_irq_read(struct dpu_kms *dpu_kms,
> > + unsigned int irq_idx)
> > {
> > struct dpu_hw_intr *intr = dpu_kms->hw_intr;
> > int reg_idx;
> > @@ -436,13 +439,7 @@ u32 dpu_core_irq_read(struct dpu_kms *dpu_kms, int irq_idx)
> > if (!intr)
> > return 0;
> >
> > - if (irq_idx < 0) {
> > - DPU_ERROR("[%pS] invalid irq_idx=%d\n",
> > - __builtin_return_address(0), irq_idx);
> > - return 0;
> > - }
> > -
> > - if (irq_idx < 0 || irq_idx >= intr->total_irqs) {
> > + if (!irq_idx || irq_idx > intr->total_irqs) {
> > pr_err("invalid IRQ index: [%d]\n", irq_idx);
> > return 0;
> > }
> > @@ -499,7 +496,7 @@ struct dpu_hw_intr *dpu_hw_intr_init(void __iomem *addr,
> >
> > intr->irq_mask |= BIT(MDP_INTFn_INTR(intf->id));
> >
> > - if (intf->intr_tear_rd_ptr != -1)
> > + if (intf->intr_tear_rd_ptr)
> > intr->irq_mask |= BIT(DPU_IRQ_REG(intf->intr_tear_rd_ptr));
> > }
> >
> > @@ -513,9 +510,10 @@ void dpu_hw_intr_destroy(struct dpu_hw_intr *intr)
> > kfree(intr);
> > }
> >
> > -int dpu_core_irq_register_callback(struct dpu_kms *dpu_kms, int irq_idx,
> > - void (*irq_cb)(void *arg),
> > - void *irq_arg)
> > +int dpu_core_irq_register_callback(struct dpu_kms *dpu_kms,
> > + unsigned int irq_idx,
> > + void (*irq_cb)(void *arg),
> > + void *irq_arg)
> > {
> > struct dpu_hw_intr_entry *irq_entry;
> > unsigned long irq_flags;
> > @@ -526,7 +524,7 @@ int dpu_core_irq_register_callback(struct dpu_kms *dpu_kms, int irq_idx,
> > return -EINVAL;
> > }
> >
> > - if (irq_idx < 0 || irq_idx >= dpu_kms->hw_intr->total_irqs) {
> > + if (!irq_idx || irq_idx > dpu_kms->hw_intr->total_irqs) {
> > DPU_ERROR("invalid IRQ index: [%d]\n", irq_idx);
> > return -EINVAL;
> > }
> > @@ -559,13 +557,14 @@ int dpu_core_irq_register_callback(struct dpu_kms *dpu_kms, int irq_idx,
> > return 0;
> > }
> >
> > -int dpu_core_irq_unregister_callback(struct dpu_kms *dpu_kms, int irq_idx)
> > +int dpu_core_irq_unregister_callback(struct dpu_kms *dpu_kms,
> > + unsigned int irq_idx)
> > {
> > struct dpu_hw_intr_entry *irq_entry;
> > unsigned long irq_flags;
> > int ret;
> >
> > - if (irq_idx < 0 || irq_idx >= dpu_kms->hw_intr->total_irqs) {
> > + if (!irq_idx || irq_idx > dpu_kms->hw_intr->total_irqs) {
> > DPU_ERROR("invalid IRQ index: [%d]\n", irq_idx);
> > return -EINVAL;
> > }
> > @@ -600,7 +599,7 @@ static int dpu_debugfs_core_irq_show(struct seq_file *s, void *v)
> > int i, irq_count;
> > void *cb;
> >
> > - for (i = 0; i < dpu_kms->hw_intr->total_irqs; i++) {
> > + for (i = 1; i <= dpu_kms->hw_intr->total_irqs; i++) {
> > spin_lock_irqsave(&dpu_kms->hw_intr->irq_lock, irq_flags);
> > irq_entry = dpu_core_irq_get_entry(dpu_kms, i);
> > irq_count = atomic_read(&irq_entry->count);
> > @@ -635,7 +634,7 @@ void dpu_core_irq_preinstall(struct msm_kms *kms)
> > dpu_disable_all_irqs(dpu_kms);
> > pm_runtime_put_sync(&dpu_kms->pdev->dev);
> >
> > - for (i = 0; i < dpu_kms->hw_intr->total_irqs; i++) {
> > + for (i = 1; i <= dpu_kms->hw_intr->total_irqs; i++) {
> > irq_entry = dpu_core_irq_get_entry(dpu_kms, i);
> > atomic_set(&irq_entry->count, 0);
> > }
> > @@ -651,7 +650,7 @@ void dpu_core_irq_uninstall(struct msm_kms *kms)
> > return;
> >
> > pm_runtime_get_sync(&dpu_kms->pdev->dev);
> > - for (i = 0; i < dpu_kms->hw_intr->total_irqs; i++) {
> > + for (i = 1; i <= dpu_kms->hw_intr->total_irqs; i++) {
> > irq_entry = dpu_core_irq_get_entry(dpu_kms, i);
> > if (irq_entry->cb)
> > DPU_ERROR("irq_idx=%d still enabled/registered\n", i);
> > diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_interrupts.h b/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_interrupts.h
> > index 59bde8ab50c8..41a020fd1288 100644
> > --- a/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_interrupts.h
> > +++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_hw_interrupts.h
> > @@ -36,9 +36,9 @@ enum dpu_hw_intr_reg {
> >
> > #define MDP_INTFn_INTR(intf) (MDP_INTF0_INTR + (intf - INTF_0))
> >
> > -#define DPU_IRQ_IDX(reg_idx, offset) (reg_idx * 32 + offset)
> > -#define DPU_IRQ_REG(irq_idx) ((irq_idx) / 32)
> > -#define DPU_IRQ_OFFSET(irq_idx) ((irq_idx) % 32)
> > +#define DPU_IRQ_IDX(reg_idx, offset) (reg_idx * 32 + offset + 1)
> > +#define DPU_IRQ_REG(irq_idx) ((irq_idx - 1) / 32)
> > +#define DPU_IRQ_OFFSET(irq_idx) ((irq_idx - 1) % 32)
> >
> > struct dpu_hw_intr_entry {
> > void (*cb)(void *arg);
> > --
> > 2.39.2
> >
--
With best wishes
Dmitry
More information about the dri-devel
mailing list