drm/msm/dpu: Convert to a chained irq chip
authorStephen Boyd <swboyd@chromium.org>
Thu, 3 Jan 2019 19:06:02 +0000 (11:06 -0800)
committerSean Paul <seanpaul@chromium.org>
Fri, 1 Feb 2019 15:35:55 +0000 (10:35 -0500)
Devices that make up DPU, i.e. graphics card, request their interrupts
from this "virtual" interrupt chip. The interrupt chip builds upon a GIC
SPI interrupt that raises high when any of the interrupts in the DPU's
irq status register are triggered. From the kernel's perspective this is
a chained irq chip, so requesting a flow handler for the GIC SPI and
then calling generic IRQ handling code from that irq handler is not
completely proper. It's better to convert this to a chained irq so that
the GIC SPI irq doesn't appear in /proc/interrupts, can't have CPU
affinity changed, and won't be accounted for with irq stats. Doing this
also silences a recursive lockdep warning because we can specify a
different lock class for the chained interrupts, silencing a warning
that is easy to see with 'threadirqs' on the kernel commandline.

 WARNING: inconsistent lock state
 4.19.10 #76 Tainted: G        W
 --------------------------------
 inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage.
 irq/40-dpu_mdss/203 [HC0[0]:SC0[2]:HE1:SE0] takes:
 0000000053ea9021 (&irq_desc_lock_class){?.-.}, at: handle_level_irq+0x34/0x26c
 {IN-HARDIRQ-W} state was registered at:
   lock_acquire+0x244/0x360
   _raw_spin_lock+0x64/0xa0
   handle_fasteoi_irq+0x54/0x2ec
   generic_handle_irq+0x44/0x5c
   __handle_domain_irq+0x9c/0x11c
   gic_handle_irq+0x208/0x260
   el1_irq+0xb4/0x130
   arch_cpu_idle+0x178/0x3cc
   default_idle_call+0x3c/0x54
   do_idle+0x1a8/0x3dc
   cpu_startup_entry+0x24/0x28
   rest_init+0x240/0x270
   start_kernel+0x5a8/0x6bc
 irq event stamp: 18
 hardirqs last  enabled at (17): [<ffffff9042385e80>] _raw_spin_unlock_irq+0x40/0xc0
 hardirqs last disabled at (16): [<ffffff904237a1f4>] __schedule+0x20c/0x1bbc
 softirqs last  enabled at (0): [<ffffff9040f318d0>] copy_process+0xb50/0x3964
 softirqs last disabled at (18): [<ffffff9041036364>] local_bh_disable+0x8/0x20

 other info that might help us debug this:
  Possible unsafe locking scenario:

        CPU0
        ----
   lock(&irq_desc_lock_class);
   <Interrupt>
     lock(&irq_desc_lock_class);

  *** DEADLOCK ***

 no locks held by irq/40-dpu_mdss/203.

 stack backtrace:
 CPU: 0 PID: 203 Comm: irq/40-dpu_mdss Tainted: G        W         4.19.10 #76
 Call trace:
  dump_backtrace+0x0/0x2f8
  show_stack+0x20/0x2c
  __dump_stack+0x20/0x28
  dump_stack+0xcc/0x10c
  mark_lock+0xbe0/0xe24
  __lock_acquire+0x4cc/0x2708
  lock_acquire+0x244/0x360
  _raw_spin_lock+0x64/0xa0
  handle_level_irq+0x34/0x26c
  generic_handle_irq+0x44/0x5c
  dpu_mdss_irq+0x64/0xec
  irq_forced_thread_fn+0x58/0x9c
  irq_thread+0x120/0x1dc
  kthread+0x248/0x260
  ret_from_fork+0x10/0x18
 ------------[ cut here ]------------
 irq 169 handler irq_default_primary_handler+0x0/0x18 enabled interrupts

Cc: Sean Paul <seanpaul@chromium.org>
Cc: Jordan Crouse <jcrouse@codeaurora.org>
Cc: Jayant Shekhar <jshekhar@codeaurora.org>
Cc: Rajesh Yadav <ryadav@codeaurora.org>
Cc: Jeykumar Sankaran <jsanka@codeaurora.org>
Signed-off-by: Stephen Boyd <swboyd@chromium.org>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
drivers/gpu/drm/msm/disp/dpu1/dpu_mdss.c

index cb307a2abf06c6136976f861e28525a244a34882..7316b4ab1b85e6ee03c0d855fbdd7045b897b3dd 100644 (file)
@@ -23,11 +23,14 @@ struct dpu_mdss {
        struct dpu_irq_controller irq_controller;
 };
 
-static irqreturn_t dpu_mdss_irq(int irq, void *arg)
+static void dpu_mdss_irq(struct irq_desc *desc)
 {
-       struct dpu_mdss *dpu_mdss = arg;
+       struct dpu_mdss *dpu_mdss = irq_desc_get_handler_data(desc);
+       struct irq_chip *chip = irq_desc_get_chip(desc);
        u32 interrupts;
 
+       chained_irq_enter(chip, desc);
+
        interrupts = readl_relaxed(dpu_mdss->mmio + HW_INTR_STATUS);
 
        while (interrupts) {
@@ -39,20 +42,20 @@ static irqreturn_t dpu_mdss_irq(int irq, void *arg)
                                           hwirq);
                if (mapping == 0) {
                        DRM_ERROR("couldn't find irq mapping for %lu\n", hwirq);
-                       return IRQ_NONE;
+                       break;
                }
 
                rc = generic_handle_irq(mapping);
                if (rc < 0) {
                        DRM_ERROR("handle irq fail: irq=%lu mapping=%u rc=%d\n",
                                  hwirq, mapping, rc);
-                       return IRQ_NONE;
+                       break;
                }
 
                interrupts &= ~(1 << hwirq);
        }
 
-       return IRQ_HANDLED;
+       chained_irq_exit(chip, desc);
 }
 
 static void dpu_mdss_irq_mask(struct irq_data *irqd)
@@ -83,16 +86,16 @@ static struct irq_chip dpu_mdss_irq_chip = {
        .irq_unmask = dpu_mdss_irq_unmask,
 };
 
+static struct lock_class_key dpu_mdss_lock_key, dpu_mdss_request_key;
+
 static int dpu_mdss_irqdomain_map(struct irq_domain *domain,
                unsigned int irq, irq_hw_number_t hwirq)
 {
        struct dpu_mdss *dpu_mdss = domain->host_data;
-       int ret;
 
+       irq_set_lockdep_class(irq, &dpu_mdss_lock_key, &dpu_mdss_request_key);
        irq_set_chip_and_handler(irq, &dpu_mdss_irq_chip, handle_level_irq);
-       ret = irq_set_chip_data(irq, dpu_mdss);
-
-       return ret;
+       return irq_set_chip_data(irq, dpu_mdss);
 }
 
 static const struct irq_domain_ops dpu_mdss_irqdomain_ops = {
@@ -159,11 +162,13 @@ static void dpu_mdss_destroy(struct drm_device *dev)
        struct msm_drm_private *priv = dev->dev_private;
        struct dpu_mdss *dpu_mdss = to_dpu_mdss(priv->mdss);
        struct dss_module_power *mp = &dpu_mdss->mp;
+       int irq;
 
        pm_runtime_suspend(dev->dev);
        pm_runtime_disable(dev->dev);
        _dpu_mdss_irq_domain_fini(dpu_mdss);
-       free_irq(platform_get_irq(pdev, 0), dpu_mdss);
+       irq = platform_get_irq(pdev, 0);
+       irq_set_chained_handler_and_data(irq, NULL, NULL);
        msm_dss_put_clk(mp->clk_config, mp->num_clk);
        devm_kfree(&pdev->dev, mp->clk_config);
 
@@ -187,6 +192,7 @@ int dpu_mdss_init(struct drm_device *dev)
        struct dpu_mdss *dpu_mdss;
        struct dss_module_power *mp;
        int ret = 0;
+       int irq;
 
        dpu_mdss = devm_kzalloc(dev->dev, sizeof(*dpu_mdss), GFP_KERNEL);
        if (!dpu_mdss)
@@ -219,12 +225,12 @@ int dpu_mdss_init(struct drm_device *dev)
        if (ret)
                goto irq_domain_error;
 
-       ret = request_irq(platform_get_irq(pdev, 0),
-                       dpu_mdss_irq, 0, "dpu_mdss_isr", dpu_mdss);
-       if (ret) {
-               DPU_ERROR("failed to init irq: %d\n", ret);
+       irq = platform_get_irq(pdev, 0);
+       if (irq < 0)
                goto irq_error;
-       }
+
+       irq_set_chained_handler_and_data(irq, dpu_mdss_irq,
+                                        dpu_mdss);
 
        pm_runtime_enable(dev->dev);