Re: [PATCH] dmaengine: xilinx: dpdma: request_irq after initializing dma channels

From: quanyang.wang
Date: Fri Apr 30 2021 - 02:58:07 EST


Hi Michal,

On 4/30/21 2:51 PM, Michal Simek wrote:
Hi,

On 4/30/21 3:45 AM, quanyang.wang wrote:
Hi Laurent,

On 4/30/21 8:36 AM, Laurent Pinchart wrote:
Hi Quanyang,

Thank you for the patch.

On Thu, Apr 29, 2021 at 02:58:21PM +0800, quanyang.wang@xxxxxxxxxxxxx
wrote:
From: Quanyang Wang <quanyang.wang@xxxxxxxxxxxxx>

In some scenarios (kdump), dpdma hardware irqs has been enabled when
calling request_irq in probe function, and then the dpdma irq handler
xilinx_dpdma_irq_handler is invoked to access xdev->chan[i]. But at
this moment xdev->chan[i] hasn't been initialized. So let's call the
request_irq after initializing dma channels to avoid kdump kernel
crash as below:

[    3.696128] Unable to handle kernel NULL pointer dereference at
virtual address 000000000000012c
[    3.696710] xilinx-zynqmp-dpdma fd4c0000.dma-controller: Xilinx
DPDMA engine is probed
[    3.704900] Mem abort info:
[    3.704902]   ESR = 0x96000005
[    3.704905]   EC = 0x25: DABT (current EL), IL = 32 bits
[    3.704907]   SET = 0, FnV = 0
[    3.704912]   EA = 0, S1PTW = 0
[    3.713800] ahci-ceva fd0c0000.ahci: supply ahci not found, using
dummy regulator
[    3.715585] Data abort info:
[    3.715587]   ISV = 0, ISS = 0x00000005
[    3.715589]   CM = 0, WnR = 0
[    3.715592] [000000000000012c] user address but active_mm is swapper
[    3.715596] Internal error: Oops: 96000005 [#1] SMP
[    3.715599] Modules linked in:
[    3.715608] CPU: 0 PID: 0 Comm: swapper/0 Not tainted
5.10.0-12170-g60894882155f-dirty #77
[    3.723937] Hardware name: ZynqMP ZCU102 Rev1.0 (DT)
[    3.723942] pstate: 80000085 (Nzcv daIf -PAN -UAO -TCO BTYPE=--)
[    3.723956] pc : xilinx_dpdma_irq_handler+0x418/0x560
[    3.793049] lr : xilinx_dpdma_irq_handler+0x3d8/0x560
[    3.798089] sp : ffffffc01186bdf0
[    3.801388] x29: ffffffc01186bdf0 x28: ffffffc011836f28
[    3.806692] x27: ffffff8023e0ac80 x26: 0000000000000080
[    3.811996] x25: 0000000008000408 x24: 0000000000000003
[    3.817300] x23: ffffffc01186be70 x22: ffffffc011291740
[    3.822604] x21: 0000000000000000 x20: 0000000008000408
[    3.827908] x19: 0000000000000000 x18: 0000000000000010
[    3.833212] x17: 0000000000000000 x16: 0000000000000000
[    3.838516] x15: 0000000000000000 x14: ffffffc011291740
[    3.843820] x13: ffffffc02eb4d000 x12: 0000000034d4d91d
[    3.849124] x11: 0000000000000040 x10: ffffffc0112d2d48
[    3.854428] x9 : ffffffc0112d2d40 x8 : ffffff8021c00268
[    3.859732] x7 : 0000000000000000 x6 : ffffffc011836000
[    3.865036] x5 : 0000000000000003 x4 : 0000000000000000
[    3.870340] x3 : 0000000000000001 x2 : 0000000000000000
[    3.875644] x1 : 0000000000000000 x0 : 000000000000012c
[    3.880948] Call trace:
[    3.883382]  xilinx_dpdma_irq_handler+0x418/0x560
[    3.888079]  __handle_irq_event_percpu+0x5c/0x178
[    3.892774]  handle_irq_event_percpu+0x34/0x98
[    3.897210]  handle_irq_event+0x44/0xb8
[    3.901030]  handle_fasteoi_irq+0xd0/0x190
[    3.905117]  generic_handle_irq+0x30/0x48
[    3.909111]  __handle_domain_irq+0x64/0xc0
[    3.913192]  gic_handle_irq+0x78/0xa0
[    3.916846]  el1_irq+0xc4/0x180
[    3.919982]  cpuidle_enter_state+0x134/0x2f8
[    3.924243]  cpuidle_enter+0x38/0x50
[    3.927810]  call_cpuidle+0x1c/0x40
[    3.931290]  do_idle+0x20c/0x270
[    3.934502]  cpu_startup_entry+0x28/0x58
[    3.938410]  rest_init+0xbc/0xcc
[    3.941631]  arch_call_rest_init+0x10/0x1c
[    3.945718]  start_kernel+0x51c/0x558

Fixes: 7cbb0c63de3f ("dmaengine: xilinx: dpdma: Add the Xilinx
DisplayPort DMA engine driver")
Signed-off-by: Quanyang Wang <quanyang.wang@xxxxxxxxxxxxx>
---
  drivers/dma/xilinx/xilinx_dpdma.c | 30 +++++++++++++++---------------
  1 file changed, 15 insertions(+), 15 deletions(-)

diff --git a/drivers/dma/xilinx/xilinx_dpdma.c
b/drivers/dma/xilinx/xilinx_dpdma.c
index 70b29bd079c9..0b599402c53f 100644
--- a/drivers/dma/xilinx/xilinx_dpdma.c
+++ b/drivers/dma/xilinx/xilinx_dpdma.c
@@ -1622,19 +1622,6 @@ static int xilinx_dpdma_probe(struct
platform_device *pdev)
      if (IS_ERR(xdev->reg))
          return PTR_ERR(xdev->reg);
  -    xdev->irq = platform_get_irq(pdev, 0);
-    if (xdev->irq < 0) {
-        dev_err(xdev->dev, "failed to get platform irq\n");
-        return xdev->irq;
-    }
-
-    ret = request_irq(xdev->irq, xilinx_dpdma_irq_handler, IRQF_SHARED,
-              dev_name(xdev->dev), xdev);
-    if (ret) {
-        dev_err(xdev->dev, "failed to request IRQ\n");
-        return ret;
-    }
-
      ddev = &xdev->common;
      ddev->dev = &pdev->dev;
  @@ -1688,6 +1675,19 @@ static int xilinx_dpdma_probe(struct
platform_device *pdev)
          goto error_of_dma;
      }
  +    xdev->irq = platform_get_irq(pdev, 0);
+    if (xdev->irq < 0) {
+        dev_err(xdev->dev, "failed to get platform irq\n");
As reported by the kbuild bot, ret isn't initialized here.
I will fix this in V2.
+        goto error_irq;
+    }
This part could stay where it was, just after
devm_platform_ioremap_resource().
OK,I will move it back.
+
+    ret = request_irq(xdev->irq, xilinx_dpdma_irq_handler, IRQF_SHARED,
+              dev_name(xdev->dev), xdev);
+    if (ret) {
+        dev_err(xdev->dev, "failed to request IRQ\n");
+        goto error_irq;
+    }
Ideally we should reset the device before requesting the IRQ, to ensure
we start in a consistent state. There doesn't seem to be any global
reset unfortunately. Shouldn't we at least disable interrupts ? It may
be a bit pointless though as we reenable them right below, so I suppose
this would be OK.
Yes, we should start in a consistent and clean state. I will disable
interrupts

and clean up registers state in V2.

+
      xilinx_dpdma_enable_irq(xdev);
        xilinx_dpdma_debugfs_init(xdev);
@@ -1696,6 +1696,8 @@ static int xilinx_dpdma_probe(struct
platform_device *pdev)
        return 0;
  +error_irq:
+    of_dma_controller_free(pdev->dev.of_node);
Why not free_irq() ? This change isn't described in the commit message.
I will add it in V2.
When I was checking it that request_irq was called last. It means it
wasn't necessary to free irq in error path. But double check it.

Thank you very much for your reminder.

Yes, free_irq is not necessary, and I have abandoned all modifications of V1 and

make a new V2 patch.

Thanks,

Quanyang


M