Read register ioctl causes crash

[ShaneEverittM]

The following code causes a crash

  int fifo_fd = open("/dev/axis_fifo_0x43c00000", O_RDWR);
  if (fifo_fd < 0)
      return 1;

  struct axis_fifo_kern_regInfo reg_info{XLLF_RLR_OFFSET, 0};
  int rc = ioctl(fifo_fd, AXIS_FIFO_GET_REG, &reg_info);

  if (rc) {
      return EXIT_FAILURE;
  } else {
      return EXIT_SUCCESS;
  }

Reads and writes are successful, other ioctl operations work fine, and the provided test app functions correctly. When I try the above code I get:

Internal error: Oops - BUG: 1406 [#1] PREEMPT SMP ARM
Modules linked in: axis_fifo(O) uio_pdrv_genirq
CPU: 1 PID: 492 Comm: fifo_benchmark Tainted: G           O      5.10.0-xilinx-v2021.2 #1
Hardware name: Xilinx Zynq Platform
PC is at axis_fifo_ioctl+0x1c0/0x29c [axis_fifo]
LR is at 0x0
pc : [<bf012960>]    lr : [<00000000>]    psr: 60010013
sp : c1c93ef0  ip : 00000018  fp : 00000000
r10: c1c92000  r9 : c18a2f00  r8 : c1c92000
r7 : 80085100  r6 : befffab4  r5 : 00000000  r4 : c1a0f540
r3 : 00000000  r2 : 00000024  r1 : befffabc  r0 : 00000000
Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment none
Control: 18c5387d  Table: 01a1404a  DAC: 00000051
Process fifo_benchmark (pid: 492, stack limit = 0xc7d15f33)
Stack: (0xc1c93ef0 to 0xc1c94000)
3ee0:                                     00000000 00000024 00000000 4fb846c8
3f00: 80085100 befffab4 c18a2f00 c19d58d0 00000003 c0210d60 80085100 c0211738
3f20: c1c92000 00000000 00000000 c1c92000 c1c93f40 ffffe000 00000000 c0108704
3f40: 00000000 00000005 00000000 00000001 00401544 00000000 00000000 00000000
3f60: 00000005 00000000 00000001 00401544 00000000 00000000 00000000 4fb846c8
3f80: 00000000 00414ed4 00414ed4 004016ad 00000036 c0100264 c1c92000 00000036
3fa0: 00000000 c0100060 00414ed4 00414ed4 00000003 80085100 befffab4 00000003
3fc0: 00414ed4 00414ed4 004016ad 00000036 00000000 00000000 00414ed4 00000000
3fe0: 00414f4c befffaa4 00403531 b6e10228 40010030 00000003 00000000 00000000
[<bf012960>] (axis_fifo_ioctl [axis_fifo]) from [<c0210d60>] (vfs_ioctl+0x20/0x38)
[<c0210d60>] (vfs_ioctl) from [<c0211738>] (sys_ioctl+0xc0/0x770)
[<c0211738>] (sys_ioctl) from [<c0100060>] (ret_fast_syscall+0x0/0x54)
Exception stack(0xc1c93fa8 to 0xc1c93ff0)
3fa0:                   00414ed4 00414ed4 00000003 80085100 befffab4 00000003
3fc0: 00414ed4 00414ed4 004016ad 00000036 00000000 00000000 00414ed4 00000000
3fe0: 00414f4c befffaa4 00403531 b6e10228
Code: e59d2004 e0833002 e5933000 f57ff04f (e28d1004) 
---[ end trace 5d0f1f8cb441a56f ]---

I am using PetaLinux 2021.2 with the following FIFO configuration Let me know if you need any more information.

[ShaneEverittM]

Further investigation showed that ioctl operations requesting register values work for all the register offsets below XLLF_RDFD_OFFSET.

[jacobfeder]

@ShaneEverittM not sure I've never tried this. Are you sure you're accessing something that's in the address range of the IP?

[ShaneEverittM]

Also not sure, I just assumed the #define block labeled "IP register offsets" were the valid values for the regNo field in the axis_fifo_kern_regInfo. Can you shed some light on the expected behavior for the AXIS_FIFO_GET_REG ioctl?

Also thanks for the quick reply!

[jacobfeder]

It's been several years since I've looked at this in detail, so can't be of much help unfortunately. I would look at the IP datasheet and make sure you're addressing valid memory regions of the IP.

[ShaneEverittM]

Alright, is this not something you would recommend for production use then? I saw it was in the staging area for the kernel.

[jacobfeder]

Yeah it's been there in limbo for awhile. It seems there are several people using it successfully, but I'm not actively developing it.

[ShaneEverittM]

Gotcha, also the same crash happens when running: cat /sys/class/axis_fifo/axis_fifo_0x43c00000/ip_registers/rlr. Is it likely something I'm messing up? I'm using 4.2 and maybe that doesn't expose the same registers?

[ShaneEverittM]

Ok, it's likely I'm just reading in invalid ways, the data sheet has rules for when registers are valid to read, and in what order.

[jacobfeder]

You definitely have to be careful about accessing the registers directly - if you access them in the wrong order it can break things. IIRC this is somewhat explained in the datasheet (about the order in which you have to access the registers).

[jacobfeder]

Yeah exactly.

[ShaneEverittM]

Cool, other than me messing up, driver has been working great, love the helpful error messages, I'll go ahead and close this issue.