Error while loading dw-i3c-master: UBSAN: shift-out-of-bounds in drivers/i3c/master/dw-i3c-master.c:885:12

Wed Dec 10 07:20:19 PST 2025

On Tue, Dec 09, 2025 at 08:52:46PM +0100, Louis Sautier wrote:
> On 12/9/25 18:24, Frank Li wrote:
> > On Tue, Dec 09, 2025 at 04:36:30PM +0100, Louis Sautier wrote:
> > > On 12/9/25 13:48, Alexandre Belloni wrote:
> > > > On 09/12/2025 12:37:12+0100, Louis Sautier wrote:
> > > > > On 12/8/25 21:58, Alexandre Belloni wrote:
> > > > > > Hello Louis,
> > > > > >
> > > > > > On 02/12/2025 21:17:31+0100, Louis Sautier wrote:
> > > > > > > Hello,
> > > > > > >
> > > > > > > I'm running into a bug when loading the dw-i3c-master module on kernel 6.18
> > > > > > > on one specific server. I suspect it has to do with the large number of
> > > > > > > CPUs on the machine (768 threads, from 2 AMD EPYC 9965 processors) but I am
> > > > > > > not sure.
> > > > > > >
> > > > > > > The system is on Ubuntu 25.10 and a 6.18 kernel with https://gist.githubusercontent.com/sbraz/a6f37fafbcf9354bbe4eace9e9eb48cb/raw/115da594dc9d7ea99b06754847571e6fd76d9da5/config
> > > > > > > (basically Ubuntu's).
> > > > > > Just to be sure, does this also happen with v6.17?
> > > > > >
> > > > > > The only change is the shutdown handling so I would guess yes.
> > > > > >
> > > > > Hello,
> > > > >
> > > > > It does happen with 6.17. I initially discovered this while running Ubuntu
> > > > > 25.10's stock kernel (6.17.0).
> > > > >
> > > > > > What is the behavior when you build the dw-i3c-master as a static driver?
> > > > > I'll try CONFIG_DW_I3C_MASTER=y and report back.
> > > > >
> > > > > Someone also suggested (they didn't reply to the list though) that I add a
> > > > > printk to see what the value of maxdevs is. I'll provide the log as soon as
> > > > > I have rebuilt with:
> > > > >
> > > > > --- linux-6.18.orig/drivers/i3c/master/dw-i3c-master.c	2025-11-30 22:42:10.000000000 +0000
> > > > > +++ linux-6.18/drivers/i3c/master/dw-i3c-master.c	2025-12-08 18:17:33.151567225 +0000
> > > > > @@ -1588,6 +1588,7 @@
> > > > >    	ret = readl(master->regs + DEVICE_ADDR_TABLE_POINTER);
> > > > >    	master->datstartaddr = ret;
> > > > >    	master->maxdevs = ret >> 16;
> > > > > +	printk("maxdevs: %d\n",  master->maxdevs);
> > > > >    	master->free_pos = GENMASK(master->maxdevs - 1, 0);
> > > > >    	master->quirks = (unsigned long)device_get_match_data(&pdev->dev);
> > > > >
> > > > Yes, that was going to be my suggestion.
> > > >
> > > I haven't tried with the driver built-in yet. This is what the printk shows:
> > > dw-i3c-master AMDI0015:00: probe with driver dw-i3c-master failed with error
> > > -110
> > > maxdevs: 65535
> > Maybe ret is bigger 0x8000_0000, and ret is sign int. so >>16 because -1.
> >
> > unsigned int val = readl(master->regs + DEVICE_ADDR_TABLE_POINTER);
> >
> > master->maxdevs = val >> 16;
> >
> > Frank
>
> I tried this and CONFIG_DW_I3C_MASTER=y:
>
> --- linux-6.18.orig/drivers/i3c/master/dw-i3c-master.c 2025-11-30
> 22:42:10.000000000 +0000
> +++ linux-6.18/drivers/i3c/master/dw-i3c-master.c       2025-12-09
> 19:21:52.735366616 +0000
> @@ -1585,9 +1585,10 @@
>         ret = readl(master->regs + DATA_BUFFER_STATUS_LEVEL);
>         master->caps.datafifodepth = DATA_BUFFER_STATUS_LEVEL_TX(ret);
>
> -       ret = readl(master->regs + DEVICE_ADDR_TABLE_POINTER);
> -       master->datstartaddr = ret;
> -       master->maxdevs = ret >> 16;
> +       unsigned int val = readl(master->regs + DEVICE_ADDR_TABLE_POINTER);
> +       master->datstartaddr = val;
> +       master->maxdevs = val >> 16;
> +       printk("maxdevs (unsigned): %d\n",  master->maxdevs);
>         master->free_pos = GENMASK(master->maxdevs - 1, 0);
>
>         master->quirks = (unsigned long)device_get_match_data(&pdev->dev);
>
> And I get this log, so no change, really. I assume there's only one
> "maxdevs" log because there is only one attempt to load the built-in driver?

It may have dependence missed at drivers. such as clock. when built-in,
this driver probe first before clock ready.

If build as module, other driver help enable this clock. So it can get
correct value.

Frank

>
> maxdevs (unsigned): 65535
> ------------[ cut here ]------------
> UBSAN: shift-out-of-bounds in drivers/i3c/master/dw-i3c-master.c:1592:21
> usb 1-1: new high-speed USB device number 2 using xhci_hcd
> shift exponent 18446744073709486145 is too large for 64-bit type 'long
> unsigned int'
> CPU: 0 UID: 0 PID: 1 Comm: swapper/0 Not tainted 6.18.0 #4
> PREEMPT(voluntary)
> Hardware name: Giga Computing MZ73-LM2-000/MZ73-LM2-000, BIOS R23_F43
> 11/28/2025
> Call Trace:
>  <TASK>
>  dump_stack_lvl+0x5f/0x90
>  dump_stack+0x10/0x18
>  ubsan_epilogue+0x9/0x39
>  __ubsan_handle_shift_out_of_bounds.cold+0xdd/0x1c9
>  dw_i3c_common_probe.cold+0x16/0x1b
>  dw_i3c_probe+0x30/0x50
>  platform_probe+0x42/0xc0
>  ? driver_sysfs_add+0x63/0xd0
>  really_probe+0xf9/0x370
>  ? pm_runtime_barrier+0x56/0xa0
>  __driver_probe_device+0x8b/0x160
>  driver_probe_device+0x24/0xd0
>  ? __pfx___driver_attach+0x10/0x10
>  __driver_attach+0xef/0x220
>  ? __pfx_dw_i3c_driver_init+0x10/0x10
>  bus_for_each_dev+0x8a/0xe0
>  driver_attach+0x1e/0x30
>  bus_add_driver+0x13e/0x230
>  ? __pfx_dw_i3c_driver_init+0x10/0x10
>  driver_register+0x75/0xf0
>  __platform_driver_register+0x1e/0x30
>  dw_i3c_driver_init+0x17/0x30
>  do_one_initcall+0x59/0x330
>  kernel_init_freeable+0x2bd/0x340
>  ? __pfx_kernel_init+0x10/0x10
>  kernel_init+0x1b/0x160
>  ? __pfx_kernel_init+0x10/0x10
>  ret_from_fork+0x202/0x230
>  ? __pfx_kernel_init+0x10/0x10
>  ret_from_fork_asm+0x1a/0x30
>  </TASK>
> ---[ end trace ]---
>
>
> --
> linux-i3c mailing list
> linux-i3c at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/linux-i3c