Re: [linux-6.6.y bugreport] riscv: kprobe crash as some patchs lost

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Fri, Apr 25, 2025 at 04:03:41PM +0800, Kai Zhang wrote:
> On 4/22/2025 4:46 PM, Greg Kroah-Hartman wrote:
> > On Tue, Apr 22, 2025 at 10:58:42AM +0800, Kai Zhang wrote:
> > > In most recent linux-6.6.y tree,
> > > `arch/riscv/kernel/probes/kprobes.c::arch_prepare_ss_slot` still has the
> > > obsolete code:
> > > 
> > >      u32 insn = __BUG_INSN_32;
> > >      unsigned long offset = GET_INSN_LENGTH(p->opcode);
> > >      p->ainsn.api.restore = (unsigned long)p->addr + offset;
> > >      patch_text_nosync(p->ainsn.api.insn, &p->opcode, 1);
> > >      patch_text_nosync((void *)p->ainsn.api.insn + offset, &insn, 1);
> > > 
> > > The last two 1s are wrong size of written instructions , which would lead to
> > > kernel crash, like `insmod kprobe_example.ko` gives:
> > > 
> > > [  509.812815][ T2734] kprobe_init: Planted kprobe at 00000000c5c46130
> > > [  509.837606][    C5] handler_pre: <kernel_clone> p->addr =
> > > 0x00000000c5c46130, pc = 0xffffffff80032ee2, status = 0x200000120
> > > [  509.839315][    C5] Oops - illegal instruction [#1]
> > > 
> > > 
> > > I've tried two patchs from torvalds tree and it didn't crash again:
> > > 
> > > 51781ce8f448 riscv: Pass patch_text() the length in bytes (rebased)
> > > 13134cc94914 riscv: kprobes: Fix incorrect address calculation
> > 
> > Neither of these apply cleanly.  Please provide working backports if you
> > wish to see them added to the tree.
> > 
> > thanks,
> > 
> > greg k-h
> 
> revert 03753bfacbc6
> apply  51781ce8f448
> apply  13134cc94914

Thanks, but that's not how we take patches for the stable tree.  Please
submit these all in a tested patch series and we will be glad to queue
them up.

greg k-h