Hi Mike,

Unfortunately, I'm still able to reproduce the panic.

In case this is relevant.  With #667 the panic followed by the debugger
interface message happened right after the image was uncompressed.  With
current versions, I have to run startx to get a first single line
message panic, and then reboot again to get a second panic with the
debugger message.  I can send you pictures if you want.


On Sat, Feb 07, 2026 at 10:59:54AM -0800, Mike Larkin wrote:
> On Sat, Feb 07, 2026 at 09:22:15AM -0800, Mike Larkin wrote:
> > On Sat, Feb 07, 2026 at 09:09:13AM -0800, Mike Larkin wrote:
> > > On Sat, Jan 17, 2026 at 07:53:14PM -0800, Mike Larkin wrote:
> > > > On Sat, Jan 17, 2026 at 12:18:28PM +0100, Walter Alejandro Iglesias 
> > > > wrote:
> > > > > This machine, apparently, hibernates correctly.  Then, booting after 
> > > > > the
> > > > > hibernation is complete, also apparently, unhibernates correctly.  
> > > > > But,
> > > > > sometimes right after the image is uncompressed, sometimes after some
> > > > > command is executed in the console, sometimes later, you always end up
> > > > > with this same kernel panic (photos included in tarball):
> > > > >
> > > > >    https://en.roquesor.com/Downloads/panic.tar.gz
> > > > >
> > > > > As I mentioned to mlarkin@ and krw@ in private, I knew for certain 
> > > > > that
> > > > > this didn't happen with this machine.  I used to test how it 
> > > > > hibernated
> > > > > because I had it connected to a UPS when I used it (for many years) 
> > > > > as a
> > > > > home mail-web server.  And even though it's an old machine, I still 
> > > > > use
> > > > > it occasionally, which is why I took the time to do the trace back.
> > > > >
> > > > > After three days installing snapshots I could find out in which one 
> > > > > the
> > > > > panic starts happening and which modification causes it.
> > > > >
> > > > > This is the last snapshot working fine (the number of the beast :-):
> > > > >
> > > > >   https://openbsd.cs.toronto.edu/archive/2025-05-22/amd64
> > > > >
> > > > >   kern.version=OpenBSD 7.7-current (GENERIC.MP) #666: Wed May 21 
> > > > > 00:12:25 MDT 2025
> > > > >     
> > > > > [email protected]:/usr/src/sys/arch/amd64/compile/GENERIC.MP
> > > > >
> > > > > And this is the first when the issue appears:
> > > > >
> > > > >   https://openbsd.cs.toronto.edu/archive/2025-05-23/amd64
> > > > >
> > > > >   kern.version=OpenBSD 7.7-current (GENERIC.MP) #667: Thu May 22 
> > > > > 22:13:35 MDT 2025
> > > > >     
> > > > > [email protected]:/usr/src/sys/arch/amd64/compile/GENERIC.MP
> > > > >
> > > > > The modification that causes the issue is this:
> > > > >
> > > > >   https://marc.info/?l=openbsd-cvs&m=174779996216651&w=2
> > > > >
> > > > > After reverting those Mike's diffs in current, the panic disappears.
> > > > >
> > > >
> > > > I'll take a look at those diff again but I don't see how this could be 
> > > > the
> > > > problem. That diff just moved the allocation earlier. Maybe there's a 
> > > > double
> > > > free. I'll look.
> > > >
> > > > -ml
> > > >
> > >
> > > Hi Walter
> > >
> > > Can you try this diff on -current and let me know if this fixes it?
> > >
> > > diff /export/bin/src/OpenBSD/hib
> > > path + /export/bin/src/OpenBSD/hib
> > > commit - 12762e4337611beea64f8029adc3932766410745
> > > blob - 2b7924651df0c61efc89543ceacf5e7e5881e4ca
> > > file + sys/kern/subr_hibernate.c
> > > --- sys/kern/subr_hibernate.c
> > > +++ sys/kern/subr_hibernate.c
> > > @@ -2012,7 +2012,7 @@ hibernate_free(void)
> > >   pmap_activate(curproc);
> > >
> > >   if (hibernate_temp_page) {
> > > -         pmap_kremove(hibernate_temp_page, PAGE_SIZE);
> > > +//               pmap_kremove(hibernate_temp_page, PAGE_SIZE);
> > >           km_free((void *)hibernate_temp_page, PAGE_SIZE,
> > >               &kv_any, &kp_none);
> > >   }
> > >
> >
> > actually disregard, this isn't quit right either. I'll let you know when I 
> > have
> > something fully baked. stay tuned.
> >
> 
> You can try this one instead. It replaces the km_alloc/km_free management of 
> the
> temp page with a fixed stolen low VA (like we do for the other pages). I have
> run this through about a dozen cycles here and it seems ok but since you can
> repro, I'd like to know if I'm on the right track or not.
> 
> Apply to -current on the affected machine and LMK.
> 
> -ml
> 
> 
> diff /export/bin/src/OpenBSD/hib
> path + /export/bin/src/OpenBSD/hib
> commit - 12762e4337611beea64f8029adc3932766410745
> blob - ec4e1904b741e23ee1df50ab3e807e90283c2fbc
> file + sys/arch/amd64/include/hibernate_var.h
> --- sys/arch/amd64/include/hibernate_var.h
> +++ sys/arch/amd64/include/hibernate_var.h
> @@ -52,8 +52,9 @@
>  #define HIBERNATE_STACK_PAGE (PAGE_SIZE * 32)
> 
>  #define HIBERNATE_INFLATE_PAGE       (PAGE_SIZE * 33)
> +#define HIBERNATE_TEMP_PAGE  (PAGE_SIZE * 34)
>  /* HIBERNATE_HIBALLOC_PAGE must be the last stolen page (see machdep.c) */
> -#define HIBERNATE_HIBALLOC_PAGE      (PAGE_SIZE * 34)
> +#define HIBERNATE_HIBALLOC_PAGE      (PAGE_SIZE * 35)
> 
>  /* Use 4MB hibernation chunks */
>  #define HIBERNATE_CHUNK_SIZE         0x400000
> commit - 12762e4337611beea64f8029adc3932766410745
> blob - 2b7924651df0c61efc89543ceacf5e7e5881e4ca
> file + sys/kern/subr_hibernate.c
> --- sys/kern/subr_hibernate.c
> +++ sys/kern/subr_hibernate.c
> @@ -66,7 +66,6 @@ CTASSERT((offsetof(union hibernate_info, sec_size) + s
>   */
> 
>  /* Temporary vaddr ranges used during hibernate */
> -vaddr_t hibernate_temp_page;
>  vaddr_t hibernate_copy_page;
>  vaddr_t hibernate_rle_page;
> 
> @@ -1527,13 +1526,14 @@ hibernate_write_chunks(union hibernate_info *hib)
>                                       case -1:
>                                               return EIO;
>                                       case 0:
> -                                             
> pmap_kenter_pa(hibernate_temp_page,
> +                                             
> pmap_kenter_pa(HIBERNATE_TEMP_PAGE,
>                                                       inaddr & PMAP_PA_MASK,
>                                                       PROT_READ);
> 
> -                                             
> bcopy((caddr_t)hibernate_temp_page,
> +                                             
> bcopy((caddr_t)HIBERNATE_TEMP_PAGE,
>                                                       
> (caddr_t)hibernate_copy_page,
>                                                       PAGE_SIZE);
> +
>                                               inaddr += hibernate_deflate(hib,
>                                                       temp_inaddr,
>                                                       &out_remaining);
> @@ -1972,8 +1972,6 @@ hibernate_suspend(void)
>  int
>  hibernate_alloc(void)
>  {
> -     KASSERT(hibernate_temp_page == 0);
> -
>       /*
>        * If we weren't able to early allocate a piglet, don't proceed
>        */
> @@ -1984,23 +1982,7 @@ hibernate_alloc(void)
>       pmap_kenter_pa(HIBERNATE_HIBALLOC_PAGE, HIBERNATE_HIBALLOC_PAGE,
>           PROT_READ | PROT_WRITE);
> 
> -     /*
> -      * Allocate VA for the temp page.
> -      *
> -      * This will become part of the suspended kernel and will
> -      * be freed in hibernate_free, upon resume (or hibernate
> -      * failure)
> -      */
> -     hibernate_temp_page = (vaddr_t)km_alloc(PAGE_SIZE, &kv_any,
> -         &kp_none, &kd_nowait);
> -     if (!hibernate_temp_page)
> -             goto unmap;
> -
>       return (0);
> -unmap:
> -     pmap_kremove(HIBERNATE_HIBALLOC_PAGE, PAGE_SIZE);
> -     pmap_update(pmap_kernel());
> -     return (ENOMEM);
>  }
> 
>  /*
> @@ -2011,13 +1993,7 @@ hibernate_free(void)
>  {
>       pmap_activate(curproc);
> 
> -     if (hibernate_temp_page) {
> -             pmap_kremove(hibernate_temp_page, PAGE_SIZE);
> -             km_free((void *)hibernate_temp_page, PAGE_SIZE,
> -                 &kv_any, &kp_none);
> -     }
> -
> -     hibernate_temp_page = 0;
> +     pmap_kremove(HIBERNATE_TEMP_PAGE, PAGE_SIZE);
>       pmap_kremove(HIBERNATE_HIBALLOC_PAGE, PAGE_SIZE);
>       pmap_update(pmap_kernel());
>  }

-- 
Walter

Reply via email to