Olaf Hering wrote:
> Upcoming glibc header files have no visible definition of HZ anymore. A
> glibc macro exist to query the current tick rate. This macro will likely
> work for all other apps, Unfortunately crash wants to know the HZ value
> of the crashed kernel.
Right...
>
> My patch makes the assumption that the dump file is analyzed on a system
> similar to the crashed one. I dont know how accurate that is.
> If I read the source right, HZ is currently forced to be 1000 which is
> not true either. Newer kernels can be configured for 100, 250 and 1000.
>
Right... (for the most part)
>
> Ideally, the core file exports HZ in some way. I see cfq_slice_async is
> used to get the HZ value in task_init().
>
That's right -- if it exists, it ends up over-writing whatever earlier
(potentially bogus) machdep->hz setting was put in place during
an earlier call to machdep_init().
So that being the case, I don't see using sysconf() as any more
useful than the hardcoded HZ on systems that have it available
when crash is built. I think it might be simpler to keep the current
code in place, except that everywhere "HZ" is used, it should
be surrounded by "#ifdef HZ". That keeps backwards-compatibility
in place, while in newer environments/kernels the cfq_slice_async
check determines the correct value anyway. Also, for example,
I often use 2.4-based machines to look at 2.6 dumpfiles that are
"pre-cfq_slice_async", and using sysconf() would break something
that works now.
As another alternative, perhaps Troy's new IKCONFIG parser could
also be used to read CONFIG_HZ?
Dave
>
> ---
> alpha.c | 2 +-
> ia64.c | 2 +-
> ppc.c | 4 +---
> ppc64.c | 4 +---
> s390.c | 2 +-
> s390x.c | 2 +-
> x86.c | 4 +---
> x86_64.c | 4 +---
> 8 files changed, 8 insertions(+), 16 deletions(-)
>
> Index: crash-4.0-2.33/alpha.c
> ===================================================================
> --- crash-4.0-2.33.orig/alpha.c
> +++ crash-4.0-2.33/alpha.c
> @@ -186,7 +186,7 @@ alpha_init(int when)
> "irq_desc", NULL, 0);
> else
> machdep->nr_irqs = 0;
> - machdep->hz = HZ;
> + machdep->hz = sysconf(_SC_CLK_TCK);
> break;
>
> case POST_INIT:
> Index: crash-4.0-2.33/ppc.c
> ===================================================================
> --- crash-4.0-2.33.orig/ppc.c
> +++ crash-4.0-2.33/ppc.c
> @@ -135,9 +135,7 @@ ppc_init(int when)
> "irq_desc", NULL, 0);
> else
> machdep->nr_irqs = 0;
> - machdep->hz = HZ;
> - if (THIS_KERNEL_VERSION >= LINUX(2,6,0))
> - machdep->hz = 1000;
> + machdep->hz = sysconf(_SC_CLK_TCK);
> machdep->section_size_bits = _SECTION_SIZE_BITS;
> machdep->max_physmem_bits = _MAX_PHYSMEM_BITS;
> break;
> Index: crash-4.0-2.33/ppc64.c
> ===================================================================
> --- crash-4.0-2.33.orig/ppc64.c
> +++ crash-4.0-2.33/ppc64.c
> @@ -237,9 +237,7 @@ ppc64_init(int when)
> */
> BZERO(&machdep->machspec->hwintrstack,
> NR_CPUS*sizeof(ulong));
> - machdep->hz = HZ;
> - if (THIS_KERNEL_VERSION >= LINUX(2,6,0))
> - machdep->hz = 1000;
> + machdep->hz = sysconf(_SC_CLK_TCK);
> /*
> * IRQ stacks are introduced in 2.6 and also configurable.
> */
> Index: crash-4.0-2.33/s390.c
> ===================================================================
> --- crash-4.0-2.33.orig/s390.c
> +++ crash-4.0-2.33/s390.c
> @@ -158,7 +158,7 @@ s390_init(int when)
> machdep->nr_irqs = 0; /* TBD */
> machdep->vmalloc_start = s390_vmalloc_start;
> machdep->dump_irq = s390_dump_irq;
> - machdep->hz = HZ;
> + machdep->hz = sysconf(_SC_CLK_TCK);
> break;
>
> case POST_INIT:
> Index: crash-4.0-2.33/s390x.c
> ===================================================================
> --- crash-4.0-2.33.orig/s390x.c
> +++ crash-4.0-2.33/s390x.c
> @@ -173,7 +173,7 @@ s390x_init(int when)
> machdep->nr_irqs = 0; /* TBD */
> machdep->vmalloc_start = s390x_vmalloc_start;
> machdep->dump_irq = s390x_dump_irq;
> - machdep->hz = HZ;
> + machdep->hz = sysconf(_SC_CLK_TCK);
> break;
>
> case POST_INIT:
> Index: crash-4.0-2.33/x86.c
> ===================================================================
> --- crash-4.0-2.33.orig/x86.c
> +++ crash-4.0-2.33/x86.c
> @@ -1775,9 +1775,7 @@ x86_init(int when)
> "irq_desc", NULL, 0);
> else
> machdep->nr_irqs = 224; /* NR_IRQS */
> - machdep->hz = HZ;
> - if (THIS_KERNEL_VERSION >= LINUX(2,6,0))
> - machdep->hz = 1000;
> + machdep->hz = sysconf(_SC_CLK_TCK);
>
> if (machdep->flags & PAE){
> machdep->section_size_bits = _SECTION_SIZE_BITS_PAE;
> Index: crash-4.0-2.33/x86_64.c
> ===================================================================
> --- crash-4.0-2.33.orig/x86_64.c
> +++ crash-4.0-2.33/x86_64.c
> @@ -238,9 +238,7 @@ x86_64_init(int when)
> machdep->nr_irqs = 224; /* NR_IRQS (at least) */
> machdep->vmalloc_start = x86_64_vmalloc_start;
> machdep->dump_irq = x86_64_dump_irq;
> - machdep->hz = HZ;
> - if (THIS_KERNEL_VERSION >= LINUX(2,6,0))
> - machdep->hz = 1000;
> + machdep->hz = sysconf(_SC_CLK_TCK);
> machdep->section_size_bits = _SECTION_SIZE_BITS;
> machdep->max_physmem_bits = _MAX_PHYSMEM_BITS;
> if (XEN()) {
> Index: crash-4.0-2.33/ia64.c
> ===================================================================
> --- crash-4.0-2.33.orig/ia64.c
> +++ crash-4.0-2.33/ia64.c
> @@ -205,7 +205,7 @@ ia64_init(int when)
> else if (symbol_exists("_irq_desc"))
> ARRAY_LENGTH_INIT(machdep->nr_irqs, irq_desc,
> "_irq_desc", NULL, 0);
> - machdep->hz = 1024;
> + machdep->hz = sysconf(_SC_CLK_TCK);
> ia64_create_memmap();
> break;
>
>
> --
> Crash-utility mailing list
> Crash-utility@xxxxxxxxxx
> https://www.redhat.com/mailman/listinfo/crash-utility
--
Crash-utility mailing list
Crash-utility@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/crash-utility
