Hi Tao,
On 05/09/24 03:40, Tao Liu wrote:
Hi Aditya,
On Wed, Sep 4, 2024 at 10:27 PM Aditya Gupta <adityag(a)linux.ibm.com> wrote:
> Hi Tao,
>
> Is this a separate independent patch or related to this series ?
>
This is an independent patch, but discovered when doing a regression
test against the stack unwinding patchset. Since we have done the
cpumask_t modification in [PATCH v6 10/14], the modification will
continue in v7 as well. So I posted it with v7 together.
Lianbo suggested that I separate some "less strongly connected with
stack unwinding" patches from v7, which formed the part 1 as mentioned
in [1]. Part 1 can be merged prior to the rest of part 2 & 3. So I
think part 1 is also suitable for the cpumask_t regression patch.
Yes, I agree. We can move 'Fix gdb_interface' patch also as independent.
It originally was required for the patch series to work, since 'info
threads' used to break, but now that can go independently.
Thanks,
Aditya Gupta
> [1]:
https://www.mail-archive.com/devel@lists.crash-utility.osci.io/msg01079.html
>
> Thanks,
> Tao Liu
>
>> - Aditya Gupta
>>
>>
>> On 04/09/24 13:19, Tao Liu wrote:
>>> There is a regression been found for xen hyper due to the commit:
>>> f615f8fab7bf ("Fix "irq -a" exceeding the memory range
issue").
>>>
>>> The reason is for xen hyper, kt->cpu is not initialized due to
>>> kernel_init() won't be called. So 0 would be assigned to cpulen and
>>> fails the GETBUF().
>>>
>>> Before:
>>> crash> bt -c 2
>>> bt: zero-size memory allocation! (called from 51f168)
>>>
>>> After:
>>> crash> bt -c 2
>>> PCPU: 0 VCPU: ffff8300001b8080
>>> #0 [ffff8300001bfe00] machine_crash_kexec at ffff83000010de72
>>> #1 [ffff8300001bfe10] do_kexec_op at ffff83000010e3cb
>>> #2 [ffff8300001bfe50] do_console_io at ffff83000011aff4
>>> #3 [ffff8300001bfe90] mod_l1_entry at ffff830000129045
>>> #4 [ffff8300001bfea0] toggle_guest_mode at ffff8300001641bf
>>> #5 [ffff8300001bfeb0] do_iret at ffff830000164888
>>> #6 [ffff8300001bff20] syscall_enter at ffff8300001633d2
>>>
>>> Since xen hyper will initialize its own cpumask_t, this patch will reuse
>>> it for XEN_HYPER_MODE. In addition, this patch won't modify the similar
>>> code hunk in generic_get_irq_affinity(), because for xen, there is no
>>> "irq" cmd, so the regression won't be triggered.
>>>
>>> Cc: Sourabh Jain <sourabhjain(a)linux.ibm.com>
>>> Cc: Hari Bathini <hbathini(a)linux.ibm.com>
>>> Cc: Mahesh J Salgaonkar <mahesh(a)linux.ibm.com>
>>> Cc: Naveen N. Rao <naveen.n.rao(a)linux.vnet.ibm.com>
>>> Cc: Lianbo Jiang <lijiang(a)redhat.com>
>>> Cc: HAGIO KAZUHITO(萩尾 一仁) <k-hagio-ab(a)nec.com>
>>> Cc: Tao Liu <ltao(a)redhat.com>
>>> Cc: Alexey Makhalov <alexey.makhalov(a)broadcom.com>
>>> Cc: Aditya Gupta <adityag(a)linux.ibm.com>
>>> Signed-off-by: Tao Liu <ltao(a)redhat.com>
>>> ---
>>> tools.c | 6 ++++++
>>> 1 file changed, 6 insertions(+)
>>>
>>> diff --git a/tools.c b/tools.c
>>> index 2b78b95..dec6e5b 100644
>>> --- a/tools.c
>>> +++ b/tools.c
>>> @@ -16,6 +16,7 @@
>>> */
>>>
>>> #include "defs.h"
>>> +#include "xen_hyper_defs.h"
>>> #include <ctype.h>
>>>
>>> #ifdef VALGRIND
>>> @@ -6720,6 +6721,11 @@ get_cpumask_buf(void)
>>> {
>>> int cpulen, len_cpumask;
>>>
>>> +#ifdef XEN_HYPERVISOR_ARCH
>>> + if (XEN_HYPER_MODE() && XEN_HYPER_VALID_SIZE(cpumask_t))
>>> + return (ulong *)GETBUF(XEN_HYPER_SIZE(cpumask_t));
>>> +#endif
>>> +
>>> cpulen = DIV_ROUND_UP(kt->cpus, BITS_PER_LONG) * sizeof(ulong);
>>> len_cpumask = STRUCT_SIZE("cpumask_t");
>>> if (len_cpumask > 0)