[Crash-utility] RISCV64: Use va_kernel_pa_offset in VTOP()

Song Shuai suagrfillet at gmail.com
Mon Jul 24 01:48:51 PDT 2023



在 2023/7/24 16:13, Alexandre Ghiti 写道:
> Hi Song,
> 
> 
> On 24/07/2023 06:06, Song Shuai wrote:
>> Since RISC-V Linux v6.4, the commit 3335068f8721 ("riscv: Use
>> PUD/P4D/PGD pages for the linear mapping") changes the
>> phys_ram_base from the kernel_map.phys_addr to the start of DRAM.
> 
> 
> Maybe we could be more explicit here, kernel_map.phys_addr actually 
> points to the physical start of the kernel so maybe something like that:
> 
> "changes phys_ram_base from the physical start of the kernel to the 
> actual start of the DRAM."
> 
ok,
> 
>>
>> The Crash's VTOP() still uses phys_ram_base and kernel_map.virt_addr
>> to translate kernel virtual address, that made Crash boot failed with
>> Linux v6.4 and later version.
>>
>> Let Linux export kernel_map.va_kernel_pa_offset in v6.5 and Crash can
>> use "va_kernel_pa_offset" to translate the kernel virtual address in
>> VTOP() correctly.
>>
>> Signed-off-by: Song Shuai <suagrfillet at gmail.com>
>> ---
>> You can check/test the Linux changes from this link:
>> https://github.com/sugarfillet/linux/commits/6.5-rc3-crash
>>
>> And I'll send the Linux changes to riscv/for-next If you're ok with 
>> this patch.
>> ---
>>   defs.h    |  4 ++--
>>   riscv64.c | 22 ++++++++++++++++++++++
>>   2 files changed, 24 insertions(+), 2 deletions(-)
>>
>> diff --git a/defs.h b/defs.h
>> index 358f365..46b9857 100644
>> --- a/defs.h
>> +++ b/defs.h
>> @@ -3662,8 +3662,7 @@ typedef signed int s32;
>>       ulong _X = X;                                    \
>>       (THIS_KERNEL_VERSION >= LINUX(5,13,0) &&                    \
>>           (_X) >= machdep->machspec->kernel_link_addr) ?                \
>> -        (((unsigned long)(_X)-(machdep->machspec->kernel_link_addr)) 
>> +        \
>> -         machdep->machspec->phys_base):                        \
>> +        ((unsigned 
>> long)(_X)-(machdep->machspec->va_kernel_pa_offset)):     \
>>           (((unsigned long)(_X)-(machdep->kvbase)) +                \
>>            machdep->machspec->phys_base);                        \
>>       })
>> @@ -7021,6 +7020,7 @@ struct machine_specific {
>>       ulong modules_vaddr;
>>       ulong modules_end;
>>       ulong kernel_link_addr;
>> +    ulong va_kernel_pa_offset;
>>       ulong _page_present;
>>       ulong _page_read;
>> diff --git a/riscv64.c b/riscv64.c
>> index 6b9a688..b9e50b4 100644
>> --- a/riscv64.c
>> +++ b/riscv64.c
>> @@ -418,6 +418,27 @@ error:
>>       error(FATAL, "cannot get vm layout\n");
>>   }
>> +static void
>> +riscv64_get_va_kernel_pa_offset(struct machine_specific *ms)
>> +{
>> +    unsigned long kernel_version = riscv64_get_kernel_version();
>> +
>> +    /*
>> +     * va_kernel_pa_offset is defined in Linux kernel since 6.5.
>> +     */
>> +    if (kernel_version >= LINUX(6,5,0)) {
>> +        char *string;
>> +        if ((string = 
>> pc->read_vmcoreinfo("NUMBER(va_kernel_pa_offset)"))) {
>> +            ms->va_kernel_pa_offset = htol(string, QUIET, NULL);
>> +            free(string);
>> +        } else
>> +            error(FATAL, "cannot read va_kernel_pa_offset\n");
>> +    } else if (kernel_version >= LINUX(6,4,0))
>> +        error(FATAL, "cannot determine va_kernel_pa_offset since 
>> Linux 6.4\n");
>> +    else
>> +        ms->va_kernel_pa_offset = ms->kernel_link_addr - ms->phys_base;
>> +}
>> +
>>   static int
>>   riscv64_is_kvaddr(ulong vaddr)
>>   {
>> @@ -1352,6 +1373,7 @@ riscv64_init(int when)
>>           riscv64_get_struct_page_size(machdep->machspec);
>>           riscv64_get_va_bits(machdep->machspec);
>>           riscv64_get_va_range(machdep->machspec);
>> +        riscv64_get_va_kernel_pa_offset(machdep->machspec);
>>           pt_level_alloc(&machdep->pgd, "cannot malloc pgd space.");
>>           pt_level_alloc(&machdep->machspec->p4d, "cannot malloc p4d 
>> space.");
> 
> 
> Would you mind giving me the instructions on how to reproduce the issue 
> please? So that I can add that to our internal CI and avoid this type of 
> breakage in the future.
> 
You can reproduce this issue via :

1. compile the Linux v6.4 or later version with Kdump support
2. generate the vmcore file via sysrq-trigger
3. start the Crash (crash-utility/crash:master) with namelist(vmlinux) 
and vmcore with optional "-d" option

Crash would boot failed with some incorrect infos (like: empty 
cpu_*_mask,utsname ) and some error like:
`crash: read error: kernel virtual address: ffffffff80ecb498  type: 
"linux_banner"`

> Thanks,
> 
> Alex
> 

-- 
Thanks
Song Shuai



More information about the linux-riscv mailing list