Re: [PATCH] kdump, vmcoreinfo: Export sme_me_mask value to vmcoreinfo

[lijiang <lijiang@redhat.com>]

在 2018年10月31日 10:47, Dave Young 写道:
> Add Kazu, the makedumpfile maintainer in cc.
> 
> On 10/31/18 at 10:26am, lijiang wrote:
>> 在 2018年10月30日 17:23, Baoquan He 写道:
>>>
>>> Hi Boris, DaveY and Lianbo,
>>>
>>> On 10/30/18 at 10:15am, Borislav Petkov wrote:
>>>> On Tue, Oct 30, 2018 at 01:09:00PM +0800, Dave Young wrote:
>>>>> It is not the content, I think it is a good catch from Boris, it would
>>>>> be good to document the exported things in somewhere eg.
>>>>> Documentation/kdump/vmcoreinfo.txt
>>>
>>> For the vmcoreinfo variables document, I personally think it might be
>>> not necessary. The reason is that all the old varialbles are exported
>>> with the name of themselves. We know what they are or what they
>>> represent since they are all kernel symbols or macro. Only this me_mask,
>>> it's a local variable and store the value of sme_me_mask for now, may
>>> store more info later like Petr mentioned, and also will store the memory
>>> encryption information of TME (which is intel's transparent memory encryption).
>>> We can add code comment around to tell these.
>>>
>> Thank you, everyone.
>>
>> I personally agree with Baoquan's opinion. What do you think about? Boris and other reviewer?
>>
>> If the vmcoreinfo document is necessary, would you like to help me to provide an outline?
>> I can try my best to write the document.
> 
> It is true like what Baoquan said, these information are mainly used by
> makedumpfile tool for creating a filtered vmcore.  But these vmcoreinfo
> hide in a lot of different code paths:
> kernel/crash_core.c,  printk code, arch code etc. 
> 
> It is a mist only a few kdump people know them, documenting them will help
> people to understand and review. It will also be clearer instead of
> digging into code?
> 
> The document can briefly explain what is vmcoreinfo, why we need it, and
> some background info.  Then list the exported values with some
> classifying by core kernel, arch related,  string or number etc.  For
> most of them like Baoquan said no need more explanation, but add
> explanations for something if needed like this sme mask.
> 
> But I think this can be done separately instead of blocking this patch.
For this patch, i think that it had been reached an agreement. So i will
update my patch log and post v2 if there is no objection.

diff --git a/arch/x86/kernel/machine_kexec_64.c b/arch/x86/kernel/machine_kexec_64.c
index 4c8acdfdc5a7..ca9bdf46b8e7 100644
--- a/arch/x86/kernel/machine_kexec_64.c
+++ b/arch/x86/kernel/machine_kexec_64.c
@@ -352,10 +352,23 @@ void machine_kexec(struct kimage *image)
 
 void arch_crash_save_vmcoreinfo(void)
 {
+       u64 sme_mask = sme_me_mask;
+
        VMCOREINFO_NUMBER(phys_base);
        VMCOREINFO_SYMBOL(init_top_pgt);
        vmcoreinfo_append_str("NUMBER(pgtable_l5_enabled)=%d\n",
                        pgtable_l5_enabled());
+       /*
+        * Currently, the local variable 'sme_mask' stores the value of
+        * sme_me_mask(bit 47), and also write the value of sme_mask to
+        * the vmcoreinfo.
+        * If need, the bit(sme_mask) might be redefined in the future.
+        * For example:
+        * [ misc          ][ enc bit  ][ other misc SME info       ]
+        * 0000_0000_0000_0000_1000_0000_0000_0000_0000_0000_..._0000
+        * 63   59   55   51   47   43   39   35   31   27   ... 3
+        */
+       VMCOREINFO_NUMBER(sme_mask);
 
 #ifdef CONFIG_NUMA
        VMCOREINFO_SYMBOL(node_data);

> Maybe Kazu knows more about it, so I would like to ask Kazu about his opinion
> and if he want to do it.
> Ok, thank you, Dave Young.

About the vmcoreinfo document issue, once make a decision, i will try my best to
do this based on your outline.

Regards,
Lianbo

>>
>> Anyway, we should make a choice.
>>
>> Regards,
>> Lianbo
>>
>>> Personal opinion.
>>>
>>> Thanks
>>> Baoquan
>>>
> 
> Thanks
> Dave
>