Re: [PATCH v19 2/7] crash: add generic infrastructure for crash hotplug support

[Eric DeVolder]

On 3/17/23 04:04, Baoquan He wrote:
> On 03/16/23 at 09:44am, Eric DeVolder wrote:
> > On 3/16/23 05:11, Baoquan He wrote:
> > > On 03/06/23 at 11:22am, Eric DeVolder wrote:
> > > ......
> > > > +static void crash_handle_hotplug_event(unsigned int hp_action, unsigned int cpu)
> > > > +{
> > > > +	/* Obtain lock while changing crash information */
> > > > +	if (kexec_trylock()) {
> > > > +
> > > > +		/* Check kdump is loaded */
> > > > +		if (kexec_crash_image) {
> > > > +			struct kimage *image = kexec_crash_image;
> > > > +
> > > > +			if (hp_action == KEXEC_CRASH_HP_ADD_CPU ||
> > > > +				hp_action == KEXEC_CRASH_HP_REMOVE_CPU)
> > > > +				pr_debug("hp_action %u, cpu %u\n", hp_action, cpu);
> > > > +			else
> > > > +				pr_debug("hp_action %u\n", hp_action);
> > > > +
> > > > +			/*
> > > > +			 * When the struct kimage is allocated, the elfcorehdr_index
> > > > +			 * is set to -1. Find the segment containing the elfcorehdr,
> > > > +			 * if not already found. This works for both the kexec_load
> > > > +			 * and kexec_file_load paths.
> > > > +			 */
> > > > +			if (image->elfcorehdr_index < 0) {
> > > > +				unsigned long mem;
> > > > +				unsigned char *ptr;
> > > > +				unsigned int n;
> > > > +
> > > > +				for (n = 0; n < image->nr_segments; n++) {
> > > > +					mem = image->segment[n].mem;
> > > > +					ptr = kmap_local_page(pfn_to_page(mem >> PAGE_SHIFT));
> > > > +					if (ptr) {
> > > > +						/* The segment containing elfcorehdr */
> > > > +						if (memcmp(ptr, ELFMAG, SELFMAG) == 0) {
> > > > +							image->elfcorehdr_index = (int)n;
> > > > +						}
> > > > +						kunmap_local(ptr);
> > > > +					}
> > > > +				}
> > > > +			}
> > > > +
> > > > +			if (image->elfcorehdr_index < 0) {
> > > > +				pr_err("unable to locate elfcorehdr segment");
> > > > +				goto out;
> > > > +			}
> > > > +
> > > > +			/* Needed in order for the segments to be updated */
> > > > +			arch_kexec_unprotect_crashkres();
> > > > +
> > > > +			/* Differentiate between normal load and hotplug update */
> > > > +			image->hp_action = hp_action;
> > > > +
> > > > +			/* Now invoke arch-specific update handler */
> > > > +			arch_crash_handle_hotplug_event(image);
> > > > +
> > > > +			/* No longer handling a hotplug event */
> > > > +			image->hp_action = KEXEC_CRASH_HP_NONE;
> > > > +			image->elfcorehdr_updated = true;
> > >
> > > It's good to initialize the image->hp_action here, however where do
> > > you check it? Do you plan to add some check somewhere?
> >
> > Hi Baoquan,
> > The hp_action member is initialized to 0 in do_image_alloc_init(). I've
> > mapped KEXEC_CRASH_HP_NONE onto 0 on purpose.
> >
> > But the use of image->hp_action = KEXEC_CRASH_HP_NONE is to actually
> > delineate that a hotplug event handling has completed. You can see
> > imae->hp_action set to hp_action to capture what the triggering event
> > was, as passed into this function.
> >
> > I will go ahead and set image->hp_action = KEXEC_CRASH_HP_NONE; explicitly
> > in do_kimage_alloc_init(), as that is done for the other crash hotplug members.
>
> Yeah, setting image->hp_action = KEXEC_CRASH_HP_NONE in
> do_kimage_alloc_init() will make code clearer. While I am wondering if
> we don't initialie image->hp_action to KEXEC_CRASH_HP_NONE, and don't
> set image->hp_action to KEXEC_CRASH_HP_NONE to actually delineate that a
> hotplug event handling has completed, what will happen?

Baoquan,
The KEXEC_CRASH_HP_NONE is the value 0, intentionally so that upon a alloc
of the kimage, the struct kimage is automatically zeroed and it was initialized
properly that way. I am explicitly setting hp_action to KEXEC_CRASH_HP_NONE now.

> I mean you set image->hp_action to KEXEC_CRASH_HP_NONE explicitly, where
> do you check if it should not be KEXEC_CRASH_HP_NONE? In
> crash_handle_hotplug_event(), we took __kexec_lock and assign the passed
> hp_action anyway.
The cpuhp callbacks and memory notifiers invoke crash_handle_hotplug_event()
with an appropriate hp_action. That hp_action is then stored in image->hp_action
within crash_handle_hotplug_event() for use by the arch-specific handler.

For x86, for example, the image->hp_action is used to short-circuit the arch-
specific handler if the event is a CPU plug/unplug (see patch x86/crash:
optimize cpu changes). For PPC, for example, the image->hp_action is used to
determine the appropriate actions for its FDT updates.

To summarize, the image->hp_action will be initalized to KEXEC_CRASH_HP_NONE
during do_kimage_alloc_init(). Then upon a cpu or memory plug/unplug/online/offline
event, the appropriate hp_action is stored in image->hp_action and then the
arch-specific handler called. Upon returning from the arch-specific handler,
the image->hp_action is reset back to KEXEC_CRASH_HP_NONE.

Hope this helps. I'll be posting v20 soon.
Thanks!
eric