lkml.org 
[lkml]   [2009]   [Aug]   [12]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [Patch 8/8] kexec: allow to shrink reserved memory
    Eric W. Biederman wrote:
    > Amerigo Wang <amwang@redhat.com> writes:
    >
    >
    >> This patch implements shrinking the reserved memory for crash kernel,
    >> if it is more than enough.
    >>
    >> For example, if you have already reserved 128M, now you just want 100M,
    >> you can do:
    >>
    >> # echo $((100*1024*1024)) > /sys/kernel/kexec_crash_size
    >>
    >
    > Getting closer (comments inline)
    >
    > Semantically this patch is non-contriversial and pretty
    > simple, but still needs a fair amount of review. Can
    > you put this patch at the front of your patch set.
    >
    >

    Sure, I will do it when I resend them next time.

    I add mm people into Cc.
    >> Index: linux-2.6/kernel/kexec.c
    >> ===================================================================
    >> --- linux-2.6.orig/kernel/kexec.c
    >> +++ linux-2.6/kernel/kexec.c
    >> @@ -1083,6 +1083,76 @@ void crash_kexec(struct pt_regs *regs)
    >> }
    >> }
    >>
    >> +int kexec_crash_kernel_loaded(void)
    >> +{
    >> + int ret;
    >> + if (!mutex_trylock(&kexec_mutex))
    >> + return 1;
    >>
    >
    > We don't need trylock on this code path
    >

    OK.

    >
    >> + ret = kexec_crash_image != NULL;
    >> + mutex_unlock(&kexec_mutex);
    >> + return ret;
    >> +}
    >> +
    >> +size_t get_crash_memory_size(void)
    >> +{
    >> + size_t size;
    >> + if (!mutex_trylock(&kexec_mutex))
    >> + return 1;
    >>
    >
    > We don't need trylock on this code path
    >
    >

    Hmm, crashk_res is a global struct, so other process can also
    change it... but currently no process does that, right?

    >> + size = crashk_res.end - crashk_res.start + 1;
    >> + mutex_unlock(&kexec_mutex);
    >> + return size;
    >> +}
    >> +
    >> +int shrink_crash_memory(unsigned long new_size)
    >> +{
    >> + struct page **pages;
    >> + int ret = 0;
    >> + int npages, i;
    >> + unsigned long addr;
    >> + unsigned long start, end;
    >> + void *vaddr;
    >> +
    >> + if (!mutex_trylock(&kexec_mutex))
    >> + return -EBUSY;
    >>
    >
    > We don't need trylock on this code path
    >
    > We are missing the check to see if the crash_kernel is loaded
    > under this lock instance. So I please move the kexec_crash_image != NULL
    > test inline here and kill the kexec_crash_kernel_loaded function.
    >

    Ok, no problem.

    >
    >> + start = crashk_res.start;
    >> + end = crashk_res.end;
    >> +
    >> + if (new_size >= end - start + 1) {
    >> + ret = -EINVAL;
    >> + if (new_size == end - start + 1)
    >> + ret = 0;
    >> + goto unlock;
    >> + }
    >> +
    >> + start = roundup(start, PAGE_SIZE);
    >> + end = roundup(start + new_size, PAGE_SIZE) - 1;
    >> + npages = (end + 1 - start ) / PAGE_SIZE;
    >> +
    >> + pages = kmalloc(sizeof(struct page *) * npages, GFP_KERNEL);
    >> + if (!pages) {
    >> + ret = -ENOMEM;
    >> + goto unlock;
    >> + }
    >> + for (i = 0; i < npages; i++) {
    >> + addr = end + 1 + i * PAGE_SIZE;
    >> + pages[i] = virt_to_page(addr);
    >> + }
    >> +
    >> + vaddr = vm_map_ram(pages, npages, 0, PAGE_KERNEL);
    >>
    >
    > This is the wrong kernel call to use. I expect this needs to look
    > like a memory hotplug event. This does not put the pages into the
    > free page pool.
    >

    Well, I also wanted to use an memory-hotplug API, but that will make the
    code depend on memory-hotplug, which certainly is not what we want...

    I checked the mm code, actually what I need is an API which is similar
    to add_active_range(), but add_active_range() can't be used here since
    it is marked as "__init".

    Do we have that kind of API in mm? I can't find one.

    Thanks!




    \
     
     \ /
      Last update: 2009-08-13 05:33    [from the cache]
    ©2003-2014 Jasper Spaans. hosted at Digital Ocean