lkml.org 
[lkml]   [2015]   [Dec]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    From
    SubjectRe: [PATCH] memory-hotplug: don't BUG() in register_memory_resource()
    On Mon, 21 Dec 2015 11:13:15 +0100 Vitaly Kuznetsov <vkuznets@redhat.com> wrote:

    > Andrew Morton <akpm@linux-foundation.org> writes:
    >
    > > On Fri, 18 Dec 2015 15:50:24 +0100 Vitaly Kuznetsov <vkuznets@redhat.com> wrote:
    > >
    > >> Out of memory condition is not a bug and while we can't add new memory in
    > >> such case crashing the system seems wrong. Propagating the return value
    > >> from register_memory_resource() requires interface change.
    > >>
    > >> --- a/mm/memory_hotplug.c
    > >> +++ b/mm/memory_hotplug.c
    > >> +static int register_memory_resource(u64 start, u64 size,
    > >> + struct resource **resource)
    > >> {
    > >> struct resource *res;
    > >> res = kzalloc(sizeof(struct resource), GFP_KERNEL);
    > >> - BUG_ON(!res);
    > >> + if (!res)
    > >> + return -ENOMEM;
    > >>
    > >> res->name = "System RAM";
    > >> res->start = start;
    > >> @@ -140,9 +142,10 @@ static struct resource *register_memory_resource(u64 start, u64 size)
    > >> if (request_resource(&iomem_resource, res) < 0) {
    > >> pr_debug("System RAM resource %pR cannot be added\n", res);
    > >> kfree(res);
    > >> - res = NULL;
    > >> + return -EEXIST;
    > >> }
    > >> - return res;
    > >> + *resource = res;
    > >> + return 0;
    > >> }
    > >
    > > Was there a reason for overwriting the request_resource() return
    > > value?
    > > Ordinarily it should be propagated back to callers.
    > >
    > > Please review.
    > >
    >
    > This is a nice-to-have addition but it will break at least ACPI
    > memhotplug: request_resource() has the following:
    >
    > conflict = request_resource_conflict(root, new);
    > return conflict ? -EBUSY : 0;
    >
    > so we'll end up returning -EBUSY from register_memory_resource() and
    > add_memory(), at the same time acpi_memory_enable_device() counts on
    > -EEXIST:
    >
    > result = add_memory(node, info->start_addr, info->length);
    >
    > /*
    > * If the memory block has been used by the kernel, add_memory()
    > * returns -EEXIST. If add_memory() returns the other error, it
    > * means that this memory block is not used by the kernel.
    > */
    > if (result && result != -EEXIST)
    > continue;
    >
    > So I see 3 options here:
    > 1) Keep the overwrite
    > 2) Change the request_resource() return value to -EEXIST
    > 3) Adapt all add_memory() call sites to -EBUSY.
    >
    > Please let me know your preference.

    urgh, what a mess. We should standardize on EBUSY or EEXIST, I don't
    see that it matter much which is chosen. And for robustness the
    callers should be checking for (err < 0) unless there's a very good
    reason otherwise.

    But it doesn't seem terribly important.


    \
     
     \ /
      Last update: 2015-12-22 00:21    [W:3.682 / U:0.004 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site