Re: [PATCH V2] mm/gup: Fix longterm pin on slow gup regression

[Yang Shi <shy828301@gmail.com>]

On Fri, Jun 28, 2024 at 1:42 PM Andrew Morton <akpm@linux-foundation.org> wrote:
>
> On Fri, 28 Jun 2024 14:01:58 +0800 yangge1116@126.com wrote:
>
> > From: yangge <yangge1116@126.com>
> >
> > If a large number of CMA memory are configured in system (for
> > example, the CMA memory accounts for 50% of the system memory),
> > starting a SEV virtual machine will fail. During starting the SEV
> > virtual machine, it will call pin_user_pages_fast(..., FOLL_LONGTERM,
> > ...) to pin memory. Normally if a page is present and in CMA area,
> > pin_user_pages_fast() will first call __get_user_pages_locked() to
> > pin the page in CMA area, and then call
> > check_and_migrate_movable_pages() to migrate the page from CMA area
> > to non-CMA area. But the current code calling __get_user_pages_locked()
> > will fail, because it call try_grab_folio() to pin page in gup slow
> > path.
> >
> > The commit 57edfcfd3419 ("mm/gup: accelerate thp gup even for "pages
> > != NULL"") uses try_grab_folio() in gup slow path, which seems to be
> > problematic because try_grap_folio() will check if the page can be
> > longterm pinned. This check may fail and cause __get_user_pages_lock()
> > to fail. However, these checks are not required in gup slow path,
> > seems we can use try_grab_page() instead of try_grab_folio(). In
> > addition, in the current code, try_grab_page() can only add 1 to the
> > page's refcount. We extend this function so that the page's refcount
> > can be increased according to the parameters passed in.
> >
> > The following log reveals it:
> >
> > [  464.325306] WARNING: CPU: 13 PID: 6734 at mm/gup.c:1313 __get_user_pages+0x423/0x520
> > [  464.325464] CPU: 13 PID: 6734 Comm: qemu-kvm Kdump: loaded Not tainted 6.6.33+ #6
> > [  464.325477] RIP: 0010:__get_user_pages+0x423/0x520
> > [  464.325515] Call Trace:
> > [  464.325520]  <TASK>
> > [  464.325523]  ? __get_user_pages+0x423/0x520
> > [  464.325528]  ? __warn+0x81/0x130
> > [  464.325536]  ? __get_user_pages+0x423/0x520
> > [  464.325541]  ? report_bug+0x171/0x1a0
> > [  464.325549]  ? handle_bug+0x3c/0x70
> > [  464.325554]  ? exc_invalid_op+0x17/0x70
> > [  464.325558]  ? asm_exc_invalid_op+0x1a/0x20
> > [  464.325567]  ? __get_user_pages+0x423/0x520
> > [  464.325575]  __gup_longterm_locked+0x212/0x7a0
> > [  464.325583]  internal_get_user_pages_fast+0xfb/0x190
> > [  464.325590]  pin_user_pages_fast+0x47/0x60
> > [  464.325598]  sev_pin_memory+0xca/0x170 [kvm_amd]
> > [  464.325616]  sev_mem_enc_register_region+0x81/0x130 [kvm_amd]
> >
>
> Well, we also have Yang Shi's patch
> (https://lkml.kernel.org/r/20240627231601.1713119-1-yang@os.amperecomputing.com)
> which takes a significantly different approach.  Which way should we
> go?

IMO, my patch is more complete, it should be sent to the mainline.
This patch can be considered if it is hard to backport my patch to the
stable tree.

>