Hi,
Thanks for bringing up the issue.
We clearly will need device-specific vma types (in plural!),
and I suspect that the way it will work is that devices will
register their vma types at boot time or dynamically (for
kernel modules).
On 10/20/2010 02:56 PM, Nathan Lynch wrote:
> This is intended to be used by drivers that allow userspace to map
> hardware resources via mmap(2). The mapping is restored but we don't
> worry about the contents. This is pretty much how we treat shared
> file mappings, where it is the user's responsibility to ensure file
> (device) availability and integrity for restart. I think this is a
> reasonable approach for the timer implementations in drivers/char as
> well as the bsr driver (for which a patch follows).
The given use case is for a specific device, and assumes special
userspace behavior and awareness (if not the application, at
least the admin that runs the c/r).
A similar solution will also apply for memory regions marked by
the process as "scratch" - a way for programmers to tell the kernel
that certain memory need not be saved/restored beyond the mapping.
I'm a bit concerned about tying this behavior permanently to a
drivers. (Although, it may make sense for some drivers). What do
you think ?
Otherwise, the patch looks correct. Perhaps modify the types to
something more generic, like CKPT_VMA_VANILLA ?
Thanks,
Oren.
>
> Signed-off-by: Nathan Lynch <ntl@xxxxxxxxx>
> ---
> include/linux/checkpoint.h | 3 ++-
> include/linux/checkpoint_hdr.h | 2 ++
> mm/checkpoint.c | 37 +++++++++++++++++++++++++++++++++++++
> 3 files changed, 41 insertions(+), 1 deletions(-)
>
> diff --git a/include/linux/checkpoint.h b/include/linux/checkpoint.h
> index 49cf01e..a02f27a 100644
> --- a/include/linux/checkpoint.h
> +++ b/include/linux/checkpoint.h
> @@ -290,7 +290,8 @@ extern int shmem_vma_checkpoint(struct ckpt_ctx *ctx,
> struct vm_area_struct *vma,
> enum vma_type type,
> int ino_objref);
> -
> +extern int device_vma_checkpoint(struct ckpt_ctx *ctx,
> + struct vm_area_struct *vma);
> extern int checkpoint_obj_mm(struct ckpt_ctx *ctx, struct task_struct *t);
> extern int restore_obj_mm(struct ckpt_ctx *ctx, int mm_objref);
>
> diff --git a/include/linux/checkpoint_hdr.h b/include/linux/checkpoint_hdr.h
> index 049bb82..6a3e309 100644
> --- a/include/linux/checkpoint_hdr.h
> +++ b/include/linux/checkpoint_hdr.h
> @@ -914,6 +914,8 @@ enum vma_type {
> #define CKPT_VMA_SHM_IPC CKPT_VMA_SHM_IPC
> CKPT_VMA_SHM_IPC_SKIP, /* shared sysvipc (skip contents) */
> #define CKPT_VMA_SHM_IPC_SKIP CKPT_VMA_SHM_IPC_SKIP
> + CKPT_VMA_DEVICE, /* c/r mapping only, skip contents */
> +#define CKPT_VMA_DEVICE CKPT_VMA_DEVICE
> CKPT_VMA_MAX,
> #define CKPT_VMA_MAX CKPT_VMA_MAX
> };
> diff --git a/mm/checkpoint.c b/mm/checkpoint.c
> index c30a195..38c8b1f 100644
> --- a/mm/checkpoint.c
> +++ b/mm/checkpoint.c
> @@ -591,6 +591,20 @@ static int anonymous_checkpoint(struct ckpt_ctx *ctx,
> return private_vma_checkpoint(ctx, vma, CKPT_VMA_ANON, 0);
> }
>
> +int device_vma_checkpoint(struct ckpt_ctx *ctx, struct vm_area_struct *vma)
> +{
> + int vma_objref;
> + int ret;
> +
> + vma_objref = checkpoint_obj(ctx, vma->vm_file, CKPT_OBJ_FILE);
> + if (vma_objref < 0)
> + return vma_objref;
> +
> + ret = generic_vma_checkpoint(ctx, vma, CKPT_VMA_DEVICE, vma_objref, 0);
> +
> + return ret;
> +}
> +
> static int checkpoint_vmas(struct ckpt_ctx *ctx, struct mm_struct *mm)
> {
> struct vm_area_struct *vma, *next;
> @@ -1114,6 +1128,23 @@ static int bad_vma_restore(struct ckpt_ctx *ctx,
> return -EINVAL;
> }
>
> +static int device_vma_restore(struct ckpt_ctx *ctx,
> + struct mm_struct *mm,
> + struct ckpt_hdr_vma *h)
> +{
> + unsigned long addr;
> + struct file *file;
> +
> + file = ckpt_obj_fetch(ctx, h->vma_objref, CKPT_OBJ_FILE);
> + if (IS_ERR(file))
> + return PTR_ERR(file);
> +
> + addr = generic_vma_restore(mm, file, h);
> +
> + return IS_ERR((void *)addr) ? PTR_ERR((void *)addr) : 0;
> +}
> +
> +
> /* callbacks to restore vma per its type: */
> struct restore_vma_ops {
> char *vma_name;
> @@ -1180,6 +1211,12 @@ static struct restore_vma_ops restore_vma_ops[] = {
> .vma_type = CKPT_VMA_SHM_IPC_SKIP,
> .restore = ipcshm_restore,
> },
> + /* device memory - restore mapping, not contents */
> + {
> + .vma_name = "DEVICE",
> + .vma_type = CKPT_VMA_DEVICE,
> + .restore = device_vma_restore,
> + },
> };
>
> /**
_______________________________________________
Containers mailing list
Containers@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linux-foundation.org/mailman/listinfo/containers
