[PATCH] access user space addresses/switch process context

[Keith Owens]

On Wed, 16 Apr 2003 17:31:33 +0530, 
"Vamsi Krishna S ." <vamsi at in.ibm.com> wrote:
>diff -urN -X /home/vamsi/.dontdiff 2420-kdb4.1-pure/kdb/kdbsupport.c 2420-kdb4.1/kdb/kdbsupport.c
>--- 2420-kdb4.1-pure/kdb/kdbsupport.c	2003-04-16 11:38:14.000000000 +0530
>+++ 2420-kdb4.1/kdb/kdbsupport.c	2003-04-16 12:56:42.000000000 +0530
>+/*
>+ * from mm/memory.c, adapted to run without any locks to work within kdb
>+ */
>+static struct page * kdb_follow_page(struct mm_struct *mm, unsigned long address, int write) 

AFAICT this function is identical to follow_page().  Instead of
duplicating that code and possibly getting out of sync with the real
follow_page(), change follow_page() so it is extern for CONFIG_KDB=y,
otherwise it is static.

>+static struct page * kdb_get_one_user_page(struct task_struct *tsk, unsigned long start,
>+		int len, int write)

Why have a write flag?  KDB will only read user pages, or are you
planning more changes that will require write access?

>+	/* shouldn't cross a page boundary. temporary restriction. */
>+	if ((from & PAGE_MASK) != ((from+size) & PAGE_MASK)) {
>+		kdb_printf("%s: crosses page boundary: from=%08lx, size=%d\n", 
>+			__FUNCTION__, from, size);
>+		return size;
>+	}

'return size' changes the semantics of __kdba_getarea_size.  Currently
it returns 0 or a negative error code, now you are returning a positive
value on an error.  Why?  Any failure to get a user page should return
-EFAULT, the same as a kernel page.