Skip to contentSkip to conversation

Problem with accessing page tables

DISCUSSION Ameya Mitragotri 2 participants 6 messages Nov 8, 2003 – Nov 11, 2003

lore
Share & tools
lore
#0 lore raw

Hi,

I wrote following piece of code to traverse the page table for a
particular
Kernel variable. But the PMD access is always a failure.

I'm using Itanium 2 with linux-2.6.0-test5.
Any pointers will be really helpful.
Thanks and regards,
Ameya

	pgd_t *pgd;
	pmd_t *pmd;
	pte_t *ptep, pte;
	unsigned long pfn;

	printk(KERN_EMERG "The address is %lx\n\n\n",address);

	pgd = pgd_offset_k(address);
	if (pgd_none(*pgd) || pgd_bad(*pgd))
		goto out;
Quoted 1 line
Quoted 1 line
	pmd = pmd_offset(pgd, address); <<<

One of the checks below fails because the call pmd_offset gives pmd With value o.

	if (pmd_none(*pmd))
		goto out;
	if (pmd_bad(*pmd))
		goto out;

	ptep = pte_offset_kernel(pmd, address);
	if (!ptep)
		goto out;
	pte = *ptep;
	pte_unmap(ptep);
	if (pte_present(pte)) {
		if (!write || (pte_write(pte) && pte_dirty(pte))) {
			pfn = pte_pfn(pte);
			if (pfn_valid(pfn)) {
				struct page *page = pfn_to_page(pfn);

				//mark_page_accessed(page);
				return page;
			}
		}
	}
	else
	{
		printk(KERN_EMERG "pte entry not present \n");
	}

out:

	printk(KERN_EMERG "At out \n");
#1 lore raw
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 1 line
Quoted 1 line · Ameya Mitragotri

pmd = pmd_offset(pgd, address); <<<

Ameya> One of the checks below fails because the call pmd_offset gives pmd Ameya> With value o.

That means the PTE directory hasn't been allocated yet. Perfectly normal for memory that hasn't been accessed yet.

	--david
#2 lore raw
In reply to Ameya Mitragotri
Quoted 6 lines

Ameya> One of the checks below fails because the call
pmd_offset gives pmd
Ameya> With value o.

That means the PTE directory hasn't been allocated yet. Perfectly normal for memory that hasn't been accessed yet.

We passed the address of swapper_pg_dir and this is what we got.

Nov  8 12:29:35 i2 kernel: The swapper_pg_dir address is
a0000001008d6000
Nov  8 12:29:35 i2 kernel: The pgd offset is 280 address of pgd entry is
a0000001008d6000 value @ pgd entry is 2146033664
Nov  8 12:29:35 i2 kernel: The pmd offset is 201 pmd address is

e00000007fe9f008
value @ pmd is 0

Now I think the memory for the swapper_pg_dir has already been allocated.
I still get PMD as 0.

Am I doing something fundamentally wrong?

Thanks
ameya

#3 lore raw
In reply to Ameya Mitragotri
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Ameya Mitragotri

Ameya> We passed the address of swapper_pg_dir and this is what we got.

Ameya> Nov 8 12:29:35 i2 kernel: The swapper_pg_dir address is Ameya> a0000001008d6000
Ameya> Nov 8 12:29:35 i2 kernel: The pgd offset is 280 address of pgd entry is Ameya> a0000001008d6000 value @ pgd entry is 2146033664
Ameya> Nov 8 12:29:35 i2 kernel: The pmd offset is 201 pmd address is Ameya> e00000007fe9f008
Ameya> value @ pmd is 0

Ameya> Now I think the memory for the swapper_pg_dir has already been Ameya> allocated.
Ameya> I still get PMD as 0.

Again, that's normal for memory that hasn't been allocated yet.

Ameya> Am I doing something fundamentally wrong?

I don't know. What exactly are you trying to do?

	--david
#4 lore raw
In reply to Ameya Mitragotri
Quoted 10 lines

Ameya> Now I think the memory for the swapper_pg_dir has
already been
Ameya> allocated.
Ameya> I still get PMD as 0.

Again, that's normal for memory that hasn't been allocated yet.

Ameya> Am I doing something fundamentally wrong?

I don't know. What exactly are you trying to do?

Our assumption is that the addresses for the global variables will surely be allocated.
(e.g. &panic_timeout should always resolve to an available PGD,PMD,PTE and
in the end to a page.)
But we see that it is not happening in our case. (In case of swapper_pg_dir
one page is already allocated so it should point to a valid page). Any valid address which we have taken from System.map does not resolve down to a
valid PTE. (??)

The whole purpose of this being, We want to simulate the page tables at user level
to analyse a crash dump (LKCD) on IA64. So we dump the page tables and when
Lcrash (the user mode tool for analysing dumps) reads this dump, it can seek the
page tables and access global variables accordingly. This also will help us eventually
in unwinding the dumped stack.

Thanks
Ameya

#5 lore raw
In reply to Ameya Mitragotri
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Ameya Mitragotri

Ameya> Our assumption is that the addresses for the global variables Ameya> will surely be allocated. (e.g. &panic_timeout should always Ameya> resolve to an available PGD,PMD,PTE and in the end to a Ameya> page.)

Nope, that's a bad assumption. The kernel's page-table is strictly needed only for vmalloc'd memory. Whether or not it's used to map the kernel itself, is architecture-specific. In the case of ia64, we use an identity-mapping for kernel memory (the kernel itself is pinned into the data TLB), so no page-tables are used.

Ameya> The whole purpose of this being, We want to simulate the page Ameya> tables at user level to analyse a crash dump (LKCD) on Ameya> IA64. So we dump the page tables and when Lcrash (the user Ameya> mode tool for analysing dumps) reads this dump, it can seek Ameya> the page tables and access global variables accordingly. This Ameya> also will help us eventually in unwinding the dumped stack.

In the 2.6 kernel, the kernel image is pinned into the D TLB at address 0xa000000100000000. This _usually_ maps to physical address 64MB, but under certain circumstances (e.g., lack of memory at 64MB, kernel image replication for better NUMA-locality), it may map to some other physical address.

	--david