Re: [PATCHv2 2/3] arm64: Add support for ARCH_SUPPORTS_DEBUG_PAGEALLOC

From: Laura Abbott
Date: Mon Feb 01 2016 - 16:24:55 EST


On 02/01/2016 04:29 AM, Mark Rutland wrote:
Hi,

On Fri, Jan 29, 2016 at 03:46:57PM -0800, Laura Abbott wrote:

ARCH_SUPPORTS_DEBUG_PAGEALLOC provides a hook to map and unmap
pages for debugging purposes. This requires memory be mapped
with PAGE_SIZE mappings since breaking down larger mappings
at runtime will lead to TLB conflicts. Check if debug_pagealloc
is enabled at runtime and if so, map everyting with PAGE_SIZE
pages. Implement the functions to actually map/unmap the
pages at runtime.


Signed-off-by: Laura Abbott <labbott@xxxxxxxxxxxxxxxxx>

I tried to apply atop of the arm64 for-next/pgtable branch, but git
wasn't very happy about that -- which branch/patches is this based on?

I'm not sure if I'm missing something, have something I shouldn't, or if
my MTA is corrupting patches again...


Hmmm, I based it off of your arm64-pagetable-rework-20160125 tag and
Ard's patch for vmalloc and set_memory_* . The patches seem to apply
on the for-next/pgtable branch as well so I'm guessing you are missing
Ard's patch.

---
arch/arm64/Kconfig | 3 +++
arch/arm64/mm/mmu.c | 10 +++++++---
arch/arm64/mm/pageattr.c | 40 +++++++++++++++++++++++++++++++++-------
3 files changed, 43 insertions(+), 10 deletions(-)

diff --git a/arch/arm64/Kconfig b/arch/arm64/Kconfig
index 8cc6228..0f33218 100644
--- a/arch/arm64/Kconfig
+++ b/arch/arm64/Kconfig
@@ -537,6 +537,9 @@ config HOTPLUG_CPU
source kernel/Kconfig.preempt
source kernel/Kconfig.hz

+config ARCH_SUPPORTS_DEBUG_PAGEALLOC
+ def_bool y
+
config ARCH_HAS_HOLES_MEMORYMODEL
def_bool y if SPARSEMEM

diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
index 103ebc0..5abbd30 100644
--- a/arch/arm64/mm/mmu.c
+++ b/arch/arm64/mm/mmu.c
@@ -181,7 +181,8 @@ static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
do {
next = pmd_addr_end(addr, end);
/* try section mapping first */
- if (((addr | next | phys) & ~SECTION_MASK) == 0) {
+ if (((addr | next | phys) & ~SECTION_MASK) == 0 &&
+ (!debug_pagealloc_enabled() || !pgtable_alloc)) {
pmd_t old_pmd =*pmd;
set_pmd(pmd, __pmd(phys |
pgprot_val(mk_sect_prot(prot))));
@@ -208,8 +209,11 @@ static void alloc_init_pmd(pud_t *pud, unsigned long addr, unsigned long end,
}

static inline bool use_1G_block(unsigned long addr, unsigned long next,
- unsigned long phys)
+ unsigned long phys, phys_addr_t (*pgtable_alloc)(void))
{
+ if (pgtable_alloc && debug_pagealloc_enabled())
+ return false;
+
if (PAGE_SHIFT != 12)
return false;

@@ -241,7 +245,7 @@ static void alloc_init_pud(pgd_t *pgd, unsigned long addr, unsigned long end,
/*
* For 4K granule only, attempt to put down a 1GB block
*/
- if (use_1G_block(addr, next, phys)) {
+ if (use_1G_block(addr, next, phys, pgtable_alloc)) {
pud_t old_pud = *pud;
set_pud(pud, __pud(phys |
pgprot_val(mk_sect_prot(prot))));

The checks for pgtable_alloc is to allow create_mapping_noalloc to use
sections regardless, right?

I got a little confused by that initially, but that seems to make sense.


Yeah, that was what I was going for. I went for the implication with pgtable_alloc
over adding a bool variable to create_mapping. I'll see if I can put a comment
in somewhere to clarify the intentions.
diff --git a/arch/arm64/mm/pageattr.c b/arch/arm64/mm/pageattr.c
index 1360a02..69a8a7d 100644
--- a/arch/arm64/mm/pageattr.c
+++ b/arch/arm64/mm/pageattr.c
@@ -38,7 +38,8 @@ static int change_page_range(pte_t *ptep, pgtable_t token, unsigned long addr,
}

static int change_memory_common(unsigned long addr, int numpages,
- pgprot_t set_mask, pgprot_t clear_mask)
+ pgprot_t set_mask, pgprot_t clear_mask,
+ bool ignore_vma_check)
{
unsigned long start = addr;
unsigned long size = PAGE_SIZE*numpages;
@@ -65,11 +66,14 @@ static int change_memory_common(unsigned long addr, int numpages,
*
* So check whether the [addr, addr + size) interval is entirely
* covered by precisely one VM area that has the VM_ALLOC flag set.
+ *
+ * The one exception to this is is we're forcing everything to be
+ * page mapped with DEBUG_PAGEALLOC
*/
area = find_vm_area((void *)addr);
- if (!area ||
+ if (!ignore_vma_check && (!area ||
end > (unsigned long)area->addr + area->size ||
- !(area->flags & VM_ALLOC))
+ !(area->flags & VM_ALLOC)))
return -EINVAL;

Perhaps we could pull out everything but the VMA checks into a static
__change_memory_common. Then we can call that directly in
__kernel_map_pages, and don't need to pass a confusing boolean
parameter.

Sounds fine.


Otherwise, this looks good to me.
Mark.


Thanks,
Laura

data.set_mask = set_mask;
@@ -86,21 +90,24 @@ int set_memory_ro(unsigned long addr, int numpages)
{
return change_memory_common(addr, numpages,
__pgprot(PTE_RDONLY),
- __pgprot(PTE_WRITE));
+ __pgprot(PTE_WRITE),
+ false);
}

int set_memory_rw(unsigned long addr, int numpages)
{
return change_memory_common(addr, numpages,
__pgprot(PTE_WRITE),
- __pgprot(PTE_RDONLY));
+ __pgprot(PTE_RDONLY),
+ false);
}

int set_memory_nx(unsigned long addr, int numpages)
{
return change_memory_common(addr, numpages,
__pgprot(PTE_PXN),
- __pgprot(0));
+ __pgprot(0),
+ false);
}
EXPORT_SYMBOL_GPL(set_memory_nx);

@@ -108,6 +115,25 @@ int set_memory_x(unsigned long addr, int numpages)
{
return change_memory_common(addr, numpages,
__pgprot(0),
- __pgprot(PTE_PXN));
+ __pgprot(PTE_PXN),
+ false);
}
EXPORT_SYMBOL_GPL(set_memory_x);
+
+#ifdef CONFIG_DEBUG_PAGEALLOC
+void __kernel_map_pages(struct page *page, int numpages, int enable)
+{
+ unsigned long addr = (unsigned long) page_address(page);
+
+ if (enable)
+ change_memory_common(addr, numpages,
+ __pgprot(PTE_VALID),
+ __pgprot(0),
+ true);
+ else
+ change_memory_common(addr, numpages,
+ __pgprot(0),
+ __pgprot(PTE_VALID),
+ true);
+}
+#endif
--
2.5.0