[PATCH] arm64: remove page granularity limitation from KFENCE

[Liu Shixin]

Currently if KFENCE is enabled in arm64, the entire linear map will be
mapped at page granularity which seems overkilled. Actually only the
kfence pool requires to be mapped at page granularity. We can remove the
restriction from KFENCE and force the linear mapping of the kfence pool
at page granularity later in arch_kfence_init_pool().

Signed-off-by: Liu Shixin <liush...@huawei.com>
---
arch/arm64/include/asm/kfence.h | 69 ++++++++++++++++++++++++++++++++-
arch/arm64/mm/mmu.c | 4 +-
2 files changed, 70 insertions(+), 3 deletions(-)

diff --git a/arch/arm64/include/asm/kfence.h b/arch/arm64/include/asm/kfence.h
index aa855c6a0ae6..bee101eced0b 100644
--- a/arch/arm64/include/asm/kfence.h
+++ b/arch/arm64/include/asm/kfence.h
@@ -8,9 +8,76 @@
#ifndef __ASM_KFENCE_H
#define __ASM_KFENCE_H

+#include <linux/kfence.h>
#include <asm/set_memory.h>
+#include <asm/pgalloc.h>

-static inline bool arch_kfence_init_pool(void) { return true; }
+static inline int split_pud_page(pud_t *pud, unsigned long addr)
+{
+ int i;
+ pmd_t *pmd = pmd_alloc_one(&init_mm, addr);
+ unsigned long pfn = PFN_DOWN(__pa(addr));
+
+ if (!pmd)
+ return -ENOMEM;
+
+ for (i = 0; i < PTRS_PER_PMD; i++)
+ set_pmd(pmd + i, pmd_mkhuge(pfn_pmd(pfn + i * PTRS_PER_PTE, PAGE_KERNEL)));
+
+ smp_wmb(); /* See comment in __pte_alloc */
+ pud_populate(&init_mm, pud, pmd);
+ flush_tlb_kernel_range(addr, addr + PUD_SIZE);
+ return 0;
+}
+
+static inline int split_pmd_page(pmd_t *pmd, unsigned long addr)
+{
+ int i;
+ pte_t *pte = pte_alloc_one_kernel(&init_mm);
+ unsigned long pfn = PFN_DOWN(__pa(addr));
+
+ if (!pte)
+ return -ENOMEM;
+
+ for (i = 0; i < PTRS_PER_PTE; i++)
+ set_pte(pte + i, pfn_pte(pfn + i, PAGE_KERNEL));
+
+ smp_wmb(); /* See comment in __pte_alloc */
+ pmd_populate_kernel(&init_mm, pmd, pte);
+
+ flush_tlb_kernel_range(addr, addr + PMD_SIZE);
+ return 0;
+}
+
+static inline bool arch_kfence_init_pool(void)
+{
+ unsigned long addr;
+ pgd_t *pgd;
+ p4d_t *p4d;
+ pud_t *pud;
+ pmd_t *pmd;
+
+ for (addr = (unsigned long)__kfence_pool; is_kfence_address((void *)addr);
+ addr += PAGE_SIZE) {
+ pgd = pgd_offset(&init_mm, addr);
+ if (pgd_leaf(*pgd))
+ return false;
+ p4d = p4d_offset(pgd, addr);
+ if (p4d_leaf(*p4d))
+ return false;
+ pud = pud_offset(p4d, addr);
+ if (pud_leaf(*pud)) {
+ if (split_pud_page(pud, addr & PUD_MASK))
+ return false;
+ }
+ pmd = pmd_offset(pud, addr);
+ if (pmd_leaf(*pmd)) {
+ if (split_pmd_page(pmd, addr & PMD_MASK))
+ return false;
+ }
+ }
+ return true;
+}

static inline bool kfence_protect_page(unsigned long addr, bool protect)
{
diff --git a/arch/arm64/mm/mmu.c b/arch/arm64/mm/mmu.c
index cfd9deb347c3..b2c79ccfb1c5 100644
--- a/arch/arm64/mm/mmu.c
+++ b/arch/arm64/mm/mmu.c
@@ -516,7 +516,7 @@ static void __init map_mem(pgd_t *pgdp)
*/
BUILD_BUG_ON(pgd_index(direct_map_end - 1) == pgd_index(direct_map_end));

- if (can_set_direct_map() || crash_mem_map || IS_ENABLED(CONFIG_KFENCE))
+ if (can_set_direct_map() || crash_mem_map)
flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;

/*
@@ -1485,7 +1485,7 @@ int arch_add_memory(int nid, u64 start, u64 size,
* KFENCE requires linear map to be mapped at page granularity, so that
* it is possible to protect/unprotect single pages in the KFENCE pool.
*/
- if (can_set_direct_map() || IS_ENABLED(CONFIG_KFENCE))
+ if (can_set_direct_map())
flags |= NO_BLOCK_MAPPINGS | NO_CONT_MAPPINGS;

__create_pgd_mapping(swapper_pg_dir, start, __phys_to_virt(start),
--
2.18.0.huawei.25

[Alexander Potapenko]

On Sat, Sep 18, 2021 at 10:10 AM Liu Shixin <liush...@huawei.com> wrote:
>
> Currently if KFENCE is enabled in arm64, the entire linear map will be
> mapped at page granularity which seems overkilled. Actually only the
> kfence pool requires to be mapped at page granularity. We can remove the
> restriction from KFENCE and force the linear mapping of the kfence pool
> at page granularity later in arch_kfence_init_pool().

There was a previous patch by Jisheng Zhang intended to remove this
requirement: https://lore.kernel.org/linux-arm-kernel/20210524180...@xhacker.debian/
Which of the two is more preferable?

> --
> You received this message because you are subscribed to the Google Groups "kasan-dev" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to kasan-dev+...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/kasan-dev/20210918083849.2696287-1-liushixin2%40huawei.com.



--
Alexander Potapenko
Software Engineer

Google Germany GmbH
Erika-Mann-Straße, 33
80636 München

Geschäftsführer: Paul Manicle, Halimah DeLaine Prado
Registergericht und -nummer: Hamburg, HRB 86891
Sitz der Gesellschaft: Hamburg

[Mark Rutland]

On Sat, Sep 18, 2021 at 04:38:49PM +0800, Liu Shixin wrote:
> Currently if KFENCE is enabled in arm64, the entire linear map will be
> mapped at page granularity which seems overkilled. Actually only the
> kfence pool requires to be mapped at page granularity. We can remove the
> restriction from KFENCE and force the linear mapping of the kfence pool
> at page granularity later in arch_kfence_init_pool().
>
> Signed-off-by: Liu Shixin <liush...@huawei.com>

We'd made this apply to the entire linear map because it was simpler to
do so, there are other reasons to want the linear map at page
granularity (e.g. RODATA_FULL), and is also the default behaviour (since
RODATA_FULL_DEFAULT_ENABLED is `default y`).

We also avoid live changes from block<->table mappings, since the
archtitecture gives us very weak guarantees there and generally requires
a Break-Before-Make sequence (though IIRC this was tightened up
somewhat, so maybe going one way is supposed to work). Unless it's
really necessary, I'd rather not split these block mappings while
they're live.

The bigger question is does this actually matter in practice? I
understand that in theory this can result in better TLB usage, but does
this actually affect a workload in a meaningful way? Without numbers,
I'd rather leave this as-is so that we're not adding complexity and an
ongoing maintenance burden.

Thanks,
Mark.

[Liu Shixin]

On 2021/9/18 19:50, Alexander Potapenko wrote:

On Sat, Sep 18, 2021 at 10:10 AM Liu Shixin <liush...@huawei.com> wrote:

Currently if KFENCE is enabled in arm64, the entire linear map will be
mapped at page granularity which seems overkilled. Actually only the
kfence pool requires to be mapped at page granularity. We can remove the
restriction from KFENCE and force the linear mapping of the kfence pool
at page granularity later in arch_kfence_init_pool().

There was a previous patch by Jisheng Zhang intended to remove this
requirement: https://lore.kernel.org/linux-arm-kernel/20210524180...@xhacker.debian/
Which of the two is more preferable?

The previous patch by Jisheng Zhang guarantees kfence pool to be mapped at
page granularity by allocating KFENCE pool before paging_init(), and then map it
at page granularity during map_mem().

The previous patch has a problem: Even If kfence is disabled in cmdline, kfence_pool
is still allocated, which is a waste.

thanks,

[Liu Shixin]

On 2021/9/18 19:50, Alexander Potapenko wrote:

On Sat, Sep 18, 2021 at 10:10 AM Liu Shixin <liush...@huawei.com> wrote:

Currently if KFENCE is enabled in arm64, the entire linear map will be
mapped at page granularity which seems overkilled. Actually only the
kfence pool requires to be mapped at page granularity. We can remove the
restriction from KFENCE and force the linear mapping of the kfence pool
at page granularity later in arch_kfence_init_pool().

There was a previous patch by Jisheng Zhang intended to remove this
requirement: 
Which of the two is more preferable?

The previous patch by Jisheng Zhang guarantees kfence pool to be mapped at

page granularity by allocating KFENCE pool before paging_init(), and then mapping it

at page granularity during map_mem().

The previous patch has a problem: Even If kfence is disabled in cmdline, kfence_pool

is still allocated, which is a waste of memory.

thanks,

[Liu Shixin]

The previous patch by Jisheng Zhang guarantees kfence pool to be mapped at

page granularity by allocating KFENCE pool before paging_init(), and then map it

at page granularity during map_mem().

The previous patch has a problem: Even If kfence is disabled in cmdline, kfence_pool

is still allocated, which is a waste.

thanks,

[Liu Shixin]

On 2021/9/18 19:50, Alexander Potapenko wrote:

> On Sat, Sep 18, 2021 at 10:10 AM Liu Shixin <liush...@huawei.com> wrote:
>> Currently if KFENCE is enabled in arm64, the entire linear map will be
>> mapped at page granularity which seems overkilled. Actually only the
>> kfence pool requires to be mapped at page granularity. We can remove the
>> restriction from KFENCE and force the linear mapping of the kfence pool
>> at page granularity later in arch_kfence_init_pool().
> There was a previous patch by Jisheng Zhang intended to remove this
> requirement: https://lore.kernel.org/linux-arm-kernel/20210524180...@xhacker.debian/
> Which of the two is more preferable?

The previous patch by Jisheng Zhang guarantees kfence pool to be mapped at
page granularity by allocating kfence pool before paging_init(), and then map it

at page granularity during map_mem().

The previous patch has a problem: Even if kfence is disabled in cmdline, kfence pool
is still allocated, which is a waste of memory.

I'm sorry for sending it repeatly, and I have no idea how to limit the email format
to TEXT/PLAIN.

thanks.