[PATCH v9] powerpc/mm: Only read faulting instruction when necessary in do_page_fault()
Christophe Leroy
christophe.leroy at c-s.fr
Wed May 23 18:53:22 AEST 2018
Commit a7a9dcd882a67 ("powerpc: Avoid taking a data miss on every
userspace instruction miss") has shown that limiting the read of
faulting instruction to likely cases improves performance.
This patch goes further into this direction by limiting the read
of the faulting instruction to the only cases where it is likely
needed.
On an MPC885, with the same benchmark app as in the commit referred
above, we see a reduction of about 3900 dTLB misses (approx 3%):
Before the patch:
Performance counter stats for './fault 500' (10 runs):
683033312 cpu-cycles ( +- 0.03% )
134538 dTLB-load-misses ( +- 0.03% )
46099 iTLB-load-misses ( +- 0.02% )
19681 faults ( +- 0.02% )
5.389747878 seconds time elapsed ( +- 0.06% )
With the patch:
Performance counter stats for './fault 500' (10 runs):
682112862 cpu-cycles ( +- 0.03% )
130619 dTLB-load-misses ( +- 0.03% )
46073 iTLB-load-misses ( +- 0.05% )
19681 faults ( +- 0.01% )
5.381342641 seconds time elapsed ( +- 0.07% )
The proper work of the huge stack expansion was tested with the
following app:
int main(int argc, char **argv)
{
char buf[1024 * 1025];
sprintf(buf, "Hello world !\n");
printf(buf);
exit(0);
}
Signed-off-by: Christophe Leroy <christophe.leroy at c-s.fr>
---
v9: Handling the semaphore drop and the retry directly in __do_page_fault(),
using fault_in_pages_readable() then perform a full retry, as suggested by Nicholas
v8: Back to a single patch as it now makes no sense to split the first part in two. The third patch has no
dependencies with the ones before, so it will be resend independantly. As suggested by Nicholas, the
patch now does the get_user() stuff inside bad_stack_expansion(), that's a mid way between v5 and v7.
v7: Following comment from Nicholas on v6 on possibility of the page getting removed from the pagetables
between the fault and the read, I have reworked the patch in order to do the get_user() in
__do_page_fault() directly in order to reduce complexity compared to version v5
v6: Rebased on latest powerpc/merge branch ; Using __get_user_inatomic() instead of get_user() in order
to move it inside the semaphored area. That removes all the complexity of the patch.
v5: Reworked to fit after Benh do_fault improvement and rebased on top of powerpc/merge (65152902e43fef)
v4: Rebased on top of powerpc/next (f718d426d7e42e) and doing access_ok() verification before __get_user_xxx()
v3: Do a first try with pagefault disabled before releasing the semaphore
v2: Changes 'if (cond1) if (cond2)' by 'if (cond1 && cond2)'
arch/powerpc/mm/fault.c | 49 +++++++++++++++++++++++++++++++++----------------
1 file changed, 33 insertions(+), 16 deletions(-)
diff --git a/arch/powerpc/mm/fault.c b/arch/powerpc/mm/fault.c
index 0c99f9b45e8f..8e3f440a5b75 100644
--- a/arch/powerpc/mm/fault.c
+++ b/arch/powerpc/mm/fault.c
@@ -66,15 +66,11 @@ static inline bool notify_page_fault(struct pt_regs *regs)
}
/*
- * Check whether the instruction at regs->nip is a store using
+ * Check whether the instruction inst is a store using
* an update addressing form which will update r1.
*/
-static bool store_updates_sp(struct pt_regs *regs)
+static bool store_updates_sp(unsigned int inst)
{
- unsigned int inst;
-
- if (get_user(inst, (unsigned int __user *)regs->nip))
- return false;
/* check for 1 in the rA field */
if (((inst >> 16) & 0x1f) != 1)
return false;
@@ -234,8 +230,8 @@ static bool bad_kernel_fault(bool is_exec, unsigned long error_code,
}
static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
- struct vm_area_struct *vma,
- bool store_update_sp)
+ struct vm_area_struct *vma, unsigned int flags,
+ bool *must_retry)
{
/*
* N.B. The POWER/Open ABI allows programs to access up to
@@ -247,6 +243,7 @@ static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
* expand to 1MB without further checks.
*/
if (address + 0x100000 < vma->vm_end) {
+ unsigned int __user *nip = (unsigned int __user *)regs->nip;
/* get user regs even if this fault is in kernel mode */
struct pt_regs *uregs = current->thread.regs;
if (uregs == NULL)
@@ -264,8 +261,22 @@ static bool bad_stack_expansion(struct pt_regs *regs, unsigned long address,
* between the last mapped region and the stack will
* expand the stack rather than segfaulting.
*/
- if (address + 2048 < uregs->gpr[1] && !store_update_sp)
- return true;
+ if (address + 2048 >= uregs->gpr[1])
+ return false;
+
+ if ((flags & FAULT_FLAG_WRITE) && (flags & FAULT_FLAG_USER) &&
+ access_ok(VERIFY_READ, nip, sizeof(*nip))) {
+ unsigned int inst;
+ int res;
+
+ pagefault_disable();
+ res = __get_user_inatomic(inst, nip);
+ pagefault_enable();
+ if (!res)
+ return !store_updates_sp(inst);
+ *must_retry = true;
+ }
+ return true;
}
return false;
}
@@ -403,7 +414,7 @@ static int __do_page_fault(struct pt_regs *regs, unsigned long address,
int is_user = user_mode(regs);
int is_write = page_fault_is_write(error_code);
int fault, major = 0;
- bool store_update_sp = false;
+ bool must_retry = false;
if (notify_page_fault(regs))
return 0;
@@ -454,9 +465,6 @@ static int __do_page_fault(struct pt_regs *regs, unsigned long address,
* can result in fault, which will cause a deadlock when called with
* mmap_sem held
*/
- if (is_write && is_user)
- store_update_sp = store_updates_sp(regs);
-
if (is_user)
flags |= FAULT_FLAG_USER;
if (is_write)
@@ -503,8 +511,17 @@ static int __do_page_fault(struct pt_regs *regs, unsigned long address,
return bad_area(regs, address);
/* The stack is being expanded, check if it's valid */
- if (unlikely(bad_stack_expansion(regs, address, vma, store_update_sp)))
- return bad_area(regs, address);
+ if (unlikely(bad_stack_expansion(regs, address, vma, flags,
+ &must_retry))) {
+ if (!must_retry)
+ return bad_area(regs, address);
+
+ up_read(&mm->mmap_sem);
+ if (fault_in_pages_readable((const char __user *)regs->nip,
+ sizeof(unsigned int)))
+ return bad_area_nosemaphore(regs, address);
+ goto retry;
+ }
/* Try to expand it */
if (unlikely(expand_stack(vma, address)))
--
2.13.3
More information about the Linuxppc-dev
mailing list