[PATCH v2 10/11] x86,tlb: really leave mm on shootdown

From: Rik van Riel
Date: Sun Jul 29 2018 - 15:55:15 EST


On Sat, 28 Jul 2018 21:21:17 -0700
Andy Lutomirski <luto@xxxxxxxxxx> wrote:

> On Sat, Jul 28, 2018 at 2:53 PM, Rik van Riel <riel@xxxxxxxxxxx> wrote:
> > Conditionally skip lazy TLB mm refcounting. When an architecture has
> > CONFIG_ARCH_NO_ACTIVE_MM_REFCOUNTING enabled, an mm that is used in
> > lazy TLB mode anywhere will get shot down from exit_mmap, and there
> > in no need to incur the cache line bouncing overhead of refcounting
> > a lazy TLB mm.
>
> Unless I've misunderstood something, this patch results in idle tasks
> whose active_mm has been freed still having active_mm pointing at
> freed memory.

Below (plus the next email) should fix the bug you pointed
out, in a somewhat non-invasive way. Patches have survived
a few simple tests on my test system, I have not thrown a
full load at them yet.

I would like to save the full rewrite to remove ->active_mm
for a later series, because this is already as much churn
as I am comfortable with for this code :)

---8<---

Author: Rik van Riel <riel@xxxxxxxxxxx>
Subject: [PATCH 10/11] x86,tlb: really leave mm on shootdown

When getting an mm shot down from under us in lazy TLB mode, don't
just switch the TLB over to the init_mm page tables, but really drop
our references to the lazy TLB mm.

This allows for faster (instant) freeing of a lazy TLB mm, which is
a precondition to getting rid of the refcounting of mms in lazy TLB mode.

Signed-off-by: Rik van Riel <riel@xxxxxxxxxxx>
---
arch/x86/mm/tlb.c | 4 ++++
1 file changed, 4 insertions(+)

diff --git a/arch/x86/mm/tlb.c b/arch/x86/mm/tlb.c
index 7b1add904396..425cb9fa2640 100644
--- a/arch/x86/mm/tlb.c
+++ b/arch/x86/mm/tlb.c
@@ -140,6 +140,8 @@ void leave_mm(void *dummy)
WARN_ON(!this_cpu_read(cpu_tlbstate.is_lazy));

switch_mm(NULL, &init_mm, NULL);
+ current->active_mm = &init_mm;
+ mmdrop(loaded_mm);
}
EXPORT_SYMBOL_GPL(leave_mm);

@@ -483,6 +485,8 @@ static void flush_tlb_func_common(const struct flush_tlb_info *f,
* IPIs to lazy TLB mode CPUs.
*/
switch_mm_irqs_off(NULL, &init_mm, NULL);
+ current->active_mm = &init_mm;
+ mmdrop(loaded_mm);
return;
}

--
2.14.4