diff mbox

[3/9] cpu: Unconditionally cleanup TLBs on P9 in opal_reinit_cpus()

Message ID 20170707210846.24825-3-benh@kernel.crashing.org
State Accepted
Headers show

Commit Message

Benjamin Herrenschmidt July 7, 2017, 9:08 p.m. UTC 
This can work around problems where Linux fails to properly
cleanup part or all of the TLB on kexec.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
---
 core/cpu.c | 13 +++++++++++--
 1 file changed, 11 insertions(+), 2 deletions(-)

Comments

Stewart Smith July 12, 2017, 3:19 a.m. UTC | #1 
Benjamin Herrenschmidt <benh@kernel.crashing.org> writes:
> This can work around problems where Linux fails to properly
> cleanup part or all of the TLB on kexec.
>
> Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> ---
>  core/cpu.c | 13 +++++++++++--
>  1 file changed, 11 insertions(+), 2 deletions(-)

merged to master as of 537b7a705df795d888852b2cb4fbe535b4d534f4
diff mbox

Patch

diff --git a/core/cpu.c b/core/cpu.c
index a0c395f..6c00bca 100644
--- a/core/cpu.c
+++ b/core/cpu.c
@@ -1191,12 +1191,21 @@  static int64_t opal_reinit_cpus(uint64_t flags)
 			else
 				req.clr_bits |= SPR_HID0_POWER9_RADIX;
 
-			cleanup_global_tlb();
 			current_radix_mode = radix;
 		}
 	}
 
-	/* Apply HID bits changes if any */
+	/* Cleanup the TLB. We do that unconditionally, this works
+	 * around issues where OSes fail to invalidate the PWC in Radix
+	 * mode for example. This only works on P9 and later, but we
+	 * also know we don't have a problem with Linux cleanups on
+	 * P8 so this isn't a problem. If we wanted to cleanup the
+	 * TLB on P8 as well, we'd have to use jobs to do it locally
+	 * on each CPU.
+	 */
+	 cleanup_global_tlb();
+
+	 /* Apply HID bits changes if any */
 	if (req.set_bits || req.clr_bits)
 		cpu_change_all_hid0(&req);