[PATCH] vgacon.c: add cond reschedule points in vgacon_do_font_op

[PATCH] vgacon.c: add cond reschedule points in vgacon_do_font_op

[Marcelo Tosatti]

Booting a 64-vcpu KVM guest, with CONFIG_PREEMPT_VOLUNTARY,
can result in a soft lockup:

BUG: soft lockup - CPU#41 stuck for 67s! [setfont:1505]
 RIP: 0010:[<ffffffff812c48da>]
[<ffffffff812c48da>] vgacon_do_font_op.clone.0+0x1ba/0x550

This is due to the 8192 (cmapsz) IO operations taking longer than expected
due to lock contention in QEMU.

Add conditional resched points in between writes allowing other tasks to
execute.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>

diff --git a/drivers/video/console/vgacon.c b/drivers/video/console/vgacon.c
index 5855d17..0dd6d96 100644
--- a/drivers/video/console/vgacon.c
+++ b/drivers/video/console/vgacon.c
@@ -1124,11 +1124,15 @@ static int vgacon_do_font_op(struct vgastate *state,char *arg,int set,int ch512)
 
 	if (arg) {
 		if (set)
-			for (i = 0; i < cmapsz; i++)
+			for (i = 0; i < cmapsz; i++) {
 				vga_writeb(arg[i], charmap + i);
+				cond_resched();
+			}
 		else
-			for (i = 0; i < cmapsz; i++)
+			for (i = 0; i < cmapsz; i++) {
 				arg[i] = vga_readb(charmap + i);
+				cond_resched();
+			}
 
 		/*
 		 * In 512-character mode, the character map is not contiguous if
@@ -1139,11 +1143,15 @@ static int vgacon_do_font_op(struct vgastate *state,char *arg,int set,int ch512)
 			charmap += 2 * cmapsz;
 			arg += cmapsz;
 			if (set)
-				for (i = 0; i < cmapsz; i++)
+				for (i = 0; i < cmapsz; i++) {
 					vga_writeb(arg[i], charmap + i);
+					cond_resched();
+				}
 			else
-				for (i = 0; i < cmapsz; i++)
+				for (i = 0; i < cmapsz; i++) {
 					arg[i] = vga_readb(charmap + i);
+					cond_resched();
+				}
 		}
 	}
 

Re: [PATCH] vgacon.c: add cond reschedule points in vgacon_do_font_op

[Dave Airlie]

> can result in a soft lockup:
>
> BUG: soft lockup - CPU#41 stuck for 67s! [setfont:1505]
>  RIP: 0010:[<ffffffff812c48da>]
> [<ffffffff812c48da>] vgacon_do_font_op.clone.0+0x1ba/0x550
>
> This is due to the 8192 (cmapsz) IO operations taking longer than expected
> due to lock contention in QEMU.
>
> Add conditional resched points in between writes allowing other tasks to
> execute.

I've applied this to drm-next against my best judgement as it steers
me close to being the console maintainer.

Dave.