Skip to contentSkip to conversation

vt.c in 2.5.65-ac1

DISCUSSION Petr Vandrovec 2 participants 3 messages Mar 21, 2003 – Mar 22, 2003 Char and misc drivers

lore
Share & tools
lore
#0 lore raw
Quoted 3 lines · Osamu Tomita

I have a aquestion about patch in patch-2.5.65-ac1 for vt.c. Here is a extracted patch from patch-2.5.65-ac1.
I think it's no need for 2.5.65.

There should be none of these two resize_screen calls. If you'll resize non-foreground VT, they'll trigger fbcon_resize with con != visible_con, resizing your display even if they should not.

Only the "if (IS_VISIBLE) err = resize_screen(...);" resize should be there (AFAIK), if con_resize follows other con_* APIs: call it only if con is visible, like it is done with putcs and others.
Petr Vandrovec [email protected]

Quoted 28 lines
drivers/char/vt.c 2003-03-18 16:58:38.000000000 +0000 renamed from drivers/char/vt.c 2003-03-18 16:46:47.000000000 +0000 +6 -6
@@ -732,6 +732,12 @@
    if (new_cols == video_num_columns && new_rows == video_num_lines)
        return 0;
 
+   err = resize_screen(currcons, new_cols, new_rows);
+   if (err) {
+       kfree(newscreen);
+       return err;
+   }
+
    newscreen = (unsigned short *) kmalloc(new_screen_size, GFP_USER);
    if (!newscreen)
        return -ENOMEM;
@@ -746,12 +752,6 @@
    video_size_row = new_row_size;
    screenbuf_size = new_screen_size;
 
-   err = resize_screen(currcons, new_cols, new_rows);
-   if (err) {
-       kfree(newscreen);
-       return err;
-   }
-
    rlth = min(old_row_size, new_row_size);
    rrem = new_row_size - rlth;
    old_origin = origin;
#1 lore raw
Quoted 35 lines · Petr Vandrovec
Quoted 3 lines · Osamu Tomita

I have a aquestion about patch in patch-2.5.65-ac1 for vt.c. Here is a extracted patch from patch-2.5.65-ac1.
I think it's no need for 2.5.65.

There should be none of these two resize_screen calls. If you'll resize non-foreground VT, they'll trigger fbcon_resize with con != visible_con, resizing your display even if they should not.

Only the "if (IS_VISIBLE) err = resize_screen(...);" resize should be there (AFAIK), if con_resize follows other con_* APIs: call it only if con is visible, like it is done with putcs and others.
Petr Vandrovec [email protected]

Quoted 19 lines
drivers/char/vt.c 2003-03-18 16:58:38.000000000 +0000 renamed from drivers/char/vt.c 2003-03-18 16:46:47.000000000 +0000 +6 -0
@@ -732,6 +732,12 @@
    if (new_cols == video_num_columns && new_rows == video_num_lines)
        return 0;
 
+   err = resize_screen(currcons, new_cols, new_rows);
+   if (err) {
+       kfree(newscreen);
+       return err;
+   }
+
    newscreen = (unsigned short *) kmalloc(new_screen_size, GFP_USER);
    if (!newscreen)
        return -ENOMEM;
@@ -746,12 +752,6 @@
    video_size_row = new_row_size;
    screenbuf_size = new_screen_size;
 
Quoted 9 lines

drivers/char/vt.c 2003-03-18 16:58:38.000000000 +0000 line 757

Quoted 9 lines
drivers/char/vt.c 2003-03-18 16:58:38.000000000 +0000 renamed from drivers/char/vt.c 2003-03-18 16:46:47.000000000 +0000 +0 -6
-   err = resize_screen(currcons, new_cols, new_rows);
-   if (err) {
-       kfree(newscreen);
-       return err;
-   }
-
    rlth = min(old_row_size, new_row_size);
    rrem = new_row_size - rlth;
    old_origin = origin;

I understand. But if resize_screen() failed this attempt to kfree before kmalloc. Is there case resize_screen success but kmalloc fail?
How about patch bellow.

drivers/char/vt.c renamed from drivers/char/vt.c.orig +6 -6
@@ -736,6 +736,12 @@
 	if (!newscreen)
 		return -ENOMEM;
 
+	err = resize_screen(currcons, new_cols, new_rows);
+	if (err) {
+		kfree(newscreen);
+		return err;
+	}
+
 	old_rows = video_num_lines;
 	old_cols = video_num_columns;
 	old_row_size = video_size_row;
@@ -746,12 +752,6 @@
 	video_size_row = new_row_size;
 	screenbuf_size = new_screen_size;
 
-	err = resize_screen(currcons, new_cols, new_rows);
-	if (err) {
-		kfree(newscreen);
-		return err;
-	}
-
 	rlth = min(old_row_size, new_row_size);
 	rrem = new_row_size - rlth;
 	old_origin = origin;

Regards,
Osamu Tomita

#2 lore raw
Quoted 11 lines · Osamu Tomita
Quoted 10 lines · Petr Vandrovec
Quoted 3 lines · Osamu Tomita

I have a aquestion about patch in patch-2.5.65-ac1 for vt.c. Here is a extracted patch from patch-2.5.65-ac1.
I think it's no need for 2.5.65.

There should be none of these two resize_screen calls. If you'll resize non-foreground VT, they'll trigger fbcon_resize with con != visible_con, resizing your display even if they should not.

Quoted 8 lines
Quoted 4 lines

Only the "if (IS_VISIBLE) err = resize_screen(...);" resize should be there (AFAIK), if con_resize follows other con_* APIs: call it only if con is visible, like it is done with putcs and others.

Quoted 1 line
    old_origin = origin;

I understand. But if resize_screen() failed this attempt to kfree before kmalloc. Is there case resize_screen success but kmalloc fail?
How about patch bellow.

As I said: currently there is one unconditional call to resize_screen and one conditional (depending on IS_VISIBLE). Unconditional one should not be moved around, but removed completely. Look at fbcon_resize (only con_resize user) implementation, it will do wrong things if called with non-visible currcons. I believe that James has this correct in his latest patches.
Petr Vandrovec [email protected]

Quoted 33 lines
drivers/char/vt.c renamed from drivers/char/vt.c.orig +6 -6
@@ -736,6 +736,12 @@
 	if (!newscreen)
 		return -ENOMEM;
 
+	err = resize_screen(currcons, new_cols, new_rows);
+	if (err) {
+		kfree(newscreen);
+		return err;
+	}
+
 	old_rows = video_num_lines;
 	old_cols = video_num_columns;
 	old_row_size = video_size_row;
@@ -746,12 +752,6 @@
 	video_size_row = new_row_size;
 	screenbuf_size = new_screen_size;
 
-	err = resize_screen(currcons, new_cols, new_rows);
-	if (err) {
-		kfree(newscreen);
-		return err;
-	}
-
 	rlth = min(old_row_size, new_row_size);
 	rrem = new_row_size - rlth;
 	old_origin = origin;

Regards,
Osamu Tomita