Skip to contentSkip to conversation

Video mode fixes in linux loader

PATCH In review Pavel Roskin 7 participants 30 messages Apr 6, 2009 – May 13, 2009

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Video mode fixes in linux loader

Hello!

This is an attempt to fix all issues with the video mode handling in the new Linux loader.

First of all, free_page() doesn't belong to grub_linux_unload(). The later function is called after the new kernel has been loaded, just before the boot. Thus it obliterates the data set up by the last "linux" command if the previous Linux boot failed.

Instead, free_page() should be called from allocate_pages() to reset not only real_mode_mem and prot_mode_mem, but also initrd_mem.

Next problem is that grub_linux_boot() can access linux_vesafb_modes with a wrong index. I believe we should threat modes that don't fit the array as text modes.

I introduced GRUB_LINUX_VID_MODE_VESA_START for the first VESA mode we support. Also, I introduced a macro ARRAY_SIZE to calculate the array size.

The default VGA mode is now GRUB_LINUX_VID_MODE_NORMAL, not the mode of the kernel we tried to load before. Finally, "vga=ask" is now recognized.

ChangeLog:

* include/grub/misc.h (ARRAY_SIZE): New macro.
* include/grub/i386/linux.h (GRUB_LINUX_VID_MODE_VESA_START):
New macro.
* loader/i386/linux.c (allocate_pages): Use free_pages(). (grub_linux_unload): Don't use free_pages(). (grub_linux_boot): Prevent accessing linux_vesafb_modes with a wrong index. Treat all other modes as text modes. (grub_cmd_linux): Initialize vid_mode unconditionally to GRUB_LINUX_VID_MODE_NORMAL. Recognize and support "vga=ask".

grub/misc.h +1 -0
Index: include/grub/misc.h
===================================================================
@@ -26,6 +26,7 @@
 #include <grub/err.h>
 
 #define ALIGN_UP(addr, align) (((grub_uint64_t)addr + align - 1) & ~(align - 1))
+#define ARRAY_SIZE(array) (sizeof (array) / sizeof (array[0]))
 
 #define grub_dprintf(condition, fmt, args...) grub_real_dprintf(__FILE__, __LINE__, condition, fmt, ## args);
 /* XXX: If grub_memmove is too slow, we must implement grub_memcpy.  */
grub/i386/linux.h +1 -0
Index: include/grub/i386/linux.h
===================================================================
@@ -38,6 +38,7 @@
 #define GRUB_LINUX_VID_MODE_NORMAL	0xFFFF
 #define GRUB_LINUX_VID_MODE_EXTENDED	0xFFFE
 #define GRUB_LINUX_VID_MODE_ASK		0xFFFD
+#define GRUB_LINUX_VID_MODE_VESA_START	0x0301
 
 #define GRUB_LINUX_SETUP_MOVE_SIZE	0x9100
 #define GRUB_LINUX_CL_MAGIC		0xA33F
i386/linux.c +8 -5
Index: loader/i386/linux.c
===================================================================
@@ -206,8 +206,7 @@ allocate_pages (grub_size_t prot_size)
   prot_mode_pages = (prot_size >> 12);
   
   /* Initialize the memory pointers with NULL for convenience.  */
-  real_mode_mem = 0;
-  prot_mode_mem = 0;
+  free_pages();
   
   /* FIXME: Should request low memory from the heap when this feature is
      implemented.  */
@@ -332,7 +331,9 @@ grub_linux_boot (void)
   
   params = real_mode_mem;
 
-  if (vid_mode == GRUB_LINUX_VID_MODE_NORMAL || vid_mode == GRUB_LINUX_VID_MODE_EXTENDED)
+  if (vid_mode < GRUB_LINUX_VID_MODE_VESA_START ||
+      vid_mode >= GRUB_LINUX_VID_MODE_VESA_START +
+		  ARRAY_SIZE (linux_vesafb_modes))
     grub_video_restore ();
   else if (vid_mode)
     {
@@ -340,7 +341,7 @@ grub_linux_boot (void)
       int depth, flags;
       
       flags = 0;
-      linux_mode = &linux_vesafb_modes[vid_mode - 0x301];
+      linux_mode = &linux_vesafb_modes[vid_mode - GRUB_LINUX_VID_MODE_VESA_START];
       depth = linux_mode->depth;
       
       /* If we have 8 or less bits, then assume that it is indexed color mode.  */
@@ -443,7 +444,6 @@ grub_linux_boot (void)
 static grub_err_t
 grub_linux_unload (void)
 {
-  free_pages ();
   grub_dl_unref (my_mod);
   loaded = 0;
   return GRUB_ERR_NONE;
@@ -570,6 +570,7 @@ grub_cmd_linux (grub_command_t cmd __att
 	       (unsigned) real_size, (unsigned) prot_size);
 
   /* Detect explicitly specified memory size, if any.  */
+  vid_mode = GRUB_LINUX_VID_MODE_NORMAL;
   linux_mem_size = 0;
   for (i = 1; i < argc; i++)
     if (grub_memcmp (argv[i], "vga=", 4) == 0)
@@ -581,6 +582,8 @@ grub_cmd_linux (grub_command_t cmd __att
 	  vid_mode = GRUB_LINUX_VID_MODE_NORMAL;
 	else if (grub_strcmp (val, "ext") == 0)
 	  vid_mode = GRUB_LINUX_VID_MODE_EXTENDED;
+	else if (grub_strcmp (val, "ask") == 0)
+	  vid_mode = GRUB_LINUX_VID_MODE_ASK;
 	else
 	  vid_mode = (grub_uint16_t) grub_strtoul (val, 0, 0);
 
Signature · 2 lines
Regards,
Pavel Roskin
#1 lore raw
Quoted 66 lines · On Tuesday 07 April 2009 00:34:03 Pavel Roskin wrote:

Hello!

This is an attempt to fix all issues with the video mode handling in the new Linux loader.

First of all, free_page() doesn't belong to grub_linux_unload(). The later function is called after the new kernel has been loaded, just before the boot. Thus it obliterates the data set up by the last "linux" command if the previous Linux boot failed.

Instead, free_page() should be called from allocate_pages() to reset not only real_mode_mem and prot_mode_mem, but also initrd_mem.

Next problem is that grub_linux_boot() can access linux_vesafb_modes with a wrong index. I believe we should threat modes that don't fit the array as text modes.

I introduced GRUB_LINUX_VID_MODE_VESA_START for the first VESA mode we support. Also, I introduced a macro ARRAY_SIZE to calculate the array size.

The default VGA mode is now GRUB_LINUX_VID_MODE_NORMAL, not the mode of the kernel we tried to load before. Finally, "vga=ask" is now recognized.

ChangeLog:

* include/grub/misc.h (ARRAY_SIZE): New macro.
* include/grub/i386/linux.h (GRUB_LINUX_VID_MODE_VESA_START):
New macro.
* loader/i386/linux.c (allocate_pages): Use free_pages(). (grub_linux_unload): Don't use free_pages(). (grub_linux_boot): Prevent accessing linux_vesafb_modes with a wrong index. Treat all other modes as text modes. (grub_cmd_linux): Initialize vid_mode unconditionally to GRUB_LINUX_VID_MODE_NORMAL. Recognize and support "vga=ask".

Index: include/grub/misc.h
===================================================================

grub/misc.h +0 -1
@@ -26,6 +26,7 @@
 #include <grub/err.h>
 
 #define ALIGN_UP(addr, align) (((grub_uint64_t)addr + align - 1) & ~(align
- 1)) +#define ARRAY_SIZE(array) (sizeof (array) / sizeof (array[0]))
 
 #define grub_dprintf(condition, fmt, args...) grub_real_dprintf(__FILE__,

__LINE__, condition, fmt, ## args); /* XXX: If grub_memmove is too slow, we must implement grub_memcpy. */ Index: include/grub/i386/linux.h
===================================================================

grub/i386/linux.h +1 -0
@@ -38,6 +38,7 @@
 #define GRUB_LINUX_VID_MODE_NORMAL	0xFFFF
 #define GRUB_LINUX_VID_MODE_EXTENDED	0xFFFE
 #define GRUB_LINUX_VID_MODE_ASK		0xFFFD
+#define GRUB_LINUX_VID_MODE_VESA_START	0x0301
 
 #define GRUB_LINUX_SETUP_MOVE_SIZE	0x9100
 #define GRUB_LINUX_CL_MAGIC		0xA33F

Index: loader/i386/linux.c
===================================================================

i386/linux.c allocate_pages (grub_size_t prot_size) · line 209

Quoted 6 lines
i386/linux.c +1 -2
@@ -206,8 +206,7 @@ allocate_pages (grub_size_t prot_size)
   prot_mode_pages = (prot_size >> 12);
 
   /* Initialize the memory pointers with NULL for convenience.  */
-  real_mode_mem = 0;
-  prot_mode_mem = 0;
+  free_pages();

Please insert a space before the parenthesis.

Quoted 33 lines
i386/linux.c +0 -1
   /* FIXME: Should request low memory from the heap when this feature is
      implemented.  */
@@ -332,7 +331,9 @@ grub_linux_boot (void)
 
   params = real_mode_mem;
 
-  if (vid_mode == GRUB_LINUX_VID_MODE_NORMAL || vid_mode ==

GRUB_LINUX_VID_MODE_EXTENDED) + if (vid_mode <
GRUB_LINUX_VID_MODE_VESA_START ||

i386/linux.c +3 -1
+      vid_mode >= GRUB_LINUX_VID_MODE_VESA_START +
+		  ARRAY_SIZE (linux_vesafb_modes))
     grub_video_restore ();
   else if (vid_mode)
     {
@@ -340,7 +341,7 @@ grub_linux_boot (void)
       int depth, flags;
 
       flags = 0;
-      linux_mode = &linux_vesafb_modes[vid_mode - 0x301];
+      linux_mode = &linux_vesafb_modes[vid_mode -
 GRUB_LINUX_VID_MODE_VESA_START]; depth = linux_mode->depth;
 
       /* If we have 8 or less bits, then assume that it is indexed color

mode. */ @@ -443,7 +444,6 @@ grub_linux_boot (void)
static grub_err_t
grub_linux_unload (void)
{

i386/linux.c +0 -1
-  free_pages ();
   grub_dl_unref (my_mod);
   loaded = 0;
   return GRUB_ERR_NONE;

i386/linux.c grub_cmd_linux (grub_command_t cmd __att · line 572

Quoted 3 lines
i386/linux.c +0 -0
@@ -570,6 +570,7 @@ grub_cmd_linux (grub_command_t cmd __att
 	       (unsigned) real_size, (unsigned) prot_size);
 
   /* Detect explicitly specified memory size, if any.  */

This is not due to you, but the comment above is not synchronized with the code apparently. Can you fix it?

i386/linux.c grub_cmd_linux (grub_command_t cmd __att · line 588

Quoted 12 lines
i386/linux.c +3 -0
+  vid_mode = GRUB_LINUX_VID_MODE_NORMAL;
   linux_mem_size = 0;
   for (i = 1; i < argc; i++)
     if (grub_memcmp (argv[i], "vga=", 4) == 0)
@@ -581,6 +582,8 @@ grub_cmd_linux (grub_command_t cmd __att
 	  vid_mode = GRUB_LINUX_VID_MODE_NORMAL;
 	else if (grub_strcmp (val, "ext") == 0)
 	  vid_mode = GRUB_LINUX_VID_MODE_EXTENDED;
+	else if (grub_strcmp (val, "ask") == 0)
+	  vid_mode = GRUB_LINUX_VID_MODE_ASK;
 	else
 	  vid_mode = (grub_uint16_t) grub_strtoul (val, 0, 0);

Regards,
Okuji

#2 lore raw
Quoted 3 lines · Yoshinori K. Okuji
Quoted 1 line

+ free_pages();

Please insert a space before the parenthesis.

Done.

Quoted 2 lines

This is not due to you, but the comment above is not synchronized with the code apparently. Can you fix it?

Fixed and committed.

Signature · 2 lines
Regards,
Pavel Roskin
#3 lore raw

Does it actually work? I don't see the code which either handles GRUB_LINUX_VID_MODE_ASK or passes it to the kernel. Also it looks like vga= parameter is parsed by grub and grub passes only video mode values as resolution and color depth. Wouldn't it be better to move to a more modern video mode specification method? As gfxpayload variable previously discussed on that list

Quoted 10 lines · Pavel Roskin
Quoted 2 lines · Yoshinori K. Okuji
Quoted 1 line

+ free_pages();

Please insert a space before the parenthesis.

Done.

Quoted 1 line

This is not due to you, but the comment above is not synchronized with the code apparently. Can you fix it?

Fixed and committed.

Signature · 2 lines
Regards
Vladimir 'phcoder' Serbinenko
#4 lore raw
Quoted 1 line · phcoder

Does it actually work?

No. But at least it loads the kernel.

Quoted 2 lines

I don't see the code which either handles
GRUB_LINUX_VID_MODE_ASK or passes it to the kernel.

We could set params->vid_mode, but the mode setting in the kernel is done in the 16-bit code that we don't use.

Quoted 5 lines

Also it looks like
vga= parameter is parsed by grub and grub passes only video mode values as resolution and color depth. Wouldn't it be better to move to a more modern video mode specification method? As gfxpayload variable previously discussed on that list

OK, I just fixed the issue that was annoying me. Being a PC user, I don't really care about the graphical boot.

After touching that code, I believe the 32-bit Linux loader should not parse the vga option at all. It's irrelevant.

Instead, the loader should either load the kernel in the current mode (text of gfxterm) or it should restore the text mode. I believe not all kernels can start in graphical mode, so using the graphical mode should be enabled explicitly by some option.

If the platform doesn't have text mode, then obviously the current mode should be used.

I don't see any strong need to use different video modes in GRUB and in the kernel. If somebody needs that, the mode in GRUB can be changed before loading the kernel. Also, a kernel video driver can change the mode upon initialization using driver-specific parameters.

GRUB could implement video mode handling compatible with the Linux internal bootloader, including "vga=ask", but I'm not sure we need it.

Signature · 2 lines
Regards,
Pavel Roskin
#5 lore raw
Quoted 5 lines · Pavel Roskin
Quoted 2 lines

I don't see the code which either handles
GRUB_LINUX_VID_MODE_ASK or passes it to the kernel.

We could set params->vid_mode, but the mode setting in the kernel is done in the 16-bit code that we don't use.

Or we could do the actual asking like linux16 does

Quoted 10 lines
Quoted 3 lines

Also it looks like
vga= parameter is parsed by grub and grub passes only video mode values as resolution and color depth. Wouldn't it be better to move to a more modern video mode specification method? As gfxpayload variable previously discussed on that list

OK, I just fixed the issue that was annoying me. Being a PC user, I don't really care about the graphical boot.

After touching that code, I believe the 32-bit Linux loader should not parse the vga option at all. It's irrelevant.

This option may be kept for backward compatibility a possible fallback chain is:
gfxpayload->vga->previousvideo mode

Quoted 5 lines

Instead, the loader should either load the kernel in the current mode (text of gfxterm) or it should restore the text mode. I believe not all kernels can start in graphical mode, so using the graphical mode should be enabled explicitly by some option.

Quoted 8 lines

If the platform doesn't have text mode, then obviously the current mode should be used.

I don't see any strong need to use different video modes in GRUB and in the kernel. If somebody needs that, the mode in GRUB can be changed before loading the kernel. Also, a kernel video driver can change the mode upon initialization using driver-specific parameters.

Every OS has its specific needs. E.g. xnu needs to boot in video mode. Perhaps it's just me not knowing how to boot it in text mode but this limitation will have to be respected. And I don't feel like initialising
video console just to set a supported mode is a good idea.

Quoted 3 lines

GRUB could implement video mode handling compatible with the Linux internal bootloader, including "vga=ask", but I'm not sure we need it.

I don't think we need it either just no system should be rendered unbootable by this

Signature · 2 lines
Regards
Vladimir 'phcoder' Serbinenko
#6 lore raw
Quoted 7 lines · phcoder
Quoted 5 lines · Pavel Roskin
Quoted 2 lines

I don't see the code which either handles
GRUB_LINUX_VID_MODE_ASK or passes it to the kernel.

We could set params->vid_mode, but the mode setting in the kernel is done in the 16-bit code that we don't use.

Or we could do the actual asking like linux16 does

I don't see that in the code.

Anyway, gfxterm and graphical boot is not something I'm very interested in. It's just the breakage with vga=1 that made fix the code.

Signature · 2 lines
Regards,
Pavel Roskin
#7 lore raw
In reply to Pavel Roskin

Hi Pavel,

Quoted 4 lines · Pavel Roskin

Hello!

This is an attempt to fix all issues with the video mode handling in the new Linux loader.

Please, in general, when you modify code that has been actively worked on by someone, try to get that person involved. If I'm to busy to check the list at a given time, feel free to CC me or msg me on IRC. Some things in your patch were not right.

Quoted 7 lines

First of all, free_page() doesn't belong to grub_linux_unload(). The later function is called after the new kernel has been loaded, just before the boot. Thus it obliterates the data set up by the last "linux" command if the previous Linux boot failed.

Instead, free_page() should be called from allocate_pages() to reset not only real_mode_mem and prot_mode_mem, but also initrd_mem.

Ok.

Quoted 7 lines

Next problem is that grub_linux_boot() can access linux_vesafb_modes with a wrong index. I believe we should threat modes that don't fit the array as text modes.

I introduced GRUB_LINUX_VID_MODE_VESA_START for the first VESA mode we support. Also, I introduced a macro ARRAY_SIZE to calculate the array size.

Ok.

Quoted 2 lines

The default VGA mode is now GRUB_LINUX_VID_MODE_NORMAL, not the mode of the kernel we tried to load before.

Ok, BUT if we're already in vesa mode, and we know it works (since we're using it), there's no point in wasting time only to get a worse mode.

We should just make sure subsequent calls to "linux" command override the previous one.

Quoted 1 line

Finally, "vga=ask" is now recognized.

With the new loader, Linux' 16-bit entry code is no longer responsible for setting vesa modes, therefore vga=ask can't work. There's no point in recognizing it (except to warn the user).

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#8 lore raw
Quoted 8 lines · Robert Millan
Quoted 2 lines

The default VGA mode is now GRUB_LINUX_VID_MODE_NORMAL, not the mode of the kernel we tried to load before.

Ok, BUT if we're already in vesa mode, and we know it works (since we're using it), there's no point in wasting time only to get a worse mode.

We should just make sure subsequent calls to "linux" command override the previous one.

I think this should do it. Can you confirm this works for you?

(note: this patch is intentionally missing indentation, for readability)

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."

* loader/i386/linux.c (grub_linux_boot): Only go back to text mode when `vid_mode' wasn't initialized. Otherwise leave the current mode (whatever that is).
(grub_cmd_linux): Default to vid_mode = 0' rather than GRUB_LINUX_VID_MODE_NORMAL'.

i386/linux.c +9 -7
Index: loader/i386/linux.c
===================================================================
@@ -331,12 +331,14 @@ grub_linux_boot (void)
   
   params = real_mode_mem;
 
-  if (vid_mode < GRUB_LINUX_VID_MODE_VESA_START ||
-      vid_mode >= GRUB_LINUX_VID_MODE_VESA_START +
-		  ARRAY_SIZE (linux_vesafb_modes))
-    grub_video_restore ();
-  else if (vid_mode)
+  if (vid_mode)
     {
+      if (vid_mode < GRUB_LINUX_VID_MODE_VESA_START ||
+	  vid_mode >= GRUB_LINUX_VID_MODE_VESA_START +
+	  ARRAY_SIZE (linux_vesafb_modes))
+	grub_video_restore ();
+      else
+	{
       struct linux_vesafb_mode *linux_mode;
       int depth, flags;
       
@@ -366,6 +368,7 @@ grub_linux_boot (void)
 	  return grub_errno;
 	}
     }
+    }
 
   if (! grub_linux_setup_video (params))
     params->have_vga = GRUB_VIDEO_TYPE_VLFB;
@@ -570,8 +573,7 @@ grub_cmd_linux (grub_command_t cmd __att
 	       (unsigned) real_size, (unsigned) prot_size);
 
   /* Look for memory size and video mode specified on the command line.  */
-  vid_mode = GRUB_LINUX_VID_MODE_NORMAL;
-  linux_mem_size = 0;
+  vid_mode = linux_mem_size = 0;
   for (i = 1; i < argc; i++)
     if (grub_memcmp (argv[i], "vga=", 4) == 0)
       {
#9 lore raw
Quoted 11 lines · Robert Millan
Quoted 8 lines · Robert Millan
Quoted 2 lines

The default VGA mode is now GRUB_LINUX_VID_MODE_NORMAL, not the mode of the kernel we tried to load before.

Ok, BUT if we're already in vesa mode, and we know it works (since we're using it), there's no point in wasting time only to get a worse mode.

We should just make sure subsequent calls to "linux" command override the previous one.

I think this should do it. Can you confirm this works for you?

I actually installed GRUB with gfxterm on a laptop that has Intel framebuffer support. Now the kernel starts in VESA mode and then the screen goes blank because intelfb cannot deal with it. Sure, intelfb should be fixed, but we should be liberal in what we accept. Some kernels may not support VESA modes at all.

Adding vga=0 to the kernel command line didn't fix it. That's bad. "vga=0" means text mode 80x25. Adding "vga=1" fixed the problem. The text mode was 80x25, not 80x50, so that's another issue.

"vga=ask" is not a warning now. It causes "error: You need to load the kernel first", apparently from initrd. In other words, the "linux" command fails and there is no visible warning.

Signature · 2 lines
Regards,
Pavel Roskin
#10 lore raw
Quoted 4 lines · Pavel Roskin

I actually installed GRUB with gfxterm on a laptop that has Intel framebuffer support. Now the kernel starts in VESA mode and then the screen goes blank because intelfb cannot deal with it. Sure, intelfb should be fixed, but we should be liberal in what we accept.

We could detect this situation by checking video= parameter, and setting text mode if intelfb is found. But then again do we want to prevent future versions of intelfb from gracefuly transitioning from vesa mode without screen glitch?

Quoted 2 lines

Some
kernels may not support VESA modes at all.

I don't think this is applicable; all modern versions of Linux include vesa modesetting in its 16-bit entry code, and older versions are already detected by the new loader (user is prompted to use linux16).

Quoted 3 lines

Adding vga=0 to the kernel command line didn't fix it. That's bad. "vga=0" means text mode 80x25. Adding "vga=1" fixed the problem. The text mode was 80x25, not 80x50, so that's another issue.

Shouldn't be hard to fix. Do you know how to switch to 80x50 mode?

Quoted 3 lines

"vga=ask" is not a warning now. It causes "error: You need to load the kernel first", apparently from initrd. In other words, the "linux" command fails and there is no visible warning.

Sounds like my error code is wrong, but we could turn it into a warning like you suggested.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#11 lore raw
Quoted 10 lines · Robert Millan
Quoted 4 lines · Pavel Roskin

I actually installed GRUB with gfxterm on a laptop that has Intel framebuffer support. Now the kernel starts in VESA mode and then the screen goes blank because intelfb cannot deal with it. Sure, intelfb should be fixed, but we should be liberal in what we accept.

We could detect this situation by checking video= parameter, and setting text mode if intelfb is found. But then again do we want to prevent future versions of intelfb from gracefuly transitioning from vesa mode without screen glitch?

No, that would be bad. It's even possible that intelfb would work correctly in other configurations. The laptop has resolution 1440x900 that doesn't match any VESA mode.

Quoted 6 lines
Quoted 2 lines

Some
kernels may not support VESA modes at all.

I don't think this is applicable; all modern versions of Linux include vesa modesetting in its 16-bit entry code, and older versions are already detected by the new loader (user is prompted to use linux16).

I can disable CONFIG_FB, and then the screen remains blank until X starts. It's entirely possible that some distros don't enable CONFIG_FB to save memory, and I don't always enable it in the kernels I configure myself.

Quoted 5 lines
Quoted 3 lines

Adding vga=0 to the kernel command line didn't fix it. That's bad. "vga=0" means text mode 80x25. Adding "vga=1" fixed the problem. The text mode was 80x25, not 80x50, so that's another issue.

Shouldn't be hard to fix. Do you know how to switch to 80x50 mode?

Well, load 8x8 font. It's done in the 16-bit code.

Quoted 6 lines
Quoted 3 lines

"vga=ask" is not a warning now. It causes "error: You need to load the kernel first", apparently from initrd. In other words, the "linux" command fails and there is no visible warning.

Sounds like my error code is wrong, but we could turn it into a warning like you suggested.

I was editing the command line from the menu, so I could not see the message. Waiting for input is a fair game for an option that implies waiting for input.

Signature · 2 lines
Regards,
Pavel Roskin
#12 lore raw
Quoted 6 lines · Pavel Roskin
Quoted 5 lines · Robert Millan
Quoted 4 lines · Pavel Roskin

I actually installed GRUB with gfxterm on a laptop that has Intel framebuffer support. Now the kernel starts in VESA mode and then the screen goes blank because intelfb cannot deal with it. Sure, intelfb should be fixed, but we should be liberal in what we accept.

Quoted 9 lines
Quoted 4 lines

We could detect this situation by checking video= parameter, and setting text mode if intelfb is found. But then again do we want to prevent future versions of intelfb from gracefuly transitioning from vesa mode without screen glitch?

No, that would be bad. It's even possible that intelfb would work correctly in other configurations. The laptop has resolution 1440x900 that doesn't match any VESA mode.

With proposed autodetections bootloader becomes overzealous. I think it should be configirable but in a more modern way. I think gfxpayload with the same syntax as gfxmode plus additional platform-specific keywords like text80x25 and text80x50 should be fine. If to gfxpayload variable set, then use current mode. I object against a solution of just using current mode because of the current state on EFI: code to retrieve framebuffer address isn't yet used for gfxterm. If this is fixed then I could agree with "last mode" solution.

vga= option could be deleted altogether and replaced by warning if someone passes it. If we do so it's better to do before debian switches

Signature · 2 lines
Regards
Vladimir 'phcoder' Serbinenko
#13 lore raw
Quoted 4 lines · phcoder

With proposed autodetections bootloader becomes overzealous. I think it should be configirable but in a more modern way. I think gfxpayload with the same syntax as gfxmode plus additional platform-specific keywords like text80x25 and text80x50 should be fine.

Sounds nice. Will you implement this?

Quoted 2 lines

If to gfxpayload variable
set, then use current mode.

I think based on what Pavel said, and on what you say below, that using current mode should be avoided in official GRUB, because of the need to support a wide variety of build configurations. If gfxpayload is unset I would just reset it.

Someone who's also distributing builds of Linux and has intimate knowledge about their setup, can also make similar decisions when it comes to setting up GRUB build options.

Quoted 3 lines

I object against a solution of just using current mode because of the current state on EFI: code to retrieve framebuffer address isn't yet used for gfxterm.

Can this be fixed?

Quoted 2 lines

vga= option could be deleted altogether and replaced by warning if someone passes it.

A warning is fine, but please keep the vga=. It's useful to provide backward compatibility, even if it's a deprecated interface.

(Though, since it's backward compatibility, I would only provide it on BIOS platforms)

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#14 lore raw
In reply to Pavel Roskin
Quoted 8 lines · Pavel Roskin
Quoted 4 lines

We could detect this situation by checking video= parameter, and setting text mode if intelfb is found. But then again do we want to prevent future versions of intelfb from gracefuly transitioning from vesa mode without screen glitch?

No, that would be bad. It's even possible that intelfb would work correctly in other configurations. The laptop has resolution 1440x900 that doesn't match any VESA mode.

Alright.

Quoted 11 lines
Quoted 6 lines
Quoted 2 lines

Some
kernels may not support VESA modes at all.

I don't think this is applicable; all modern versions of Linux include vesa modesetting in its 16-bit entry code, and older versions are already detected by the new loader (user is prompted to use linux16).

I can disable CONFIG_FB, and then the screen remains blank until X starts. It's entirely possible that some distros don't enable CONFIG_FB to save memory, and I don't always enable it in the kernels I configure myself.

Makes sense for official GRUB.

However, I'd still like to add a macro check that can be enabled on distros that ship Linux builds with CONFIG_FB and want to enable seamless mode transition (this will be the case for e.g. Debian).

Quoted 10 lines
Quoted 6 lines
Quoted 3 lines

"vga=ask" is not a warning now. It causes "error: You need to load the kernel first", apparently from initrd. In other words, the "linux" command fails and there is no visible warning.

Sounds like my error code is wrong, but we could turn it into a warning like you suggested.

I was editing the command line from the menu, so I could not see the message. Waiting for input is a fair game for an option that implies waiting for input.

Spot on. Will do that.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#15 lore raw
Quoted 4 lines · Robert Millan
Quoted 4 lines
Quoted 4 lines
Quoted 3 lines

"vga=ask" is not a warning now. It causes "error: You need to load the kernel first", apparently from initrd. In other words, the "linux" command fails and there is no visible warning.

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

Sounds like my error code is wrong, but we could turn it into a warning like you suggested.

I was editing the command line from the menu, so I could not see the message. Waiting for input is a fair game for an option that implies waiting for input.

Spot on. Will do that.

Fixed.

See also this new patch. It restructures the checks so that "vid_mode == 0" indicates lack of "vga=" parameter. For user requesting text mode (vga=normal or vga=0) we already have GRUB_LINUX_VID_MODE_NORMAL so there's no need to handle both values in GRUB.

It also introduces the GRUB_ASSUME_LINUX_HAS_FB_SUPPORT macro, which allows easy override of the "fallback to text mode" setting.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."

2009-05-02 Robert Millan <[email protected]>

* loader/i386/linux.c (GRUB_ASSUME_LINUX_HAS_FB_SUPPORT): New macro. (grub_linux_boot): Don't check for `linux_vesafb_modes' bounds (this is done by grub_cmd_linux() now).
[! GRUB_ASSUME_LINUX_HAS_FB_SUPPORT]: If "vga=" parameter wasn't set, restore video to text mode.
(grub_cmd_linux): Default vid_mode' initialization to 0, which indicates lack of "vga=" parameter. "vga=0" is mapped to GRUB_LINUX_VID_MODE_NORMAL'.

i386/linux.c +32 -4
Index: loader/i386/linux.c
===================================================================
@@ -39,6 +39,14 @@
 #define GRUB_LINUX_CL_OFFSET		0x1000
 #define GRUB_LINUX_CL_END_OFFSET	0x2000
 
+/* This macro is useful for distributors, who can be certain they built FB support
+   into Linux, and therefore can benefit from seamless mode transition between
+   GRUB and Linux (saving boot time and visual glitches).  Official GRUB, OTOH,
+   needs to be conservative.  */
+#ifndef GRUB_ASSUME_LINUX_HAS_FB_SUPPORT
+#define GRUB_ASSUME_LINUX_HAS_FB_SUPPORT 0
+#endif
+
 static grub_dl_t my_mod;
 
 static grub_size_t linux_mem_size;
@@ -332,9 +340,7 @@ grub_linux_boot (void)
   
   params = real_mode_mem;
 
-  if (vid_mode < GRUB_LINUX_VID_MODE_VESA_START ||
-      vid_mode >= GRUB_LINUX_VID_MODE_VESA_START +
-		  ARRAY_SIZE (linux_vesafb_modes))
+  if (vid_mode == GRUB_LINUX_VID_MODE_NORMAL || vid_mode == GRUB_LINUX_VID_MODE_EXTENDED)
     grub_video_restore ();
   else if (vid_mode)
     {
@@ -367,6 +373,12 @@ grub_linux_boot (void)
 	  return grub_errno;
 	}
     }
+#if ! GRUB_ASSUME_LINUX_HAS_FB_SUPPORT
+  else
+    /* If user didn't request a video mode, and we can't assume Linux supports FB,
+       then we go back to text mode.  */
+    grub_video_restore ();
+#endif
 
   if (! grub_linux_setup_video (params))
     params->have_vga = GRUB_VIDEO_TYPE_VLFB;
@@ -571,7 +583,7 @@ grub_cmd_linux (grub_command_t cmd __att
 	       (unsigned) real_size, (unsigned) prot_size);
 
   /* Look for memory size and video mode specified on the command line.  */
-  vid_mode = GRUB_LINUX_VID_MODE_NORMAL;
+  vid_mode = 0;
   linux_mem_size = 0;
   for (i = 1; i < argc; i++)
     if (grub_memcmp (argv[i], "vga=", 4) == 0)
@@ -596,6 +608,22 @@ grub_cmd_linux (grub_command_t cmd __att
 	else
 	  vid_mode = (grub_uint16_t) grub_strtoul (val, 0, 0);
 
+	switch (vid_mode)
+	  {
+	  case 0:
+	    vid_mode = GRUB_LINUX_VID_MODE_NORMAL;
+	    break;
+	  case 1:
+	    vid_mode = GRUB_LINUX_VID_MODE_EXTENDED;
+	    break;
+	  default:
+	    /* Ignore invalid values.  */
+	    if (vid_mode < GRUB_LINUX_VID_MODE_VESA_START ||
+		vid_mode >= GRUB_LINUX_VID_MODE_VESA_START +
+		ARRAY_SIZE (linux_vesafb_modes))
+	      vid_mode = 0;
+	  }
+
 	if (grub_errno)
 	  goto fail;
       }
#16 lore raw
Quoted 18 lines · Robert Millan
Quoted 9 lines · Robert Millan
Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

"vga=ask" is not a warning now. It causes "error: You need to load the kernel first", apparently from initrd. In other words, the "linux" command fails and there is no visible warning.

Sounds like my error code is wrong, but we could turn it into a warning like you suggested.

I was editing the command line from the menu, so I could not see the message. Waiting for input is a fair game for an option that implies waiting for input.

Spot on. Will do that.

Fixed.

See also this new patch. It restructures the checks so that "vid_mode == 0" indicates lack of "vga=" parameter. For user requesting text mode (vga=normal or vga=0) we already have GRUB_LINUX_VID_MODE_NORMAL so there's no need to handle both values in GRUB.

Quoted 6 lines

It also introduces the GRUB_ASSUME_LINUX_HAS_FB_SUPPORT macro, which allows easy override of the "fallback to text mode" setting.

------------------------------------------------------------------------

This may follow the last little problem we are having getting the splash screen to show in Zenwalk. We were in IRC last weekend trying to come up with a solution as to why we had to use linux16 & initrd16 with the vga= command in the grub.cfg, but nothing he suggested seemed to work. Here is a section from my grub.cfg:
#
# DO NOT EDIT THIS FILE
#
# It is automatically generated by /usr/sbin/grub-mkconfig using templates

# from  and settings from /etc/default/grub
#

### BEGIN /etc/grub.d/00_header ###
set default=0
set timeout=10
set root=(hd0,1)
search --fs-uuid --set 82d81803-e0a2-487d-9c83-ec6d8d35cf2a if loadfont /usr/share/grub/ascii.pf2 ; then
set gfxmode=1024x768x32
insmod gfxterm
insmod video
insmod vbe
if terminal_output gfxterm ; then true ; else
# For backward compatibility with versions of terminal.mod that don't # understand terminal_output
terminal gfxterm
fi
fi
insmod png
background_image /boot/grub/ZenBlack.png
### END /etc/grub.d/00_header ###

### BEGIN /etc/grub.d/10_linux ###

menuentry "Zenwalk 6.0 GNU/Linux" {
	set root=(hd0,1)
	search --fs-uuid --set 82d81803-e0a2-487d-9c83-ec6d8d35cf2a
	linux16	/boot/vmlinuz root=/dev/sda1 ro resume=/dev/sda4 splash=silent 

vga=794
initrd16 /boot/initrd.splash
}
-------------------------
My first question for you Robert is; will your vid_mode.diff correct the need for using vga= and linux16 to get the splash screen to display in the correct resolution? Second, what is needed in the grub.cfg to be able to not use the deprecated linux16 & vga= anymore? Does someone have a sample grub.cfg for x86 Linux that works to display the splash screen using linux, but without the vga= ? If you do, please send it to me, so I might try your solution. I can't believe this is as big a problem as it appears to be, but no one seems to have mentioned a viable solution yet!

Thanks for your help!
Pat

#17 lore raw
Quoted 2 lines · BandiPat

linux16 /boot/vmlinuz root=/dev/sda1 ro resume=/dev/sda4 splash=silent
vga=794

This means vga=0x31a, aka 16-bit 1280x1024. Does 24-bit (vga=0x31b) work?

I suspect there's some fuzzy matching here.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#18 lore raw
Quoted 8 lines · Robert Millan
Quoted 2 lines · BandiPat

linux16 /boot/vmlinuz root=/dev/sda1 ro resume=/dev/sda4 splash=silent
vga=794

This means vga=0x31a, aka 16-bit 1280x1024. Does 24-bit (vga=0x31b) work?

I suspect there's some fuzzy matching here.

=======
Thanks Robert for the suggestion. I had renewed hopes this might work, but it would not. After editing the menuentry to use either the vga=0x31b or vga=795, the machine booted to a blinking cursor. It would not go beyond that.

Does that help in determining where maybe the problem lies? Also, I thought both linux16 and vga were deprecated, no longer to be used?

Thanks,
Pat

Signature · 3 lines
       ---Zenwalk v6.0--Linux 2.6.28---
        Registered Linux User #225206
"Ever tried Zen computing?"  http://www.zenwalk.org
#19 lore raw
Quoted 7 lines · BandiPat
Quoted 6 lines · Robert Millan
Quoted 2 lines · BandiPat

linux16 /boot/vmlinuz root=/dev/sda1 ro resume=/dev/sda4
splash=silent vga=794

This means vga=0x31a, aka 16-bit 1280x1024. Does 24-bit (vga=0x31b) work?

Quoted 7 lines
Quoted 2 lines

I suspect there's some fuzzy matching here.

=======
Thanks Robert for the suggestion. I had renewed hopes this might work, but it would not. After editing the menuentry to use either the vga=0x31b or vga=795, the machine booted to a blinking cursor. It would not go beyond that.

What suitable modes does "vbeinfo" list?

Quoted 2 lines

Does that help in determining where maybe the problem lies? Also, I thought both linux16 and vga were deprecated, no longer to be used?

"vga" is deprecated, but we don't have a replacement for it yet.

"linux16" will eventually become completely obsolete, but some users (e.g. you) still need to rely on it, which is why we didn't remove it.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#20 lore raw
Quoted 5 lines · Robert Millan
Quoted 2 lines

=======
Thanks Robert for the suggestion. I had renewed hopes this might work, but it would not. After editing the menuentry to use either the vga=0x31b or vga=795, the machine booted to a blinking cursor. It would not go beyond that.

What suitable modes does "vbeinfo" list?

===========
A little help here. Where do you run this "vbeinfo" command to get a listing of suitable modes? I am going to load the vbeinfo.mod in the grub.cfg, but what after that to get a listing?

Thanks

#21 lore raw

Am Montag, den 04.05.2009, 14:16 -0400 schrieb BandiPat:

Quoted 4 lines

A little help here. Where do you run this "vbeinfo" command to get a listing of suitable modes? I am going to load the vbeinfo.mod in the grub.cfg, but what after that to get a listing?

Just press `c' to get to commandline and type vbeinfo. Normally you shouldn't ever need to insmod command modules. GRUB should load them automatically as long as you have /boot/grub/command.lst.

Signature · 1 line
Felix Zielcke
#22 lore raw
In reply to Robert Millan

Btw, I increased the mode list considerably (using documented modes from Wikipedia). Chances that your ultra-weird mode of choice is supported are much greater now.

There's still no fuzzy matching, though. I'm not sure if we'd want to do that at all for vga= modes, since we already do it properly in our video subsystem.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#23 lore raw
Quoted 8 lines · Robert Millan

Btw, I increased the mode list considerably (using documented modes from Wikipedia). Chances that your ultra-weird mode of choice is supported are much greater now.

There's still no fuzzy matching, though. I'm not sure if we'd want to do that at all for vga= modes, since we already do it properly in our video subsystem.

=========
I'll give them a look as soon as I can get something new built. I'll have to admit, I didn't expect you guys to go to so much trouble, since the vga= will be going away anyhow. We can all live with the "fuzzy matching" for now, until you get the new video subsystem operating properly. Don't put a lot of time into something that won't be used much longer, as we can live with things as they are for the moment.

Thanks for everything.
Pat

Signature · 3 lines
       ---Zenwalk v6.0--Linux 2.6.28---
        Registered Linux User #225206
"Ever tried Zen computing?"  http://www.zenwalk.org
#24 lore raw
Quoted 9 lines · BandiPat
Quoted 8 lines · Robert Millan

Btw, I increased the mode list considerably (using documented modes from Wikipedia). Chances that your ultra-weird mode of choice is supported are much greater now.

There's still no fuzzy matching, though. I'm not sure if we'd want to do that at all for vga= modes, since we already do it properly in our video subsystem.

Quoted 7 lines

=========
I'll give them a look as soon as I can get something new built. I'll have to admit, I didn't expect you guys to go to so much trouble, since the vga= will be going away anyhow. We can all live with the "fuzzy matching" for now, until you get the new video subsystem operating properly. Don't put a lot of time into something that won't be used much longer, as we can live with things as they are for the moment.

Maybe you're right :-/

Hopefuly this will be less of a problem when we can set it from a variable using a sane interface...

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#25 lore raw
In reply to Robert Millan
Quoted 8 lines · Robert Millan

See also this new patch. It restructures the checks so that "vid_mode == 0" indicates lack of "vga=" parameter. For user requesting text mode (vga=normal or vga=0) we already have GRUB_LINUX_VID_MODE_NORMAL so there's no need to handle both values in GRUB.

It also introduces the GRUB_ASSUME_LINUX_HAS_FB_SUPPORT macro, which allows easy override of the "fallback to text mode" setting.

Committed.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#26 lore raw
In reply to Robert Millan
Quoted 12 lines · Robert Millan
Quoted 11 lines
Quoted 6 lines
Quoted 2 lines

Some
kernels may not support VESA modes at all.

I don't think this is applicable; all modern versions of Linux include vesa modesetting in its 16-bit entry code, and older versions are already detected by the new loader (user is prompted to use linux16).

I can disable CONFIG_FB, and then the screen remains blank until X starts. It's entirely possible that some distros don't enable CONFIG_FB to save memory, and I don't always enable it in the kernels I configure myself.

Quoted 5 lines

Makes sense for official GRUB.

However, I'd still like to add a macro check that can be enabled on distros that ship Linux builds with CONFIG_FB and want to enable seamless mode transition (this will be the case for e.g. Debian).

Can we have GRUB2 not be distro-dependent, because I often depend on it to launch multiple different distros (and I'd rather not worry too much about which distro installed my grub2)? It [i.e., whether to use seamless mode transition] can be a setting in the grub.cfg if necessary.

-Isaac

#27 lore raw
Quoted 13 lines · Isaac Dupree
Quoted 12 lines · Robert Millan
Quoted 11 lines
Quoted 6 lines
Quoted 2 lines

Some
kernels may not support VESA modes at all.

I don't think this is applicable; all modern versions of Linux include vesa modesetting in its 16-bit entry code, and older versions are already detected by the new loader (user is prompted to use linux16).

I can disable CONFIG_FB, and then the screen remains blank until X starts. It's entirely possible that some distros don't enable CONFIG_FB to save memory, and I don't always enable it in the kernels I configure myself.

Quoted 7 lines
Quoted 5 lines

Makes sense for official GRUB.

However, I'd still like to add a macro check that can be enabled on distros that ship Linux builds with CONFIG_FB and want to enable seamless mode transition (this will be the case for e.g. Debian).

Can we have GRUB2 not be distro-dependent,

We aren't talking about making GRUB distro-dependant, only about what is the default behaviour when user didn't specify any.

Official GRUB, as discussed, will default to text mode. I don't see why downstream distributors would need to do the same, since it's so easy to override in case a user needs that.

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."
#28 lore raw
In reply to Robert Millan
Quoted 10 lines · Robert Millan

Hi Pavel,

Quoted 4 lines · Pavel Roskin

Hello!

This is an attempt to fix all issues with the video mode handling in the new Linux loader.

Please, in general, when you modify code that has been actively worked on by someone, try to get that person involved.

I tried, but probably not hard enough. Sorry.

Quoted 5 lines
Quoted 2 lines

The default VGA mode is now GRUB_LINUX_VID_MODE_NORMAL, not the mode of the kernel we tried to load before.

Ok, BUT if we're already in vesa mode, and we know it works (since we're using it), there's no point in wasting time only to get a worse mode.

Maybe we should introduce another macro, e.g. GRUB_LINUX_VID_MODE_CURRENT, recognize it in the loader and default to it.

Quoted 2 lines

We should just make sure subsequent calls to "linux" command override the previous one.

Yes, failed loading of one kernel should not affect loading of another kernel. That was the most annoying bug, and I hope it's fixed now.

Quoted 5 lines
Quoted 1 line

Finally, "vga=ask" is now recognized.

With the new loader, Linux' 16-bit entry code is no longer responsible for setting vesa modes, therefore vga=ask can't work. There's no point in recognizing it (except to warn the user).

I agree. Generally, "vga=random_string" causes an error now. Perhaps any unrecognized values of "vga" should cause a warning, not an error.

Signature · 2 lines
Regards,
Pavel Roskin
#29 lore raw
Quoted 5 lines · Pavel Roskin
Quoted 2 lines · Robert Millan

Please, in general, when you modify code that has been actively worked on by someone, try to get that person involved.

I tried, but probably not hard enough. Sorry.

No problem :-)

Quoted 6 lines
Quoted 2 lines

Ok, BUT if we're already in vesa mode, and we know it works (since we're using it), there's no point in wasting time only to get a worse mode.

Maybe we should introduce another macro, e.g. GRUB_LINUX_VID_MODE_CURRENT, recognize it in the loader and default to it.

If we're going to get creative and add new options that weren't present in the legacy interface, then I'd really prefer if we design a new one, and better a kernel-independant one like an env variable (we discussed this in another thread, but didn't agree on which name to use yet).

Quoted 2 lines

Yes, failed loading of one kernel should not affect loading of another kernel. That was the most annoying bug, and I hope it's fixed now.

Yeah I don't see anything wrong with this situation now, but I could be missing something, since there are many ways to "fail".

Quoted 8 lines
Quoted 5 lines
Quoted 1 line

Finally, "vga=ask" is now recognized.

With the new loader, Linux' 16-bit entry code is no longer responsible for setting vesa modes, therefore vga=ask can't work. There's no point in recognizing it (except to warn the user).

I agree. Generally, "vga=random_string" causes an error now. Perhaps any unrecognized values of "vga" should cause a warning, not an error.

Probably better when it comes to headless machines, yes...

Signature · 5 lines
Robert Millan

  The DRM opt-in fallacy: "Your data belongs to us. We will decide when (and
  how) you may access your data; but nobody's threatening your freedom: we
  still allow you to remove your data and not access it at all."