Skip to contentSkip to conversation

[5.10] regression: virtual consoles 2-12 unusable

DISCUSSION Ben Hutchings 3 participants 4 messages Jan 2 – Jan 28 5 messages not in the archive Stable branch

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): CAN2UaigCW-BZTifuo-ADCw=uDq85A_KwOHcceyaXDnVo8OQZiQ@mail.gmail.com

#5 lore raw

Hello stable maintainers,

Several Debian users reported a regression after updating to kernel version 5.10.247.

Commit f0982400648a ("fbdev: Add bounds checking in bit_putcs to fix vmalloc-out-of-bounds"), a backport of upstream commit 3637d34b35b2, depends on vc_data::vc_font.charcount being initialised correctly.

However, before commit a1ac250a82a5 ("fbcon: Avoid using FNTCHARCNT() and hard-coded built-in font charcount") in 5.11, this member was set to 256 for VTs initially created with a built-in font and 0 for VTs initially created with a user font.

Since Debian normally sets a user font before creating VTs 2 and up, those additional VTs became unusable. VT 1 also doesn't work correctly if the user font has > 256 characters, and the bounds check is ineffective if it has < 256 characters.

This can be fixed by backporting the following commits from 5.11:

7a089ec7d77f console: Delete unused con_font_copy() callback implementations 259a252c1f4e console: Delete dummy con_font_set() and con_font_default() callback implementations 4ee573086bd8 Fonts: Add charcount field to font_desc
4497364e5f61 parisc/sticore: Avoid hard-coding built-in font charcount a1ac250a82a5 fbcon: Avoid using FNTCHARCNT() and hard-coded built-in font charcount

These all apply without fuzz and builds cleanly for x86_64 and parisc64.

I tested on x86_64 that:

- VT 2 works again
- bit_putcs_aligned() is setting charcnt = 256
- After loading a font with 512 characters, bit_putcs_aligned() sets charcnt = 512 and is able to display characters at positions >= 256

Ben.

Signature · 3 lines
Ben Hutchings
Man invented language to satisfy his deep need to complain.
                                                          - Lily Tomlin
#6 lore raw
Quoted 27 lines · Ben Hutchings

Hello stable maintainers,

Several Debian users reported a regression after updating to kernel version 5.10.247.

Commit f0982400648a ("fbdev: Add bounds checking in bit_putcs to fix vmalloc-out-of-bounds"), a backport of upstream commit 3637d34b35b2, depends on vc_data::vc_font.charcount being initialised correctly.

However, before commit a1ac250a82a5 ("fbcon: Avoid using FNTCHARCNT() and hard-coded built-in font charcount") in 5.11, this member was set to 256 for VTs initially created with a built-in font and 0 for VTs initially created with a user font.

Since Debian normally sets a user font before creating VTs 2 and up, those additional VTs became unusable. VT 1 also doesn't work correctly if the user font has > 256 characters, and the bounds check is ineffective if it has < 256 characters.

This can be fixed by backporting the following commits from 5.11:

7a089ec7d77f console: Delete unused con_font_copy() callback implementations 259a252c1f4e console: Delete dummy con_font_set() and con_font_default() callback implementations 4ee573086bd8 Fonts: Add charcount field to font_desc
4497364e5f61 parisc/sticore: Avoid hard-coding built-in font charcount a1ac250a82a5 fbcon: Avoid using FNTCHARCNT() and hard-coded built-in font charcount

Quoted 8 lines

These all apply without fuzz and builds cleanly for x86_64 and parisc64.

I tested on x86_64 that:

- VT 2 works again
- bit_putcs_aligned() is setting charcnt = 256
- After loading a font with 512 characters, bit_putcs_aligned() sets charcnt = 512 and is able to display characters at positions >= 256

All now queued up, thanks!

greg k-h

#7 lore raw
Quoted 32 lines · Greg Kroah-Hartman
Quoted 31 lines · Ben Hutchings

Hello stable maintainers,

Several Debian users reported a regression after updating to kernel version 5.10.247.

Commit f0982400648a ("fbdev: Add bounds checking in bit_putcs to fix vmalloc-out-of-bounds"), a backport of upstream commit 3637d34b35b2, depends on vc_data::vc_font.charcount being initialised correctly.

However, before commit a1ac250a82a5 ("fbcon: Avoid using FNTCHARCNT() and hard-coded built-in font charcount") in 5.11, this member was set to 256 for VTs initially created with a built-in font and 0 for VTs initially created with a user font.

Since Debian normally sets a user font before creating VTs 2 and up, those additional VTs became unusable. VT 1 also doesn't work correctly if the user font has > 256 characters, and the bounds check is ineffective if it has < 256 characters.

This can be fixed by backporting the following commits from 5.11:

7a089ec7d77f console: Delete unused con_font_copy() callback implementations 259a252c1f4e console: Delete dummy con_font_set() and con_font_default() callback implementations 4ee573086bd8 Fonts: Add charcount field to font_desc
4497364e5f61 parisc/sticore: Avoid hard-coding built-in font charcount a1ac250a82a5 fbcon: Avoid using FNTCHARCNT() and hard-coded built-in font charcount

These all apply without fuzz and builds cleanly for x86_64 and parisc64.

I tested on x86_64 that:

Quoted 8 lines
Quoted 4 lines

- VT 2 works again
- bit_putcs_aligned() is setting charcnt = 256
- After loading a font with 512 characters, bit_putcs_aligned() sets charcnt = 512 and is able to display characters at positions >= 256

All now queued up, thanks!

greg k-h

For what it's worth, now that the above commits are queued up for 5.10.y: There are two more commits, which were previously applied to 5.15.y, that now apply to 5.10.y without merge conflicts (also without fuzz, if you apply the 5.15.y versions of the patches):

a5a923038d70
fbdev: fbcon: Properly revert changes when vc_resize() failed (previously applied to 5.15.64 and 5.19.6)

3c3bfb8586f8
fbdev: fbcon: release buffer when fbcon_do_set_font() failed (previously applied to 5.15.86, 6.0.16, and 6.1.2)

After looking at these two commits, it seems to me that they are now applicable to 5.10.y, and I think they probably should be applied (unless I'm overlooking or missing something).

By the way, for the past few days I have been running 5.10.247 + the latest 5.10.y queue + a5a923038d70 + 3c3bfb8586f8 (updating as new patches have been added to the queue). It seems to be running fine, and the virtual consoles seem to be working fine; it's also what I'm running now as I write and send this email.

(To be clear, 5.10.247 + the 5.10.y queue, without the two additional commits, also fixes the original 5.10.247 virtual console regression for me. However, applying commits a5a923038d70 and 3c3bfb8586f8 on top of 5.10.y doesn't appear to create any new regressions in my testing so far.)

Signature · 1 line
-Barry K. Nathan  <[email protected]>
#8 lore raw
Quoted 52 lines · Barry K. Nathan
Quoted 32 lines · Greg Kroah-Hartman
Quoted 31 lines · Ben Hutchings

Hello stable maintainers,

Several Debian users reported a regression after updating to kernel version 5.10.247.

Commit f0982400648a ("fbdev: Add bounds checking in bit_putcs to fix vmalloc-out-of-bounds"), a backport of upstream commit 3637d34b35b2, depends on vc_data::vc_font.charcount being initialised correctly.

However, before commit a1ac250a82a5 ("fbcon: Avoid using FNTCHARCNT() and hard-coded built-in font charcount") in 5.11, this member was set to 256 for VTs initially created with a built-in font and 0 for VTs initially created with a user font.

Since Debian normally sets a user font before creating VTs 2 and up, those additional VTs became unusable. VT 1 also doesn't work correctly if the user font has > 256 characters, and the bounds check is ineffective if it has < 256 characters.

This can be fixed by backporting the following commits from 5.11:

7a089ec7d77f console: Delete unused con_font_copy() callback implementations 259a252c1f4e console: Delete dummy con_font_set() and con_font_default() callback implementations 4ee573086bd8 Fonts: Add charcount field to font_desc
4497364e5f61 parisc/sticore: Avoid hard-coding built-in font charcount a1ac250a82a5 fbcon: Avoid using FNTCHARCNT() and hard-coded built-in font charcount

These all apply without fuzz and builds cleanly for x86_64 and parisc64.

I tested on x86_64 that:

Quoted 8 lines
Quoted 4 lines

- VT 2 works again
- bit_putcs_aligned() is setting charcnt = 256
- After loading a font with 512 characters, bit_putcs_aligned() sets charcnt = 512 and is able to display characters at positions >= 256

All now queued up, thanks!

greg k-h

For what it's worth, now that the above commits are queued up for 5.10.y:
There are two more commits, which were previously applied to 5.15.y, that now apply to 5.10.y without merge conflicts (also without fuzz, if you apply the 5.15.y versions of the patches):

a5a923038d70
fbdev: fbcon: Properly revert changes when vc_resize() failed (previously applied to 5.15.64 and 5.19.6)

Quoted 8 lines

3c3bfb8586f8
fbdev: fbcon: release buffer when fbcon_do_set_font() failed (previously applied to 5.15.86, 6.0.16, and 6.1.2)

After looking at these two commits, it seems to me that they are now applicable to 5.10.y, and I think they probably should be applied (unless I'm overlooking or missing something).

Both now applied, thanks!

greg k-h