NetBSD Problem Report #59361

From www@netbsd.org  Sat Apr 26 16:34:53 2025
Return-Path: <www@netbsd.org>
Received: from mail.netbsd.org (mail.netbsd.org [199.233.217.200])
	(using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits)
	 key-exchange X25519 server-signature RSA-PSS (2048 bits) server-digest SHA256
	 client-signature RSA-PSS (2048 bits) client-digest SHA256)
	(Client CN "mail.NetBSD.org", Issuer "mail.NetBSD.org CA" (not verified))
	by mollari.NetBSD.org (Postfix) with ESMTPS id CD8721A923C
	for <gnats-bugs@gnats.NetBSD.org>; Sat, 26 Apr 2025 16:34:52 +0000 (UTC)
Message-Id: <20250426163451.94F421A923E@mollari.NetBSD.org>
Date: Sat, 26 Apr 2025 16:34:51 +0000 (UTC)
From: campbell+netbsd@mumble.net
Reply-To: campbell+netbsd@mumble.net
To: gnats-bugs@NetBSD.org
Subject: paging up in less(1) under qemu vga console is slow
X-Send-Pr-Version: www-1.0

>Number:         59361
>Category:       port-amd64
>Synopsis:       paging up in less(1) under qemu vga console is slow
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    port-amd64-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Apr 26 16:35:00 +0000 2025
>Last-Modified:  Mon Oct 27 00:00:01 +0000 2025
>Originator:     Taylor R Campbell
>Release:        current
>Organization:
The Slow386BSD Emulation
>Environment:
>Description:
Paging _down_ in less(1) under qemu vga console is reasonably quick.  Not instantaneous -- I still see a flicker sometimes -- but quick enough.

Paging _up_ in less(1) under qemu vga console is excruciatingly slow, takes multiple seconds to go a single page in a 25x80 display.
>How-To-Repeat:
1. run NetBSD/amd64 under qemu-system-x86_64
2. dmesg | less, or whatever
3. hit SPC to page down, hit b to page up
>Fix:
Yes, please!

Someone should alert the resident graphics expert.

>Audit-Trail:
From: RVP <rvp@SDF.ORG>
To: gnats-bugs@netbsd.org
Cc: 
Subject: Re: port-amd64/59361: paging up in less(1) under qemu vga console
 is slow
Date: Sun, 26 Oct 2025 22:35:24 +0000 (UTC)

 On Sat, 26 Apr 2025, campbell+netbsd@mumble.net wrote:

 >> Description:
 > Paging _down_ in less(1) under qemu vga console is reasonably quick.  Not instantaneous -- I still see a flicker sometimes -- but quick enough.
 >
 > Paging _up_ in less(1) under qemu vga console is excruciatingly slow, takes multiple seconds to go a single page in a 25x80 display.
 >

 There are a few workarounds:

 Userspace:
 ----------
 1. `export LESS=-c' or run `less -c' to that less(1) repaints rather than scrolls
 up from the bottom. (This also works around most, not all, delays in the
 `vt100' terminfo entry.)

 2. Don't use `-accel nvmm' :(. (TCG, the default accel doesn't seem to suffer
     from this.)

 Bootloader:
 -----------
 Force use of VESA:
 boot> vesa on

 Kernel:
 -------
 Scrolling up (erasure of 1st line, followed by moving the rest of the lines up)
 seems to do a backwards copy. Use plain memmove(3) instead:

 ```
 --- sys/dev/ic/vga.c.orig	2021-08-07 16:19:12.000000000 +0000
 +++ sys/dev/ic/vga.c	2025-05-20 04:02:12.133432513 +0000
 @@ -544,8 +544,8 @@
   	    &vh->vh_ioh_6845))
   		panic("vga_init: couldn't map 6845 io");

 -	if (bus_space_map(vh->vh_memt, 0xa0000, 0x20000, 
 -	    BUS_SPACE_MAP_CACHEABLE, &vh->vh_allmemh))
 +	if (bus_space_map(vh->vh_memt, 0xa0000, 0x20000,
 +	    BUS_SPACE_MAP_PREFETCHABLE | BUS_SPACE_MAP_LINEAR, &vh->vh_allmemh))
   		panic("vga_init: couldn't map memory");

   	if (bus_space_subregion(vh->vh_memt, vh->vh_allmemh,
 @@ -1168,10 +1168,15 @@
   				    scr->pcs.cursorrow, scr->pcs.cursorcol);
   #endif
   		} else {
 +			bus_space_handle_t dst = memh + scr->pcs.dispoffset + dstoff * 2;
 +			bus_space_handle_t src = memh + scr->pcs.dispoffset + srcoff * 2;
 +			memmove((void* )dst, (void* )src, nrows * ncols * 2);
 +#if 0
   			bus_space_copy_region_2(memt, memh,
   			    scr->pcs.dispoffset + srcoff * 2,
   			    memh, scr->pcs.dispoffset + dstoff * 2,
   			    nrows * ncols);
 +#endif
   		}
   	} else
   		memcpy(&scr->pcs.mem[dstoff], &scr->pcs.mem[srcoff],
 ```

 HTH,

 -RVP

From: Taylor R Campbell <riastradh@NetBSD.org>
To: RVP <rvp@SDF.ORG>
Cc: gnats-bugs@NetBSD.org, netbsd-bugs@NetBSD.org
Subject: Re: port-amd64/59361: paging up in less(1) under qemu vga console
	is slow
Date: Sun, 26 Oct 2025 23:55:09 +0000

 > Date: Sun, 26 Oct 2025 22:35:24 +0000 (UTC)
 > From: RVP <rvp@SDF.ORG>
 >=20
 > Scrolling up (erasure of 1st line, followed by moving the rest of the lin=
 es up)
 > seems to do a backwards copy. Use plain memmove(3) instead:
 > =20
 > @@ -1168,10 +1168,15 @@
 >   				    scr->pcs.cursorrow, scr->pcs.cursorcol);
 >   #endif
 >   		} else {
 > +			bus_space_handle_t dst =3D memh + scr->pcs.dispoffset + dstoff * 2;
 > +			bus_space_handle_t src =3D memh + scr->pcs.dispoffset + srcoff * 2;
 > +			memmove((void* )dst, (void* )src, nrows * ncols * 2);
 > +#if 0
 >   			bus_space_copy_region_2(memt, memh,
 >   			    scr->pcs.dispoffset + srcoff * 2,
 >   			    memh, scr->pcs.dispoffset + dstoff * 2,
 >   			    nrows * ncols);
 > +#endif

 Thanks, I bet bus_space_copy_region_2 is slow under nvmm because it
 has to do a VM exit for every pair of bytes copied, and VM exits are
 costly:

     811 	if (x86_bus_space_is_io(t)) {
     812 		if (addr1 >=3D addr2) {
     813 			/* src after dest: copy forward */
     814 			for (; c !=3D 0; c--, addr1 +=3D 2, addr2 +=3D 2)
     815 				outw(addr2, inw(addr1));
     816 		} else {
     817 			/* dest after src: copy backwards */
     818 			for (addr1 +=3D 2 * (c - 1), addr2 +=3D 2 * (c - 1);
     819 			    c !=3D 0; c--, addr1 -=3D 2, addr2 -=3D 2)
     820 				outw(addr2, inw(addr1));
     821 		}

 https://nxr.netbsd.org/xref/src/sys/arch/x86/x86/bus_space.c?r=3D1.47#811

 I don't think there's any I/O port equivalent of REP MOV, so there
 might not be a better way to do this properly (and still support VGA
 in I/O space rather than memory-mapped, as well as machines where
 access must be 2-bytes-at-a-time) other than a larger temporary
 buffer.

Home	PR Database Search