Re: kern/57981: ld@virtio confuses F_SEG_MAX negotiation

[Tetsuya Isaki <isaki%pastel-flower.jp@localhost>]

The following reply was made to PR kern/57981; it has been noted by GNATS.

From: Tetsuya Isaki <isaki%pastel-flower.jp@localhost>
To: Taylor R Campbell <riastradh%NetBSD.org@localhost>
Cc: gnats-bugs%NetBSD.org@localhost, netbsd-bugs%NetBSD.org@localhost
Subject: Re: kern/57981: ld@virtio confuses F_SEG_MAX negotiation
Date: Sun, 10 Mar 2024 13:37:55 +0900

 At Sat, 9 Mar 2024 13:27:44 +0000,
 Taylor R Campbell wrote:
 > Does this affect netbsd-10 or earlier?  Should it be pulled up?
 
 It has been since 2015 (since netbsd-8).
 But this will not occur under normal conditions.
 So I think it's not necessary to pullup (in a hurry).
 
 Details:
 In the previous code, if the maxnsegs read from a virtio device
 at line 320 is less than 2, it falls back maxnsegs to larger
 value and go.  (By the way, I could not find any description about
 it in the virtio spec 1.0, but strictly speaking, it was a protocol
 violation, I think...)
 
 When I wrote this PR, I searched which command line option can
 specify the SEG_MAX value in QEMU a little.  But I could not find,
 so I could not fill 'how-to-repeat' field.
 If QEMU has no such option, it will never happen on QEMU.
 (It looks to use 254 (=256-2) as the SEG_MAX)
 
 I'm writing a virtio device on my emulator now.  So I could make
 it happen intentionally by rewriting the emulator's source code.
 But such small SEG_MAX like 1 or 2 is useless for our ld driver.
 We usually don't do it, too.
 In result, it will only happen on useless devices..
 
 Thanks,
 ---
 Tetsuya Isaki <isaki%pastel-flower.jp@localhost / isaki%NetBSD.org@localhost>