| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-ID: <36B40EDC960C005C+20250217070120.666907-1-wangyuli@uniontech.com>
Date: Mon, 17 Feb 2025 15:01:20 +0800
From: WangYuli <wangyuli@...ontech.com>
To: cassel@...nel.org
Cc: chenhuacai@...nel.org,
chenhuacai@...ngson.cn,
dlemoal@...nel.org,
jiaxun.yang@...goat.com,
kernel@...0n.name,
linux-ide@...r.kernel.org,
linux-kernel@...r.kernel.org,
WangYuli <wangyuli@...ontech.com>,
Jie Fan <fanjie@...ontech.com>,
Erpeng Xu <xuerpeng@...ontech.com>
Subject: Re: Re: [PATCH] ahci: Marvell controllers prefer DMA for ATAPI
Reported-by: WangYuli <wangyuli@...ontech.com>
Hi all,
During testing, we observed that reading CDs on an INSPUR CE3000F desktop
computer (with a Phytium FT2000/4 processor [1]) running deepin 23 [2]
distribution is very slow, taking over a minute.
Further testing revealed that mounting the device from the command line is
fast.
e.g., executing "time mount /dev/sr0 /mnt" returns a value of only 0.898s.
However, deepin's dde-file-manager reads more than just mounting when opening
the optical drive; it also reads disc information (such as capacity, etc.).
After using udisksctl to first mount and then unmount the disc to reset its
state, and then using the cd_sessions tool to read disc information, we used
strace to monitor system calls, testing commands are as follows:
udisksctl mount -b /dev/sr0
udisksctl unmount -b /dev/sr0
strace -Ttt cd_sessions /dev/sr0
and found that openat was blocked for 42 seconds, and at this point, the
process had entered kernel space.
[1]. https://www.phytium.com.cn/homepage/production/6/
[2]. https://cdimage.deepin.com/releases/23/arm64/deepin-desktop-community-23-arm64.iso
TEST REPORT:
For this patch, before its introduction, with original system-integrated optical
drives capable of recognizing discs, the total execution time for cd_sessions was
92.4 seconds, including 31.6 seconds for openat and 60.6 seconds for ioctl.
With LITE-ON optical drives, under the same condition of disc recognition, the
total execution time for cd_sessions was 108.2 seconds, with openat taking 45.4
seconds and ioctl taking 62.9 seconds.
With RITEK optical drives, also when discs were recognizable, the total
cd_sessions execution time was 98.2 seconds, with openat at 36.7 seconds and ioctl
at 60.8 seconds.
After the patch was introduced, the time to read disc information (using cd_sessions)
with all optical drives was reduced to less than 0.5 seconds.
I've tried different pairings of optical drives and SATA controllers, and the test
results were consistent across the board. The Marvell 88SE9215 controller consistently
emerges as the source of the problem, no matter the optical drive tested.
Devices lists are as follow:
CD-ROM DRIVER LIST:
1. original system-integrated optical drive
*-cdrom
description: DVD writer
product: DVD A DH16AFSH
vendor: ATAPI
physical id: 0
bus info: scsi@2:0.0.0
logical name: /dev/cdrom
logical name: /dev/sr0
version: DC2M
capabilities: removable audio cd-r cd-rw dvd dvd-r
configuration: ansiversion=5 status=ready
*-medium
physical id: 0
logical name: /dev/cdrom
2. a LITE-ON optical drive
*-cdrom
description: DVD writer
product: DVD-RW DH16AFSH
vendor: PLDS
physical id: 0
bus info: scsi@2:0.0.0
logical name: /dev/cdrom
logical name: /dev/sr0
version: DL3M
capabilities: removable audio cd-r cd-rw dvd dvd-r
configuration: ansiversion=5 status=nodisc
3. a RITEK optical drive
*-cdrom
description: DVD writer
product: DH-16AFSH SUPDRV
vendor: RIDATA
physical id: 0
bus info: scsi@2:0.0.0
logical name: /dev/cdrom
logical name: /dev/sr0
version: NWDN
capabilities: removable audio cd-r cd-rw dvd dvd-r
configuration: ansiversion=5 status=nodisc
SATA Controllers LIST:
1. INSPUR CE3000F
*-sata
description: SATA controller
product: 88SE9215 PCIe 2.0 x1 4-port SATA 6 Gb/s Controller
vendor: Marvell Technology Group Ltd.
physical id: 0
bus info: pci@...0:03:00.0
logical name: scsi2
logical name: scsi3
version: 11
width: 32 bits
clock: 33MHz
capabilities: sata pm msi pciexpress ahci_1.0 bus_master cap_list rom emulated
configuration: driver=ahci latency=0
resources: irq:43 ioport:2020(size=8) ioport:2030(size=4) ioport:2028(size=8) ioport:2034(size=4) ioport:2000(size=32) memory:58140000-581407ff memory:58100000-5813ffff
2. Maxsun MS-WS W680 D4
Hardware Class: storage
Model: "Intel SATA controller"
Vendor: pci 0x8086 "Intel Corporation"
Device: pci 0x7ae2
Revision: 0x11
Driver: "ahci"
Driver Modules: "ahci"
Memory Range: 0x82100000-0x82101fff (rw,non-prefetchable)
Memory Range: 0x82102800-0x821028ff (rw,non-prefetchable)
Memory Range: 0x82102000-0x821027ff (rw,non-prefetchable)
IRQ: 166 (no events)
Module Alias: "pci:v00008086d00007AE2sv00000000sd00000000bc01sc06i01"
Driver Info #0:
Driver Status: ahci is active
Driver Activation Cmd: "modprobe ahci"
Config Status: cfg=new, avail=yes, need=no, active=unknown
Tested-by: Jie Fan <fanjie@...ontech.com>
Tested-by: Erpeng Xu <xuerpeng@...ontech.com>
Tested-by: WangYuli <wangyuli@...ontech.com>
Currently, due to the limited hardware I have access to, I'm unsure if this is a Marvell
88SE9215's specific issue or a general Marvell SATA controller problem.
So, I think it's reasonable to add a quirk to handle this for the Marvell 88SE9215 now.
Thanks,
--
WangYuli
Powered by blists - more mailing lists