amiga-news.de - Articia S: Genesi and Hyperion gave information about the supposed "Bug"

05.Nov.2003
ANN Articia S: Genesi and Hyperion gave information about the supposed "Bug"
During a discussion at the thread "Pegasos Italia" at ANN Raquel Velasco and Bill Buck by Genesi as well as Ben Hermans by Hyperion gave detailed information about the supposed bug of the Articia S northbridge for the first time.

Genesi published during that discussion protocol with which the problem should be reproduceable:

The following problem is given with the current Articia S versions:

The CPU interface fails to make all needed PCI to Memory transactions transparent to the CPU. This leads to failed (missed) snoop responses from the CPU in case of a modified cache line. The result may be a data corruption. The problem can be verified by creating a huge DMA load to the system.

1) Create a Linux partition with reiserfs using the IDE channel
2) Create a Linux partition with reiserfs using a 1394 (firewire) drive
3) Create any Linux partition (any device possible)
4) Enable DMA for IDE using 'hdparm'
5) Copy some data ~1-2GB using 1) and 2)
6) unmount 1)+2)
7) run the 'reiserfsck' tool using an endless shell script for 1)+2)
8) Start a ftp of a huge file (>memory+swap) using 3) and calculate a checksum for that file using an endless shell script

Be sure that the L2 cache of the CPU is enabled during your tests. Watch the results.

Required Resources
1 - Red Hat Linux Distribution 7.x
2 - IDE Hard Drives
1 - PCI Firewire Card
1 - 1394 Firewire Drive

Notes:
Line 1) - One IDE Drive for the O/S
Line 2) - Needs PCI Firewire Card, 1394 Firewire Card and SW Drivers for same
Line 3) - Configure partition on additional IDE Drive

Requires 2 weeks minimum full time effort (probably 3 weeks) for someone with advanced Linux configuration setup and test expertise.

Hyperions answer to this is:

"This 'so called' problem is the direct result of the Articia S's ability to handle memory access by the CPU and PCI bus.

This behavior reduces latency but is unknown in the Wintel world which is why the Linux drivers don't cater for it.

In the Wintel world either the CPU or the PCI busmaster is allowed to write to memory, not both at the same time.

MAI has been reworking the Linux drivers to cope with this and Hyperion has been aware of this feature for quite some time and has taken it into account when doing DMA capable drivers for the AmigaOne hardware. [...]" (snx) (Translation: ub)

[News message: 05. Nov. 2003, 10:53] [Comments: 0]
[Send via e-mail] [Print version] [ASCII version]

< Next message

Prior message >