[BUG] bug in mmap_kmem

[BUG] bug in mmap_kmem

[Yoav Etsion]

Hi,

The problem is that mmaping a segment <addr,len> from /dev/kmem gives
differnet results than reading the same <addr,len>.

It seems that the bug is that mmap_kmem is a macro for mmap_mem (for
/dev/mem).
mmap_mem maps a _physical_ offset into a vm area vma, while mmap_kmem
should map a _virtual_ offset into a vm area. 

I hacked it to work by copying mmap_mem to mmap_kmem and adding __pa() in
the proper assginment, but I don't know how to check if the offset is
valid in the kernel virtual memory. Maybe someone who knows the mm code
better can fix this?

Thanks,

Yoav Etsion

Scsi

[Dead2]

Attached below: original mail to the 'linux-scsi' list
Please read attachment first.

It seems that the Davicom NIC's is the problem in this system.
I have tried to swap them with new davicom cards, and tried
with only one of them.

Davicom DM9102AF is the chip on all the cards.
Adaptec SCSI 19160 Ultra160, 68/50pin  (7892B, rev2)

I changed the Davicom's with Intel cards (same IRQ), and now it works
flawlessly. I think this might be an error worth looking at..

Hans K. Rosbach  aka. Dead2
CircleStorm Productions




-START ATTACHMENT-

Hi, i'm getting the following errors constantly while accessing my scsi disk
(10x per second or more)

"Kernel: scsi0: PCI error Interrupt at seqaddr = 0x*"
"Kernel: scsi0: Data Parity Error Detected during address or write data
phase"
(where * is a number, most often 8 or 9)

I'm running Kernel 2.4.6pre1 (also got it with 2.4.5)
Worked fine with the distro's kernel 2.4.0

Hardware:
  Adaptec SCSI 19160 Ultra160, 68/50pin  (7892B, rev2)
  MSI K7T-PRO2 motherboard (VIA chipset)
  Amd Duron 750Mhz cpu
  Winbond SDRAM (7ns)
  2x Davicom PCI 10/100 cards
  ATI 3D Rage IIc AGP

The computer is running SuSE 7.1 without X or other fancy
packages, as it is going to run a squid cache server.

The Scsi drivers are compiled into the kernel.

Hans K. Rosbach  aka. Dead2
CircleStorm Productions

-END ATTACHMENT-