linux-kernel - [Intel IOMMU][patch 0/8] Intel IOMMU Support.

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

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-Id: <20070424060259.426374000@intel.com>
Date:	Mon, 23 Apr 2007 23:02:59 -0700
From:	Ashok Raj <ashok.raj@...el.com>
To:	linux-kernel@...r.kernel.org
Cc:	ashok.raj@...el.com
Subject: [Intel IOMMU][patch 0/8] Intel IOMMU Support.

Hello again!

Andrew: Could you help include in -mm to give it more exposure preparing for 
mainline inclusion with more testing?

This is a resend of the patches after addressing some feedback we received.

1. As len requested, we moved most of the acpi parts to drivers/pci instead
   of leaving them in drivers/acpi, including some renaming of functions,
   using just acpi_get_table() only.
2. Made the guard page support configurable.
3. Added new CONFIG option to allocate consecutive address instead of re-using 
   free addresses as an experimental option, Not validated extensively.
   Its expected to improve in certain cases...
4. Fixed a couple minor bugs that got exposed in testing.

Other feedback: 

- Some suggested depend on ACPI, but thats not doable for several reasons.
- Graphics 1-1 maps exist only for compatibility until graphics drivers
  start calling pci map functions, including user space X that might be
  using /dev/mem.

Some more interesting possibilities. -- enhancements to work on.

- In order to ensure we dont break any driver that may not be using dma api's
  here are some suggestions to work on.
    - Create a single 1-1 map, and make sure any pci device gets this map
      when they do a pci_set_master() to enable bus mastering automatically.
    - When the device driver does a first call to do a dma mapping, then
      dissociate the device from the unity map domain, to its own. This gives
      limited protection but doesnt break drivers that do not use dma 
      mapping.
    - Creating context entries only 1 per segment.. today we create one per 
      IOMMU which is not required.

This way we can avoid doing some of the workarounds we do for some devices
today, and will function as a default container for compatibility.

On one hand, this will provide more compatibility, but we will lose 
oppertunity to identify broken device drivers that dont use dma api's 
and fix them.... Depending on who you talk to.. some like it.. some just
hate it! and would like to fix the broken ones instead.

Cheers,
Ashok Raj
-- 
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/