diff --git a/repos/os/src/drivers/acpi/README b/repos/os/src/drivers/acpi/README
index b58f6ce1be..df59a8fedb 100644
--- a/repos/os/src/drivers/acpi/README
+++ b/repos/os/src/drivers/acpi/README
@@ -3,12 +3,12 @@ This directory contains an implementation of a simple ACPI parser
 Behavior
 --------
 
-This server should be used when using a kernel (like Fiasco.OC or Nova) that
-takes advantage of x86's APIC. The server traverses the ACPI tables and sets the
-interrupt line of devices within the PCI config space to the GSIs found in the
-ACPI tables. The 'MADT' table is parsed by the server as well, enabling clients
-to use the correct IRQ when 'Interrupt Override' structures are found wihtin the
-table.
+This server should be used when using a kernel (like Fiasco.OC, Nova,
+base_hw x86_64) that takes advantage of x86's APIC. The server traverses the
+ACPI tables and sets the interrupt line of devices within the PCI config space
+to the GSIs found in the ACPI tables. The 'MADT' table is parsed by the server
+as well, enabling clients to use the correct IRQ when 'Interrupt Override'
+structures are found wihtin the table.
 
 Usage
 -----
@@ -20,15 +20,18 @@ since this will be used as a slave of the 'acpi_drv'. You still must load the
 Configuration snipped (please note that IRQ service requests of the 'timer' are
 routed to the ACPI-driver):
 
-!<start name="acpi">
-!  <resource name="RAM" quantum="2M"/>
-!    <binary name="acpi_drv"/>
+!<start name="acpi_drv">
+!  <resource name="RAM" quantum="2M" constrain_phys="yes"/>
 !    <provides>
 !      <service name="PCI"/>
 !      <service name="IRQ" />
 !    </provides>
+!    <config>
+!      <policy label="acpi_drv">
+!        <pci class="ALL"/>
+!      </policy>
+!    </config>
 !    <route>
-!      <service name="PCI"> <any-child /> </service>
 !      <any-service> <parent/> <any-child /> </any-service>
 !    </route>
 !</start>
@@ -37,14 +40,15 @@ routed to the ACPI-driver):
 !  <resource name="RAM" quantum="1M"/>
 !  <provides><service name="Timer"/></provides>
 !  <route>
-!    <service name="IRQ"><child name="acpi" /></service>
+!    <service name="IRQ"><child name="acpi_drv" /></service>
 !    <any-service> <parent /> <any-child /></any-service>
 !  </route>
 !</start>
 
-Limitations and known issues
-----------------------------
-
-Currently there is no interface to set the interrupt mode for Irq_sessions
-(e.g., level or edge triggered). This is required by Fiasco.OCs kernel
-interface. We regard this as future work.
+The 'config' node on the 'acpi_drv' requires an policy entry that permits the
+'acpi_drv' to iterate over all pci devices in order to rewrite the
+interrupt numbers according to the 'MADT' table. Furthermore the whole 'config'
+node will be passed on to the 'pci_drv' which obtains from it the required
+policy for each client. Additionally the constrain_phys attribute is a
+feature of init, which applies to the pci_drv. Please read the 'pci_drv'
+README for more details.
diff --git a/repos/os/src/drivers/pci/README b/repos/os/src/drivers/pci/README
new file mode 100644
index 0000000000..9ea33e078b
--- /dev/null
+++ b/repos/os/src/drivers/pci/README
@@ -0,0 +1,118 @@
+This directory contains the implementation of Genode's PCI driver.
+
+Behavior
+--------
+
+On startup the driver scans the PCI bus hierarchy and stores the found devices.
+Per client a policy must be configured that states which client can
+access certain devices to form a virtual pci bus per client. The client may
+iterate through the virtual pci bus using the 'first' and 'next' methods of
+the pci_session interface to discover all available devices of the virtual bus.
+Non PCI devices may be discovered by using 'device' of the pci_session
+interface.  As a result of the discovery a client obtains a device capability.
+
+With the device capability the resources of the devices can be obtained, e.g.
+io_port, io_mem and irq of the pci_device interface.
+
+Policy usage
+------------
+
+A policy may contain several nodes describing several devices. The entries of
+a policy may describe PCI devices as non PCI devices. A PCI device is
+explicitly configured by the triple 'bus', 'device', 'function':
+
+!<start name="pci_drv">
+! <resource name="RAM" quantum="8M" constrain_phys="yes"/>
+! ...
+! <config>
+!  <policy label="usb_drv">
+!   <pci bus="0" device="19" function="0"/>
+!   <pci bus="0" device="18" function="3"/>
+!  </policy>
+! </config>
+!</start>
+
+or more fuzzy by a device class alias:
+
+!<start name="pci_drv">
+! <resource name="RAM" quantum="8M" constrain_phys="yes"/>
+! ...
+! <config>
+!  <policy label="usb_drv">
+!   <pci class="USB"/>
+!  </policy>
+! </config>
+!</start>
+
+
+Non PCI devices, as the PS2 controller are named by a "device" node in the policy:
+
+!<start name="pci_drv">
+! <resource name="RAM" quantum="8M" constrain_phys="yes"/>
+! <config>
+!  <policy label="ps_drv">
+!   <device name="PS2/>
+!  </policy>
+! </config>
+!</start>
+
+
+The first entry ('pci' or 'dev') of the policy node that matches will grant
+access of a device or device class to the client. Subsequent entries will not
+be checked. If a 'bus', 'device', 'function' triple was specified in one of the
+policies and in another policy a fuzzy pci class alias which would include
+the device specified by the triple, the device will not appear during device
+discovery by the client with the fuzzy pci class policy.
+
+
+By default the driver will try to use MSIs if the device and the used kernel
+supports it. This behaviour can be overwritten:
+
+!<start name="pci_drv">
+! <resource name="RAM" quantum="8M" constrain_phys="yes"/>
+! <config>
+!  <policy label="nic_drv" irq_mode="nomsi">
+!  ...
+!  </policy>
+! </config>
+!</start>
+
+The constrain_phys attribute is evaluated by init. If set to "yes" it
+permits a component, the pci driver, to restrict the allocation of memory to
+specific physical RAM ranges. The PCI driver uses this feature to ensure that
+the allocation of DMA capable memory consider several restrictions. For
+example, some drivers, as the UHCI controller, requires a
+physical memory address below 4G. Another example is that on 32bit hosts
+physical to virtual identical mappings of DMA memory for the device_pd
+(required when IOMMU is used) must be below the kernel memory boundary (3G).
+
+Supported PCI class aliases
+---------------------------
+
+The following class names are supported which corresponds to the
+specified PCI class(C), subclass(S) and programming interface(P):
+
+alias      C    S   P
+*********************
+ALL      0x0 0x00 0x0
+AHCI     0x1 0x06 0x0
+AUDIO    0x4 0x00 0x0
+ETHERNET 0x2 0x00 0x0
+USB      0xc 0x03 0x0
+VGA      0x3 0x00 0x0
+WIFI     0x2 0x80 0x0
+
+
+Supported non PCI devices
+-------------------------
+
+The driver provides for the PS2 controller the IO_PORT and IRQ resources.
+
+!<start name="pci_drv">
+! <resource name="RAM" quantum="8M" constrain_phys="yes"/>
+! <config>
+!  <policy label="ps_drv">
+!   <dev name="PS2/>
+!  </policy>
+! </config>
+!</start>