linux-kernel - [PATCH] Extend PCIE_BUS_PEER2PEER to set MRSS=128 to fix CNS3xxx BM DMA.

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: <m3pouoqidz.fsf@t19.piap.pl>
Date:	Mon, 21 Mar 2016 10:39:52 +0100
From:	khalasa@...p.pl (Krzysztof Halasa)
To:	Bjorn Helgaas <bhelgaas@...gle.com>
Cc:	Arnd Bergmann <arnd@...db.de>, linux-pci@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: [PATCH] Extend PCIE_BUS_PEER2PEER to set MRSS=128 to fix CNS3xxx BM DMA.

The platform in question is Cavium CNS3xxx, ARMv6.

A recent patch by Arnd Bergmann (498a92d42596 "ARM: cns3xxx: pci: avoid
potential stack overflow") converted an explicit setting of
PCI_EXP_DEVCTL_READRQ = 0 (i.e., max 128 bytes for bus-mastering PCIe DMA
read request) to:
+    pcie_bus_config = PCIE_BUS_PEER2PEER;

with the following commentary:
    "The second part is how the driver sets up the Max_Read_Request_Size
    value for the first device/function on bus 1, i.e. the device
    plugged directly into the PCIe root port.
    For all I can tell, this is in fact incomplete, as it does not
    perform the same setting on devices attached to a PCIe switch,
    or multi-function devices.
    The solution for this part fortunately is even easier: if we
    just set the global pcie_bus_config variable to PCIE_BUS_PEER2PEER,
    all PCIe devices in the system are limited to 128 byte MPS, which
    in turn limits the MRRS to 128 bytes for all devices, and we
    no longer even need to touch any devices."

The problem is the MRRS setting is never written to the hardware.
I propose the following, though I'm not sure if we can do this safely,
especially given the comments in probe.c. OTOH, this change may be
required in other (all?) cases when the user requests
PCIE_BUS_PEER2PEER.

On this Laguna GW-2388 the following patch fixes BM DMA with:
0000:00:00.0 PCI bridge: Cavium Networks Device 3400 (rev 01)
0000:01:00.0 PCI bridge: Texas Instruments XIO2001 PCI Express-to-PCI Bridge
0000:02:0e.0 (PCI devices behind the bridge, these are doing actual BM xfers)
0001:00:00.0 PCI bridge: Cavium Networks Device 3400 (rev 01 - this is
	     the second lane from the CPU)

pci 0000:00:00.0: Max Payload Size set to  128/ 128 (was  128), Max Read Rq  128
pci 0000:01:00.0: Max Payload Size set to  128/ 512 (was  128), Max Read Rq  128
pci 0001:00:00.0: Max Payload Size set to  128/ 128 (was  128), Max Read Rq  128

Signed-off-by: Krzysztof Hałasa <khalasa@...p.pl>
Fixes: 498a92d42596 ("ARM: cns3xxx: pci: avoid potential stack overflow")

diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
index 6d7ab9b..91713b6 100644
--- a/drivers/pci/probe.c
+++ b/drivers/pci/probe.c
@@ -1919,7 +1919,8 @@ static void pcie_write_mrrs(struct pci_dev *dev)
 	/* In the "safe" case, do not configure the MRRS.  There appear to be
 	 * issues with setting MRRS to 0 on a number of devices.
 	 */
-	if (pcie_bus_config != PCIE_BUS_PERFORMANCE)
+	if (pcie_bus_config != PCIE_BUS_PERFORMANCE &&
+	    pcie_bus_config != PCIE_BUS_PEER2PEER)
 		return;
 
 	/* For Max performance, the MRRS must be set to the largest supported
diff --git a/include/linux/pci.h b/include/linux/pci.h
index 2771625..6f5088a 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -756,7 +756,7 @@ enum pcie_bus_config_types {
 	PCIE_BUS_DEFAULT,	/* ensure MPS matches upstream bridge */
 	PCIE_BUS_SAFE,		/* use largest MPS boot-time devices support */
 	PCIE_BUS_PERFORMANCE,	/* use MPS and MRRS for best performance */
-	PCIE_BUS_PEER2PEER,	/* set MPS = 128 for all devices */
+	PCIE_BUS_PEER2PEER,	/* set MPS and MRSS to 128 for all devices */
 };
 
 extern enum pcie_bus_config_types pcie_bus_config;