linux-kernel - About configuring the PCIe device MPS

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: <3d7bc589-76ca-f557-c18b-2c5a47969d68@broadcom.com>
Date:   Thu, 20 Oct 2016 15:54:04 -0700
From:   Ray Jui <ray.jui@...adcom.com>
To:     Bjorn Helgaas <helgaas@...nel.org>
Cc:     alex.barba@...adcom.com,
        BCM Kernel Feedback <bcm-kernel-feedback-list@...adcom.com>,
        linux-kernel@...r.kernel.org, linux-pci@...r.kernel.org
Subject: About configuring the PCIe device MPS

Hi Bjorn,

Alex Barba (CCed in the email thread) from Broadcom discovered that PCIe 
endpoint devices attached to our ARM64 based iProc platforms are not 
configured to their optimal MPS size. Through some digging in the PCIe 
stack of the kernel, I found that:

1. For ARM32 based PCIe core code, 'pcie_bus_configure_settings' is 
called from 'pci_common_init_dev' before the devices are added to be bus

2. For ARM64 based PCIe core code, 'pcie_bus_configure_settings' is 
called similarly in 'pci_acpi_scan_root' for ACPI based ARM64 platform

3. But for ARM64 platforms that have not yet supported ACPI, there does 
not appear to be a common place for this configuration. Obviously, we 
have the option of calling it in our iProc PCIe host driver (right 
before devices are added to the bus). I'd like to check with you if 
that's the expected way of handling it. We do see a couple PCIe host 
drivers already handle it this way.

Thanks,

Ray