Patchwork [v2] PCI: Fixup the RTIT_BAR of Intel TH on Denverton

login
register
mail settings
Submitter Alexander Shishkin
Date Feb. 7, 2019, 1:30 p.m.
Message ID <20190207133005.67328-1-alexander.shishkin@linux.intel.com>
Download mbox | patch
Permalink /patch/720629/
State New
Headers show

Comments

Alexander Shishkin - Feb. 7, 2019, 1:30 p.m.
On Denverton's integration of the Intel(R) Trace Hub (for a reference
and overview see Documentation/trace/intel_th.txt) the reported size of
one of its resources (RTIT_BAR) doesn't match its actual size, which
leads to overlaps with other devices' resources.

In practice, it overlaps with XHCI MMIO space, which results in the xhci
driver bailing out after seeing its registers as 0xffffffff, and perceived
disappearance of all USB devices:

> intel_th_pci 0000:00:1f.7: enabling device (0004 -> 0006)
> xhci_hcd 0000:00:15.0: xHCI host controller not responding, assume dead
> xhci_hcd 0000:00:15.0: xHC not responding in xhci_irq, assume controller is dead
> xhci_hcd 0000:00:15.0: HC died; cleaning up
> usb 1-1: USB disconnect, device number 2
...

For this reason, we need to resize the RTIT_BAR on Denverton to its
actual size, which in this case is 4MB. The corresponding erratum is
DNV36 at the link below.

Link: https://www.intel.com/content/dam/www/public/us/en/documents/specification-updates/atom-c3000-family-spec-update.pdf
Fixes: 5118ccd34780 ("intel_th: pci: Add Denverton SOC support")
Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: stable@vger.kernel.org
---
 arch/x86/pci/fixup.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
Bjorn Helgaas - Feb. 7, 2019, 3:02 p.m.
On Thu, Feb 07, 2019 at 03:30:05PM +0200, Alexander Shishkin wrote:
> On Denverton's integration of the Intel(R) Trace Hub (for a reference
> and overview see Documentation/trace/intel_th.txt) the reported size of
> one of its resources (RTIT_BAR) doesn't match its actual size, which
> leads to overlaps with other devices' resources.
> 
> In practice, it overlaps with XHCI MMIO space, which results in the xhci
> driver bailing out after seeing its registers as 0xffffffff, and perceived
> disappearance of all USB devices:
> 
> > intel_th_pci 0000:00:1f.7: enabling device (0004 -> 0006)
> > xhci_hcd 0000:00:15.0: xHCI host controller not responding, assume dead
> > xhci_hcd 0000:00:15.0: xHC not responding in xhci_irq, assume controller is dead
> > xhci_hcd 0000:00:15.0: HC died; cleaning up
> > usb 1-1: USB disconnect, device number 2
> ...
> 
> For this reason, we need to resize the RTIT_BAR on Denverton to its
> actual size, which in this case is 4MB. The corresponding erratum is
> DNV36 at the link below.
> 
> Link: https://www.intel.com/content/dam/www/public/us/en/documents/specification-updates/atom-c3000-family-spec-update.pdf
> Fixes: 5118ccd34780 ("intel_th: pci: Add Denverton SOC support")
> Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
> Cc: stable@vger.kernel.org

I applied this with the changelog below to pci/enumeration for v5.1,
thanks!

commit 68071384ca08
Author: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Date:   Thu Feb 7 15:30:05 2019 +0200

    x86/PCI: Fixup RTIT_BAR of Intel Denverton Trace Hub
    
    On Denverton's integration of the Intel(R) Trace Hub (for a reference and
    overview see Documentation/trace/intel_th.txt) the reported size of one of
    its resources (RTIT_BAR) doesn't match its actual size, which leads to
    overlaps with other devices' resources.
    
    In practice, it overlaps with XHCI MMIO space, which results in the xhci
    driver bailing out after seeing its registers as 0xffffffff, and perceived
    disappearance of all USB devices:
    
      intel_th_pci 0000:00:1f.7: enabling device (0004 -> 0006)
      xhci_hcd 0000:00:15.0: xHCI host controller not responding, assume dead
      xhci_hcd 0000:00:15.0: xHC not responding in xhci_irq, assume controller is dead
      xhci_hcd 0000:00:15.0: HC died; cleaning up
      usb 1-1: USB disconnect, device number 2
    
    For this reason, we need to resize the RTIT_BAR on Denverton to its actual
    size, which in this case is 4MB.  The corresponding erratum is DNV36 at the
    link below:
    
      DNV36.       Processor Host Root Complex May Incorrectly Route Memory
                   Accesses to Intel® Trace Hub
    
      Problem:     The Intel® Trace Hub RTIT_BAR (B0:D31:F7 offset 20h) is
                   reported as a 2KB memory range.  Due to this erratum, the
                   processor Host Root Complex will forward addresses from
                   RTIT_BAR to RTIT_BAR + 4MB -1 to Intel® Trace Hub.
    
      Implication: Devices assigned within the RTIT_BAR to RTIT_BAR + 4MB -1
                   space may not function correctly.
    
      Workaround:  A BIOS code change has been identified and may be
                   implemented as a workaround for this erratum.
    
      Status:      No Fix.
    
    Note that 5118ccd34780 ("intel_th: pci: Add Denverton SOC support") updates
    the Trace Hub driver so it claims the Denverton device, but the resource
    overlap exists regardless of whether that driver is loaded or that commit
    is included.
    
    Link: https://www.intel.com/content/dam/www/public/us/en/documents/specification-updates/atom-c3000-family-spec-update.pdf
    Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
    [bhelgaas: include erratum text, clarify relationship with 5118ccd34780]
    Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
    Cc: stable@vger.kernel.org

> ---
>  arch/x86/pci/fixup.c | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
> 
> diff --git a/arch/x86/pci/fixup.c b/arch/x86/pci/fixup.c
> index 30a5111ae5fd..527e69b12002 100644
> --- a/arch/x86/pci/fixup.c
> +++ b/arch/x86/pci/fixup.c
> @@ -635,6 +635,22 @@ static void quirk_no_aersid(struct pci_dev *pdev)
>  DECLARE_PCI_FIXUP_CLASS_EARLY(PCI_VENDOR_ID_INTEL, PCI_ANY_ID,
>  			      PCI_CLASS_BRIDGE_PCI, 8, quirk_no_aersid);
>  
> +static void quirk_intel_th_dnv(struct pci_dev *dev)
> +{
> +	struct resource *r = &dev->resource[4];
> +
> +	/*
> +	 * Denverton reports 2k of RTIT_BAR (intel_th resource 4), which
> +	 * appears to be 4 MB in reality.
> +	 */
> +	if (r->end == r->start + 0x7ff) {
> +		r->start = 0;
> +		r->end   = 0x3fffff;
> +		r->flags |= IORESOURCE_UNSET;
> +	}
> +}
> +DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x19e1, quirk_intel_th_dnv);
> +
>  #ifdef CONFIG_PHYS_ADDR_T_64BIT
>  
>  #define AMD_141b_MMIO_BASE(x)	(0x80 + (x) * 0x8)
> -- 
> 2.20.1
>
Ingo Molnar - Feb. 11, 2019, 7:41 a.m.
* Alexander Shishkin <alexander.shishkin@linux.intel.com> wrote:

> On Denverton's integration of the Intel(R) Trace Hub (for a reference
> and overview see Documentation/trace/intel_th.txt) the reported size of
> one of its resources (RTIT_BAR) doesn't match its actual size, which
> leads to overlaps with other devices' resources.

Note that in the latest kernels it's Documentation/trace/intel_th.rst, 
not .txt.

Thanks,

	Ingo
Bjorn Helgaas - Feb. 11, 2019, 2:42 p.m.
On Mon, Feb 11, 2019 at 08:41:41AM +0100, Ingo Molnar wrote:
> 
> * Alexander Shishkin <alexander.shishkin@linux.intel.com> wrote:
> 
> > On Denverton's integration of the Intel(R) Trace Hub (for a reference
> > and overview see Documentation/trace/intel_th.txt) the reported size of
> > one of its resources (RTIT_BAR) doesn't match its actual size, which
> > leads to overlaps with other devices' resources.
> 
> Note that in the latest kernels it's Documentation/trace/intel_th.rst, 
> not .txt.

I fixed this, thanks, Ingo!
Bjorn Helgaas - Feb. 11, 2019, 8:20 p.m.
On Mon, Feb 11, 2019 at 11:26 AM Sasha Levin <sashal@kernel.org> wrote:
>
> Hi,
>
> [This is an automated email]
>
> This commit has been processed because it contains a "Fixes:" tag,
> fixing commit: 5118ccd34780 intel_th: pci: Add Denverton SOC support.
>
> The bot has tested the following trees: v4.20.7, v4.19.20, v4.14.98.
>
> v4.20.7: Build OK!
> v4.19.20: Build OK!
> v4.14.98: Failed to apply! Possible dependencies:
>     fa564ad96366 ("x86/PCI: Enable a 64bit BAR on AMD Family 15h (Models 00-1f, 30-3f, 60-7f)")
>
>
> How should we proceed with this patch?

This patch should be applied to all stable kernels.  There is no
dependency on any other patch.  There will be trivial conflicts on
kernels older than v4.15, but those are easy to resolve.

Bjorn

Patch

diff --git a/arch/x86/pci/fixup.c b/arch/x86/pci/fixup.c
index 30a5111ae5fd..527e69b12002 100644
--- a/arch/x86/pci/fixup.c
+++ b/arch/x86/pci/fixup.c
@@ -635,6 +635,22 @@  static void quirk_no_aersid(struct pci_dev *pdev)
 DECLARE_PCI_FIXUP_CLASS_EARLY(PCI_VENDOR_ID_INTEL, PCI_ANY_ID,
 			      PCI_CLASS_BRIDGE_PCI, 8, quirk_no_aersid);
 
+static void quirk_intel_th_dnv(struct pci_dev *dev)
+{
+	struct resource *r = &dev->resource[4];
+
+	/*
+	 * Denverton reports 2k of RTIT_BAR (intel_th resource 4), which
+	 * appears to be 4 MB in reality.
+	 */
+	if (r->end == r->start + 0x7ff) {
+		r->start = 0;
+		r->end   = 0x3fffff;
+		r->flags |= IORESOURCE_UNSET;
+	}
+}
+DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_INTEL, 0x19e1, quirk_intel_th_dnv);
+
 #ifdef CONFIG_PHYS_ADDR_T_64BIT
 
 #define AMD_141b_MMIO_BASE(x)	(0x80 + (x) * 0x8)