@@ -196,11 +196,15 @@ static void xhci_pci_quirks(struct device *dev, struct xhci_hcd *xhci)
xhci->quirks |= XHCI_BROKEN_STREAMS;
}
if (pdev->vendor == PCI_VENDOR_ID_RENESAS &&
- pdev->device == 0x0014)
+ pdev->device == 0x0014) {
xhci->quirks |= XHCI_TRUST_TX_LENGTH;
+ xhci->quirks |= XHCI_ZERO_64B_REGS;
+ }
if (pdev->vendor == PCI_VENDOR_ID_RENESAS &&
- pdev->device == 0x0015)
+ pdev->device == 0x0015) {
xhci->quirks |= XHCI_RESET_ON_RESUME;
+ xhci->quirks |= XHCI_ZERO_64B_REGS;
+ }
if (pdev->vendor == PCI_VENDOR_ID_VIA)
xhci->quirks |= XHCI_RESET_ON_RESUME;
@@ -209,6 +209,68 @@ int xhci_reset(struct xhci_hcd *xhci)
return ret;
}
+static void xhci_zero_64b_regs(struct xhci_hcd *xhci)
+{
+ struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
+ int err, i;
+ u64 val;
+
+ /*
+ * Some Renesas controllers get into a weird state if they are
+ * reset while programmed with 64bit addresses (they will preserve
+ * the top half of the address in internal, non visible
+ * registers). You end up with half the address coming from the
+ * kernel, and the other half coming from the firmware. Also,
+ * changing the programming leads to extra accesses even if the
+ * controller is supposed to be halted. The controller ends up with
+ * a fatal fault, and is then ripe for being properly reset.
+ *
+ * Special care is taken to only apply this if the device is behind
+ * an iommu. Doing anything when there is no iommu is definitely
+ * unsafe...
+ */
+ if (!(xhci->quirks & XHCI_ZERO_64B_REGS) || !dev->iommu_group)
+ return;
+
+ xhci_info(xhci, "Zeroing 64bit base registers, expecting fault\n");
+
+ /* Clear HSEIE so that faults do not get signaled */
+ val = readl(&xhci->op_regs->command);
+ val &= ~CMD_HSEIE;
+ writel(val, &xhci->op_regs->command);
+
+ /* Clear HSE (aka FATAL) */
+ val = readl(&xhci->op_regs->status);
+ val |= STS_FATAL;
+ writel(val, &xhci->op_regs->status);
+
+ /* Now zero the registers, and brace for impact */
+ val = xhci_read_64(xhci, &xhci->op_regs->dcbaa_ptr);
+ if (upper_32_bits(val))
+ xhci_write_64(xhci, 0, &xhci->op_regs->dcbaa_ptr);
+ val = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
+ if (upper_32_bits(val))
+ xhci_write_64(xhci, 0, &xhci->op_regs->cmd_ring);
+
+ for (i = 0; i < HCS_MAX_INTRS(xhci->hcs_params1); i++) {
+ struct xhci_intr_reg __iomem *ir;
+
+ ir = &xhci->run_regs->ir_set[i];
+ val = xhci_read_64(xhci, &ir->erst_base);
+ if (upper_32_bits(val))
+ xhci_write_64(xhci, 0, &ir->erst_base);
+ val= xhci_read_64(xhci, &ir->erst_dequeue);
+ if (upper_32_bits(val))
+ xhci_write_64(xhci, 0, &ir->erst_dequeue);
+ }
+
+ /* Wait for the fault to appear. It will be cleared on reset */
+ err = xhci_handshake(&xhci->op_regs->status,
+ STS_FATAL, STS_FATAL,
+ XHCI_MAX_HALT_USEC);
+ if (!err)
+ xhci_info(xhci, "Fault detected\n");
+}
#ifdef CONFIG_USB_PCI
/*
@@ -1004,6 +1066,7 @@ int xhci_resume(struct xhci_hcd *xhci, bool hibernated)
xhci_dbg(xhci, "Stop HCD\n");
xhci_halt(xhci);
+ xhci_zero_64b_regs(xhci);
xhci_reset(xhci);
spin_unlock_irq(&xhci->lock);
xhci_cleanup_msix(xhci);
@@ -4921,6 +4984,8 @@ int xhci_gen_setup(struct usb_hcd *hcd, xhci_get_quirks_t get_quirks)
if (retval)
return retval;
+ xhci_zero_64b_regs(xhci);
+
xhci_dbg(xhci, "Resetting HCD\n");
/* Reset the internal HC memory state and registers. */
retval = xhci_reset(xhci);
@@ -1831,6 +1831,7 @@ struct xhci_hcd {
#define XHCI_HW_LPM_DISABLE BIT_ULL(29)
#define XHCI_SUSPEND_DELAY BIT_ULL(30)
#define XHCI_INTEL_USB_ROLE_SW BIT_ULL(31)
+#define XHCI_ZERO_64B_REGS BIT_ULL(32)
unsigned int num_active_eps;
unsigned int limit_active_eps;
Some Renesas controllers get into a weird state if they are reset while programmed with 64bit addresses (they will preserve the top half of the address in internal, non visible registers). You end up with half the address coming from the kernel, and the other half coming from the firmware. Also, changing the programming leads to extra accesses even if the controller is supposed to be halted. The controller ends up with a fatal fault, and is then ripe for being properly reset. On the flip side, this is completely unsafe if the defvice isn't behind an IOMMU, so we have to make sure that this is the case. Can you say "broken"? This is an alternative method to the one introduced in 8466489ef5ba ("xhci: Reset Renesas uPD72020x USB controller for 32-bit DMA issue"), which will subsequently be removed. Tested-by: Domenico Andreoli <domenico.andreoli@linux.com> Signed-off-by: Marc Zyngier <marc.zyngier@arm.com> --- drivers/usb/host/xhci-pci.c | 8 ++++-- drivers/usb/host/xhci.c | 65 +++++++++++++++++++++++++++++++++++++++++++++ drivers/usb/host/xhci.h | 1 + 3 files changed, 72 insertions(+), 2 deletions(-)