powerpc/powernv: Override dma_get_required_mask()

The dma_get_required_mask() function is used by some drivers to
query the platform about what DMA mask is needed to cover all of
memory. This is a bit of a strange semantic when we have to choose
between IOMMU translation or bypass, but essentially what it means
is "what DMA mask will give best performances".

Currently, our IOMMU backend always returns a 32-bit mask here, we
don't do anything special to it when we have bypass available. This
causes some drivers to choose a 32-bit mask, thus losing the ability
to use the bypass window, thinking this is more efficient. The problem
was reported from the driver of following device:

0004:03:00.0 0107: 1000:0087 (rev 05)
0004:03:00.0 Serial Attached SCSI controller: LSI Logic / Symbios \
Logic SAS2308 PCI-Express Fusion-MPT SAS-2 (rev 05)

This patch adds an override of that function in order to, instead,
return a 64-bit mask whenever a bypass window is available in order
for drivers to prefer this configuration.

Reported-by: Murali N. Iyer <[e-mail]>
Suggested-by: Benjamin Herrenschmidt <[e-mail]>
Signed-off-by: Gavin Shan <[e-mail]>
Signed-off-by: Michael Ellerman <[e-mail]>

arch/powerpc/include/asm/dma-mapping.h

···
 
 extern int dma_set_mask(struct device *dev, u64 dma_mask);
 extern int __dma_set_mask(struct device *dev, u64 dma_mask);
+extern u64 __dma_get_required_mask(struct device *dev);
 
 #define dma_alloc_coherent(d,s,h,f)	dma_alloc_attrs(d,s,h,f,NULL)
 

arch/powerpc/kernel/dma.c

···
 	*dev->dma_mask = dma_mask;
 	return 0;
 }
+
 int dma_set_mask(struct device *dev, u64 dma_mask)
 {
 	if (ppc_md.dma_set_mask)
···
 }
 EXPORT_SYMBOL(dma_set_mask);
 
-u64 dma_get_required_mask(struct device *dev)
+u64 __dma_get_required_mask(struct device *dev)
 {
 	struct dma_map_ops *dma_ops = get_dma_ops(dev);
-
-	if (ppc_md.dma_get_required_mask)
-		return ppc_md.dma_get_required_mask(dev);
 
 	if (unlikely(dma_ops == NULL))
 		return 0;
···
 		return dma_ops->get_required_mask(dev);
 
 	return DMA_BIT_MASK(8 * sizeof(dma_addr_t));
+}
+
+u64 dma_get_required_mask(struct device *dev)
+{
+	if (ppc_md.dma_get_required_mask)
+		return ppc_md.dma_get_required_mask(dev);
+
+	return __dma_get_required_mask(dev);
 }
 EXPORT_SYMBOL_GPL(dma_get_required_mask);
 

arch/powerpc/platforms/powernv/pci-ioda.c

···
 	return 0;
 }
 
+static u64 pnv_pci_ioda_dma_get_required_mask(struct pnv_phb *phb,
+					      struct pci_dev *pdev)
+{
+	struct pci_dn *pdn = pci_get_pdn(pdev);
+	struct pnv_ioda_pe *pe;
+	u64 end, mask;
+
+	if (WARN_ON(!pdn || pdn->pe_number == IODA_INVALID_PE))
+		return 0;
+
+	pe = &phb->ioda.pe_array[pdn->pe_number];
+	if (!pe->tce_bypass_enabled)
+		return __dma_get_required_mask(&pdev->dev);
+
+
+	end = pe->tce_bypass_base + memblock_end_of_DRAM();
+	mask = 1ULL << (fls64(end) - 1);
+	mask += mask - 1;
+
+	return mask;
+}
+
 static void pnv_ioda_setup_bus_dma(struct pnv_ioda_pe *pe,
 				   struct pci_bus *bus,
 				   bool add_to_iommu_group)
···
 	/* Setup TCEs */
 	phb->dma_dev_setup = pnv_pci_ioda_dma_dev_setup;
 	phb->dma_set_mask = pnv_pci_ioda_dma_set_mask;
+	phb->dma_get_required_mask = pnv_pci_ioda_dma_get_required_mask;
 
 	/* Setup shutdown function for kexec */
 	phb->shutdown = pnv_pci_ioda_shutdown;

arch/powerpc/platforms/powernv/pci.c

···
 	return __dma_set_mask(&pdev->dev, dma_mask);
 }
 
+u64 pnv_pci_dma_get_required_mask(struct pci_dev *pdev)
+{
+	struct pci_controller *hose = pci_bus_to_host(pdev->bus);
+	struct pnv_phb *phb = hose->private_data;
+
+	if (phb && phb->dma_get_required_mask)
+		return phb->dma_get_required_mask(phb, pdev);
+
+	return __dma_get_required_mask(&pdev->dev);
+}
+
 void pnv_pci_shutdown(void)
 {
 	struct pci_controller *hose;

arch/powerpc/platforms/powernv/pci.h

···
 	void (*dma_dev_setup)(struct pnv_phb *phb, struct pci_dev *pdev);
 	int (*dma_set_mask)(struct pnv_phb *phb, struct pci_dev *pdev,
 			    u64 dma_mask);
+	u64 (*dma_get_required_mask)(struct pnv_phb *phb,
+				     struct pci_dev *pdev);
 	void (*fixup_phb)(struct pci_controller *hose);
 	u32 (*bdfn_to_pe)(struct pnv_phb *phb, struct pci_bus *bus, u32 devfn);
 	void (*shutdown)(struct pnv_phb *phb);

arch/powerpc/platforms/powernv/powernv.h

···
 extern void pnv_pci_init(void);
 extern void pnv_pci_shutdown(void);
 extern int pnv_pci_dma_set_mask(struct pci_dev *pdev, u64 dma_mask);
+extern u64 pnv_pci_dma_get_required_mask(struct pci_dev *pdev);
 #else
 static inline void pnv_pci_init(void) { }
 static inline void pnv_pci_shutdown(void) { }
···
 static inline int pnv_pci_dma_set_mask(struct pci_dev *pdev, u64 dma_mask)
 {
 	return -ENODEV;
+}
+
+static inline u64 pnv_pci_dma_get_required_mask(struct pci_dev *pdev)
+{
+	return 0;
 }
 #endif
 

arch/powerpc/platforms/powernv/setup.c

···
 	return __dma_set_mask(dev, dma_mask);
 }
 
+static u64 pnv_dma_get_required_mask(struct device *dev)
+{
+	if (dev_is_pci(dev))
+		return pnv_pci_dma_get_required_mask(to_pci_dev(dev));
+
+	return __dma_get_required_mask(dev);
+}
+
 static void pnv_shutdown(void)
 {
 	/* Let the PCI code clear up IODA tables */
···
 	.power_save             = power7_idle,
 	.calibrate_decr		= generic_calibrate_decr,
 	.dma_set_mask		= pnv_dma_set_mask,
+	.dma_get_required_mask	= pnv_dma_get_required_mask,
 #ifdef CONFIG_KEXEC
 	.kexec_cpu_down		= pnv_kexec_cpu_down,
 #endif