From: Jerome Brunet jbrunet@baylibre.com
[ Upstream commit 5ad865862a0fd349163243e1834ed98ba9b81905 ]
The NTB epf host driver assumes the BAR number associated with a memory window is just incremented from the BAR number associated with MW1. This seems to have been enough so far but this is not really how the endpoint side work and the two could easily become mis-aligned.
ntb_epf_mw_to_bar() even assumes that the BAR number is the memory window index + 2, which means the function only returns a proper result if BAR_2 is associated with MW1.
Instead, fully describe and allow arbitrary NTB BAR mapping.
Signed-off-by: Jerome Brunet jbrunet@baylibre.com Signed-off-by: Jon Mason jdmason@kudzu.us Signed-off-by: Sasha Levin sashal@kernel.org ---
LLM Generated explanations, may be completely bogus:
YES - The patch fixes a real bug: `ntb_epf_mw_to_bar()` no longer hard-codes `idx + 2` but queries the per-platform BAR map (`drivers/ntb/hw/epf/ntb_hw_epf.c:110`), so code that computes MW alignment and size now points at the actual resource instead of the doorbell BAR (previously breaking i.MX8, whose MW1 lives at BAR4). - The BAR map itself is formalized via `barno_map` and explicit arrays for each supported platform (`drivers/ntb/hw/epf/ntb_hw_epf.c:52` and `drivers/ntb/hw/epf/ntb_hw_epf.c:578`), eliminating the prior “MW1 must be BAR2 and everything is sequential” assumption that caused host/endpoint mismatches in real deployments. - Memory-window helpers now reuse that mapping and bail out cleanly if the endpoint reports a window without a backing BAR (`drivers/ntb/hw/epf/ntb_hw_epf.c:141`, `drivers/ntb/hw/epf/ntb_hw_epf.c:413`), and the new guard on `mw_count` prevents overrunning the table (`drivers/ntb/hw/epf/ntb_hw_epf.c:451`). Together these changes let the existing NTB EPF hardware actually expose its windows safely. - Risk is contained: the refactor stays inside the NTB EPF host driver, keeps the public NTB API untouched, and the new data is static per- device, so stable backport fallout should be low compared to the very real malfunction the current code has on non-TI layouts.
Given it repairs functionality that never worked correctly for at least the NXP i.MX8 configuration, this is a good candidate for stable.
drivers/ntb/hw/epf/ntb_hw_epf.c | 103 ++++++++++++++++---------------- 1 file changed, 53 insertions(+), 50 deletions(-)
diff --git a/drivers/ntb/hw/epf/ntb_hw_epf.c b/drivers/ntb/hw/epf/ntb_hw_epf.c index 00f0e78f685bf..2b51156e01b0f 100644 --- a/drivers/ntb/hw/epf/ntb_hw_epf.c +++ b/drivers/ntb/hw/epf/ntb_hw_epf.c @@ -49,6 +49,7 @@ #define NTB_EPF_COMMAND_TIMEOUT 1000 /* 1 Sec */
enum pci_barno { + NO_BAR = -1, BAR_0, BAR_1, BAR_2, @@ -57,16 +58,26 @@ enum pci_barno { BAR_5, };
+enum epf_ntb_bar { + BAR_CONFIG, + BAR_PEER_SPAD, + BAR_DB, + BAR_MW1, + BAR_MW2, + BAR_MW3, + BAR_MW4, + NTB_BAR_NUM, +}; + +#define NTB_EPF_MAX_MW_COUNT (NTB_BAR_NUM - BAR_MW1) + struct ntb_epf_dev { struct ntb_dev ntb; struct device *dev; /* Mutex to protect providing commands to NTB EPF */ struct mutex cmd_lock;
- enum pci_barno ctrl_reg_bar; - enum pci_barno peer_spad_reg_bar; - enum pci_barno db_reg_bar; - enum pci_barno mw_bar; + const enum pci_barno *barno_map;
unsigned int mw_count; unsigned int spad_count; @@ -85,17 +96,6 @@ struct ntb_epf_dev {
#define ntb_ndev(__ntb) container_of(__ntb, struct ntb_epf_dev, ntb)
-struct ntb_epf_data { - /* BAR that contains both control region and self spad region */ - enum pci_barno ctrl_reg_bar; - /* BAR that contains peer spad region */ - enum pci_barno peer_spad_reg_bar; - /* BAR that contains Doorbell region and Memory window '1' */ - enum pci_barno db_reg_bar; - /* BAR that contains memory windows*/ - enum pci_barno mw_bar; -}; - static int ntb_epf_send_command(struct ntb_epf_dev *ndev, u32 command, u32 argument) { @@ -144,7 +144,7 @@ static int ntb_epf_mw_to_bar(struct ntb_epf_dev *ndev, int idx) return -EINVAL; }
- return idx + 2; + return ndev->barno_map[BAR_MW1 + idx]; }
static int ntb_epf_mw_count(struct ntb_dev *ntb, int pidx) @@ -413,7 +413,9 @@ static int ntb_epf_mw_set_trans(struct ntb_dev *ntb, int pidx, int idx, return -EINVAL; }
- bar = idx + ndev->mw_bar; + bar = ntb_epf_mw_to_bar(ndev, idx); + if (bar < 0) + return bar;
mw_size = pci_resource_len(ntb->pdev, bar);
@@ -455,7 +457,9 @@ static int ntb_epf_peer_mw_get_addr(struct ntb_dev *ntb, int idx, if (idx == 0) offset = readl(ndev->ctrl_reg + NTB_EPF_MW1_OFFSET);
- bar = idx + ndev->mw_bar; + bar = ntb_epf_mw_to_bar(ndev, idx); + if (bar < 0) + return bar;
if (base) *base = pci_resource_start(ndev->ntb.pdev, bar) + offset; @@ -560,6 +564,11 @@ static int ntb_epf_init_dev(struct ntb_epf_dev *ndev) ndev->mw_count = readl(ndev->ctrl_reg + NTB_EPF_MW_COUNT); ndev->spad_count = readl(ndev->ctrl_reg + NTB_EPF_SPAD_COUNT);
+ if (ndev->mw_count > NTB_EPF_MAX_MW_COUNT) { + dev_err(dev, "Unsupported MW count: %u\n", ndev->mw_count); + return -EINVAL; + } + return 0; }
@@ -596,14 +605,15 @@ static int ntb_epf_init_pci(struct ntb_epf_dev *ndev, dev_warn(&pdev->dev, "Cannot DMA highmem\n"); }
- ndev->ctrl_reg = pci_iomap(pdev, ndev->ctrl_reg_bar, 0); + ndev->ctrl_reg = pci_iomap(pdev, ndev->barno_map[BAR_CONFIG], 0); if (!ndev->ctrl_reg) { ret = -EIO; goto err_pci_regions; }
- if (ndev->peer_spad_reg_bar) { - ndev->peer_spad_reg = pci_iomap(pdev, ndev->peer_spad_reg_bar, 0); + if (ndev->barno_map[BAR_PEER_SPAD] != ndev->barno_map[BAR_CONFIG]) { + ndev->peer_spad_reg = pci_iomap(pdev, + ndev->barno_map[BAR_PEER_SPAD], 0); if (!ndev->peer_spad_reg) { ret = -EIO; goto err_pci_regions; @@ -614,7 +624,7 @@ static int ntb_epf_init_pci(struct ntb_epf_dev *ndev, ndev->peer_spad_reg = ndev->ctrl_reg + spad_off + spad_sz; }
- ndev->db_reg = pci_iomap(pdev, ndev->db_reg_bar, 0); + ndev->db_reg = pci_iomap(pdev, ndev->barno_map[BAR_DB], 0); if (!ndev->db_reg) { ret = -EIO; goto err_pci_regions; @@ -659,12 +669,7 @@ static void ntb_epf_cleanup_isr(struct ntb_epf_dev *ndev) static int ntb_epf_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id) { - enum pci_barno peer_spad_reg_bar = BAR_1; - enum pci_barno ctrl_reg_bar = BAR_0; - enum pci_barno db_reg_bar = BAR_2; - enum pci_barno mw_bar = BAR_2; struct device *dev = &pdev->dev; - struct ntb_epf_data *data; struct ntb_epf_dev *ndev; int ret;
@@ -675,18 +680,10 @@ static int ntb_epf_pci_probe(struct pci_dev *pdev, if (!ndev) return -ENOMEM;
- data = (struct ntb_epf_data *)id->driver_data; - if (data) { - peer_spad_reg_bar = data->peer_spad_reg_bar; - ctrl_reg_bar = data->ctrl_reg_bar; - db_reg_bar = data->db_reg_bar; - mw_bar = data->mw_bar; - } + ndev->barno_map = (const enum pci_barno *)id->driver_data; + if (!ndev->barno_map) + return -EINVAL;
- ndev->peer_spad_reg_bar = peer_spad_reg_bar; - ndev->ctrl_reg_bar = ctrl_reg_bar; - ndev->db_reg_bar = db_reg_bar; - ndev->mw_bar = mw_bar; ndev->dev = dev;
ntb_epf_init_struct(ndev, pdev); @@ -730,30 +727,36 @@ static void ntb_epf_pci_remove(struct pci_dev *pdev) ntb_epf_deinit_pci(ndev); }
-static const struct ntb_epf_data j721e_data = { - .ctrl_reg_bar = BAR_0, - .peer_spad_reg_bar = BAR_1, - .db_reg_bar = BAR_2, - .mw_bar = BAR_2, +static const enum pci_barno j721e_map[NTB_BAR_NUM] = { + [BAR_CONFIG] = BAR_0, + [BAR_PEER_SPAD] = BAR_1, + [BAR_DB] = BAR_2, + [BAR_MW1] = BAR_2, + [BAR_MW2] = BAR_3, + [BAR_MW3] = BAR_4, + [BAR_MW4] = BAR_5 };
-static const struct ntb_epf_data mx8_data = { - .ctrl_reg_bar = BAR_0, - .peer_spad_reg_bar = BAR_0, - .db_reg_bar = BAR_2, - .mw_bar = BAR_4, +static const enum pci_barno mx8_map[NTB_BAR_NUM] = { + [BAR_CONFIG] = BAR_0, + [BAR_PEER_SPAD] = BAR_0, + [BAR_DB] = BAR_2, + [BAR_MW1] = BAR_4, + [BAR_MW2] = BAR_5, + [BAR_MW3] = NO_BAR, + [BAR_MW4] = NO_BAR };
static const struct pci_device_id ntb_epf_pci_tbl[] = { { PCI_DEVICE(PCI_VENDOR_ID_TI, PCI_DEVICE_ID_TI_J721E), .class = PCI_CLASS_MEMORY_RAM << 8, .class_mask = 0xffff00, - .driver_data = (kernel_ulong_t)&j721e_data, + .driver_data = (kernel_ulong_t)j721e_map, }, { PCI_DEVICE(PCI_VENDOR_ID_FREESCALE, 0x0809), .class = PCI_CLASS_MEMORY_RAM << 8, .class_mask = 0xffff00, - .driver_data = (kernel_ulong_t)&mx8_data, + .driver_data = (kernel_ulong_t)mx8_map, }, { }, };