tpm_tis: fix stall after iowrite*()s

ioread8() operations to TPM MMIO addresses can stall the CPU when
immediately following a sequence of iowrite*()'s to the same region.

For example, cyclitest measures ~400us latency spikes when a non-RT
usermode application communicates with an SPI-based TPM chip (Intel Atom
E3940 system, PREEMPT_RT kernel). The spikes are caused by a
stalling ioread8() operation following a sequence of 30+ iowrite8()s to
the same address. I believe this happens because the write sequence is
buffered (in CPU or somewhere along the bus), and gets flushed on the
first LOAD instruction (ioread*()) that follows.

The enclosed change appears to fix this issue: read the TPM chip's
access register (status code) after every iowrite*() operation to
amortize the cost of flushing data to chip across multiple instructions.

Signed-off-by: Haris Okanovic <haris.okanovic@ni.com>
Link: https://lore.kernel.org/r/20230323153436.B2SATnZV@linutronix.de
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Tested-by: Jarkko Sakkinen <jarkko@kernel.org>
Reviewed-by: Jarkko Sakkinen <jarkko@kernel.org>
Signed-off-by: Jarkko Sakkinen <jarkko@kernel.org>

1 files changed, 41 insertions, 2 deletions

diff --git a/drivers/char/tpm/tpm_tis.c b/drivers/char/tpm/tpm_tis.c
index ba0402202bee3..7af3898066433 100644
--- a/drivers/char/tpm/tpm_tis.c
+++ b/drivers/char/tpm/tpm_tis.c
@@ -50,6 +50,45 @@ static inline struct tpm_tis_tcg_phy *to_tpm_tis_tcg_phy(struct tpm_tis_data *da
 	return container_of(data, struct tpm_tis_tcg_phy, priv);
 }
 
+#ifdef CONFIG_PREEMPT_RT
+/*
+ * Flush previous write operations with a dummy read operation to the
+ * TPM MMIO base address.
+ */
+static inline void tpm_tis_flush(void __iomem *iobase)
+{
+	ioread8(iobase + TPM_ACCESS(0));
+}
+#else
+#define tpm_tis_flush(iobase) do { } while (0)
+#endif
+
+/*
+ * Write a byte word to the TPM MMIO address, and flush the write queue.
+ * The flush ensures that the data is sent immediately over the bus and not
+ * aggregated with further requests and transferred later in a batch. The large
+ * write requests can lead to unwanted latency spikes by blocking the CPU until
+ * the complete batch has been transferred.
+ */
+static inline void tpm_tis_iowrite8(u8 b, void __iomem *iobase, u32 addr)
+{
+	iowrite8(b, iobase + addr);
+	tpm_tis_flush(iobase);
+}
+
+/*
+ * Write a 32-bit word to the TPM MMIO address, and flush the write queue.
+ * The flush ensures that the data is sent immediately over the bus and not
+ * aggregated with further requests and transferred later in a batch. The large
+ * write requests can lead to unwanted latency spikes by blocking the CPU until
+ * the complete batch has been transferred.
+ */
+static inline void tpm_tis_iowrite32(u32 b, void __iomem *iobase, u32 addr)
+{
+	iowrite32(b, iobase + addr);
+	tpm_tis_flush(iobase);
+}
+
 static int interrupts = -1;
 module_param(interrupts, int, 0444);
 MODULE_PARM_DESC(interrupts, "Enable interrupts");
@@ -186,12 +225,12 @@ static int tpm_tcg_write_bytes(struct tpm_tis_data *data, u32 addr, u16 len,
 	switch (io_mode) {
 	case TPM_TIS_PHYS_8:
 		while (len--)
-			iowrite8(*value++, phy->iobase + addr);
+			tpm_tis_iowrite8(*value++, phy->iobase, addr);
 		break;
 	case TPM_TIS_PHYS_16:
 		return -EINVAL;
 	case TPM_TIS_PHYS_32:
-		iowrite32(le32_to_cpu(*((__le32 *)value)), phy->iobase + addr);
+		tpm_tis_iowrite32(le32_to_cpu(*((__le32 *)value)), phy->iobase, addr);
 		break;
 	}