[Skiboot] [PATCH v2 02/15] opal/hmi: Remove races in clearing HMER

[Mahesh J Salgaonkar]

From: Benjamin Herrenschmidt <benh at kernel.crashing.org>

Writing to HMER acts as an "AND". The current code writes back the
value we originally read with the bits we handled cleared. This is
racy, if a new bit gets set in HW after the original read, we'll end
up clearing it without handling it.

Instead, use an all 1's mask with only the bit handled cleared.

Signed-off-by: Benjamin Herrenschmidt <benh at kernel.crashing.org>
---
 core/hmi.c |   22 ++++++++++++----------
 1 file changed, 12 insertions(+), 10 deletions(-)

diff --git a/core/hmi.c b/core/hmi.c
index 8c100ade5..f4cdbd57f 100644
--- a/core/hmi.c
+++ b/core/hmi.c
@@ -1139,7 +1139,7 @@ int handle_hmi_exception(uint64_t hmer, struct OpalHMIEvent *hmi_evt)
 {
 	struct cpu_thread *cpu = this_cpu();
 	int recover = 1;
-	uint64_t tfmr;
+	uint64_t tfmr, handled = 0;
 
 	/*
 	 * In case of split core, some of the Timer facility errors need
@@ -1174,7 +1174,7 @@ int handle_hmi_exception(uint64_t hmer, struct OpalHMIEvent *hmi_evt)
 			}
 		}
 
-		hmer &= ~SPR_HMER_PROC_RECV_DONE;
+		handled |= SPR_HMER_PROC_RECV_DONE;
 		if (hmi_evt) {
 			hmi_evt->severity = OpalHMI_SEV_NO_ERROR;
 			hmi_evt->type = OpalHMI_ERROR_PROC_RECOV_DONE;
@@ -1182,7 +1182,7 @@ int handle_hmi_exception(uint64_t hmer, struct OpalHMIEvent *hmi_evt)
 		}
 	}
 	if (hmer & SPR_HMER_PROC_RECV_ERROR_MASKED) {
-		hmer &= ~SPR_HMER_PROC_RECV_ERROR_MASKED;
+		handled |= SPR_HMER_PROC_RECV_ERROR_MASKED;
 		if (hmi_evt) {
 			hmi_evt->severity = OpalHMI_SEV_NO_ERROR;
 			hmi_evt->type = OpalHMI_ERROR_PROC_RECOV_MASKED;
@@ -1191,7 +1191,7 @@ int handle_hmi_exception(uint64_t hmer, struct OpalHMIEvent *hmi_evt)
 		hmi_print_debug("Processor recovery Done (masked).", hmer);
 	}
 	if (hmer & SPR_HMER_PROC_RECV_AGAIN) {
-		hmer &= ~SPR_HMER_PROC_RECV_AGAIN;
+		handled |= SPR_HMER_PROC_RECV_AGAIN;
 		if (hmi_evt) {
 			hmi_evt->severity = OpalHMI_SEV_NO_ERROR;
 			hmi_evt->type = OpalHMI_ERROR_PROC_RECOV_DONE_AGAIN;
@@ -1202,7 +1202,7 @@ int handle_hmi_exception(uint64_t hmer, struct OpalHMIEvent *hmi_evt)
 	}
 	/* Assert if we see malfunction alert, we can not continue. */
 	if (hmer & SPR_HMER_MALFUNCTION_ALERT) {
-		hmer &= ~SPR_HMER_MALFUNCTION_ALERT;
+		handled |= SPR_HMER_MALFUNCTION_ALERT;
 
 		hmi_print_debug("Malfunction Alert", hmer);
 		if (hmi_evt)
@@ -1211,7 +1211,7 @@ int handle_hmi_exception(uint64_t hmer, struct OpalHMIEvent *hmi_evt)
 
 	/* Assert if we see Hypervisor resource error, we can not continue. */
 	if (hmer & SPR_HMER_HYP_RESOURCE_ERR) {
-		hmer &= ~SPR_HMER_HYP_RESOURCE_ERR;
+		handled |= SPR_HMER_HYP_RESOURCE_ERR;
 
 		hmi_print_debug("Hypervisor resource error", hmer);
 		recover = 0;
@@ -1228,10 +1228,10 @@ int handle_hmi_exception(uint64_t hmer, struct OpalHMIEvent *hmi_evt)
 	 */
 	if (hmer & SPR_HMER_TFAC_ERROR) {
 		tfmr = mfspr(SPR_TFMR);		/* save original TFMR */
+		handled |= SPR_HMER_TFAC_ERROR;
 
 		hmi_print_debug("Timer Facility Error", hmer);
 
-		hmer &= ~SPR_HMER_TFAC_ERROR;
 		recover = chiptod_recover_tb_errors();
 		if (hmi_evt) {
 			hmi_evt->severity = OpalHMI_SEV_ERROR_SYNC;
@@ -1242,7 +1242,7 @@ int handle_hmi_exception(uint64_t hmer, struct OpalHMIEvent *hmi_evt)
 	}
 	if (hmer & SPR_HMER_TFMR_PARITY_ERROR) {
 		tfmr = mfspr(SPR_TFMR);		/* save original TFMR */
-		hmer &= ~SPR_HMER_TFMR_PARITY_ERROR;
+		handled |= SPR_HMER_TFMR_PARITY_ERROR;
 
 		hmi_print_debug("TFMR parity Error", hmer);
 		recover = chiptod_recover_tb_errors();
@@ -1259,9 +1259,11 @@ int handle_hmi_exception(uint64_t hmer, struct OpalHMIEvent *hmi_evt)
 	/*
 	 * HMER bits are sticky, once set to 1 they remain set to 1 until
 	 * they are set to 0. Reset the error source bit to 0, otherwise
-	 * we keep getting HMI interrupt again and again.
+	 * we keep getting HMI interrupt again and again. Writing to HMER
+	 * acts as an AND, so we write mask of all 1's except for the bits
+	 * we want to clear.
 	 */
-	mtspr(SPR_HMER, hmer);
+	mtspr(SPR_HMER, ~handled);
 	hmi_exit();
 	/* Set the TB state looking at TFMR register before we head out. */
 	cpu->tb_invalid = !(mfspr(SPR_TFMR) & SPR_TFMR_TB_VALID);