In my opinion, changing the type to the conventional type used for socket_id seems like a much better solution than just changing the signedness, as proposed in another RFC [1]. (If we used more specialized types, like socket_id_t, we wouldn't even have this discussion. It is the DPDK convention to avoid specialized types, and I'm not against this convention; I'm only mentioning it to support changing the type here to int.)
And SOCKET_ID_ANY (-1) being used for multiple purposes, as discussed in the RFC, is another issue, to be discussed separately.
[1]: https://patches.dpdk.org/project/dpdk/patch/20230117101646.2521875-1-didier.pallard@6wind.com/
Acked-by: Morten Brørup
Acked-by: Kai Ji
________________________________
From: Power, Ciara
Sent: 29 June 2023 14:21
To: dev@dpdk.org
Cc: Ji, Kai ; Power, Ciara ; Matz, Olivier ; Akhil Goyal ; Fan Zhang
Subject: [PATCH] cryptodev: fix device socket ID type
The socket ID field for a cryptodev device data was unsigned int.
Due to recent changes to the default device socket ID,
this caused an issue when the socket ID was unknown and set to -1.
The device socket ID wrapped around to 255,
and caused errors when allocating memory.
Changing this field type to int fixes the issue, as it now takes the
correct -1 value.
Fixes: 7dcd73e37965 ("drivers/bus: set device NUMA node to unknown by default")
Cc: olivier.matz@6wind.com
Signed-off-by: Ciara Power
---
lib/cryptodev/cryptodev_pmd.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/lib/cryptodev/cryptodev_pmd.h b/lib/cryptodev/cryptodev_pmd.h
index 8710ed7558..4c98cedca6 100644
--- a/lib/cryptodev/cryptodev_pmd.h
+++ b/lib/cryptodev/cryptodev_pmd.h
@@ -65,7 +65,7 @@ struct rte_cryptodev_data {
/** Device ID for this instance */
uint8_t dev_id;
/** Socket ID where memory is allocated */
- uint8_t socket_id;
+ int socket_id;
/** Unique identifier name */
char name[RTE_CRYPTODEV_NAME_MAX_LEN];
--
2.25.1