BernardMetzler
commented
7 years ago ...interesting side effect! right - will do so, even if it increases the stack size for the siw_tx_hdt() function. maybe i should consider better changing the array allocations in that function to dynamic kmalloc or keep the memory at the QP.
Thanks! Bernard.
From: Patrick MacArthur notifications@github.com To: zrlio/softiwarp softiwarp@noreply.github.com Cc: Subscribed subscribed@noreply.github.com Date: 05/04/2017 05:33 PM Subject: [zrlio/softiwarp] Regression: SIW_MAX_INLINE decreased from 72 to 64 (#13)
openmpi fails to run against softiwarp with default parameters, as commit f1a754b decreased the max inline limit from 72 to 64, because 8 bytes were removed from struct siw_sge. openmpi has the inline limit for softiwarp set to 72 in its configuration file, so unless the user specifically overrides this, openmpi will always fail trying to create a QP since the max_inline_data it passes is greater than the limit.
The easiest solution that I can see would be to increase SIW_MAX_SGE from 5 to 6, which would allow for 80 bytes of inline data.
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.
openmpi fails to run against softiwarp with default parameters, as commit f1a754b6d27b90b1aacd3912a023cb11dc1ba25a decreased the max inline limit from 72 to 64, because 8 bytes were removed from
struct siw_sge. openmpi has the inline limit for softiwarp set to 72 in its configuration file, so unless the user specifically overrides this, openmpi will always fail trying to create a QP since the max_inline_data it passes is greater than the limit.The easiest solution that I can see would be to increase SIW_MAX_SGE from 5 to 6, which would allow for 80 bytes of inline data.