Home
last modified time | relevance | path

Searched hist:"5 c2ffce1e9496477720966e70d79f2da3e4b84e6" (Results 1 – 2 of 2) sorted by relevance

/openbmc/linux/arch/um/os-Linux/
H A Dsignal.cdiff 5c2ffce1e9496477720966e70d79f2da3e4b84e6 Fri Jan 04 09:38:21 CST 2019 Anton Ivanov <anton.ivanov@cambridgegreys.com> um: Revert to using stack for pt_regs in signal handling

Reverts commit b6024b21fec8367ef961a771cc9dde31f1831965 and
adjusts default stack sizing to cope with larger size of
floating point save registers on the newer Intel CPUs.

b6024b21fec8367ef961a771cc9dde31f1831965 replaced storing the
register state on the stack with kmalloc-ed storage. That has
a number of issues and a panic if that fails.
1. kmalloc/ATOMIC can fail. There was a latent hard crash
in all interrupt and fault handling as a result.
2. kmalloc in the interrupt path introduces a considerable
performance penalty for networking ~ 14% on iperf.

This commit restores uml to a stable state until a better
solution is found.

Signed-off-by: Anton Ivanov <anton.ivanov@cambridgegreys.com>
Signed-off-by: Richard Weinberger <richard@nod.at>
/openbmc/linux/arch/um/
H A DKconfigdiff 5c2ffce1e9496477720966e70d79f2da3e4b84e6 Fri Jan 04 09:38:21 CST 2019 Anton Ivanov <anton.ivanov@cambridgegreys.com> um: Revert to using stack for pt_regs in signal handling

Reverts commit b6024b21fec8367ef961a771cc9dde31f1831965 and
adjusts default stack sizing to cope with larger size of
floating point save registers on the newer Intel CPUs.

b6024b21fec8367ef961a771cc9dde31f1831965 replaced storing the
register state on the stack with kmalloc-ed storage. That has
a number of issues and a panic if that fails.
1. kmalloc/ATOMIC can fail. There was a latent hard crash
in all interrupt and fault handling as a result.
2. kmalloc in the interrupt path introduces a considerable
performance penalty for networking ~ 14% on iperf.

This commit restores uml to a stable state until a better
solution is found.

Signed-off-by: Anton Ivanov <anton.ivanov@cambridgegreys.com>
Signed-off-by: Richard Weinberger <richard@nod.at>