- Queries
- All Stories
- Search
- Advanced Search
- Transactions
- Transaction Logs
Advanced Search
Nov 25 2024
Nov 20 2024
Nov 19 2024
Nov 8 2024
I was running on a Neoverse-N2 CPU. I ran with your benchmark and received mixed results, some faster some slower.
Have you tried any other benchmarks? I tried testing against the glibc memcmp using their benchmarking tool as should be the same code as on FreeBSD and found this version was slower in most cases.
Nov 7 2024
Nov 4 2024
Does scmi_buf_get need to be updated with SCMI_MAX_MSG_REPLY_SIZE changing?
Nov 1 2024
How do you plan to handle the GICv3 ITS? It needs to use the same physical pages in both kernels. We already handle this when kexecing from Linux, but will need to pass the info needed to the rescue kernel.
Oct 31 2024
Oct 30 2024
It seems like an unneeded level of indirection in the non-KMSAN case
- Move kmsan calls to asm
- Add kmsan_check and kmsan_mark to arm_smccc_1_2_*
Oct 29 2024
Don't we need to set GPIO_PIN_INPUT for interrupts?
Oct 28 2024
Oct 25 2024
In D47283#1078259, @cperciva wrote:Should I be testing this on Graviton (with my _AEI patches minus the bogus resource activation one)?
I think the other drivers can use bus_generic_rman_activate_resource directly, however I can't easily test them.
We don't build the kernel with -fshort-enums, and even if we did then the assembly for arm and arm64 would be identical to the current code as the enum is passed in via a register to intr_irq_handler
See D47282 for teaching bus_generic_rman_activate_resource about INTRNG
We can assume the size of an enum on arm and arm64 as the ABI requires it to be either an int or unsigned int (unless it needs to hold values that are too large so can't be expressed as either of these). This means we can assume the size of this enum is 32-bit on arm and arm64. I assume riscv has a similar requirement.
Oct 24 2024
Oct 23 2024
Is there a reason to not do it in bus_generic_rman_activate_resource on those archs? We could then replace nexus_activate_resource with bus_generic_rman_activate_resource on arm and riscv (it's missing non-posted memory support needed on arm64)
Also check I and F flags before enabling interrupts
Remove __predict_true
The issue is we need to call into subr_intr.c to activate the interrupt. Currently this is done in nexus on arm, arm64, and riscv but this requires calling the parent bus_activate_resource in each bus driver.