diff options
| author | Subrata Banik <subratabanik@google.com> | 2024-03-16 18:06:02 +0530 |
|---|---|---|
| committer | Subrata Banik <subratabanik@google.com> | 2024-03-17 11:55:08 +0000 |
| commit | 9355f318faceee026abc2dd7b57f8927df286ff7 (patch) | |
| tree | 29198c189e93e5e196f9a42d2299effa5b5f97a8 /src/soc/intel/common/basecode/ramtop | |
| parent | a4c91e15f879eb2b6cf0984f97c6efdc6c8fdca8 (diff) | |
soc/intel/cmn/ramtop: Refactor MTRR handling for RAMTOP range
This patch refactors RAMTOP MTRR type selection to address a critical
NEM logic bug on SoCs with non-power-of-two cache sets. This bug can
cause runtime hangs when Write Back (WB) caching is enabled.
Workaround: Force MTRR type to WC (Write Combining) on affected SoCs
when the cache set count is not a power of two.
BUG=b:306677879
BRANCH=firmware-rex-15709.B
TEST=Verified boot on google/ovis and google/rex (including Ovis with
non-power-of-two cache configuration).
Change-Id: Ia9a8f0d37d581b05c19ea7f9b1a07933caa956d4
Signed-off-by: Subrata Banik <subratabanik@google.com>
Reviewed-on: https://review.coreboot.org/c/coreboot/+/81269
Reviewed-by: Stefan Reinauer <stefan.reinauer@coreboot.org>
Tested-by: build bot (Jenkins) <no-reply@coreboot.org>
Diffstat (limited to 'src/soc/intel/common/basecode/ramtop')
| -rw-r--r-- | src/soc/intel/common/basecode/ramtop/ramtop.c | 18 |
1 files changed, 17 insertions, 1 deletions
diff --git a/src/soc/intel/common/basecode/ramtop/ramtop.c b/src/soc/intel/common/basecode/ramtop/ramtop.c index ec326bb1c0..9cef9b1358 100644 --- a/src/soc/intel/common/basecode/ramtop/ramtop.c +++ b/src/soc/intel/common/basecode/ramtop/ramtop.c @@ -2,6 +2,7 @@ #include <commonlib/bsd/ipchksum.h> #include <console/console.h> +#include <cpu/cpu.h> #include <cpu/x86/mtrr.h> #include <intelbasecode/ramtop.h> #include <pc80/mc146818rtc.h> @@ -125,9 +126,24 @@ void early_ramtop_enable_cache_range(void) printk(BIOS_WARNING, "ramtop_table update failure due to no free MTRR available!\n"); return; } + + /* + * Background: Some SoCs have a critical bug inside the NEM logic which is responsible + * for mapping cached memory to physical memory during tear down and + * eventually malfunctions if the number of cache sets is not a power of two. + * This can lead to runtime hangs. + * + * Workaround: To mitigate this issue on affected SoCs, we force the MTRR type to + * WC (Write Combining) unless the cache set count is a power of two. + * This change alters caching behavior but prevents the runtime failures. + */ + unsigned int mtrr_type = MTRR_TYPE_WRCOMB; /* * We need to make sure late romstage (including FSP-M post mem) will be run * cached. Caching 16MB below ramtop is a safe to cover late romstage. */ - set_var_mtrr(mtrr, ramtop - 16 * MiB, 16 * MiB, MTRR_TYPE_WRBACK); + if (is_cache_sets_power_of_two()) + mtrr_type = MTRR_TYPE_WRBACK; + + set_var_mtrr(mtrr, ramtop - 16 * MiB, 16 * MiB, mtrr_type); } |