Attention is currently required from: Matt DeVillier, Tim Wawrzynczak, Angel Pons, Patrick Rudolph. Hello build bot (Jenkins), Nico Huber, Patrick Rudolph, Matt DeVillier, Benjamin Doron, Tim Wawrzynczak, Michael Niewöhner, Patrick Rudolph,
I'd like you to reexamine a change. Please visit
https://review.coreboot.org/c/coreboot/+/50754
to look at the new patch set (#15).
Change subject: soc/intel/common: Prevent SMI storm when setting SPI WPD bit ......................................................................
soc/intel/common: Prevent SMI storm when setting SPI WPD bit
From Skylake/Sunrise Point onwards, there are two BIOS_CNTL registers: one on the LPC/eSPI PCI device, and another on the SPI PCI device. When the WPD bit changes from 0 to 1 and the LE bit is set, the PCH raises a TCO SMI with the BIOSWR_STS bit set. However, the BIOSWR_STS bit is not set when the TCO SMI comes from the SPI or eSPI controller instead, but a status bit in the BIOS_CNTL register gets set. If the SMI cause is not handled, another SMI will happen immediately after returning from the SMI handler, which results in a deadlock.
Prevent deadlocks by clearing the SPI synchronous SMI status bit in the SMI handler. When SPI raises a synchronous SMI, the TCO_STS bit in the SMI_STS register is continously set until the SPI synchronous SMI status bit is cleared. To not risk missing any other TCO SMIs, do not clear the TCO_STS bit again in the same SMI handler invocation. If the TCO_STS bit remains set when returning from SMM, another SMI immediately happens and clears the TCO_STS bit, handling any pending events.
SPI can also generate asynchronous SMIs when the WPD bit is cleared and one attempts to write to flash using SPI hardware sequencing. This patch does not account for SPI asynchronous SMIs, because they are disabled by default and cannot be enabled once the BIOS Interface Lock-Down bit in the BIOS_CNTL register has been set, which coreboot already does. These asynchronous SMIs set the SPI_STS bit of the SMI_STS register. Clearing the SPI asynchronous SMI source should be done inside the SPI_STS SMI handler, which is currently not implemented. All of this goes out of the scope of this patch, and is currently not necessary anyway.
This patch does not handle eSPI because I cannot test it, and knowing if a board uses LPC or eSPI from common code is currently not possible, and this is beyond the scope of what this commit tries to achieve (fix SPI).
Tested on HP 280 G2, no longer deadlocks when SMM BIOS write protection is on. Write protection will be enforced in a follow-up.
Change-Id: Iec498674ae70f6590c33a6bf4967876268f2b0c8 Signed-off-by: Angel Pons th3fanbus@gmail.com --- M src/soc/intel/common/block/fast_spi/fast_spi.c M src/soc/intel/common/block/fast_spi/fast_spi_def.h M src/soc/intel/common/block/include/intelblocks/fast_spi.h M src/soc/intel/common/block/smm/smihandler.c 4 files changed, 31 insertions(+), 0 deletions(-)
git pull ssh://review.coreboot.org:29418/coreboot refs/changes/54/50754/15