hi , >I assume it's the BIOS's driver and it's choosing a range in low memory, >but still I'm not sure why its DMA is racing against the PAM update >(which is done very early). Felix, do you know the answer?
This bug is triggered by make_bios_readonly() in seabios. Make_bios_readonly() will write pam register. And ehci_setup() is executed before make_bios_readonly(). After initializing EHCI in seabios, it will continuously send dma cycles. So when write pam register, this bug appeared. In addition, before write pam registers, flatview_simplify() has merged a very large range.For example, this large range is 0xc0000-0xbfffffff. So even if EHCI is configured to not allocate buffers in low memory, this bug will still occur.Thanks. Best regards Felixcui-oc ________________________________ 发件人: Paolo Bonzini <pbonz...@redhat.com> 发送时间: 2020年10月22日 3:50:15 收件人: Alex Williamson 抄送: FelixCui-oc; Richard Henderson; Eduardo Habkost; qemu-devel@nongnu.org; RockCui-oc; Tony W Wang-oc; CobeChen-oc 主题: Re: [PATCH 1/1] Skip flatview_simplify() for cpu vendor zhaoxin On 21/10/20 20:49, Alex Williamson wrote: > I can understand the general benefit of flatview_simplify(), but I > wonder if the best short term solution is to skip operating on the x86 > PAM range, which I understand to be a small number of memory chunks > below 1MB. I'd rather remove flatview_simplify altogether, it probably triggers relatively rarely. Possibly do not let it operate on RAM/ROM regions, only on I/O regions. > I might also wonder why the EHCI controller on this > platform is choosing that range for DMA. I assume it's the BIOS's driver and it's choosing a range in low memory, but still I'm not sure why its DMA is racing against the PAM update (which is done very early). Felix, do you know the answer? Paolo
