aarch64: introduce user_rootpgt param to read user space memory #77
Conversation
Commit dfa510f (aarch64: Fix bottom VA range translation in Linux) introduced METH_UPGT for user-space bottom addresses that cover TTBR0_EL1 addresses. Unfortunately, it breaks the case where root pgt from user land process was passed and used to translate vmcore memory to get specified user land process virtual memory reads. libkdumpfile based gdbserver was using such functionality before: https://github.com/AlexanderKamensky/libkdumpfile/tree/new_20230805/gdbserver#process This patch introduces new keyword parameter user_rootpgt for System.os_init, that would be filled into METH_UPGT method param.pgt.root. As result any kernel or user land virtual addresses will be readable. Signed-off-by: Alexander Kamensky <[email protected]>
|
Hi Alexander! Thank you for your PR. I am trying to understand why a user-space page table hierarchy is needed on AArch64. The address translation library is based on the idea of address spaces, and AFAIK AArch64 has a single page table hierarchy in hardware, so one root page table should be sufficient to represent any existing address space. The concept of user-space page table root was added to cover the case where two page table hierarchies are used to cover a single address space. This happens with Linux on IBM POWER without hardware page tables. If the operating system implements page table isolation between kernel-space and user-space, libaddrxlat users are expected to build two translation systems and always use the one that is appropriate for the situation (kernel-space or user-space). It seems that you want to represent an address space which does not actually exist in the real world, using a "kernel page table root" for one range of virtual addresses (kernel-space) and a "user page table root" for another range (user-space). I can see how such an aggregate system would be useful on architectures where kernel-space addresses are in a different range than user-space addresses, but I'm not 100% sure about the proposed API, because "user page table root" then has a different meaning for POWER and AArch64. Please, can you confirm that I have understood the intended use case correctly? |
|
Hi Alexander, |
| @@ -829,6 +829,11 @@ typedef enum _addrxlat_optidx { | |||
| */ | |||
| ADDRXLAT_OPT_rootpgt, | |||
|
|
|||
| /** User land root page table address. Used *only* in aarch64 case. | |||
| * Type: full address | |||
There was a problem hiding this comment.
The comment is confusing. I don't see anything architecture-specific here, and in fact the option should also be very useful on IBM POWER.
Hi Petr,
My libkdumpfile based gdbserver
https://github.com/AlexanderKamensky/libkdumpfile/tree/new_20230805/gdbserver#process
does not work with your latest code.
The issue that now I cannot set rootpgt to read user-land
memory. This patch addresses it by introducing user_rootpgt.
Thanks,
Alexander