| Age | Commit message (Collapse) | Author |
|---|
|
Some T-Head based processors, for example the Sophgo SG2044, have a
JTLB errata that requires special handling of certain TLB maintenance
instructions (sfence.vma).
Introduce a sbi_tlb_local_operations structure so platforms can provide
custom local TLB flush implementations. Use it to implement the JTLB
workaround on affected SoCs (currently Sophgo SG2044).
Signed-off-by: Xiang W <[email protected]>
Signed-off-by: Han Gao <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Anup Patel <[email protected]>
|
|
The __sbi_sfence_vma_all() can be shared by different parts of
OpenSBI so rename __tlb_flush_all() to __sbi_sfence_vma_all()
and make it global function.
Signed-off-by: Andrew Waterman <[email protected]>
Signed-off-by: Anup Patel <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Anup Patel <[email protected]>
|
|
Extend sbi_fifo_enqueue() to allow forceful queueing by droping
data from the tail.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-By: Himanshu Chauhan <[email protected]>
|
|
Let us reduce the size of struct sbi_tlb_info by doing the
following:
1) Change the data type of asid and vmid fields to uint16_t
2) Replace local_fn() function pointer with an enum
Based on the above, the size of struct sbi_tlb_info is reduced
by 16 bytes on RV64 and 4 bytes on RV32.
Signed-off-by: Anup Patel <[email protected]>
|
|
The tlb_update() callback is called for each destination hart.
Move the size check earlier, so it is executed only once.
Signed-off-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Per the SBI specification, the effects of these functions are limited to
a specific ASID and/or VMID. This applies even when flushing the entire
address space.
Signed-off-by: Samuel Holland <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Let us prefer hartindex over hartid in IPI framework which in-turn
forces IPI users to also prefer hartindex.
Signed-off-by: Anup Patel <[email protected]>
|
|
The sbi_hartid_to_scratch() involves translating hartid to hartindex
which is expensive so let's use sbi_hartindex_to_scratch() instead.
Signed-off-by: Anup Patel <[email protected]>
|
|
Currently, the sbi_hartmask is indexed by hartid which puts a
limit on hartid to be less than SBI_HARTMASK_MAX_BITS.
We extend the sbi_hartmask implementation to use hartindex and
support updating sbi_hartmask using hartid. This removes the
limit on hartid and existing code works largely unmodified.
Signed-off-by: Xiang W <[email protected]>
Signed-off-by: Anup Patel <[email protected]>
|
|
If the system is defined from tlb_fifo_num_entries, the scratch may be
too small to hold the fifo, so it is alloc through the heap.
Signed-off-by: Xiang W <[email protected]>
Signed-off-by: Xing Xiaoguang <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
For some platforms with a particularly high number of harts, if the
tlb fifo is too small, it case harts to wait. Platforms should be
allowed to specify the size of the tlb fifo.
Signed-off-by: Xiang W <[email protected]>
Signed-off-by: Xing Xiaoguang <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
When tlb_fifo is full, it will wait and affect the ipi update to
other harts. This patch is optimized.
Signed-off-by: Xiang W <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
Tested-by: Anup Patel <[email protected]>
|
|
tlb_process_count is only used when count=1, so refactor to
tlb_process_once and add the return value to be reused in
tlb_process
Signed-off-by: Xiang W <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
Tested-by: Anup Patel <[email protected]>
|
|
Originally, the process and sync of sbi_tlb need to wait for each other.
Evasion by atomic addition and subtraction.
Signed-off-by: Xiang W <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
Tested-by: Anup Patel <[email protected]>
|
|
The parameter passed to HFENCE.GVMA instruction in rs1 register
is guest physical address right shifted by 2 (i.e. divided by 4).
Unfortunately, we overlooked the semantics of rs1 registers for
HFENCE.GVMA instruction and never right shifted guest physical
address by 2. This issue did not manifest for hypervisors till
now because all H-extension implementations (such as QEMU, Spike,
Rocket Core FPGA, etc) we tried till now were conservatively
flushing everything upon any HFENCE.GVMA instruction.
This patch fixes GPA passed to __sbi_hfence_gvma_vmid_gpa()
and __sbi_hfence_gvma_gpa() functions.
Fixes: 331ff6a162c1 ("lib: Support stage1 and stage2 tlb flushing")
Reported-by: Ian Huang <[email protected]>
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Xiang W <[email protected]>
Reviewed-by: Dong Du <[email protected]>
|
|
Follow the standard conventon for static function names:
All global functions should be start with sbi_<module name>_
All static functions should be start with <module name>_
Reviewed-by: Anup Patel <[email protected]>
Signed-off-by: Atish Patra <[email protected]>
|
|
RISC-V SBI v0.3 specification defines a set of firmware events that can
provide additional information about the current firmware context. All
of the firmware event monitoring are enabled now. The firmware
events must be defined as raw perf event with MSB set as specified in the
specification.
Reviewed-by: Anup Patel <[email protected]>
Signed-off-by: Atish Patra <[email protected]>
|
|
Compiling sbi_tlb_entry_process() with GCC 11 and -Wextra yields:
lib/sbi/sbi_tlb.c: In function ‘sbi_tlb_process_count’:
lib/sbi/sbi_tlb.c:206:31:
error: comparison of integer expressions of different signedness:
‘u32’ {aka ‘unsigned int’} and ‘int’ [-Werror=sign-compare]
206 | if (deq_count > count)
Signed-off-by: Heinrich Schuchardt <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
Reviewed-by: Xiang W <[email protected]>
|
|
The parameter owner of function sbi_scratch_alloc_offset() is never used.
The scratch memory is small. We should not use it for debug information in
future. Hence eliminate the parameter.
Signed-off-by: Heinrich Schuchardt <[email protected]>
Reviewed-by: Xiang W <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Currently, we have fixed TLB flush types supported by the
remote TLB library. This approach is not flexible and does
not allow custom local TLB flush function. For example,
after updating PMP entries on a set of HARTs at runtime,
we have to flush TLB on these HARTs as well.
To support custom local TLB flush function, we replace the
"type" field of "struct sbi_tlb_info" with a local TLB flush
function pointer. We also provide definitions of standard TLB
flush operations (such as fence_i, sfence.vma, hfence.vvma,
hfence.gvma, etc).
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
The arguments/parameters of __sbi_hfence_gvma_vmid_gpa() and
__sbi_hfence_vvma_asid_va() functions are swapped so we fix it.
Currently, we did not face any issues because QEMU does a full
TLB flush for all HFENCE instructions.
We also improve documentation of HFENCE.GVMA and HFENCE.VVMA
instruction encoding.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
The HFENCE VVMA instructions flushes TLB based on the VMID
present in HGATP CSR. To handle this, we get the current
VMID for SBI HFENCE VVMA call and we use this current VMID
to do remote TLB HFENCE VVMA on desired set of HARTs.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
This patch removes scratch parameter from sbi_dprintf() function
because sbi_dprintf() can use sbi_scratch_thishart_ptr() to get
current HART scratch space.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
This patch removes scratch parameter from most sbi_ipi functions.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
The sbi_ipi_send_many() should get current HART scratch pointer
on it's own using eventually hence removing scratch parameter from
sbi_tlb_request().
Signed-off-by: Anup Patel <[email protected]>
|
|
The sbi_hartid_to_scratch() API can fail for non-existent HARTs so
all uses of sbi_hartid_to_scratch() API should check return value.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Bin Meng <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
We remove redundant scratch parameter from sbi_tlb_entry_process()
function.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Bin Meng <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
This patch simplify HART id to scratch macro as follows:
1. Remove current "scratch" pointer argument because now we
use HART id to scratch table
2. Rename sbi_hart_id_to_scratch() to sbi_hartid_to_scratch()
to have macro name consistent with the name of callback
in struct sbi_scratch
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Bin Meng <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
The sbi_current_hartid() being a regular function is quite
expensive because for callers it is a function call instead
of a direct CSR read. This patch converts sbi_current_hartid()
into a macro in riscv_asm.h.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
Instead of using single ulong as source mask for sbi_tlb_info,
we use sbi_hartmask. This way sbi_tlb_info can easily scale
for large number of HARTs.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Bin Meng <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
This fixes various coding style issues found in the SBI codes.
No functional changes.
Signed-off-by: Bin Meng <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
This patch drops _fifo from the name of various sbi_tlb_fifo_xyz()
functions because all these functions deal with remote TLB managment
and FIFO is the per-HART data structure used internally by remote
TLB implementation.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
This patch introduces sbi_ipi_event_create/destroy() APIs and
struct sbi_ipi_event_ops for creating/destroying IPI events
at runtime based of event operations.
This new APIs will help platform code and utils code to create
custom IPI events which are not part of generic OpenSBI library.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
Instead of directly calling sbi_ipi_send_many(), we introduce
sbi_tlb_fifo_request() for halting a set of HARTs.
This way in future we can assign any IPI event number for remote
FENCE within sbi_tlb.c only.
Signed-off-by: Anup Patel <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
The hypervisor specification support hfence calls which can be used
issue tlb flush requests at both level of address translation. Currently,
these requests are issued only via SBI which are defined in v0.2.
Signed-off-by: Atish Patra <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Currently, the tlb range flush threshold is fixed and set to 4k for
all platforms. However, it should be platform specific as it completely
depends upon how platform actually implements sfence instruction.
Define a platform feature that allows every individual platform to set
different values. If a platform doesn't define it, just use a page size as
the threshold.
Signed-off-by: Atish Patra <[email protected]>
Reviewed-by: Bin Meng <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Currently, we upgrade to a full tlb flush only If a tlb flush request
size is greater than the threshold. This is done as sfence in RISC-V
can only flush 4KB at a time. Doing a large number of flushes page by
page impacts the performance. It is better to do a full tlbflush if the
request size is at least equal to the threshold size.
Signed-off-by: Atish Patra <[email protected]>
Reviewed-by: Anup Patel <[email protected]>
|
|
Linux kernel expects tlb flush SBI call to be completely synchronous i.e.
the SBI call should only return once corresponding *fence* instruction is
executed.
OpenSBI manages the outstanding TLB flush requests by keeping them in a
per hart based fifo. However, there are few corner cases that may lead to
race conditions while updating the fifo.
Currently, the caller hart waits for IPI acknowledgement via clint
address which is not a very good method as synchronization on MMIO may not
be supported in every platform. Moreover, the waiter doesn't have any way of
identifying if the IPI is received for specific tlb flush request or any
other IPI. This may lead to unpredictable behavior in supervisor/user space.
Fix this by waiting on individual fifo entries rather than MMIO address.
Currently, a relaxed loop is being used because wfi again involves MMIO write
which would be slower compared to relaxed loop. To avoid deadlock, fifo
is processed every time a hart loops for fifo enqueue or fifo sync to consume
the tlb flush requests sent by other harts.
Signed-off-by: Anup Patel <[email protected]>
Signed-off-by: Atish Patra <[email protected]>
|
|
In order to prevent a possible null pointer dereference, return early if
either one of 'in' or 'data' is null.
Signed-off-by: Jacob Garber <[email protected]>
Reviewed-by: Bin Meng <[email protected]>
Reviewed-by: Atish Patra <[email protected]>
|
|
All string functions are part of libsbi. It makes more sense
to rename them to sbi_string.x as the libsbi can be linked
with external libraries that can have similar implementation.
Signed-off-by: Atish Patra <[email protected]>
Acked-by: Anup Patel <[email protected]>
|
|
libsbi needs some of the custom libc functions. It should be directly
included in libsbi instead of platform specific libraries.
Signed-off-by: Atish Patra <[email protected]>
Acked-by: Anup Patel <[email protected]>
|
|
Signed-off-by: Atish Patra <[email protected]>
Acked-by: Anup Patel <[email protected]>
|
