linux.git/kernel/dma, branch v6.6-rc2

Revert "dma-contiguous: check for memory region overlap"

2023-09-08T08:58:32+00:00

This reverts commit 3fa6456ebe13adab3ba1817c8e515a5b88f95dce.

The Commit broke the CMA region creation through DT on arm64,
as showed below logs with "memblock=debug":
[    0.000000] memblock_phys_alloc_range: 41943040 bytes align=0x200000
from=0x0000000000000000 max_addr=0x00000000ffffffff
early_init_dt_alloc_reserved_memory_arch+0x34/0xa0
[    0.000000] memblock_reserve: [0x00000000fd600000-0x00000000ffdfffff]
memblock_alloc_range_nid+0xc0/0x19c
[    0.000000] Reserved memory: overlap with other memblock reserved region

>From call flow, region we defined in DT was always reserved before entering
into rmem_cma_setup. Also, rmem_cma_setup has one routine cma_init_reserved_mem
to ensure the region was reserved. Checking the region not reserved here seems
not correct.

early_init_fdt_scan_reserved_mem:
    fdt_scan_reserved_mem
        __reserved_mem_reserve_reg
		early_init_dt_reserve_memory
			memblock_reserve(using “reg” prop case)
        fdt_init_reserved_mem
		__reserved_mem_alloc_size
			*early_init_dt_alloc_reserved_memory_arch*
				memblock_reserve(dynamic alloc case)
        __reserved_mem_init_node
		rmem_cma_setup(region overlap check here should always fail)

Example DT can be used to reproduce issue:

    dump_mem: mem_dump_region {
            compatible = "shared-dma-pool";
            alloc-ranges = <0x0 0x00000000 0x0 0xffffffff>;
            reusable;
            size = <0 0x2800000>;
    };

Signed-off-by: Zhenhua Huang

dma-pool: remove a __maybe_unused label in atomic_pool_expand

2023-08-31T12:12:37+00:00

Move the #endif a line so that free_page label is only seen by the
compile pass when actually used.

Signed-off-by: Christoph Hellwig 
Reviewed-by: Chunhui He 
Reviewed-by: Robin Murphy

dma-contiguous: fix the Kconfig entry for CONFIG_DMA_NUMA_CMA

2023-08-30T11:52:53+00:00

It makes no sense to expose CONFIG_DMA_NUMA_CMA if CONFIG_NUMA is not
enabled, and random config options shouldn't be default unless there
is a good reason.  Replace the default NUMA with a depends on to fix both
issues.

Signed-off-by: Christoph Hellwig 
Reviewed-by: Robin Murphy

dma-debug: don't call __dma_entry_alloc_check_leak() under free_entries_lock

2023-08-30T09:29:08+00:00

__dma_entry_alloc_check_leak() calls into printk -> serial console
output (qcom geni) and grabs port->lock under free_entries_lock
spin lock, which is a reverse locking dependency chain as qcom_geni
IRQ handler can call into dma-debug code and grab free_entries_lock
under port->lock.

Move __dma_entry_alloc_check_leak() call out of free_entries_lock
scope so that we don't acquire serial console's port->lock under it.

Trimmed-down lockdep splat:

 The existing dependency chain (in reverse order) is:

               -> #2 (free_entries_lock){-.-.}-{2:2}:
        _raw_spin_lock_irqsave+0x60/0x80
        dma_entry_alloc+0x38/0x110
        debug_dma_map_page+0x60/0xf8
        dma_map_page_attrs+0x1e0/0x230
        dma_map_single_attrs.constprop.0+0x6c/0xc8
        geni_se_rx_dma_prep+0x40/0xcc
        qcom_geni_serial_isr+0x310/0x510
        __handle_irq_event_percpu+0x110/0x244
        handle_irq_event_percpu+0x20/0x54
        handle_irq_event+0x50/0x88
        handle_fasteoi_irq+0xa4/0xcc
        handle_irq_desc+0x28/0x40
        generic_handle_domain_irq+0x24/0x30
        gic_handle_irq+0xc4/0x148
        do_interrupt_handler+0xa4/0xb0
        el1_interrupt+0x34/0x64
        el1h_64_irq_handler+0x18/0x24
        el1h_64_irq+0x64/0x68
        arch_local_irq_enable+0x4/0x8
        ____do_softirq+0x18/0x24
        ...

               -> #1 (&port_lock_key){-.-.}-{2:2}:
        _raw_spin_lock_irqsave+0x60/0x80
        qcom_geni_serial_console_write+0x184/0x1dc
        console_flush_all+0x344/0x454
        console_unlock+0x94/0xf0
        vprintk_emit+0x238/0x24c
        vprintk_default+0x3c/0x48
        vprintk+0xb4/0xbc
        _printk+0x68/0x90
        register_console+0x230/0x38c
        uart_add_one_port+0x338/0x494
        qcom_geni_serial_probe+0x390/0x424
        platform_probe+0x70/0xc0
        really_probe+0x148/0x280
        __driver_probe_device+0xfc/0x114
        driver_probe_device+0x44/0x100
        __device_attach_driver+0x64/0xdc
        bus_for_each_drv+0xb0/0xd8
        __device_attach+0xe4/0x140
        device_initial_probe+0x1c/0x28
        bus_probe_device+0x44/0xb0
        device_add+0x538/0x668
        of_device_add+0x44/0x50
        of_platform_device_create_pdata+0x94/0xc8
        of_platform_bus_create+0x270/0x304
        of_platform_populate+0xac/0xc4
        devm_of_platform_populate+0x60/0xac
        geni_se_probe+0x154/0x160
        platform_probe+0x70/0xc0
        ...

               -> #0 (console_owner){-...}-{0:0}:
        __lock_acquire+0xdf8/0x109c
        lock_acquire+0x234/0x284
        console_flush_all+0x330/0x454
        console_unlock+0x94/0xf0
        vprintk_emit+0x238/0x24c
        vprintk_default+0x3c/0x48
        vprintk+0xb4/0xbc
        _printk+0x68/0x90
        dma_entry_alloc+0xb4/0x110
        debug_dma_map_sg+0xdc/0x2f8
        __dma_map_sg_attrs+0xac/0xe4
        dma_map_sgtable+0x30/0x4c
        get_pages+0x1d4/0x1e4 [msm]
        msm_gem_pin_pages_locked+0x38/0xac [msm]
        msm_gem_pin_vma_locked+0x58/0x88 [msm]
        msm_ioctl_gem_submit+0xde4/0x13ac [msm]
        drm_ioctl_kernel+0xe0/0x15c
        drm_ioctl+0x2e8/0x3f4
        vfs_ioctl+0x30/0x50
        ...

 Chain exists of:
   console_owner --> &port_lock_key --> free_entries_lock

  Possible unsafe locking scenario:

        CPU0                    CPU1
        ----                    ----
   lock(free_entries_lock);
                                lock(&port_lock_key);
                                lock(free_entries_lock);
   lock(console_owner);

                *** DEADLOCK ***

 Call trace:
  dump_backtrace+0xb4/0xf0
  show_stack+0x20/0x30
  dump_stack_lvl+0x60/0x84
  dump_stack+0x18/0x24
  print_circular_bug+0x1cc/0x234
  check_noncircular+0x78/0xac
  __lock_acquire+0xdf8/0x109c
  lock_acquire+0x234/0x284
  console_flush_all+0x330/0x454
  console_unlock+0x94/0xf0
  vprintk_emit+0x238/0x24c
  vprintk_default+0x3c/0x48
  vprintk+0xb4/0xbc
  _printk+0x68/0x90
  dma_entry_alloc+0xb4/0x110
  debug_dma_map_sg+0xdc/0x2f8
  __dma_map_sg_attrs+0xac/0xe4
  dma_map_sgtable+0x30/0x4c
  get_pages+0x1d4/0x1e4 [msm]
  msm_gem_pin_pages_locked+0x38/0xac [msm]
  msm_gem_pin_vma_locked+0x58/0x88 [msm]
  msm_ioctl_gem_submit+0xde4/0x13ac [msm]
  drm_ioctl_kernel+0xe0/0x15c
  drm_ioctl+0x2e8/0x3f4
  vfs_ioctl+0x30/0x50
  ...

Reported-by: Rob Clark 
Signed-off-by: Sergey Senozhatsky 
Acked-by: Robin Murphy 
Signed-off-by: Christoph Hellwig

swiotlb: optimize get_max_slots()

2023-08-08T17:29:21+00:00

Use a simple logical shift and increment to calculate the number of slots
taken by the DMA segment boundary.

At least GCC-13 is not able to optimize the expression, producing this
horrible assembly code on x86:

	cmpq	$-1, %rcx
	je	.L364
	addq	$2048, %rcx
	shrq	$11, %rcx
	movq	%rcx, %r13
.L331:
	// rest of the function here...

	// after function epilogue and return:
.L364:
	movabsq $9007199254740992, %r13
	jmp	.L331

After the optimization, the code looks more reasonable:

	shrq	$11, %r11
	leaq	1(%r11), %rbx

Signed-off-by: Petr Tesarik 
Signed-off-by: Christoph Hellwig

swiotlb: move slot allocation explanation comment where it belongs

2023-08-08T17:29:06+00:00

Move the comment down in front of the loop that actually sets the list
member of struct io_tlb_slot to zero.

Fixes: 26a7e094783d ("swiotlb: refactor swiotlb_tbl_map_single")
Signed-off-by: Petr Tesarik 
Signed-off-by: Christoph Hellwig

swiotlb: search the software IO TLB only if the device makes use of it

2023-08-01T16:02:32+00:00

Skip searching the software IO TLB if a device has never used it, making
sure these devices are not affected by the introduction of multiple IO TLB
memory pools.

Additional memory barrier is required to ensure that the new value of the
flag is visible to other CPUs after mapping a new bounce buffer. For
efficiency, the flag check should be inlined, and then the memory barrier
must be moved to is_swiotlb_buffer(). However, it can replace the existing
barrier in swiotlb_find_pool(), because all callers use is_swiotlb_buffer()
first to verify that the buffer address belongs to the software IO TLB.

Signed-off-by: Petr Tesarik 
Signed-off-by: Christoph Hellwig

swiotlb: allocate a new memory pool when existing pools are full

2023-08-01T16:02:27+00:00

When swiotlb_find_slots() cannot find suitable slots, schedule the
allocation of a new memory pool. It is not possible to allocate the pool
immediately, because this code may run in interrupt context, which is not
suitable for large memory allocations. This means that the memory pool will
be available too late for the currently requested mapping, but the stress
on the software IO TLB allocator is likely to continue, and subsequent
allocations will benefit from the additional pool eventually.

Keep all memory pools for an allocator in an RCU list to avoid locking on
the read side. For modifications, add a new spinlock to struct io_tlb_mem.

The spinlock also protects updates to the total number of slabs (nslabs in
struct io_tlb_mem), but not reads of the value. Readers may therefore
encounter a stale value, but this is not an issue:

- swiotlb_tbl_map_single() and is_swiotlb_active() only check for non-zero
  value. This is ensured by the existence of the default memory pool,
  allocated at boot.

- The exact value is used only for non-critical purposes (debugfs, kernel
  messages).

Signed-off-by: Petr Tesarik 
Signed-off-by: Christoph Hellwig

swiotlb: determine potential physical address limit

2023-08-01T16:02:24+00:00

The value returned by default_swiotlb_limit() should be constant, because
it is used to decide whether DMA can be used. To allow allocating memory
pools on the fly, use the maximum possible physical address rather than the
highest address used by the default pool.

For swiotlb_init_remap(), this is either an arch-specific limit used by
memblock_alloc_low(), or the highest directly mapped physical address if
the initialization flags include SWIOTLB_ANY. For swiotlb_init_late(), the
highest address is determined by the GFP flags.

Signed-off-by: Petr Tesarik 
Signed-off-by: Christoph Hellwig

swiotlb: if swiotlb is full, fall back to a transient memory pool

2023-08-01T16:02:20+00:00

Try to allocate a transient memory pool if no suitable slots can be found
and the respective SWIOTLB is allowed to grow. The transient pool is just
enough big for this one bounce buffer. It is inserted into a per-device
list of transient memory pools, and it is freed again when the bounce
buffer is unmapped.

Transient memory pools are kept in an RCU list. A memory barrier is
required after adding a new entry, because any address within a transient
buffer must be immediately recognized as belonging to the SWIOTLB, even if
it is passed to another CPU.

Deletion does not require any synchronization beyond RCU ordering
guarantees. After a buffer is unmapped, its physical addresses may no
longer be passed to the DMA API, so the memory range of the corresponding
stale entry in the RCU list never matches. If the memory range gets
allocated again, then it happens only after a RCU quiescent state.

Since bounce buffers can now be allocated from different pools, add a
parameter to swiotlb_alloc_pool() to let the caller know which memory pool
is used. Add swiotlb_find_pool() to find the memory pool corresponding to
an address. This function is now also used by is_swiotlb_buffer(), because
a simple boundary check is no longer sufficient.

The logic in swiotlb_alloc_tlb() is taken from __dma_direct_alloc_pages(),
simplified and enhanced to use coherent memory pools if needed.

Note that this is not the most efficient way to provide a bounce buffer,
but when a DMA buffer can't be mapped, something may (and will) actually
break. At that point it is better to make an allocation, even if it may be
an expensive operation.

Signed-off-by: Petr Tesarik 
Reviewed-by: Greg Kroah-Hartman 
Signed-off-by: Christoph Hellwig