Re: [PATCH] esp: Make driver SMP-correct
From: Andrew Morton (akpm_at_osdl.org)
Date: 12/31/04
- Previous message: Andrew Morton: "Re: waiting 10s before mounting root filesystem?"
- In reply to: James Nelson: "[PATCH] esp: Make driver SMP-correct"
- Next in thread: Russell King: "Re: [PATCH] esp: Make driver SMP-correct"
- Reply: Russell King: "Re: [PATCH] esp: Make driver SMP-correct"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Fri, 31 Dec 2004 01:46:11 -0800 To: James Nelson <james4765@verizon.net>
James Nelson <james4765@verizon.net> wrote:
>
> This is an attempt to make the esp serial driver SMP-correct. It also removes
> some cruft left over from the serial_write() conversion.
From a quick scan:
- startup() does multiple sleeping allocations and request_irq() under
spin_lock_irqsave(). Maybe fixed by this:
--- 25/drivers/char/esp.c~esp-make-driver-smp-correct-fixes 2004-12-31 01:40:57.987232152 -0800
+++ 25-akpm/drivers/char/esp.c 2004-12-31 01:42:21.444544712 -0800
@@ -851,16 +851,14 @@ static int startup(struct esp_struct * i
int retval=0;
unsigned int num_chars;
- spin_lock_irqsave(&info->irq_lock, flags);
-
if (info->flags & ASYNC_INITIALIZED)
- goto out;
+ goto out_unlocked;
if (!info->xmit_buf) {
info->xmit_buf = (unsigned char *)get_zeroed_page(GFP_KERNEL);
retval = -ENOMEM;
if (!info->xmit_buf)
- goto out;
+ goto out_unlocked;
}
#ifdef SERIAL_DEBUG_OPEN
@@ -907,7 +905,7 @@ static int startup(struct esp_struct * i
&info->tty->flags);
retval = 0;
}
- goto out;
+ goto out_unocked;
}
if (!(info->stat_flags & ESP_STAT_USE_PIO) && !dma_buffer) {
@@ -926,6 +924,8 @@ static int startup(struct esp_struct * i
}
+ spin_lock_irqsave(&info->irq_lock, flags);
+
info->MCR = UART_MCR_DTR | UART_MCR_RTS | UART_MCR_OUT2;
serial_out(info, UART_ESI_CMD1, ESI_WRITE_UART);
serial_out(info, UART_ESI_CMD2, UART_MCR);
@@ -965,7 +965,9 @@ static int startup(struct esp_struct * i
info->flags |= ASYNC_INITIALIZED;
retval = 0;
-out: spin_unlock_irqrestore(&info->irq_lock, flags);
+out:
+ spin_unlock_irqrestore(&info->irq_lock, flags);
+out_unocked:
return retval;
}
_
- startup() calls change_speed() under info->irq_lock, but change_speed()
also takes info->irq_lock. Instant deadlock on driver initialisation.
The driver needs more serious surgery than this.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
- Previous message: Andrew Morton: "Re: waiting 10s before mounting root filesystem?"
- In reply to: James Nelson: "[PATCH] esp: Make driver SMP-correct"
- Next in thread: Russell King: "Re: [PATCH] esp: Make driver SMP-correct"
- Reply: Russell King: "Re: [PATCH] esp: Make driver SMP-correct"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
|
